Medical threatened event estimation method and externally worn defibrillator

Abstract

The invention provides a medical threatened event estimation method and an externally worn defibrillator. The medical threatened event estimation system comprises the following steps: acquiring a first physiological information set and a second physiological information set of a detected person; calculating a first risk score and a second risk score associated with estimating the risk of the arrhythmia event based on applying the first set of physiological information and the second set of physiological information to a machine learning classifier model; providing at least the first risk score and the second risk score associated with the potential arrhythmic event as a time-varying series of risk scores; and classifying the first risk score and the second risk score associated with estimating a risk of the potential arrhythmic event for the subject based on the one or more thresholds.

Description

[0001] This application is a divisional application of the invention patent application with application date of November 13, 2015, application number 201580073217.8, entitled “Medical Precursor Event Estimation” (with application date of November 13, 2015, application number 202110841247.6, entitled “Medical Precursor Event Estimation System and Externally Worn Defibrillator”).

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0003] This application claims priority under 35 U.S.C. §119(e) to U.S. Provisional Application Serial No. 62 / 080,083, filed on November 14, 2014, entitled “MEDICAL PREMONITORY EVENT ESTIMATION,” U.S. Provisional Application Serial No. 62 / 096,140, ​​filed on December 23, 2014, entitled “MEDICAL PREMONITORY EVENT ESTIMATION,” and U.S. Provisional Application Serial No. 62 / 235,911, filed on October 1, 2015, entitled “MEDICAL PREMONITORY EVENT ESTIMATION,” the entire contents of all three of which are hereby incorporated by reference. Background Art

[0004] The present invention relates to medical premonitory event estimation, detection systems and methods, and in some embodiments, to systems and methods for calculating an event estimation of risk score associated with a potential medical event of a subject. Background Art

[0006] Many patients are at risk of sudden death as a result of cardiac arrest or asystole caused by cardiac tachyarrhythmias. For example, patients with a decreased left ventricular ejection fraction and diagnosed with myocardial infarction, non-ischemic cardiomyopathy, dilated cardiomyopathy, and congestive heart failure have been identified as being at high risk compared to the general population. Effective methods for determining risk levels within these populations that would allow for the identification of warning signs before sudden cardiac death occurs are lacking. In the absence of reliable methods for classifying high-risk subjects, the treatment recommendations available to these patients often involve hospitalization and observation followed by medication. Despite the availability of medication, sudden death rates have only slightly decreased, and once a subject experiences cardiac arrest, the only proven treatment is defibrillation. In fact, the probability that a subject will survive cardiac arrest decreases by approximately 10% for every minute that defibrillation treatment is delayed. Consequently, subjects would benefit from systems and methods for detecting warning signs of high-risk sudden cardiac death that enable proactive decision-making, such as the application of advanced medical interventions.

[0007] During use, a medical device that monitors the physiological condition of the subject (such as a portable defibrillator monitor, a wearable defibrillator, a wearable insulin pump, a wearable ECG monitor or a monitor of other physiological parameters, etc.) can send an audible alarm, a voice prompt or message or a visual message to warn the subject or bystander of the medical condition, instruct the subject or bystander to take some action to correct the problem, call for assistance, request information from the subject or bystander, or provide feedback to the device so that the device can continue to function properly. For example, if the medical device is a wearable defibrillator, if the device determines that the subject is experiencing an arrhythmia, the device can send an alarm and a voice message to provide instructions to the subject. The alarm and message can also be directed to a bystander, thereby instructing the bystander to call for medical assistance, or to warn any bystander that a defibrillation shock is about to be delivered and to stay away from the subject. Summary of the Invention

[0008] In some embodiments, a medical precursor event estimation system is provided, the system comprising: a non-transitory computer-readable storage medium that communicates with one or more processors and stores instructions, wherein the instructions, when executed by the one or more processors, cause the one or more processors to perform operations including: calculating, for multiple time periods, an event estimation risk score associated with a potential medical event of the subject occurring within the associated time period based at least in part on the subject's physiological parameter data.

[0009] Preferred and non-limiting embodiments or aspects of the present invention will now be described in the following numbered clauses:

[0010] Item 1. A medical precursor event estimation system, comprising: a non-transitory computer-readable storage medium that communicates with one or more processors and stores instructions, wherein the instructions, when executed by the one or more processors, cause the one or more processors to perform operations including: calculating, for multiple time periods, an event estimation risk score associated with a potential medical event of the subject occurring within the associated time period based at least in part on the subject's physiological parameter data.

[0011] Clause 2. The medical precursor event estimation system of clause 1, wherein the physiological parameter data comprises ECG data.

[0012] Clause 3. A medical precursor event estimation system according to clause 1 or 2, wherein the physiological parameter data includes at least one of blood pressure data, heart rate data, chest impedance data, pulse oximetry data, respiratory rate data, heart sound data, lung sound data and activity level data.

[0013] Clause 4. The medical precursor event estimation system of any one of clauses 1-3, wherein the potential medical event comprises a cardiac event.

[0014] Clause 5. The medical precursor event estimation system according to clause 4, wherein the cardiac event comprises at least one of an ectopic beat, a series of ectopic beats, ventricular tachycardia, bradycardia, cardiac arrest, and T wave abnormality.

[0015] Clause 6. The medical precursor event estimation system of any one of clauses 1-5, wherein the potential medical event comprises at least one of a plurality of medical events, an increase in the rate of a medical event, and / or an increase in the intensity of a medical event.

[0016] Clause 7. A medical precursor event estimation system according to any one of clauses 1-6, wherein the potential medical event is defined in a multidimensional parameter space including the physiological parameter data and at least one other type of physiological parameter data and / or demographic data of the subject.

[0017] Clause 8. A medical precursor event estimation system according to any one of clauses 1-7, wherein the one or more processors perform operations including: calculating multiple different event estimation risk scores associated with the potential medical event of the subject within an associated time period based at least in part on the physiological parameter data.

[0018] Clause 9. A medical precursor event estimation system according to any one of clauses 1-8, wherein the one or more processors perform operations including: calculating multiple different event estimation risk scores associated with multiple different potential medical events of the subject within an associated time period based at least in part on the physiological parameter data.

[0019] Clause 10. The medical precursor event estimation system of any one of clauses 1-9, wherein calculating the event estimated risk score comprises calculating an urgency score representing the importance of the potential medical event relative to at least one other potential medical event.

[0020] Clause 11. The medical precursor event estimation system of any one of clauses 1-10, wherein calculating the event estimation risk score comprises calculating a confidence score comprising a probability of the potential medical event occurring within an associated time period.

[0021] Clause 12. A medical precursor event estimation system according to any one of clauses 1-11, wherein the one or more processors perform the following operations: determining that the event estimated risk score within an associated time period satisfies one or more event estimated risk thresholds; and determining a response to the potential medical event based at least in part on the one or more event estimated risk thresholds determined to be satisfied.

[0022] Clause 13. A medical precursor event estimation system according to clause 12, wherein the determined response to the potential medical event includes at least one of the following: notifying the subject of an advanced diagnosis; advising the subject not to remove the device; advising the subject of a behavioral modification; alerting a medical professional; and preparing the device for treatment.

[0023] Clause 14. A medical precursor event estimation system according to clause 12 or 13, wherein the one or more event estimation risk thresholds each include at least one confidence threshold and at least one criticality threshold, wherein the at least one confidence threshold includes the required probability of the potential medical event occurring within an associated time period, and the at least one criticality threshold includes the required importance of the potential medical event relative to at least one other potential medical event.

[0024] Clause 15. The medical precursor event estimation system according to any one of clauses 12-14, wherein the one or more event estimated risk thresholds comprise a plurality of different event estimated risk thresholds for associated time periods.

[0025] Clause 16. The medical precursor event estimation system of any of clauses 12-15, wherein the one or more estimated event risk thresholds for the first time period are different from the one or more estimated event risk thresholds for the second time period.

[0026] Clause 17. A medical precursor event estimation system according to any one of clauses 12-16, wherein the response determined for the potential medical event of the subject occurring within a first associated time period is different from the response determined for the potential medical event of the subject occurring within a second associated time period.

[0027] Clause 18. A medical precursor event estimation system according to any one of clauses 1-17, wherein the one or more processors perform the following operations: determining that the event estimation risk score within the associated time period fails to meet at least one event estimation risk threshold; receiving additional data of the subject; and calculating an enhanced event estimation risk score associated with the potential medical event of the subject occurring within the associated time period based at least in part on the physiological parameter data and the additional data.

[0028] Clause 19. The medical precursor event estimation system according to Clause 18, wherein the additional data includes at least one of the following data: image data of the subject; audio data containing the subject's voice; and data based on the subject's galvanic skin response.

[0029] Clause 20. The medical precursor event estimation system of clause 18 or 19, wherein the one or more processors perform operations comprising setting the one or more event estimation risk thresholds based at least in part on historical patient data collected from a plurality of patients.

[0030] Clause 21. The medical precursor event estimation system of any of clauses 18-20, wherein the one or more processors perform operations comprising setting the one or more event estimation risk thresholds based at least in part on input from a user.

[0031] Clause 22. The medical precursor event estimation system of any one of clauses 1-21, wherein the one or more processors perform operations comprising calculating the event estimated risk score at periodic time intervals.

[0032] Clause 23. A medical precursor event estimation system according to any one of clauses 1-21, wherein the one or more processors perform operations including: calculating the event estimation risk score at dynamic time intervals, wherein the duration of the dynamic time intervals is at least partially based on the event estimation risk score.

[0033] Clause 24. The medical precursor event estimation system of any one of clauses 1-21, wherein the one or more processors perform operations comprising: continuously calculating the event estimated risk score.

[0034] Clause 25. The medical precursor event estimation system according to any one of clauses 1-24, wherein calculating the estimated risk score for the event comprises applying a logistic regression model to the physiological parameter data to determine the estimated risk score for the event.

[0035] Clause 26. A medical precursor event estimation system according to any one of clauses 1-24, wherein calculating the event estimation risk score includes: generating at least two substantially orthogonal vectors based at least in part on the physiological parameter data; processing the at least two substantially orthogonal vectors to determine a cyclic trajectory of the physiological parameter data; and identifying trajectory bifurcation by the following operations: characterizing a control loop trajectory group including one or more cyclic trajectories obtained in a first time period; characterizing a test loop trajectory group including one or more cyclic trajectories obtained in a second time period after the first time period; comparing the characterization value of the control loop trajectory group with the characterization value of the test loop trajectory group; measuring the degree of trajectory bifurcation between the control loop trajectory group and the test loop trajectory group; and calculating the event estimation risk score based at least in part on the measured value of the degree of trajectory bifurcation.

[0036] Clause 27. A medical precursor event estimation system according to any one of clauses 1-24, wherein calculating the event estimated risk score comprises: calculating a first event estimated risk score comprising a first criticality score for a first potential medical event; and calculating a second event estimated risk score comprising a second criticality score for a second potential medical event based at least in part on the first event estimated risk score, wherein the first criticality score indicates that the importance of the first potential medical event is different from the importance of the second potential medical event.

[0037] Clause 28. A medical precursor event estimation system according to any one of clauses 1-24, wherein calculating the event estimation risk score includes: calculating a first event estimation risk score associated with the potential medical event of the subject occurring within an associated time period based on a first shockable rhythm detection algorithm; and calculating a second event estimation risk score associated with the potential medical event of the subject occurring within an associated time period based on a second rhythm detection algorithm, wherein the second rhythm detection algorithm is adjusted to achieve higher sensitivity to the physiological parameter data compared to the first shockable rhythm detection algorithm.

[0038] Clause 29. The medical precursor event estimation system according to any one of clauses 1-24, wherein calculating the event estimation risk score comprises applying at least two different rhythm detection algorithms to different time periods of the physiological parameter data.

[0039] Clause 30. The medical precursor event estimation system of any one of clauses 1-24, wherein calculating the event estimated risk score comprises: receiving data representing a patient's survivability; and determining a response to the potential medical event based at least in part on the patient's survivability.

[0040] Clause 31. The medical precursor event estimation system according to any one of clauses 1-30, further comprising: a medical device comprising one or more sensors configured to sense the physiological parameter data of the subject.

[0041] Clause 32. A medical precursor event estimation system according to any one of clauses 1-31, wherein the medical device includes a wearable medical device, the one or more sensors include multiple ECG sensors, the physiological parameter data of the subject includes ECG data, and the potential medical event includes a cardiac event.

[0042] Clause 33. The medical precursor event estimation system according to clause 31 or 32, further comprising: a communication network configured to communicate at least one of the physiological parameter data and the event estimation risk score from the medical device to another computing device.

[0043] Clause 34. The medical precursor event estimation system according to any one of clauses 1-33, further comprising: a display for displaying a time-based visual indication of the estimated risk score of the event for multiple time periods.

[0044] Clause 35. The medical precursor event estimation system of any one of clauses 1-34, wherein the one or more processors perform operations comprising determining a response to the potential medical event based at least in part on the event estimated risk score.

[0045] Clause 36. The medical precursor event estimation system of Clause 35, wherein the determined response to the potential medical event comprises providing the subject with instructions to contact a medical professional.

[0046] Clause 37. The medical precursor event estimation system according to clause 35 or 36, further comprising: a wearable medical device, wherein the determined response to the potential medical event includes providing an instruction to the subject to check a battery of the wearable medical device.

[0047] Clause 38. The medical precursor event estimation system of any one of clauses 35-37, further comprising: a wearable medical device, wherein the determined response to the potential medical event includes charging an electric shock mechanism of the wearable medical device.

[0048] Clause 39. A medical precursor event estimation system according to any of clauses 35-38, wherein the one or more processors perform operations including: determining a response to the potential medical event based at least in part on the sensitivity and specificity of the event estimation risk score.

[0049] Clause 40. The medical precursor event estimation system of any of clauses 35-39, wherein the determined response based on the first sensitivity and the first specificity is different from the determined response based on a second, different sensitivity and a second, different specificity.

[0050] Clause 41. A medical precursor event estimation system according to any of clauses 35-40, wherein the determined response to the potential medical event includes at least one of: notifying the subject of an advanced diagnosis; advising the subject not to remove the device; advising the subject of a behavioral modification; alerting a medical professional; and preparing the device for treatment.

[0051] Clause 42. The medical precursor event estimation system of any one of clauses 35-41, wherein the one or more processors perform operations comprising modifying the sensitivity of an algorithm used to determine the event estimated risk score based on the subject's risk level.

[0052] Clause 43. A medical precursor event estimation system according to any one of clauses 1-42, wherein the multiple time periods include at least one time period of less than about 10 minutes, at least one time period of less than about 1 hour, at least one time period of less than about 3 hours, at least one time period of less than about 1 day, at least one time period of less than about 1 week, and at least one time period of less than about 1 month.

[0053] Clause 44. The medical precursor event estimation system according to any one of clauses 1-43, wherein calculating the estimated risk score for the event comprises calculating a confidence band for the estimated risk score for the event.

[0054] Clause 45. The medical precursor event estimation system according to any one of clauses 1-44, wherein calculating the estimated risk score for the event comprises calculating an error band for the estimated risk score for the event.

[0055] Clause 46. The medical precursor event estimation system of any one of clauses 1-45, wherein the plurality of time periods comprises a plurality of time periods of less than 4 hours.

[0056] Clause 47. A method for estimating medical precursor events, comprising the following steps: utilizing one or more processors to receive physiological parameter data of a subject; and for multiple time periods, utilizing the one or more processors to calculate an event estimation risk score associated with a potential medical event of the subject occurring within an associated time period based at least in part on the physiological parameter data of the subject.

[0057] Clause 48. The method of clause 47, wherein the physiological parameter data comprises ECG data.

[0058] Clause 49. A method according to clause 47 or 48, wherein the physiological parameter data includes at least one of blood pressure data, heart rate data, chest impedance data, pulse oximetry data, respiratory rate data, heart sound data, lung sound data and activity level data.

[0059] Clause 50. The method of any one of Clauses 47-49, wherein the potential medical event comprises a cardiac event.

[0060] Clause 51. The method of clause 50, wherein the cardiac event comprises at least one of an ectopic beat, a series of ectopic beats, ventricular tachycardia, bradycardia, cardiac arrest, and a T wave abnormality.

[0061] Clause 52. The method of any one of clauses 47-51, wherein the potential medical event comprises at least one of a plurality of medical events, an increase in the rate of a medical event, and / or an increase in the intensity of a medical event.

[0062] Clause 53. The method of any one of clauses 47-52, wherein the potential medical event is defined in a multidimensional parameter space comprising the physiological parameter data and at least one other type of physiological parameter data and / or demographic data of the subject.

[0063] Clause 54. A method according to any one of clauses 47-53, further comprising the following steps: using the one or more processors to calculate, at least in part based on the physiological parameter data, a plurality of different event estimated risk scores associated with the potential medical event of the subject within an associated time period.

[0064] Clause 55. A method according to any one of clauses 47-54, further comprising the following steps: utilizing the one or more processors to calculate, at least in part based on the physiological parameter data, a plurality of different event estimated risk scores associated with a plurality of different potential medical events of the subject within an associated time period.

[0065] Clause 56. The method of any one of clauses 47-55, wherein calculating the estimated risk score for the event comprises calculating a criticality score representing the importance of the potential medical event relative to at least one other potential medical event.

[0066] Clause 57. The method of any one of clauses 47-56, wherein calculating the estimated risk score for the event comprises calculating a confidence score comprising a probability of the potential medical event occurring within an associated time period.

[0067] Clause 58. A method according to any one of clauses 47-57, further comprising the following steps: using the one or more processors to determine that the event estimated risk score within the associated time period satisfies one or more event estimated risk thresholds; and using the one or more processors to determine a response to the potential medical event based at least in part on the one or more event estimated risk thresholds determined to be satisfied.

[0068] Clause 59. A method according to clause 58, wherein the determined response to the potential medical event includes at least one of: notifying the subject of an advanced diagnosis; advising the subject not to remove the device; advising the subject of a behavioral modification; alerting a medical professional; and preparing the device for treatment.

[0069] Clause 60. A method according to clause 58 or 59, wherein the one or more event estimated risk thresholds each include at least one confidence threshold and at least one criticality threshold, wherein the at least one confidence threshold includes a required probability of the potential medical event occurring within an associated time period, and the at least one criticality threshold includes a required importance of the potential medical event relative to at least one other potential medical event.

[0070] Clause 61. The method of any one of clauses 58-60, wherein the one or more event estimated risk thresholds comprise a plurality of different event estimated risk thresholds for associated time periods.

[0071] Clause 62. The method of any of clauses 58-61, wherein the one or more estimated risk thresholds of an event for the first time period are different from the one or more estimated risk thresholds of an event for the second time period.

[0072] Clause 63. A method according to any of clauses 58-62, wherein the determined response to the potential medical event of the subject occurring within a first associated time period is different from the determined response to the potential medical event of the subject occurring within a second associated time period.

[0073] Clause 64. A method according to any one of clauses 47-63, further comprising the following steps: using the one or more processors to determine that the event estimated risk score within the associated time period fails to meet at least one event estimated risk threshold; using the one or more processors to receive additional data of the subject; and using the one or more processors to calculate an enhanced event estimated risk score associated with the potential medical event of the subject occurring within the associated time period based at least in part on the physiological parameter data and the additional data.

[0074] Clause 65. The method according to clause 64, wherein the additional data comprises at least one of the following data: image data of the subject; audio data containing the subject's voice; and data based on the subject's galvanic skin response.

[0075] Clause 66. The method of any one of clauses 58-65, further comprising the step of setting, using the one or more processors, the one or more event estimated risk thresholds based at least in part on historical patient data collected from a plurality of patients.

[0076] Clause 67. The method of any one of clauses 58-66, further comprising the step of setting, using the one or more processors, the one or more event estimated risk thresholds based at least in part on input from a user.

[0077] Clause 68. The method of any one of clauses 47-67, further comprising the step of calculating, using the one or more processors, the estimated risk score for the event at periodic time intervals.

[0078] Clause 69. A method according to any one of clauses 47-67, further comprising the following steps: using the one or more processors to calculate the event estimated risk score at dynamic time intervals, wherein the duration of the dynamic time interval is at least partially based on the event estimated risk score.

[0079] Clause 70. The method according to any one of clauses 47-67, further comprising the step of: continuously calculating the event estimated risk score using the one or more processors.

[0080] Clause 71. The method of any one of clauses 47-70, wherein calculating the estimated risk score for an event comprises applying a logistic regression model to the physiological parameter data to determine the estimated risk score for an event.

[0081] Clause 72. A method according to any one of clauses 47-70, wherein calculating the event estimated risk score includes: generating at least two substantially orthogonal vectors based at least in part on the physiological parameter data; processing the at least two substantially orthogonal vectors to determine the cyclic trajectory of the physiological parameter data; and identifying trajectory bifurcation by: characterizing a control loop trajectory group including one or more cyclic trajectories obtained in a first time period; characterizing a test loop trajectory group including one or more cyclic trajectories obtained in a second time period after the first time period; comparing the characterization value of the control loop trajectory group with the characterization value of the test loop trajectory group; measuring the degree of trajectory bifurcation between the control loop trajectory group and the test loop trajectory group; and calculating the event estimated risk score based at least in part on the measured value of the degree of trajectory bifurcation.

[0082] Clause 73. A method according to any one of clauses 47-70, wherein calculating the event estimated risk score comprises: calculating a first event estimated risk score comprising a first criticality score for a first potential medical event; and calculating a second event estimated risk score comprising a second criticality score for a second potential medical event based at least in part on the first event estimated risk score, wherein the first criticality score indicates that the importance of the first potential medical event is different from the importance of the second potential medical event.

[0083] Clause 74. A method according to any one of clauses 47-70, wherein calculating the event estimated risk score comprises: calculating a first event estimated risk score associated with the potential medical event of the subject occurring within an associated time period based on a first shockable rhythm detection algorithm; and calculating a second event estimated risk score associated with the potential medical event of the subject occurring within an associated time period based on a second rhythm detection algorithm, wherein the second rhythm detection algorithm is adjusted to achieve higher sensitivity to the physiological parameter data compared to the first shockable rhythm detection algorithm.

[0084] Clause 75. The method of any one of clauses 47-70, wherein calculating the estimated risk score for an event comprises applying at least two different rhythm detection algorithms to different time periods of the physiological parameter data.

[0085] Clause 76. The method of any one of clauses 47-70, wherein calculating the estimated risk score for the event comprises: receiving data representing the survivability of a patient; and determining a response to the potential medical event based at least in part on the survivability of the patient.

[0086] Clause 77. The method according to any one of clauses 47-70, further comprising the step of sensing the physiological parameter data of the subject using a medical device comprising one or more sensors.

[0087] Clause 78. The method of clause 77, wherein the medical device comprises a wearable medical device, the one or more sensors comprise a plurality of ECG sensors, the physiological parameter data of the subject comprises ECG data, and the potential medical event comprises a cardiac event.

[0088] Clause 79. The method of any one of clauses 47-78, further comprising the step of communicating at least one of the physiological parameter data and the estimated event risk score from the medical device to another computing device using a communication network.

[0089] Clause 80. The method of any one of clauses 47-79, further comprising the step of displaying, using a display controlled by the one or more processors, a time-based visual indication of the estimated risk score of the event for a plurality of time periods.

[0090] Clause 81. The method of any one of clauses 47-80, wherein the plurality of time periods comprises a plurality of time periods of less than 4 hours.

[0091] Clause 82. A method according to any one of clauses 47-80, wherein the multiple time periods include at least one time period of less than about 10 minutes, at least one time period of less than about 1 hour, at least one time period of less than about 3 hours, at least one time period of less than about 1 day, at least one time period of less than about 1 week, and at least one time period of less than about 1 month.

[0092] Clause 83. The method of any one of clauses 47-80, wherein calculating the estimated risk score for the event comprises calculating a confidence band for the estimated risk score for the event.

[0093] Clause 84. The method of any one of clauses 47-80, wherein calculating the estimated risk score for the event comprises calculating an error band for the estimated risk score for the event.

[0094] Clause 85. The method of any one of clauses 47-84, further comprising the step of determining a response to the potential medical event based at least in part on the estimated risk score for the event.

[0095] Clause 86. The method of Clause 85, wherein the determined response to the potential medical event comprises providing the subject with instructions to contact a medical professional.

[0096] Clause 87. The system of clause 85 or 86, wherein the determined response to the potential medical event comprises providing an instruction to the subject to check a battery of a wearable medical device.

[0097] Clause 88. The system of any of clauses 85-87, wherein the determined response to the potential medical event comprises charging a shock mechanism of a wearable medical device.

[0098] Clause 89. The system of any one of clauses 85-88, further comprising the step of determining a response to the potential medical event based at least in part on the sensitivity and specificity of the event estimated risk score.

[0099] Clause 90. The system of clause 89, wherein the determined response based on the first sensitivity and the first specificity is different from the determined response based on a second, different sensitivity and a second, different specificity.

[0100] Clause 91. A system according to any of clauses 85-90, wherein the determined response to the potential medical event includes at least one of: notifying the subject of an advanced diagnosis; advising the subject not to remove the device; advising the subject of a behavioral modification; alerting a medical professional; and preparing the device for treatment.

[0101] Clause 92. The system of any one of clauses 85-91, further comprising the step of modifying the sensitivity of an algorithm used to determine the estimated risk score for the event based on the risk level of the subject.

[0102] Item 93. A medical event estimation system comprising: at least one non-transitory computer-readable storage medium in communication with at least one processor and storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising: receiving at least one signal comprising physiological parameter data of a subject; and calculating, based at least in part on the physiological parameter data, a time estimate of a potential event associated with at least two different potential medical events of the subject.

[0103] Item 94. A medical event estimation system comprising: at least one non-transitory computer-readable storage medium in communication with at least one processor and storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising: receiving at least one signal comprising physiological parameter data of a subject; and calculating, based at least in part on the physiological parameter data, a risk score for a potential event based on a time estimate associated with at least two different potential medical events of the subject.

[0104] Item 95. A medical event estimation system comprising: a medical device configured to measure physiological parameter data of a subject, send a signal comprising the physiological parameter data of the subject, and perform a plurality of different actions in response to a plurality of different medical events; and at least one non-transitory computer-readable storage medium in communication with at least one processor and storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising: receiving at least one signal comprising the physiological parameter data of the subject; calculating a risk score based on a time estimate of a potential event associated with at least two different potential medical events of the subject based at least in part on the physiological parameter data; and controlling the medical device to perform at least one of the plurality of different actions based at least in part on at least one of the calculated risk scores.

[0105] Item 96. A method for medical event estimation, comprising the following steps: measuring physiological parameter data of a subject using a medical device; sending a signal including the physiological parameter data of the subject using the medical device; receiving at least one signal including the physiological parameter data of the subject using one or more processors; calculating, using the one or more processors, a risk score based on a time estimate of a potential event associated with at least two different potential medical events of the subject based at least in part on the physiological parameter data; and performing, using the medical device, at least in part on at least one of a plurality of different actions based at least in part on at least one of the calculated risk scores.

[0106] Item 97. A medical precursor event estimation system, comprising: a non-transitory computer-readable storage medium that communicates with one or more processors and stores instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: calculating an event estimation risk score associated with a potential medical event of the subject occurring within an associated time period based at least in part on a physiological parameter signal of the subject, wherein calculating the event estimation risk score comprises: extracting a plurality of physiological measurement values ​​from the physiological parameter signal of the subject; combining at least a portion of the plurality of physiological measurement values ​​into a multivariate parameter signal; applying change point analysis to the multivariate parameter signal to determine at least one change point; determining the distance of the at least one change point relative to a baseline; applying anomaly detection to the multivariate parameter signal to determine at least one anomaly score; and determining the event estimation risk score based at least in part on at least one of the distances and at least one of the anomaly scores.

[0107] Clause 98. The medical precursor event estimation system according to Clause 97, wherein the physiological parameter signal includes an ECG signal of the subject.

[0108] Clause 99. A medical precursor event estimation system according to clause 97 or 98, wherein the multiple physiological measurements extracted from the ECG signal include at least one of the following: heart rate, heart rate variability, PVC burden or count, activity, noise quantization, atrial fibrillation, transient pauses, heart rate oscillations, QRS height, QRS width, changes in morphological size or shape, cosine RT, artificial pacing, corrected QT interval, QT variability, T wave width, T wave alternans, T wave variability, ST segment changes, early repolarization, late potentials, fragmented QRS / HF content and fragmented T wave / HF content.

[0109] Clause 100. The medical precursor event estimation system of any one of clauses 97-99, wherein the physiological parameter signal further comprises an accelerometer signal.

[0110] Clause 101. The medical precursor event estimation system of clause 100, wherein calculating the event estimated risk score further comprises removing one or more ECG signal artifacts from the ECG signal based at least in part on the accelerometer signal.

[0111] Clause 102. A medical precursor event estimation system according to any one of clauses 97-101, wherein calculating the event estimation risk score further comprises: calculating a statistical value of the multivariate parameter signal for at least one interval of the multivariate parameter signal associated with the at least one change point; and determining a difference between at least one of the statistical values ​​and a corresponding statistical value of the baseline.

[0112] Clause 103. The medical precursor event estimation system of any one of clauses 97-102, wherein applying the anomaly detection further comprises determining a plurality of raw single parameter outputs corresponding to the plurality of physiological parameter measurements.

[0113] Clause 104. The medical precursor event estimation system of any one of clauses 97-103, wherein the anomaly detection comprises a neural network, and wherein the neural network is trained with the baseline of the multivariate parameter signal.

[0114] Clause 105. A medical precursor event estimation system according to any of clauses 97-104, wherein determining the event estimated risk score based at least in part on at least one of the distances and at least one of the anomaly scores includes: using one or more machine learning processes to classify at least one of the distances and at least one of the anomaly scores.

[0115] Clause 106. A medical precursor event estimation system according to any one of clauses 97-105, wherein calculating the event estimation risk score further comprises: applying change point analysis to the multivariate parameter signal to determine a plurality of change points; determining a plurality of distances of the plurality of change points relative to a baseline; applying anomaly detection to the multivariate parameter signal to determine a plurality of anomaly scores corresponding to the plurality of change points; and determining the event estimation risk score for the plurality of change points based on the plurality of distances and the plurality of anomaly scores.

[0116] Clause 107. The medical precursor event estimation system according to any one of clauses 97-106, further comprising a wearable medical device.

[0117] Item 108. A method for estimating a medical precursor event, the method comprising the following steps: utilizing one or more processors to calculate an event estimation risk score associated with a potential medical event of the subject occurring within an associated time period based at least in part on a physiological parameter signal of the subject, wherein calculating the event estimation risk score comprises: extracting a plurality of physiological measurement values ​​from the physiological parameter signal of the subject; combining at least a portion of the plurality of physiological measurement values ​​into a multivariate parameter signal; applying change point analysis to the multivariate parameter signal to determine at least one change point; determining a distance of the at least one change point relative to a baseline; applying anomaly detection to the multivariate parameter signal to determine at least one anomaly score; and determining the event estimation risk score based at least in part on at least one of the distances and at least one of the anomaly scores.

[0118] Clause 109. The method of clause 108, wherein the physiological parameter signal comprises an ECG signal of the subject.

[0119] Clause 110. A method according to clause 108 or 109, wherein the multiple physiological measurements extracted from the ECG signal include at least one of the following: heart rate, heart rate variability, PVC burden or count, activity, noise quantization, atrial fibrillation, transient pauses, heart rate oscillations, QRS height, QRS width, changes in size or shape of morphology, cosine RT, artificial pacing, corrected QT interval, QT variability, T wave width, T wave alternans, T wave variability, ST segment changes, early repolarization, late potentials, fragmented QRS / HF content and fragmented T wave / HF content.

[0120] Clause 111. The method of any one of clauses 108-110, wherein the physiological parameter signal further comprises an accelerometer signal.

[0121] Clause 112. The method of clause 111, wherein calculating the estimated risk score for the event further comprises removing one or more ECG signal artifacts from the ECG signal based at least in part on the accelerometer signal.

[0122] Clause 113. A method according to any one of clauses 108-112, wherein calculating the event estimated risk score further comprises: calculating a statistical value of the multivariate parameter signal for at least one interval of the multivariate parameter signal associated with the at least one change point; and determining a difference between at least one of the statistical values ​​and a corresponding statistical value of the baseline.

[0123] Clause 114. The method of any one of clauses 108-113, wherein applying the anomaly detection further comprises determining a plurality of raw single parameter outputs corresponding to the plurality of physiological parameter measurements.

[0124] Clause 115. The method of any one of clauses 108-114, wherein the anomaly detection comprises a neural network, and wherein the neural network is trained with the baseline of the multivariate parameter signal.

[0125] Clause 116. A method according to any of clauses 108-115, wherein determining the event estimated risk score based at least in part on at least one of the distances and at least one of the anomaly scores includes: using one or more machine learning processes to classify at least one of the distances and at least one of the anomaly scores.

[0126] Clause 117. A method according to any one of clauses 108-116, wherein calculating the event estimated risk score further comprises: applying change point analysis to the multivariate parameter signal to determine a plurality of change points; determining a plurality of distances of the plurality of change points relative to a baseline; applying anomaly detection to the multivariate parameter signal to determine a plurality of anomaly scores corresponding to the plurality of change points; and determining the event estimated risk score for the plurality of change points based on the plurality of distances and the plurality of anomaly scores.

[0127] Clause 118. A wearable medical device configured to perform the method of any of clauses 108-117.

[0128] Clause 119. A medical precursor event estimation system, comprising: a non-transitory computer-readable storage medium that communicates with one or more processors and stores instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: obtaining a first set of physiological information of the subject received during a first time period and at least partially based on a first ECG signal of the subject, and a second set of physiological information of the subject received during a second time period; calculating a first risk assessment associated with estimating the risk of a potential arrhythmic event of the subject based on applying the first set of physiological information and the second set of physiological information to one or more machine learning classifier models trained with a training metric; and a second risk score associated with the potential arrhythmic event, wherein the training metrics include at least one of: i) cardiac electrophysiological metrics of a plurality of first subjects; and ii) at least one of demographic metrics and medical history metrics of the plurality of first subjects, wherein the one or more machine learning classifier models are validated with validation metrics of a plurality of second subjects, and one or more thresholds of the one or more machine learning classifier models are set based on the validation; at least the first risk score and the second risk score associated with the potential arrhythmic event are provided as a time-varying series of risk scores; and the first risk score and the second risk score associated with estimating the risk of the potential arrhythmic event for the subject are classified based on the one or more thresholds.

[0129] Clause 120. A medical precursor event estimation system according to Clause 119, wherein the one or more thresholds include at least a high risk threshold and an immediate risk threshold, and the classification includes: for each of the first risk score and the second risk score, classifying the risk of the potential arrhythmic event of the subject as high risk based on the first risk score or the second risk score exceeding the high risk threshold; and classifying the risk of the potential arrhythmic event of the subject as immediate risk based on the first risk score or the second risk score exceeding the immediate risk threshold.

[0130] Clause 121. The medical precursor event estimation system of clause 119 or 120, wherein the classification comprises a time-varying classification of the subject's risk of the potential arrhythmic event based on the time-varying series of risk scores.

[0131] Clause 122. A medical precursor event estimation system according to any of clauses 119-121, wherein the classification includes adjusting the base specificity of the one or more machine learning classifier models to reduce type 1 errors (false positives) in the base classification of the risk of the potential arrhythmic event in the subject.

[0132] Clause 123. The medical precursor event estimation system of any one of clauses 119-122, wherein the validation metrics include one or more of cardiac electrophysiological metrics, demographic metrics, and medical history metrics of the plurality of second subjects.

[0133] Clause 124. A medical precursor event estimation system according to any one of clauses 119-123, wherein the one or more processors perform operations including: updating the verification metric by at least one of the following operations: 1) adjusting one or more metrics in the verification metric and 2) expanding the verification metric based on adding one or more additional subjects to the multiple second subjects; and refining the one or more thresholds based on the updated verification metric.

[0134] Clause 125. A medical precursor event estimation system according to any one of clauses 119-124, wherein the one or more processors perform operations including: updating the training metrics by at least one of the following operations: 1) adjusting one or more metrics in the training metrics and 2) expanding the training metrics based on adding one or more additional subjects to the multiple first subjects; and retraining the one or more machine learning classifier models based on the updated training metrics.

[0135] Clause 126. The medical precursor event estimation system according to any one of clauses 119-125, wherein the validation metrics of the plurality of second subjects are independent of the training metrics of the plurality of first subjects.

[0136] Clause 127. A medical precursor event estimation system according to any one of clauses 119-126, wherein the one or more machine learning classifier models are validated with the validation metric, and the validation metric includes an indication of the presence or absence of ectopic beats in one or more base validation metrics of the validation metric.

[0137] Clause 128. A medical precursor event estimation system according to any one of clauses 119-127, wherein the one or more processors perform operations including: distinguishing between normal beats and ectopic beats in the first physiological information set of the subject received during the first time period and based at least in part on the first ECG signal.

[0138] Clause 129. A medical precursor event estimation system according to any of clauses 119-128, wherein at least one of the training metric and the validation metric includes a metric based on at least one of heart rate, heart rate variability, number of non-sustained ventricular tachycardia (VT) episodes, and premature ventricular contraction (PVC) counts.

[0139] Clause 130. A medical precursor event estimation system according to any of clauses 119-129, wherein at least one of the training metric and the validation metric includes a metric based on heart rate variability, and the metric based on heart rate variability includes a standard deviation of normal intervals over time.

[0140] Clause 131. A medical precursor event estimation system according to any of clauses 119-130, wherein at least one of the training metric and the validation metric includes a metric based on at least one of QRS width, QRS height, single-lead QRS morphology and dual-lead QRS morphology.

[0141] Clause 132. A medical precursor event estimation system according to any one of clauses 119-131, wherein at least one of the training metric and the validation metric includes a metric based on a single-lead QRS morphology, and the metric based on a single-lead QRS morphology includes an average over time of similarity scores on each of the lateral (SS) channel and the front-back (FB) channel.

[0142] Clause 133. A medical precursor event estimation system according to any of clauses 119-132, wherein at least one of the training metric and the validation metric includes a metric based on QRS width, and the QRS width-based metric includes at least one of a standard deviation of the estimated width of the QRS complex wave over time and a mean of the estimated width of the QRS complex wave over time.

[0143] Clause 134. A medical precursor event estimation system according to any of clauses 119-133, wherein at least one of the training metric and the validation metric includes a QRS height-based metric, and the QRS height-based metric includes a standard deviation of the estimated height of the QRS complex wave over time.

[0144] Clause 135. A medical precursor event estimation system according to any of clauses 119-134, wherein at least one of the training metric and the validation metric includes a metric based on at least one of: QT variability; ST depression, elevation and / or slope; T wave alternans; T wave variability; and dual-lead T wave morphology.

[0145] Clause 136. The medical precursor event estimation system of any one of clauses 119-135, wherein at least one of the training metric and the validation metric comprises a heart sound metric.

[0146] Clause 137. The medical precursor event estimation system of clause 136, wherein the heart sound metric comprises an S3 heart sound metric and an S4 heart sound metric.

[0147] Clause 138. A medical precursor event estimation system according to any one of clauses 119-137, wherein at least one of the training metric and the validation metric includes an electromechanical activation time metric, wherein the electromechanical activation time metric describes the interval from a first predetermined reference time point of an electrocardiogram (ECG) to a second predetermined reference time point of a subsequent mechanical activity of the heart.

[0148] Clause 139. A medical precursor event estimation system according to any one of clauses 119-138, wherein the first predetermined reference time point of the ECG includes the starting points of the P wave and the QRS complex wave, wherein the starting points of the P wave and the QRS complex wave include time points related to at least one of a) a P wave, b) a Q wave, c) an R wave and d) an S wave.

[0149] Clause 140. The medical precursor event estimation system of any one of clauses 119-139, wherein the subsequent mechanical activity of the heart comprises left ventricular wall motion.

[0150] Clause 141. A medical precursor event estimation system according to any one of clauses 119-140, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart includes at least one of: a) the time point of maximum left ventricular wall motion; and b) the relaxation state of the left ventricular wall motion.

[0151] Clause 142. The medical precursor event estimation system according to any one of clauses 119-141, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart comprises a time point of a peak intensity of an S1 heart sound.

[0152] Clause 143. The medical precursor event estimation system according to any one of clauses 119-142, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart is based on an ultrasound measurement of the heart.

[0153] Clause 144. The medical precursor event estimation system of any of clauses 119-143, wherein the electromechanical activation time metric comprises a percentage electromechanical activation time metric.

[0154] Clause 145. The medical precursor event estimation system of any one of clauses 119-144, wherein the classifying comprises calculating at least one of an area under a plotted curve of the time-varying series of risk scores and a mean of the time-varying series of risk scores.

[0155] Clause 146. The medical precursor event estimation system of any of clauses 119-145, wherein the first risk score and the second risk score are classified based on an amount by which the first risk score and the second risk score exceed the one or more thresholds.

[0156] Clause 147. The medical precursor event estimation system of any of clauses 119-146, wherein the first risk score and the second risk score are classified based on a number of times the risk score of the time-varying series exceeds the one or more thresholds.

[0157] Clause 148. The medical precursor event estimation system of any one of clauses 119-147, wherein the one or more processors perform operations comprising notifying at least one of the subject and a third party based on the classification of the first risk score and the second risk score.

[0158] Clause 149. The medical precursor event estimation system of any one of clauses 119-148, wherein the classification of the first risk score and the second risk score represents at least one of high risk and immediate risk.

[0159] Clause 150. The medical precursor event estimation system of any of clauses 119-149, wherein the notifying comprises sending a notification to at least one member of the subject's medical team.

[0160] Clause 151. A medical precursor event estimation system according to any of clauses 119-150, wherein the one or more processors perform operations including adjusting a time interval between detection of a cardiac event and treatment of the cardiac event based on a classification of the first risk score and the second risk score.

[0161] Clause 152. An external medical device comprising the medical precursor event estimation system according to any one of clauses 119-151, wherein the external medical device is configured to monitor a cardiac condition of the subject.

[0162] Clause 153. The external medical device of clause 152, wherein the external medical device comprises a wearable medical device.

[0163] Clause 154. An external medical device according to clause 152 or 153, wherein the one or more processors perform operations including modifying one or more functions or features of a user interface of the external medical device based on the classification of the first risk score and the second risk score.

[0164] Item 155. A medical precursor event estimation system, comprising: a non-transitory computer-readable storage medium that communicates with one or more processors and stores instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: calculating an event estimation risk score associated with a potential medical event for a subject, wherein the event estimation risk score is calculated based on a machine learning classifier, wherein the machine learning classifier is trained with training metrics including at least one of the following: i) cardiac electrophysiological metrics of multiple subjects; and ii) at least one of demographic metrics and medical history metrics of multiple subjects.

[0165] Clause 156. A medical precursor event estimation system according to clause 155, wherein the machine learning classifier is validated based on a validation metric independent of the training metric, and at least one threshold of the machine learning classifier is set based on the validation.

[0166] Clause 157. The medical precursor event estimation system of clause 155 or 156, wherein the one or more processors perform operations comprising classifying the event estimated risk score based on at least one threshold.

[0167] Clause 158. The medical precursor event estimation system of any of clauses 155-157, wherein the at least one threshold is substantially continuously refined based on registration of substantially continuously updated metrics.

[0168] Clause 159. A medical precursor event estimation system according to any one of clauses 155-158, wherein the one or more processors perform operations including: calculating multiple event estimated risk scores associated with the potential medical event of the subject within a time series; and analyzing the multiple event estimated risk scores to determine at least one trend associated with the potential medical event of the subject within the time series.

[0169] Clause 160. The medical precursor event estimation system of any one of clauses 155-159, wherein at least one of the training metric and the validation metric comprises a heart sound metric.

[0170] Clause 161. The medical precursor event estimation system of any one of clauses 155-160, wherein the heart sound metric comprises an S3 heart sound metric and an S4 heart sound metric.

[0171] Clause 162. A medical precursor event estimation system according to any one of clauses 155-161, wherein at least one of the training metric and the validation metric includes an electromechanical activation time metric, which describes the interval from a first predetermined reference time point of an electrocardiogram (ECG) to a second predetermined reference time point of a subsequent mechanical activity of the heart.

[0172] Clause 163. A medical precursor event estimation system according to any one of clauses 155-162, wherein the first predetermined reference time point of the ECG includes the starting points of the P wave and the QRS complex wave, wherein the starting points of the P wave and the QRS complex wave include time points related to at least one of a) a P wave, b) a Q wave, c) an R wave and d) an S wave.

[0173] Clause 164. The medical precursor event estimation system of any one of clauses 155-163, wherein the subsequent mechanical activity of the heart comprises left ventricular wall motion.

[0174] Clause 165. A medical precursor event estimation system according to any one of clauses 155-164, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart includes at least one of: a) the time point of maximum left ventricular wall motion; and b) the relaxation state of the left ventricular wall motion.

[0175] Clause 166. The medical precursor event estimation system according to any one of clauses 155-165, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart comprises a time point of a peak intensity of an S1 heart sound.

[0176] Clause 167. The medical precursor event estimation system according to any one of clauses 155-166, wherein the second predetermined reference time point of the subsequent mechanical activity of the heart is based on an ultrasound measurement of the heart.

[0177] Clause 168. The medical precursor event estimation system of any of clauses 155-167, wherein the electromechanical activation time metric comprises a percentage electromechanical activation time metric.

[0178] Item 169. A method for estimating a medical precursor event, comprising the following steps: obtaining a first set of physiological information of the subject received during a first time period and at least partially based on a first ECG signal of the subject, and a second set of physiological information of the subject received during a second time period; calculating a first risk score and a second risk score associated with estimating the risk of a potential arrhythmic event for the subject based on applying the first set of physiological information and the second set of physiological information to one or more machine learning classifier models trained with a training metric, wherein the training metric includes at least one of the following: i) cardiac electrophysiological metrics of multiple first subjects; and ii) at least one of demographic metrics and medical history metrics of the multiple first subjects, wherein the one or more machine learning classifier models are validated with validation metrics of multiple second subjects, and one or more thresholds of the one or more machine learning classifier models are set based on the validation; providing at least the first risk score and the second risk score associated with the potential arrhythmic event as a time-varying series of risk scores; and classifying the first risk score and the second risk score associated with estimating the risk of the potential arrhythmic event for the subject based on the one or more thresholds.

[0179] Clause 170. A method for estimating a medical precursor event according to Clause 169, wherein the one or more thresholds include at least a high-risk threshold and an immediate-onset risk threshold, and the classification includes: for each of the first risk score and the second risk score, classifying the risk of the potential arrhythmic event of the subject as high risk based on the first risk score or the second risk score exceeding the high-risk threshold; and classifying the risk of the potential arrhythmic event of the subject as immediate-onset risk based on the first risk score or the second risk score exceeding the immediate-onset risk threshold.

[0180] Clause 171. A method for estimating a medical precursor event according to clause 169 or 170, wherein the classification comprises a time-varying classification of the risk of the potential arrhythmic event of the subject based on the time-varying series of risk scores.

[0181] Clause 172. A method for estimating a medical precursor event according to any of clauses 169-171, wherein the classification includes adjusting the base specificity of the one or more machine learning classifier models to reduce type 1 errors (false positives) in the base classification of the risk of the potential arrhythmic event in the subject.

[0182] Clause 173. The medical precursor event estimation method according to any one of clauses 169-172, wherein the validation metric comprises one or more of cardiac electrophysiological metrics, demographic metrics, and medical history metrics of the plurality of second subjects.

[0183] Clause 174. The medical precursor event estimation method of any one of clauses 169-173, wherein at least one of the training metric and the validation metric comprises a heart sound metric.

[0184] Clause 175. The method for estimating a medical precursor event according to clauses 169-164, wherein the heart sound metric includes an S3 heart sound metric and an S4 heart sound metric. BRIEF DESCRIPTION OF THE DRAWINGS

[0185] These and other features and characteristics of the present invention, as well as the methods of operation and function of the related elements of construction and the combination of parts and economies of manufacture, will become more apparent upon consideration of the following description and appended claims, all of which form a part hereof, with reference to the accompanying drawings, in which like reference numerals designate corresponding parts throughout the several figures. It should be expressly understood, however, that the drawings are for purposes of illustration and description only and are not intended as a definition of the limits of the present invention.

[0186] Further features of the present invention as well as other objects and advantages will become apparent from the following detailed description made with reference to the accompanying drawings, in which:

[0187] Figure 1 is a graph of an example electrocardiogram (ECG) signal of a heart in sinus rhythm;

[0188] Figure 2 A wearable medical device is shown;

[0189] Figure 3 is a block diagram of a medical device;

[0190] Figure 4 is a flow chart of a method for estimating and detecting medical precursor events;

[0191] Figure 5 A system for responding to a medical condition is shown;

[0192] Figure 6A and 6B is a plot of example estimated risk scores;

[0193] Figure 7 This is a flow chart of another method for estimating medical precursor events;

[0194] Figure 8A shows the distribution of estimated risk scores for events for patients;

[0195] Figure 8B An example ECG signal for determining a single-lead QRS morphology metric is shown;

[0196] Figure 8C An example ECG signal for determining a heart rate variability metric is shown;

[0197] Figure 8D An example of QRS width is shown;

[0198] Figure 8E An example of QRS height is shown;

[0199] Figure 8F is a flow chart of a process for determining an estimated risk score of an event and a corresponding classification;

[0200] Figure 8G is a block diagram of a system for determining, classifying, and informing an estimated risk score of an event;

[0201] Figure 8H It is a flow chart of the state change analysis method used for estimating medical precursor events;

[0202] Figure 8I A two-lead ECG signal is shown for extracting PVC parameters from the ECG signal;

[0203] Figure 8J An example multivariate parameter signal including a change point is shown;

[0204] Figure 8K shows a multivariate summation of example abnormality scores for treatment subjects;

[0205] Figure 8L shows change-of-state results including sensitivity and specificity percentages for predicting medical precursor events based on an example test set of 200 patients;

[0206] Figure 8M shows the multivariate feature selection process;

[0207] Figure 9It is a flow chart of the vector loop bifurcation of vectorcardiography;

[0208] Figure 10 is a diagram of an example subject's ECG waveform;

[0209] Figure 11 is a graph that plots example actions based on criticality (risk) measures and confidence measures;

[0210] Figure 12 is a graph of example actions against time periods;

[0211] Figure 13 shows an example of a similarity matrix illustrating K-means clustering performed on accelerometer signals;

[0212] Figures 14A to 14D shows example screenshots for an example cardiac arrest prediction (CAP) application;

[0213] Figure 15 shows example physiological parameters that the CAP application may consider when calculating the Modified Early Warning (MEW) score;

[0214] Figure 16 shows example patient characteristics that may be considered when calculating the Glasgow Coma Scale (GCS) score; and

[0215] Figure 17 It is a flow chart of the method for estimating medical precursor events. DETAILED DESCRIPTION

[0216] Provided herein are improved systems and methods for automatically analyzing subject data to predict conditions that may occur for the subject, including medical events (e.g., adverse cardiac events). One or more event estimated risk scores based on the expected time of occurrence of the event can be generated. For example, event estimated risk scores can be calculated for multiple time periods (e.g., such that each time period has its own event estimated risk score). In this way, multiple different event estimated risk scores can be calculated for multiple different time periods (e.g., less than about 10 minutes, less than about 1 hour, less than about 3 hours, less than about 1 day, less than about 1 week, and / or less than about 1 month, etc.). The event estimated risk score provides a measure of the likelihood or probability of one or more medical events occurring within the associated time period.

[0217] The estimated risk score for an event for one or more time periods can be compared to a stored estimated risk threshold for the event associated with the corresponding time period to determine an action plan or response for the event. The action plan or response can change based on the predicted time for the event and / or the estimated risk score for the event. Thus, the action plan or response determined for an event with a more distant probability of occurrence may differ from the action plan or response determined for an event with a more recent probability of occurrence.

[0218] As used herein, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise.

[0219] As used herein, the terms "end," "upper," "lower," "right," "left," "vertical," "horizontal," "top," "bottom," "lateral," "longitudinal," and their derivatives shall refer to the present invention as it is directed in the accompanying drawings. However, it should be understood that the present invention may assume various alternative orientations, and thus these terms should not be considered limiting. Furthermore, it should be understood that the present invention may assume various alternative variations and stage sequences, unless expressly indicated to the contrary. It should also be understood that the specific devices and processes shown in the drawings and described in the following specification are merely typical embodiments of the present invention. Thus, specific dimensions and other physical characteristics related to the embodiments disclosed herein should not be considered limiting.

[0220] For the purposes of this specification, unless otherwise indicated, all numbers used in this specification and claims indicating amounts of ingredients, reaction conditions, dimensions, and physical properties are to be understood as being modified in all instances by the term "about". Therefore, unless otherwise indicated, the numerical parameters set forth in the following specification and the appended claims are approximate values ​​that may vary depending on the desired properties sought to be obtained by the present invention. At least rather than attempting to limit the scope of the application of the doctrine of equivalents to the claims, each numerical parameter should at least be interpreted in light of the number of reported significant figures and by applying ordinary rounding techniques.

[0221] Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the invention are approximations, the numerical values ​​set forth in the specific examples are reported as precisely as possible. Any numerical value, however, inherently contains certain errors necessarily resulting from the standard deviation found in their respective testing measurements.

[0222] In addition, it should be understood that any numerical range described herein is intended to include all subranges contained within the numerical range. For example, the range "1 to 10" is intended to include any and all subranges between the stated minimum value of 1 and the stated maximum value of 10 (including endpoints), that is, all subranges starting from a minimum value equal to or greater than 1 and ending with a maximum value equal to or less than 10, as well as all subranges such as 1 to 6.3, or 5.5 to 10, or 2.7 to 6.1.

[0223] As used herein, the terms "communication" and "communicating" refer to the reception or transmission of one or more signals, messages, commands or other types of data. A unit or component communicates with another unit or component, which means that one unit or component is able to directly or indirectly receive data and / or send data relative to another unit or component. This can refer to a direct or indirect connection that can be wired and / or wireless in nature. In addition, even if the data sent can be modified, processed and sent between the first unit or component and the second unit or component, the two units or components can communicate with each other. For example, even if the first unit passively receives data and does not actively send data to the second unit, the first unit can communicate with the second unit. As another example, if an intermediate unit processes data from one unit and sends the processed data to the second unit, the first unit can communicate with the second unit. It should be understood that many other configurations are also possible.

[0224] In general, systems and methods are provided for automatically analyzing subject data to predict conditions including medical events (e.g., adverse cardiac events). In some embodiments, systems and methods are provided for generating one or more event estimation risk scores based on the expected time of the event. Here, the event estimation risk score is sometimes referred to as a cardiac arrest prediction (CAP) score. In some embodiments, the following systems and methods are provided: for multiple different time periods (e.g., less than about 10 minutes, less than about 1 hour, less than about 3 hours, less than about 1 day, less than about 1 week and / or less than about 1 month, etc.), multiple different event estimation risk scores are calculated. The event estimation risk score provides a measurement of the likelihood or probability of one or more medical events occurring within the associated time period.

[0225] Systems and methods according to some embodiments can compare different event estimated risk scores for time periods with stored event estimated risk thresholds associated with those time periods to determine an action plan or response for an event that changes based on the predicted time of the event and / or the event estimated risk score. Thus, the action plan or response determined for an event that is more likely to occur in the distant future may be different from the action plan or response determined for an event that is more likely to occur in the near future.

[0226] Non-limiting example methods for predicting treatment efficacy in patients, generally based on various vital signs, are described in the following patents: U.S. Patent No. 8,668,644, filed April 23, 2013, entitled “METHOD OF PREDICTING ACUTE CARDIOPULMONARY EVENTS AND SURVIVABILITY OF A PATIENT”; U.S. Patent No. 8,932,220, filed March 6, 2014, entitled “METHOD OF PREDICTING ACUTE CARDIOPULMONARY EVENTS AND SURVIVABILITY OF A PATIENT”; U.S. Patent No. 8,951,193, filed March 6, 2014, entitled “METHOD OF PREDICTING ACUTE CARDIOPULMONARY EVENTS AND SURVIVABILITY OF A PATIENT”; U.S. Patent No. 8,951,193, filed December 12, 2014, entitled “PREDICTING ACUTE CARDIOPULMONARY EVENTS AND and PCT Patent Publication No. WO 2011 / 115576, filed March 14, 2011, entitled “Method of predicting the survivability of a patient,” all of which are incorporated herein by reference in their entirety.

[0227] The time period may include any duration. For example, the time period may include a time period from the current time to a future time point, a time period between two future time points, a certain instant, and / or any combination thereof.

[0228] Systems and methods for detecting and estimating medical precursor events based on various time periods are described herein. Systems and methods according to some embodiments provide different suggested action plans or responses based on the detection and estimation of medical precursor events. Systems for estimating and detecting medical precursor events may be referred to herein as "early warning systems."

[0229] The Oxford English Dictionary defines "premonition" as "a strong feeling that something is about to happen, especially something unpleasant." The Latin roots of "premonition" are prae (before) + monere (warning). As used herein, "premonition" means an indication that there is a possibility or probability of something happening, and "medical premonitory event" means a medical event that there is a possibility or probability of happening to a subject. Thus, the detection and estimation of medical premonitory events can be used as an early warning system to provide individuals and / or medical professionals with time to prepare for a medical event, such as a potentially adverse or fatal deterioration in the subject's or patient's medical condition, potentially mitigate or avoid the adverse effects of the deterioration, or even potentially avoid the deterioration or event altogether through timely appropriate treatment.

[0230] Non-limiting examples of these medical events may include cardiac events such as myocardial infarction or cardiac arrest, profound bradycardia due to acute decompensated heart failure, acute coronary syndrome, etc. Non-limiting examples of deterioration of a medical condition may include the onset of a disease state, progression or worsening of a disease state, and / or an adverse medical event such as an irregular heartbeat, a heart attack, a subject suffering a traumatic injury who experiences a potentially fatal rapid drop in blood pressure due to difficult-to-detect internal bleeding, etc. Other possible medical events or deterioration of a subject's medical condition may be caused by, for example, physical injury, diabetes, septic shock, seizures, or epilepsy.

[0231] Non-limiting examples of cardiac precursor events may include tangible events that can be seen by an experienced clinician, such as ectopic beats, a series of ectopic beats, ventricular tachycardia, bradycardia, and / or irregularities or abnormalities in P waves, QRS complexes, T waves, and U waves. For example, irregularities or abnormalities in the electrical activity of the heart may include flat T waves, inverted T waves, hyperacute T waves or peaked T waves, beat-to-beat T wave variations, shortened QT intervals, prolonged QT intervals, wide QRS, prominent U waves, etc. Alternatively or in addition, medical precursor events may include intermediate events, such as the detection of clusters of events, an acceleration in the rate of events, an increase in the intensity or criticality of events, etc. Alternatively or in addition, medical precursor events may include higher-order events, where these higher-order events may be defined, for example, in a multidimensional parameter space (e.g., parameters including electrocardiogram ("ECG") data and / or other relevant physiological parameters and / or subject demographics and other health history). Non-limiting examples of precursor events are discussed in more detail below.

[0232] The systems and methods described herein are often described with respect to subjects who may be at risk for a cardiac event; however, the disclosed embodiments are not limited to cardiac events. The systems and methods described herein can be used to detect and estimate any medical precursor event based on any monitored physiological parameter of the subject.

[0233] Subjects who may be at risk for a cardiac event (e.g., cardiac arrest, ventricular tachycardia ("VT"), ventricular fibrillation ("VF"), pulseless electrical activity ("PEA"), cardiac arrest, etc.) may be monitored for indications of an impending cardiac event so that action may be taken to reduce the probability of a cardiac event occurring and / or mitigate harm to the subject resulting from the cardiac event. An event estimated risk score may be determined for any combination of these cardiac events. Subjects to be monitored may include subjects who have experienced a cardiac event in the past, subjects who are recovering from cardiac surgery or other procedures, and / or subjects who have demonstrated other signs of possible cardiac dysfunction (e.g., otherwise unexplained loss of consciousness, rapid heartbeat, or chest pain).

[0234] The external medical device as described herein may include, but is not limited to, one or more cardiac sensing electrodes, a controller, and a user interface. The cardiac sensing electrodes monitor cardiac electrophysiological phenomena and can be used to obtain electrocardiographic records such as p-waves, q-waves, r-waves, s-waves, t-waves, and u-waves, as well as premature ventricular contractions, tachyarrhythmias, and cardiac waveform variations. These electrocardiographic records and related information derived from the cardiac rhythm data can be used alone or in combination with demographic data and medical history to classify subjects at high risk of sudden cardiac death that may be caused by cardiac arrest or cardiac asystole. The controller can be adjusted to analyze a combination of cardiac electrophysiological phenomena and subject data, wherein the combination includes information obtained from electrocardiographic records from several hours to as little as 45 seconds or less. Electrocardiographic records can be obtained for each subject once or multiple times while the subject is wearing the device. For example, in an implementation involving multiple recordings, time-dependent measurements of changes in risk classification for sudden cardiac death can be obtained. The user interface may provide a visual display with audio and vibration components, as well as cellular and wireless Internet connections, any or all of which may be used as a way to provide a summary of risk information to the subject, first responders, or medical professionals.

[0235] In implementations, the systems, methods, and devices described herein can be used to classify subjects by risk of sudden cardiac death, for example, based on an estimated event risk score. For example, the principles described herein can be implemented in a cardiac monitoring and / or treatment device. For example, at least one embodiment generally relates to a cardiac monitoring device, such as for use in mobile cardiac telemetry (MCT) and / or continuous event monitoring (CEM) applications.

[0236] In some implementations, at least one embodiment relates generally to wearable therapeutic devices, and more specifically to wearable therapeutic devices configured to monitor cardiac electrophysiological phenomena in a subject. Embodiments include an electrocardiogram of the subject as well as demographic information and medical history (such as gender, age, left ventricular ejection fraction, comorbidities, and the cardiac indication that led to the prescription of the wearable medical device). In some embodiments, the subject is evaluated for the likelihood that one or more risk metrics associated with sudden cardiac death will remain for an extended period of months to years. For example, the subject can be evaluated for the long-term risk of maintaining a reduced left ventricular ejection fraction (e.g., ≤35%) or the likelihood that he or she will continue to exhibit premature ventricular contractions, tachyarrhythmias, or changes in cardiac waveform morphology.

[0237] The parameters or metrics of the subject that can be monitored include various parameters of the subject's ECG. ECG parameters may include, for example, T-wave changes such as T-wave alternans or T-wave instability (e.g., not limited to alternating beats but changes in the morphology of the T-wave that occur for each (e.g., third, fourth, fifth, etc.) beat), PR interval, QT interval, QRS complex, heart rate variability (HRV), and / or other parameters of the subject's ECG. For example, in Figure 1 and Figures 8B to 8E A portion of the ECG signal corresponding to these parameters or metrics is shown in FIG, and this portion of the ECG signal can be non-invasively monitored, such as by an ECG monitor or an automatic external defibrillator (AED) when the device includes an ECG monitor or similar functionality.

[0238] A non-exhaustive list of other measurable parameters or metrics of a subject that may predict an impending cardiac event may include a low and decreasing heart rate (i.e., bradycardia, collectively referred to as "Bradying down"), an increasing heart rate, blood pressure, expiratory CO2 (i.e., the concentration or partial pressure of carbon dioxide in the subject's breathing gas), SpO2 (i.e., a measure of blood oxygen saturation), SMO2 (i.e., a measure of muscle oxygen saturation), cerebral blood flow, electroencephalogram (EEG) signals, electromechanical activation time (EMAT) (as described in further detail below), heart sounds (e.g., S3 and S4 sounds), brain oxygen level (i.e., cerebral blood oxygen saturation), tissue pH, ultrasound images of the subject's heart, and the response of the subject's heart rate to the subject's tilt.

[0239] Parameters of the subject that can be monitored for indications of an impending cardiac event can be monitored invasively or non-invasively, and many parameters of the subject can be monitored using optical parameters. For example, sublingual CO2, brain oxygen levels, and tissue pH can be measured optically.

[0240] Measurable parameters or metrics based on heart sounds may include acoustic cardiographic metrics derived from the acquisition and quantitative measurement of combined ECG and cardiac acoustic data. These metrics may include metrics that identify and quantify normal and abnormal heart sounds, such as those associated with the left ventricle, and indicate the timing of heart sounds in each cardiac cycle relative to the onset of the P wave and QRS complex in a roughly contemporaneously recorded ECG. For example, these metrics may indicate the presence and intensity of heart sounds (such as S3 and S4, etc.) and the duration of systolic intervals.

[0241] In some examples, this acoustic and / or combined acoustic and ECG metric can comprise electromechanical activation time (EMAT) metric.For example, as used herein, EMAT metric describes the interval from some reference time points in electrocardiogram (ECG) until some reference time points in the subsequent mechanical activity of the heart.An example of the reference time point in ECG is the starting point of the QRS complex wave such as P wave, Q wave, R wave or S wave etc.

[0242] An example of mechanical activity of the heart is valve closure, such as closure of the aortic valve. In some examples, the mechanical activity may be left ventricular wall motion, and the reference time point may be the time point of maximum left ventricular wall motion, or the reference time point may be the onset of diastole. As illustrated in the example of the present invention, heart sounds can be used to measure the reference time point of the subsequent mechanical activity of the heart, or ultrasound measurements of ventricular or valve motion can be used to measure the reference time point of the subsequent mechanical activity of the heart. In some implementations, an example of the reference time point in the subsequent mechanical activity of the heart can be the time point of the peak intensity of the S1 heart sound.

[0243] For example, these EMAT metrics can provide a measure of the amount of time that the left ventricle may need to generate sufficient force to close the mitral valve. In cases involving systolic heart failure, for example, EMAT measurements may be prolonged.

[0244] For example, an EMAT metric includes a percentage EMAT (%EMAT), which can be calculated as dividing the EMAT time measurement by the measurement of the dominant RR interval. The %EMAT metric is related to the efficiency of the heart's pump function. For example, in some cases, as described in Chao T et al., Prediction of Discharge Outcomes in Patients Hospitalized with AHFS, Internal Med. 2010, 49: 2031-2037, %EMAT>15% can predict the likelihood of a patient developing heart failure-related complications (e.g., the possibility of rehospitalization).

[0245] In addition, as described by Efstratiadis S et al. in Computerized Acoustic Cardiographic Electromechanical Activation Time Correlates with Invasive and Echocardiogrpahic Parameters of LV Contractility in J of Card. Failure. 2008, 14(7): 577-582, a shortened EMAT measurement can be associated with increased contractility and short electromechanical delay. As described above, the S3 and S4 heart sound metrics can be used in specific prediction models described herein, such as estimating a risk score associated with a specific treatment effect. For example, the S3 and S4 heart sound metrics can predict worsening of heart failure. Therefore, the S3 and S4 metrics can be included in addition to other metrics input into the prediction model described herein. According to the principles described herein, S3 and / or S4 can predict the need for increasing medication or preparing for hospitalization related to myocardial infarction and / or heart failure complications.

[0246] For example, in adults (e.g., over the age of approximately 40), the S3 metric can be used to indicate a correlation with ventricular dysfunction or volume overload. Examples of related conditions include valvular regurgitation, high-output states, left-right intracardiac shunts, complete atrioventricular block, renal failure, and volume overload from water intake or blood transfusions. In an implementation, S3 can represent congestive heart failure caused by volume overload (water intake). For example, an S3 heart sound can be detected in situations where volume overload is severe. In some implementations, the presence of S3 can predict an increased risk of sudden death, such as due to myocardial stretch. In some situations involving a high risk score and the presence of S3, for example, a caregiver may wish to respond by adding or increasing a diuretic.

[0247] The S4 metric can indicate an increase in cardiac filling pressures, such as increased resistance to filling the left or right ventricle due to decreased ventricular wall compliance, and can be accompanied by a disproportionate increase in ventricular end-diastolic pressure. For example, an S4 in individuals aged approximately 50 years or older can predict patients with systemic hypertension, aortic stenosis, hypertrophic cardiomyopathy, and coronary artery disease, or worsening of these conditions.

[0248] For example, during an episode of angina, S4 may become apparent or the intensity of S4 may increase. For example, the presence of S4 may indicate the early stages of an acute myocardial infarction that may or may not be accompanied by signs of heart failure. In situations involving a high risk score and the presence of S4, for example, a caregiver may wish to respond by treating the angina (ischemia) with increasing daily or situational medications (e.g., nitrates).

[0249] In some implementations, the predictive models described herein can use heart sound metrics, such as by comparing a patient's heart sounds to a baseline model, to estimate a risk score associated with a particular treatment effect. For example, if a patient's heart sounds deviate from a baseline model, the predictive model can determine that the heart sounds are abnormal, and thus can predict an adverse event (e.g., predicting that a shock is being delivered). The degree to which the patient's heart sounds deviate from the baseline model can correspond to an impact on the patient's risk score. For example, a heart score that deviates significantly from the baseline model may have a relatively greater impact on the patient's risk score than a heart sound that deviates minimally from the baseline model.

[0250] In some implementations, heart sounds can be sampled as 10-second snapshots. 10-second snapshots can be sampled every 5 minutes over a 4-hour period. For example, 10-second snapshots can be sampled every 5 minutes during the nighttime hours of each day (e.g., from midnight until 4 AM), resulting in a total of 48 snapshots collected each day. The snapshots can include lateral (SS) and anteroposterior (FB) ECG signals. In some implementations, the snapshots can also include x-accelerometer signals, y-accelerometer signals, and z-accelerometer signals.

[0251] A template (e.g., a general representation of sampled data) can be constructed based on snapshots measured on the first day of use of the medical device described herein. A baseline model can be based on the patient's heart sounds (e.g., the patient's heart sounds measured on the first day of use of the medical device). For example, the baseline model can include a subset of the constructed templates. Templates can be constructed based on individual snapshots; 48 templates can be constructed based on 48 snapshots. An algorithm (e.g., a heart sound template algorithm) can be used to construct templates based on snapshots. The heart sound template algorithm can construct templates by considering heart sound samples, SS and FB ECG signals, and accelerometer signals. As described in more detail below, the heart sound template algorithm can also calculate the similarity of one template relative to another template, where the similarity can be expressed as a scalar value from 0 to 1. In some implementations, a subset of templates (e.g., representative templates) can be selected from the 48 templates so that these representative templates constitute the baseline model.

[0252] In some implementations, a representative template can be selected from the 48 templates based on the signal-to-noise ratio (SNR) of the template. For example, templates with low SNR can be discarded. The remaining templates can then be divided into multiple clusters. For example, the remaining templates can be divided into three clusters, where each cluster corresponds to a body position (e.g., supine, left side lying, right side lying). The body position can be determined based on the x accelerometer signal, the y accelerometer signal, and the z accelerometer signal. For example, referring to Figure 13 , K-means clustering can be performed on the accelerometer signals corresponding to each template to separate the templates into three clusters (eg, C1, C2, and C3). Figure 13 The similarity matrix 1300 shows three clusters C1, C2 and C3 represented by groups of rows and columns. The numerical identifiers of the rows and columns each identify a possible template including multiple samples. The similarity between the various possible templates is represented as a scalar value from 0 to 1 and is represented as a color in the similarity matrix 1300, so that relatively darker colors in the similarity matrix 1300 correspond to templates with relatively higher similarity. Each possible template can be compared with the samples that constitute each other template. For example, when the template with the numerical identifier "1" on the x-axis is compared with the sample of the template with the numerical identifier "33" on the y-axis, the similarity is relatively low (for example, the corresponding color is light gray). On the other hand, when the template with the numerical identifier "1" on the x-axis is compared with the sample of the template with the numerical identifier "2" on the y-axis, the similarity is relatively high (for example, the corresponding color is dark gray). When a template having a numerical identifier of "1" on the x-axis is compared with a sample of the template having a numerical identifier of "1" on the y-axis (for example, when a template is compared with the same sample that constitutes the template), the similarity has the highest possible value (for example, the corresponding color is black).

[0253] A representative template can be selected for each cluster. In some implementations, the template with the greatest mean similarity to other templates in the cluster can be selected as the representative template of the cluster. For example, for cluster C i (i=1,2,3), the representative template M={τ1′,τ2′,τ3′} can be selected according to the following equations (1), (2) and (3):

[0254] τ i '=argmax Q(C i |τ)

[0255] τ∈Ci(1)

[0256]

[0257] g(v|τ): similarity function (3)

[0258] In some implementations, once the baseline model is established, an anomaly metric can be determined for each snapshot collected each day (e.g., 48 snapshots). In some implementations, snapshots with low SNRs are discarded, so that fewer than 48 anomaly metrics can be determined. For each snapshot, an anomaly metric can be determined based on the similarity between a template derived from the snapshot and the baseline model (e.g., the corresponding representative template).

[0259] If the SNR is less than 3 dB, the snapshot may be identified as having a low SNR (and, for example, may be discarded accordingly).The SNR of a snapshot may be determined based on an algorithm (eg, a heart sound template algorithm).

[0260] Sensors for monitoring any one or more of the parameters of the subject listed herein (which may provide an indication of an impending cardiac event) may be included in a medical device (e.g., a wearable medical device) or may be included as part of an AED or other monitoring modality or life support device. The medical device may be portable or wearable by the subject, or fixed or attached to a support. Figure 2 A wearable medical device according to one non-limiting embodiment is shown, which is available from ZOLL Medical, Inc., located in Chelmsford, Massachusetts. Wearable cardioverter defibrillators are similar and may include sensors for monitoring parameters of the subject that may predict an impending cardiac event. Non-limiting examples of suitable wearable defibrillators are disclosed in U.S. Patent Nos. 4,928,690, 5,078,134, 5,741,306, 5,944,669, 6,065,154, 6,253,099, 6,280,461, 6,681,003, 8,271,082, and 8,369,944; all of which are incorporated herein by reference in their entirety. However, the disclosed embodiments are not limited to wearable medical devices, and medical devices used to monitor any one or more of the parameters of the subject listed herein may include implantable medical devices, fixed devices (such as those used in hospitals or bedside settings such as defibrillators or monitors), portable medical devices (such as automated external defibrillators or mobile outpatient cardiac telemetry ("MCOT") devices), or other monitors of the subject's physiological parameters.

[0261] like Figure 2 As shown, in some embodiments, the wearable medical device 100 includes a strap 110 having a pair of shoulder pads and a waist belt worn around the torso of the subject. The wearable medical device 100 includes a plurality of ECG sensing electrodes 112, wherein the plurality of ECG sensing electrodes 112 are mounted to various locations of the strap 110 around the subject's body during use and are electrically connected to the control unit 120 via a connector 130. The plurality of ECG sensing electrodes 112, which may be dry-sensitive capacitive electrodes, are used by the control unit 120 to monitor the subject's cardiac function and may include a pair of front and back ECG sensing electrodes and a pair of side ECG sensing electrodes. Additional ECG sensing electrodes and / or other sensors may be provided, and the plurality of ECG sensing electrodes 112 may be configured at various locations on the patient's body.

[0262] The wearable medical device 100 includes a plurality of treatment electrodes 114a, 114b, wherein the plurality of treatment electrodes 114a, 114b are electrically connected to the control unit 120 via a connector 130 and are capable of delivering such treatment to the patient's body if it is determined that one or more treatment defibrillation shocks are warranted. The plurality of treatment electrodes 114a, 114b may include a first treatment electrode 114a disposed on the front of the subject's torso and a second treatment electrode 114b disposed on the back of the subject's torso. The second treatment electrode 114b includes a pair of treatment electrodes electrically connected together and serving as the second treatment electrode 114b. The use of two treatment electrodes 114a, 114b allows for the delivery of a biphasic shock to the subject's body, such that the first of the two treatment electrodes can deliver the first phase of the biphasic shock (with the other treatment electrode serving as a return), and the other treatment electrode can deliver the second phase of the biphasic shock (with the first treatment electrode serving as a return). The connector 130 electrically connects the plurality of ECG sensing electrodes 112 and the plurality of therapy electrodes 114a, 114b to the control unit 120 and may include various electronic circuits. For example, the connector 130 may include a signal acquisition circuit such as a plurality of differential amplifiers, wherein the signal acquisition circuit is configured to receive ECG signals from different ECG sensing electrodes in the plurality of ECG sensing electrodes 112 and provide a differential ECG signal to the control unit 120 based on the difference between these ECG signals. The connector 130 may include other electronic circuits, such as a motion sensor (e.g., an accelerometer and / or a gyroscope) that can monitor the subject's activity.

[0263] like Figure 2 As shown, the wearable medical device 100 includes a user interface 140, wherein the user interface 140 is electrically connected to the control unit 120. The user interface 140 can be attached to the subject's clothing or the harness 110, for example, via a clip (not shown) attached to a portion of the interface 140. Alternatively, the user interface 140 can be simply held in the hand of the subject or another person. The user interface 140 can be attached to the subject's clothing or the harness 110, for example, using a clip (not shown) attached to a portion of the interface 140. Alternatively, the user interface 140 can be simply held in the hand of the subject or another person. A wireless USB, ZigBee, wireless Ethernet, GSM or other type of communication interface can be used to communicate wirelessly with the control unit 120. The user interface 140 typically includes a plurality of buttons or a touch screen that the subject or bystander can use to communicate with the control unit 120, as well as a speaker and / or a display that the control unit 120 can use to communicate with the subject or bystander. For example, if the control unit 120 determines that the subject is experiencing an arrhythmia, the control unit 120 can emit an audible alarm via a speaker (not shown) on the control unit 120 and / or the user interface 140, thereby alerting the subject and any bystanders to the subject's medical condition. The control unit 120 can also instruct the subject to press and hold one or two buttons on the control unit 120 or the user interface 140 to indicate that the subject is conscious, thereby instructing the control unit 120 to stop the delivery of one or more therapeutic defibrillation shocks. If the subject does not respond, the device can infer that the subject is unconscious and continue the treatment sequence, ultimately delivering one or more defibrillation shocks to the subject's body. In some embodiments, the functionality of the user interface 140 may be integrated into the control unit 120 .

[0264] The control unit 120 includes at least one processor, microprocessor, or controller, such as processors available from companies such as Texas Instruments, Intel, AMD, Sun, IBM, Motorola, Freescale, and ARM Holdings. The at least one processor of the control unit 120 is configured to monitor the medical condition of the subject, perform medical data recording and storage, and provide medical treatment to the subject in response to detected medical events such as arrhythmias.

[0265] Although not shown, the wearable medical device 100 may further include additional sensors capable of monitoring the subject's physiological condition or activity in addition to the ECG sensing electrodes 112. For example, sensors capable of measuring blood pressure, heart rate, chest impedance, pulse oximetry, respiratory rate, heart sounds, lung sounds, expired CO2, SMO2, cerebral blood flow, EEG signals, brain oxygen level, tissue pH, an ultrasound image of the subject's heart, and the subject's activity level may also be provided.

[0266] In addition to or as an alternative to defibrillation, the wearable medical device 100 may also include functionality for administering therapy or intervention. Additional or alternative therapy or intervention may include pacing the subject's heart, utilizing a device commercially available from ZOLL Medical, Inc., located in Chelmsford, Massachusetts, or administering a therapy or intervention. Automatic management of chest compressions by devices similar to non-invasive cardiac support pumps, and / or automatic drug delivery. The portion of the wearable medical device 100 that provides defibrillation or other treatment is detachable from the control unit 120. By making the control unit and the treatment unit modular, the wearable medical device 100 can be worn without the defibrillation unit installed, thereby making the wearable medical device 100 smaller, lighter and more comfortable for the subject. A subject who is wearing the wearable medical device 100 without a defibrillation unit, for example while showering or swimming, can install the defibrillation unit if instructed by the wearable medical device 100. A wearable medical device 100 that does not include a defibrillation unit is sometimes referred to as a monitoring device.

[0267] Figure 3 is a block diagram of a medical support device 305, wherein the medical support device 305 may include sensors for monitoring parameters of a subject to estimate the risk of a medical event occurring. The medical support device 305 may include a life support device, such as at least one basic life support (BLS) device or an advanced life support (ALS) device such as a defibrillator or AED. The life support device 305 may be part of an external defibrillator or pacemaker such as the AED Pro or AED Plus defibrillators manufactured by ZOLL Medical of Chelmsford, MA, ProPack MD, X-Series Defibrillator, M-Series Defibrillator, R-Series Defibrillator BLS, or E-Series Defibrillator. The medical support device 305 need not be a wearable defibrillator or other therapeutic device, but may be an automated external defibrillator (AED).

[0268] The medical support device 305 can be designed to be operated by people with less or no medical training, prompting the user with a series of instructions for treating the subject and preventing the user from deviating from these instructions. For example, the medical support device 305 can determine the power level of the defibrillation shock to be applied to the subject and prevent the user from adjusting the power of the defibrillation shock. The user can follow the instructions (for example, by placing electrodes on the subject according to the instructions or by turning on the medical support device 305) to connect the medical support device 305 to the subject, and the medical support device 305 can automatically determine the treatment to be applied to the subject. In some embodiments, the medical support device 305 may include functions suitable for use by users with professional medical training (such as ALS providers, doctors or emergency medical technicians). For example, the medical support device 305 can respond to user input. The user can overrule the treatment suggested by the medical support device 305, or modify the power level, pulse width, pulse type (for example, monophasic or biphasic) or other parameters of the defibrillation shock or intervention to be applied to the subject.

[0269] The medical support device 305 may include a pacing unit or monitoring device for recording health-related information from the subject, such as information related to the subject's respiratory rate or cardiac activity. The medical support device 305 may include an electrocardiograph for displaying the subject's ECG. The medical support device 305 may be carried by a paramedic, an emergency healthcare provider, or a first responder such as a police officer or firefighter, or may be the type of life support device provided by a business or common carrier (e.g., an airline or bus) for use in an emergency situation. These or other users may connect the medical support device 305 to the subject to monitor the subject to diagnose conditions such as cardiac arrest, arrhythmia, etc., or to apply treatments such as defibrillation therapy, pacing therapy, etc. to the subject.

[0270] The medical support device 305 includes at least one control unit 310, wherein the at least one control unit 310 includes one or more processors for controlling the operation of the medical support device 305. For example, the control unit 310 can instruct the medical support device 305 to monitor the medical condition of the subject or apply treatment to the subject. The medical support device 305 can also include at least one memory unit 315. The memory unit 315 can be connected to the control unit 310 and can store monitoring information of the subject, such as ECG data or the subject's respiratory rate. The memory unit 315 can store current and / or historical information about the subject, information about the subject's age, weight, health, or gender, and / or information about past treatments applied to the subject by the medical support device 305. The stored information can include the date and time the treatment was applied to the subject and the nature of the treatment, such as pulse width, amplitude, energy level of the treatment, whether monophasic or biphasic pulses were delivered, etc.

[0271] The medical support device 305 may include at least one display monitor 320. The monitor 320 may be connected to the control unit 310 and the memory unit 315 and display information related to the subject, such as the subject's ECG, heart rate, and / or respiratory rate. The monitor 320 may display other information related to the subject (e.g., age, weight, etc.) and indicate the treatment being applied to the subject and the time the treatment is being applied. The monitor 320 may indicate that treatment is being applied to the subject during treatment being applied using the medical support device 305. For example, the monitor 320 may display a warning that an electric shock is about to be or is being applied to the subject using the medical support device 305. The warning may be a visual, audio, and / or tactile warning. The monitor 320 may include a user interface that a user of the medical support device 305 (e.g., the subject or first responders) may use to input information (e.g., information related to the subject's health, appearance, age, weight, and / or gender). The user input information may be stored in the memory unit 315.

[0272] The medical support device 305 includes at least one electrode 325. The at least one electrode 325 may include one or more sensing electrodes capable of monitoring the subject and detecting ECG signals, respiratory information, and / or other information related to the subject's condition. The at least one electrode 325 may include one or more treatment electrodes for applying treatment (such as administering pacing signals or defibrillation shocks) to the subject. The at least one electrode 325 may include sensing electrodes and treatment electrodes as dispersed electrodes, or combined into an integrated sensing / treatment electrode assembly. The at least one electrode 325 may be an external electrode that can be placed on the subject's body.

[0273] The medical support device 305 includes at least one power source 330. The at least one power source 330 may include a battery or other power source including an AC power source and an uninterruptible power supply. The at least one power source 330 provides power to the medical support device 305 including the control unit 310, the memory unit 315, the monitor 320, and the at least one electrode 325.

[0274] Although not shown, the medical support device 305 may include additional sensors capable of monitoring the subject's physiological condition or activity in addition to the electrodes 325. For example, sensors capable of measuring blood pressure, heart rate, chest impedance, pulse oximetry, respiratory rate, heart sounds, lung sounds, expired CO2, SMO2, cerebral blood flow, EEG signals, brain oxygen level, tissue pH, ultrasound images of the subject's heart, and the subject's activity level may be provided.

[0275] In addition to or as an alternative to defibrillation, the medical support device 305 may include functionality to administer treatments or interventions. These additional or alternative treatments or interventions may include, for example, pacing of the subject's heart, automatic management of chest compressions, and / or automatic drug delivery.

[0276] The medical support device 305 may include functionality to determine the type of cardiac event the subject is experiencing or the type of cardiac event that may be about to occur. The medical support device 305 may determine the appropriate treatment or intervention for that type of cardiac event and may automatically administer the appropriate treatment or intervention to the subject. If the device 305 is used in combination with other types of life support devices such as heart support pumps, the medical support device 305 can communicate the recommended intervention to the other life support devices.

[0277] If the cardiac event is one for which the wearable medical device 100 or the medical support device 305 determines that appropriate treatment or intervention cannot be administered, or if the cardiac event is one for which no effective treatment or intervention exists at the time of treatment, the wearable medical device 100 or the medical support device 305 may, for example, request assistance by issuing an audio or visual indication of a need for assistance and / or by placing a call to, for example, a physician or an emergency medical response system via a cellular network. If the wearable medical device 100 or the medical support device 305 determines that appropriate treatment or intervention cannot be administered itself, the wearable medical device 100 or the medical support device 305 may, for example, visually via a display on the user interface 140 or the monitor 320 and / or audibly via a speaker on the control unit 120 or the control unit 310, provide instructions for appropriate treatment or intervention to responding personnel.

[0278] Various parameters of the subject can predict various cardiac events and can provide various amounts of time before the predicted cardiac event, wherein within these amounts of time, a treatment provider such as a rescuer or physician can take action or provide treatment to reduce the probability of the cardiac event and / or reduce the harm caused to the subject by the cardiac event or other medical event. For example, if the QRS complex of the subject's ECG is observed to be continuously widening or the subject's heart rate is continuously decreasing, there may be a relatively long precursor time before the predicted cardiac event during which the rescuer can take action. In another example, changes in the T wave of the subject's ECG may occur within about 15 seconds to about 30 seconds before the onset of VF. T wave alternans (periodic beat-to-beat changes in the amplitude or shape of the T wave) can indicate an impending VT or arrhythmia. As described herein, alternative measures, estimates, and detection methods can also be used by the medical support device 305 to predict different cardiac events and respond to these cardiac events.

[0279] Various parameters of the subject that can predict an impending or ongoing cardiac event can invoke various actions or responses to be taken by the wearable medical device 100 or medical support device 305, the subject, and / or a bystander or medical professional. For example, if the subject's expiratory CO2 or cerebral blood flow fails to meet a threshold level, or if analysis of the subject's EEG performed using the control unit 120 or control unit 310 indicates that the subject's brain is not receiving enough oxygen, this can indicate that the subject's heart is pumping blood but is unable to deliver enough blood to the heart itself and / or other parts of the subject's body. If no treatment is given, a subject who exhibits a decrease in expiratory CO2 or cerebral blood flow may exhibit a decreased heart rate. An appropriate response to the subject's decrease in expiratory CO2 or cerebral blood flow may be to initiate CPR. If the subject exhibits changes in the ECG T wave, initiating chest compressions may be appropriate. If the subject exhibits a cardiac rhythm but no pulse (eg, pulseless electrical activity (PEA)), it may be appropriate to administer CPR and / or medications until the subject exhibits return of spontaneous circulation (ROSC).

[0280] If the action taken is not fast enough or if the action taken in response to the parameter representing the probability of an impending cardiac event is ineffective, the subject may experience a cardiac event. Some types of cardiac events (e.g., early VF and some types of VT) are appropriately treatable with an electric shock (defibrillation). Other types of cardiac events (e.g., late VF (VF that has occurred for more than about 4 or 5 minutes)) may be appropriately treatable with CPR (possibly followed by defibrillation), but cannot be appropriately treated with defibrillation from the outset. Other types of cardiac events (e.g., acute myocardial infarction) may be appropriately treated with CPR followed by the administration of thrombolytics or other types of medications. Some types of cardiac events (e.g., cardiac arrest) may simply be incurable. Thus, when a cardiac event occurs, the control unit 120 or the control unit 310 may desire to identify the type of cardiac event and determine what type of action or treatment will be most effective before initiating action in response to the occurrence of the cardiac event.

[0281] Emergency medical technicians can be trained to immediately transport a subject experiencing a cardiac event to a hospital; however, this may not always be the best course of action for the subject. For example, the subject may benefit from the administration of treatment that results in ROSC before being moved. Upon determining the type of cardiac event the subject is experiencing, defibrillation may be applied to the subject by the wearable medical device 100 or the medical support device 305 and / or the subject or a medical professional, as appropriate, or CPR and / or medication may be administered. If CPR and / or medication are administered, then (as described in U.S. Patent Application Publication No. 2013 / 0331719, the entire contents of which are incorporated herein by reference) the parameters of the subject's ECG (e.g., ASA, AMSA, and / or PSA) may be monitored using the wearable medical device 100 or the control unit 120 or the medical device 305 until the wearable medical device 100 or the medical device 305 determines that defibrillation is appropriate for the subject. If defibrillation is determined to be appropriate, defibrillation is applied using the medical device 305. If ROSC is observed in the subject and / or if improvement in SpO2 or expiratory CO2 is observed, the medical device 305 may indicate that the subject can be transported to a hospital for further treatment or monitoring. In another example, if the subject's heart rate is decreasing (i.e., bradying down), the subject may benefit from chest compressions before being transported, wherein the chest compressions may be indicated by the medical device 305 to the medical professional on site.

[0282] An estimated event risk score can be calculated by the control unit 120 or the control unit 310 for a subject at risk of experiencing a cardiac event and / or for a subject currently experiencing a cardiac event. The estimated event risk score can be calculated based on one or more measured physiological parameters of the subject, such as, for example, one or more parameters associated with the subject's ECG or EEG, blood pressure, heart rate or heart rate variability, expiratory CO2, SpO2, SMO2, cerebral blood flow, cerebral oxygen level, tissue pH, the subject's heart's response to the subject's tilt, and / or an ultrasound image of the subject's heart. If the estimated event risk score meets a threshold and / or indicates that the subject's condition is improving, the wearable medical device 100 or the medical support device 305 can recommend that the subject be transported to a hospital for further treatment and / or monitoring. If the estimated event risk score does not meet a threshold and / or indicates that the subject's condition is not improving or is worsening, the wearable medical device 100 or the medical support device 305 can recommend treatment or continue the subject's treatment before transport. If the subject exhibits a poor or increasingly poor estimated risk of an event score, the wearable medical device 100 or medical device 305 may recommend administering intervention, such as administering chest compressions manually or using an automated chest compression device, until the subject's estimated risk of an event improves to a level indicating that the subject is stable enough for transport. The calculation and analysis of the estimated risk of an event score are discussed in more detail below.

[0283] In addition to or as an alternative to the application of chest compressions, the wearable medical device 100 or the medical support device 305 may recommend the administration of epinephrine, norepinephrine, beta-blockers, nitrates, vasodilators, or other medications and / or the provision of oxygen to the subject until the estimated risk score for the event improves to a level that indicates to the wearable medical device 100 or the medical support device 305 that the subject is stable enough for transport. For example, to improve the estimated risk score for the event, the wearable medical device 100 or the medical support device 305 may recommend the provision of controlled vasodilation to the subject. Nitrates (e.g., nitroglycerin or nitrous oxide) may be administered to the subject while chest compressions are being performed. If the subject has a detectable pulse, chest compressions may be synchronized with the subject's cardiac cycle. The wearable medical device 100 or the medical support device 305 may be used to monitor tissue or blood oxygen levels to determine whether an appropriate amount of nitrates or other vasodilators has been administered to the subject. If the wearable medical device 100 or the medical support device 305 determines that the subject exhibits ROSC or the estimated risk of event improves to a level indicating that the subject is stable enough for transport, chest compressions and / or administration of vasodilators may be terminated.

[0284] Once the subject achieves ROSC, the subject can be transported to a hospital to medically stabilize the subject and treat the subject for the underlying condition that caused the cardiac arrest (e.g., heart disease). In some examples, 15% to 20% of subjects who achieve ROSC and are being transported to a hospital experience another cardiac arrest during transport to the hospital. If the risk score for a ROSC subject or event during transport meets a threshold, the wearable medical device 100 or the medical support device 305 can use an alarm or other communication to notify the ambulance or other transport driver or attendant who is riding with the subject of the increased risk, so that the attendant can intervene by providing direct treatment to the subject or driving to reach the hospital faster.

[0285] Measuring an ECG as a predictive indicator of a potential impending cardiac event or as a measure of the subject's myocardial state during a cardiac event and / or while therapy is being administered can be facilitated by using a motion sensor such as an accelerometer. An accelerometer built into the wearable medical device 100 or the medical support device 305 or a CPR assistance device (e.g., both available from ZOLL Medical, Inc., Chelmsford, Massachusetts) can be used. Real-time CPR feedback device or The accelerometer (e.g., a heart support pump) can provide a signal related to the subject's movement due to chest compressions or other movement of the subject, wherein the signal can be used to remove ECG signal artifacts caused by the subject's movement from the subject's monitored ECG signal. The accelerometer can be a 3-axis accelerometer, wherein the 3-axis accelerometer can be used to correlate posture (e.g., the degree of tilt of a supine subject) with changes in the subject's ECG. For example, if a supine subject is tilted by, for example, more than about 10 degrees, and a change occurs in the subject's ECG signal, this may be an indication that the subject should not be transported immediately. Treatment (e.g., chest compressions, medication, or other intervention) can be performed until the subject can be tilted without large or detectable changes in the subject's ECG signal, at which point it can be recommended that the subject be transported to a hospital for further treatment and / or monitoring.

[0286] In some embodiments, an accelerometer can be used to monitor the movement of the subject during transport, for example, due to the ambulance traveling on a rough road, and the subject's movement can be correlated with changes in the subject's ECG. If the changes in the subject's ECG are correlated with the subject's movement, the wearable medical device 100 or the medical support device 305 can provide a recommendation to slow the ambulance down and / or take a longer but smoother route.

[0287] Figure 4Is the wearable medical device 100 or medical support device 305 (e.g., ECG monitor / defibrillator, AED, or other device such as Flowchart of a method that can be implemented by a wearable medical device (such as a wearable cardioverter-defibrillator). In stage 410, the wearable medical device 100 or the medical support device 305 monitors one or more parameters of the subject. The one or more parameters can predict an impending cardiac event and / or can be used to calculate an event estimation risk score for a medical event for the subject. The one or more parameters may include one or more parameters associated with: the subject's ECG or electroencephalogram ("EEG"), the subject's heart rate or changes in heart rate, the subject's blood pressure, expiratory CO2, SpO2, SMO2, cerebral blood flow, cerebral oxygen level, tissue pH, the subject's heart response to the subject's tilt, and / or an ultrasound image of the subject's heart. In stage 420, the parameters are analyzed using the control unit 120 or the control unit 310 of the medical support device 305, for example to calculate an event estimation risk score or to determine whether the likelihood of an impending cardiac event is high and / or whether a cardiac event is occurring. If the control unit 120 or the control unit 310 determines in stage 430 that there is no ongoing cardiac event and the likelihood of an impending cardiac event is not high, for example, the estimated event risk score is below a threshold, the device returns to stage 410 to monitor the subject's parameters. In some embodiments, different parameters of the subject may be monitored at different times (e.g., sequentially).

[0288] If the control unit 120 or the control unit 310 determines in stage 430 that a cardiac event is occurring or that the likelihood of an impending cardiac event is high, then in stage 440 the control unit 120 or the control unit 310 determines whether the cardiac event is treatable, e.g., whether an intervention can be effective in treating the cardiac event or reducing the likelihood or severity of an impending cardiac event. If the control unit 120 or the control unit 310 determines in stage 440 that the type of cardiac event the subject is experiencing is not currently treatable by any intervention available through a medical device or by first responders, or if there is no intervention currently available through a medical device or by first responders that can reduce the likelihood or severity of an impending cardiac event, then in stage 450 the device may recommend transporting the subject to a facility, such as a hospital, for treatment and / or monitoring, and the device returns to stage 410 to monitor the subject's parameters. The subject's parameters may be recorded so that if the subject arrives at a hospital or has another source of treatment beyond the capabilities of the medical device, the subject's medical history may be reviewed to determine appropriate treatment.

[0289] If the control unit 120 or the control unit 310 determines in stage 440 that an ongoing cardiac event can be treated by an intervention available through a medical device or by first responders, or that an intervention available through a medical device or by first responders can reduce the probability or severity of an impending cardiac event, then the device determines an available action (e.g., an intervention) to provide to the subject in stage 460. In some implementations, the device considers multiple potential interventions but may provide the best type of available intervention to the subject.

[0290] In stage 470, the device performs or instructs the determined intervention, such as by administering automated chest compressions, defibrillation, and / or automated drug delivery. If the selected intervention is one that the device cannot perform on its own, the device may provide instructions to first responders, such as via a display or speaker, regarding what intervention to administer to the subject, or communicate the recommended treatment to another type of life support device that is capable of performing the recommended intervention.

[0291] Systems and methods for providing event estimated risk scores according to some embodiments are now discussed in greater detail.

[0292] The systems and methods according to some embodiments provide software analysis tools, wherein these software analysis tools automatically analyze the subject's physiological parameter data (e.g., ECG data) to detect and predict pending conditions by calculating event estimation risk scores for medical events that are occurring or may occur within a certain time period. Event estimation risk scores can be calculated for more imminent medical precursor events and longer-term medical precursor events. Event estimation risk scores can be calculated for medical events in multiple different time periods. In order to provide risk levels for various different time periods, multiple (e.g., different) event estimation risk scores can be calculated for different time periods (e.g., less than about 10 minutes, less than about 1 hour, less than about 3 hours, less than about 1 day, less than about 1 week, less than about 1 month, less than about 3 months, etc.) to provide the possibility or probability of medical events occurring in various time periods.

[0293] The various estimated event risk scores for various time periods can be compared to stored estimated event risk thresholds associated with those time periods to determine a plan of action or response that varies based on the estimated event risk scores for the various time periods. Different time periods can be associated with different thresholds, and different medical events can be associated with different thresholds. For example, a threshold for applying defibrillation in response to cardiac arrest in a more imminent time period may be different, and, for example, may be easier to meet, than a threshold for applying defibrillation in response to cardiac arrest in a longer time period.

[0294] Multiple different event estimated risk scores for multiple different medical events can be calculated for a time period. For example, an event estimated risk score for cardiac arrest and an event estimated risk score for non-sustained ventricular tachycardia can be calculated for the same time period.

[0295] In some implementations, an event estimated risk score can be calculated for a time period. For example, an event estimated risk score can be calculated for one of the following time periods: a time period of less than about 10 minutes, a time period of less than 1 hour, a time period of less than about 3 hours, a time period of less than about 1 day, a time period of less than about 1 week, a time period of less than about 1 month, a time period of less than about 3 months, etc., thereby providing a likelihood or probability of a medical event occurring within the time period. The event estimated risk score can be continuously updated over time (e.g., within a specific time period). For example, an event estimated risk score can be initially determined for a time period within 1 hour, and the event estimated risk score for the time period within 1 hour can be continuously updated over time.

[0296] The event estimation risk score includes a criticality measurement and a confidence measurement. The criticality measurement can provide an estimated measurement of the risk, severity, or importance of a potential medical precursor event. For example, medical events such as cardiac arrest can be associated with relatively high criticality measurements, while medical events such as ectopic beats or non-sustained ventricular tachycardia can be associated with relatively low criticality measurements. The criticality measurement can be a quantitative score or a qualitative assessment. The confidence measurement can provide the likelihood or probability that a specific medical event may occur within an associated time period. For example, the confidence measurement can be the percentage chance of an event occurring within an associated time period. The combination of the criticality measurement and the confidence measurement of the event estimation risk score can be used to determine an appropriate response to the medical precursor event (such as appropriate actions to be taken, etc.). For example, some treatments such as defibrillation may require a relatively high confidence value before implementation. Other treatments such as instructing the subject to sit down may require a relatively low confidence value before implementation. Defibrillation may require a relatively higher measure of criticality (eg, risk of cardiac arrest) than other treatments, such as instructing the subject to sit down, which may be adequate with a relatively lower measure of criticality (eg, elevated heart rate).

[0297] Figure 5 A system for responding to a medical event is shown. The early warning system 10 monitors the physical condition of a subject 104 and uses this information to estimate the risk of a medical event occurring. The early warning system 10 may include a wearable medical device 100 worn by the subject 104. Figure 5; however, the disclosed embodiments are not limited thereto, and the early warning system 10 may include another wearable device, a medical support device 305, an implantable medical device, a portable medical device, or other monitor of a subject's physiological parameters.

[0298] The wearable medical device 100 can measure data (such as ECG data) from the subject 104 and use this data to estimate the risk of a medical event occurring within various time periods. The control unit 120 can calculate an estimated event risk score by estimating the likelihood or probability that the subject 104 will experience a future medical event if no treatment is taken or if the treatment is unsuccessful, and / or determining that a chain of events that could potentially lead to a medical event has begun without an emergency response. The estimated event risk score may sometimes be referred to as an "event prediction" or "event prediction score" or used interchangeably with "event prediction" or "event prediction score," although the term "event risk estimation" is generally more comprehensive because, unlike the weather, which is largely unchangeable, one of the goals of using estimated event risk is to change the underlying process of the subject's condition. Estimated event risk enables the device or medical personnel to take more action than simply "carrying an umbrella" to respond to an event that occurs. Estimated event risk enables the device or medical personnel to prevent a medical event or mitigate its impact. For example, the wearable medical device 100 can provide feedback and treatment options to the subject 104 to help prevent the occurrence of a medical event and / or mitigate its impact. The wearable medical device 100 may additionally communicate information to an external entity, such as one or more medical professionals 108 , to enable the medical professionals to determine a plan of action or course of treatment for the subject 104 .

[0299] The early warning system 10 according to a non-limiting embodiment can utilize various degrees of sensitivity and specificity to predict medical events (e.g., adverse cardiac events) within seconds to years of the event. Various actions or responses can be suggested to the subject 104 based on the time for the predicted medical event. For example, if an adverse cardiac event is predicted within a time frame of 10 minutes from the current moment, the medical device 100 can warn the subject 104 to suppress strenuous activities (e.g., walking the dog) and not to remove the medical device 100, such as to enable the medical device to be used to provide defibrillation shocks when needed. In contrast, if the medical device 100 determines that the subject 104 is showing a high risk level of adverse cardiac events occurring in a time frame of 1 week from the current moment, the medical device 100 can warn the medical professional 108 or request the subject 104 to carefully follow the instructions of the medical professional 108. More details of exemplary warnings, suggested feedback, and recommended activities are described here for each time period.

[0300] As described above, the medical device 100 can predict the likelihood or probability of an adverse cardiac event occurring within a plurality of different time periods. The medical device 100 can calculate a plurality of different event estimated risk scores associated with the likelihood of an adverse cardiac event for the subject 104, wherein each different event estimated risk score is associated with a different time period, for example, each event estimated risk score represents the probability of an event occurring within each time period. The medical device 100 can calculate different event estimated risk scores for the short-term likelihood and long-term likelihood of an adverse cardiac event (such as the likelihood that an adverse cardiac event will occur within 1 minute, 10 minutes, 1 hour, 3 hours, 1 day, 1 week, 1 month, 3 months, etc.). Providing a plurality of event estimated risk scores associated with different time periods can allow different treatments and alarms to be triggered or recommended based on the likelihood of an adverse event occurring within a particular time period.

[0301] The control unit 120 of the wearable medical device 100 can monitor the ECG signals provided by the plurality of ECG electrodes 112 and (if provided) some or all of the signals provided by other sensors. The control unit 120 can calculate an event estimation risk score associated with the likelihood of an adverse cardiac event for the subject 104 within an associated time period based on the signals provided from the ECG electrodes 112 and (if provided) some or all of the signals provided by other sensors. The wearable medical device 100 can provide feedback to the subject 104 based on the calculated event estimation risk score. The feedback can be provided directly from the wearable medical device 100 and / or provided via a remote computing device 550 (such as a mobile phone, a computer or other computing device) in a wired or wireless communication with the wearable medical device 100.

[0302] like Figure 5 As shown, the wearable medical device 100 can be associated with a base unit 530. For example, using LifeVest TMA wearable cardioverter-defibrillator may provide a base unit 530 that is capable of performing a number of different functions. One of the functions that may be performed by the base unit 530 is to store and / or communicate information received from the wearable medical device 100 via a wired or wireless communication network 580. For example, information related to the medical condition of a subject over a certain period of time may be communicated by the base unit 530 to a medical professional 108, such as a doctor, so that the doctor can remotely monitor the medical condition of the subject. The information received by the base unit 530 may be communicated via the communication network 580 immediately after the information is received by the base unit 530, or alternatively, the information received by the base unit 530 may be stored in a memory of the base unit 530 and communicated later via the communication network 580. The information communicated by the base unit 530 may be maintained in the memory of the base unit.

[0303] In some embodiments, the base unit 530 or the remote computing device 550 can receive measurement data, such as ECG data of the subject, from the wearable medical device 100 and use this data to estimate the risk of a medical event occurring within various time periods. For example, because some algorithms used for estimating and detecting medical precursor events can be computationally intensive, some or all of the processing for determining an estimated event risk score and / or actions in response thereto, such as the operations performed by the machine learning classification system described below, can be performed on server hardware separate from the wearable medical device 100 itself, which is being worn by or in the vicinity of the subject 104.

[0304] The base unit 530 may include a visual display 531 for communicating visual messages to the subject 104 or a bystander; an audio output device 534, such as a speaker, for communicating audible messages to the subject or bystander; and a plurality of buttons 533, wherein the subject or bystander can utilize the plurality of buttons 533 to communicate with the base unit 530. The visual display 531 may include a touch screen display, so that the subject 104 or bystander can also communicate with the base unit 530 via the visual display 531. The visual display 531 may display information related to the predicted risk for the subject 104. The information may be displayed graphically to enable the subject 104 and / or a caregiver to assess both the current risk and the future risk of an adverse cardiac event. For example, briefly referring to Figure 6A, a bar graph can provide a visual representation of example estimated cardiac event risk scores for multiple time periods. The visual representation of the estimated event risk score for each time period can include information related to the reliability of the estimated event risk score (e.g., a confidence value associated with the corresponding estimated event risk score). For example, visual representations 602a-602h of the confidence values ​​can be superimposed on each bar of the bar graph. The visual representations 602a-602h of the confidence values ​​can represent the likelihood that an event may occur within the associated time period, such that visual representations spanning a relatively large distance (e.g., visual representations 602g and 602h) represent a relatively lower confidence in the estimated event risk score compared to visual representations spanning a relatively small distance (e.g., visual representations 602a and 602b). In some implementations, the range of the visual representations 602a-602h of the confidence values ​​can represent a possible range of error in the corresponding estimated event risk score. In some examples, the confidence value of the estimated risk score for an event corresponding to a relatively long time period (e.g., within a month, within three months) is lower than the confidence value of the estimated risk score for an event corresponding to a relatively short time period (e.g., within 10 minutes, within an hour). This may be due to the difficulty involved in making predictions in the relatively distant future, for example.

[0305] In some examples, historical information regarding the progression or change in the estimated risk score for an event may be provided. Figure 6B An example of this history information is shown in . Figure 6B A plurality of trend lines representing previously calculated estimated risk scores for an event over a period of time are included. The estimated risk scores for an event and other information related to the estimated risk scores for an event can be transmitted from the medical device 100 and / or the base unit 530 to a remote computing device 550, wherein the remote computing device 550 can include a smartphone, a personal digital assistant (PDA), a laptop, a tablet personal computer (PC), a desktop PC, a set-top box, an interactive television, and / or combinations thereof, or any other type of medical device for display and / or processing.

[0306] The base unit 530 may include at least one charging bay 532 for receiving a rechargeable auxiliary battery (e.g., a battery of the control unit 120). The base unit 530 may include one or more different communication interfaces, such as: a device communication interface 538 configured to receive information from the control unit 120 of the wearable medical device 100; a telephone network interface 536 for communicating information received from the wearable medical device 100 via a telephone network; and a network interface 537 for communicating information received from the wearable medical device 100 via a wired or wireless network connection. The base unit 530 may include an antenna 535 and (optionally) a wireless transceiver 539, wherein the antenna 535 or the wireless transceiver 539 is configured to wirelessly communicate information received from the wearable medical device 100 via a cellular (e.g., 2G, 3G, and 4G) or another wireless network. The base unit 530 may transmit the calculated estimated risk score for an event to the medical professional 108 (e.g., to enable the medical professional 108 to determine an action plan or course of treatment for the subject 104).

[0307] Now refer to Figure 7 , shows a flow chart illustrating an example method for determining a cardiac risk indicator such as an estimated event risk score. The estimated event risk score can be determined by the wearable medical device 100 ( Figure 1 and 5 ) is determined; however, example embodiments are not limited thereto, and the event estimated risk score may be determined by other devices such as the base unit 530 and the remote computing device 550 and / or a computing device of the medical professional 108 in communication with the medical device 100 and / or the base unit 530.

[0308] In stage 702, the wearable medical device 100 monitors the subject by receiving, measuring and / or recording one or more types of physiological data (e.g., signals including ECG data). The ECG data may be received from any appropriate source of subject ECG data. For example, the ECG data may be received in real time from ECG electrodes 112 mounted to the subject. The ECG data may be of any appropriate type. The ECG data may be recorded from a plurality of lead locations on the surface of the subject's body. The ECG data may be received together with additional subject data including subject statistics and / or other physiological data records. The subject statistics may be used in combination with the subject-specific ECG data for data processing and display.

[0309] In stage 704 , an event estimated risk score, such as an estimated quantitative risk value, is calculated (eg, using the control unit 120 ) for a plurality of time periods (eg, a plurality of time periods until the event).

[0310] At stage 706, the calculated estimated risk score for the event as a function of time is stored (e.g., using the control unit 120). The estimated risk score for the event (e.g., which may include both a criticality measure and a confidence measure) is stored for each time period and each medical event for which a score is calculated. For example, separate scores may be calculated and stored for time periods 10 minutes before, 1 hour before, 2 hours before, 4 hours before, 8 hours before, 12 hours before, 24 hours before, 48 hours before, and 72 hours before, and scores for multiple different medical events may be calculated and stored for each time period.

[0311] In stage 708, the calculated estimated risk of event scores for each time period are compared (eg, using control unit 120) to event thresholds, such as stored estimated risk of event thresholds, etc. Based on the comparison, a treatment plan or course of action is determined.

[0312] If estimated event risk scores are calculated for multiple different time periods, for example, if control unit 120 calculates an estimated event risk score for a first time period and an estimated event risk score for a second time period, control unit 120 may prioritize one of the estimated event risk scores over the remaining estimated event risk scores. For example, the estimated event risk scores may be prioritized based on the criticality of each score. Control unit 120 may compare estimated event risk scores with relatively high criticality measures with stored estimated event risk thresholds for the corresponding time period and, based on this comparison, determine a treatment plan or course of action before proceeding with a treatment plan or course of action for an estimated event risk score with relatively low criticality measures. In some embodiments, the prioritization of estimated event risk scores may be based on a weighting scheme, where the weighting scheme may be calculated based on one or more of a criticality measure, a confidence measure, and the risk associated with each estimated event risk score. In some embodiments, control unit 120 may prioritize estimated event risk scores for relatively more imminent time periods over estimated event risk scores for relatively more distant time periods. Similarly, if the control unit 120 calculates event estimated risk scores for multiple different medical precursor events within the same time period, the multiple event estimated risk scores for the same time period can be prioritized based on the criticality of each score or a weighting scheme, where the weighting scheme is calculated based on one or more of the criticality measurement, the confidence measurement, and the risk associated with each event estimated risk score.

[0313] In stage 710, based on the comparison result of the event estimated risk score and the event estimated risk threshold, it is determined (for example, using the control unit 120) whether to recommend an action. For example, if the event estimated risk score for a specific time period meets the stored threshold for the specific time period, the wearable medical device 100 can be used to perform an associated action and / or the wearable medical device 100 can be used to recommend an associated action to the subject. Example actions may include requesting the subject to keep wearing the wearable medical device 100, suggesting that the subject reduce physical activity, administering medication, or defibrillation. Figure 12 Various exemplary actions are shown in FIG and described in more detail here for different time periods.

[0314] If an action is recommended in stage 710, then in stage 712, an action associated with the determined event is taken (e.g., performed using the wearable medical device 100). The method then returns to stage 702 to receive physiological data. If no action is recommended in stage 710, then the method returns to stage 702 to receive physiological data.

[0315] In some embodiments, the wearable medical device 100 may display the calculated estimated risk score for an event on the interface 140, the remote computing device 550, the visual display 531 of the base unit 530, and / or the computing device of the medical professional 108. For example, Figure 6A and 6B As shown, the estimated event risk scores can be displayed on a display, for example, as a bar graph, a list, or a curve. For example, the subject or other user can select a time window for displaying previous estimated event risk scores to display a shorter or longer time range for the scores. In this way, a time-dependent estimated event risk calculation and evolution can be generated and displayed to the subject and / or medical professional.

[0316] Various processes, algorithms, scoring models, mathematical models, statistical analyses, and the like may be used to calculate an event estimated risk score. The method used to calculate the event estimated risk score may depend on the type of medical event to be predicted. For example, the control unit 120 may be configured to use a first process or algorithm for calculating an event estimated risk score for cardiac arrest, and a different second process or algorithm for calculating an event estimated risk score for ventricular fibrillation. In this way, various methods may be used to calculate the risk of an imminent rapid deterioration of the subject's medical condition to cardiac arrest or other serious cardiopulmonary conditions. Different methods and algorithms may be used to calculate event estimated risk scores for different time periods. For example, the control unit 120 may be configured to use a first process or algorithm for calculating an event estimated risk score for cardiac arrest within a first time period, and a different second process or algorithm for calculating an event estimated risk score for cardiac arrest within a different second time period.

[0317] In some implementations, a machine learning classifier as described in more detail below can be trained for a large population (e.g., a population of thousands to tens of thousands of patient records including electrophysiological phenomena, demographic and medical history information). Machine learning tools can include, but are not limited to, classification and regression tree decision models such as random forests and gradient boosting (e.g., implemented using R or any other statistical / mathematical programming language). Any other classification-based machine learning tools including neural networks (as described in more detail below) and support vector machines can be used. Because machine learning tools can be computationally intensive, some or all of the processing used by the machine learning tools can be performed on a server separate from the medical device.

[0318] An overview of how the Random Forest tool can be applied to a given data set can illustrate how the classification tool can work in interpreting a given parameter or metric. A random forest is a collection of decision trees. A decision tree is a flowchart-like structure in which each node represents a test for a metric and each branch represents the result of the test. The tree ends with a classification label, such as the decision ultimately obtained after calculating each metric. Each tree in the Random Forest tool receives "votes" when classifying a given set of metrics. There are two components of randomness involved in building a random forest. First, when each tree is created, a random subsample of the total data set is selected to grow the tree. Second, at each node of the tree, a "splitter variable" is selected and potential patients are divided into two categories. For example, one class of patients (e.g., response or occurrence of cardiac arrest) can be separated from another class of patients (e.g., non-response). The tree grows using additional splitter variables until all terminal nodes (leaves) of the tree are purely one class or the other. The "test" tree is recorded for the previously reserved patients. Each patient test record traverses the tree by proceeding down one branch or another depending on the metrics included in the record for each partition variable. A prediction result is assigned to the patient test record based on where the record lands in the tree (a vote). The entire process can be repeated with new random partitions of the base data set to generate additional trees and ultimately a "forest." In each case, a different subset of patients can be used to build the tree and test its performance.

[0319] In developing the results described in the following example implementations, a predetermined number of model variations are trained. For example, each model variation is labeled sequentially (e.g., labeled 1 to 100 for 100 runs). In each run of the model, the software randomly samples a predetermined portion of the population (e.g., 80% of the population) as a training set and retains the remainder (e.g., 20%) as a test set. The ratio of no-shock "non-responders" to shock "responders" observed in the baseline data is 2.5:1, however any other ratio of responders to non-responders or ratio of training and test sets can be selected.

[0320] In the example implementation shown below, example inputs to the model may include the following:

[0321] a) ECG measurements; and

[0322] b) Non-ECG metrics as described in more detail below.

[0323] Possible model results for training in the example implementation are as follows:

[0324] i) appropriate shock (1 = yes; 0 = no);

[0325] ii) cardiac arrest (1 = yes; 0 = no); and

[0326] iii) Ejection fraction (EF) is improved (eg, above 35%) (1 = yes; 0 = no).

[0327] In some embodiments, the risk of other endpoints (end of device use point) such as inappropriate shock, compatibility with device use, hospitalization, medication change, and death (non-sudden or total mortality) can be estimated. In some embodiments, highly correlated metrics can be removed or replaced with a representative set of metrics before running the model.

[0328] As described above, a machine learning tool can train a classifier on a first portion of a base dataset and test the classifier on a second portion of the dataset or on another separate dataset. When evaluating the performance of each classifier, the performance of the underlying decision trees within the random forest can be evaluated based on specificity and sensitivity parameters. For example, the sensitivity parameter can be based on a measure of the classifier's ability to correctly predict whether a patient is at risk of needing treatment. For example, the sensitivity parameter can be based on the proportion of patients who are appropriately treated for whom the model correctly predicted the risk of being treated. The specificity parameter can be based on the proportion of patients who were not treated at the end of device use and who were predicted by the relevant classifier to be at no risk of needing treatment. It can be advantageous to optimally balance each performance variable, such as sensitivity and specificity, at a high level. For example, by setting specificity to a relatively high value (e.g., 95%), the underlying threshold within the classifier model can be adjusted to minimize false positives. After defining specificity, the sensitivity measurement can be considered a performance measure, for example, typically in the range of 15-35% for a given model; however, smaller or larger values ​​of sensitivity are also possible.

[0329] For example, as described below, a validation protocol can be used to validate the predictive performance of a training model. In implementation, a validation phase can be used to determine appropriate threshold scores for classifying future patients (the results are currently unknown and the results are expected to be predicted), and to determine the predictive performance of each classifier model generated using a machine learning tool. In order to validate various classifiers and associated threshold scores, a second group of individuals, such as a validation population (or cohort), can be used. For example, the validation population used can be a new validation population. The results for the patients in the validation cohort are ultimately learned as these patients enter the end of device use. In an embodiment, the patients in the validation population can be different from the above-mentioned groups of training and test patients for training a classifier. For example, the validation population of a patient and its associated metric (validation metric) can be independent of the training population and associated metric (training metric) of the patient. In some implementations, there may be overlap between the validation metric and the training metric.

[0330] As shown in Table 1 below, validation data can be obtained. In some examples, a baseline ECG recording can be obtained from a potential patient at the start of wear. Baseline recordings are typically approximately 45 to 60 seconds long, although longer or shorter ECG recordings are also possible. Each subsequent week, a new 45 to 60 second ECG recording can be obtained from the patient, although the frequency of ECG recordings can be increased or decreased. For example, the frequency of ECG recordings can be increased to obtain a more precisely timed event risk estimate score for each patient.

[0331] Table 1

[0332]

[0333] For each patient in the validation population, an event estimate risk score can be generated based on each machine learning classifier model. For example, a "predict" command in a statistical computing language such as R (available from the R base) can be called for each classifier. In R, the "predict" command takes as input the patient's metrics and the corresponding classifier model of the machine learning algorithm for which the patient's metrics are to be performed. For a random forest classifier model, the output can be a probability score with a value ranging from 0 to 1. Therefore, since the base model is based on appropriate shock "response" and no shock (non-response), the example output score is in the range of 0 (risk of shock equivalent to most patients) to 1 (because the risk of shock is elevated relative to most patients, the risk is defined as immediate). In this way, the patient's metrics can be used for each pre-built model to identify the best performing machine learning classifier. As described above, when using the validation population, the patient's outcome is ultimately known. In this way, during post-analysis, it is possible to evaluate how well each classifier model performs. This process can be repeated for each patient in the validation population.

[0334] In some implementations, the validation population can be updated by at least one of the following operations: 1) adjusting one or more degrees in the validation metric; and 2) expanding the validation metric based on adding one or more additional subjects to the population of subjects that constitute the validation population. The thresholds used to classify future patients can be refined based on the updated validation metric. For example, one or more metrics in the validation metric can be adjusted using metrics of a patient who is currently being treated or monitored or who has not progressed through the end of device use, or the patient's metrics can be added to the validation population as metrics from a new subject. The validation metric that can be adjusted to a new metric for the patient is determined during the monitoring or treatment of the patient. In some examples, the patient's metrics can be added to the validation population as the monitored patient progresses through the end of device use, and / or the patient's metrics can be used to adjust the metrics in the validation metric after the patient progresses through the end of device use.

[0335] In some implementations, the training population can be updated by at least one of: 1) adjusting one or more metrics in the training metrics; and 2) expanding the training metrics based on adding one or more additional subjects to the first plurality of subjects. The machine learning classifier model can be retrained based on the updated training metrics. For example, as additional patient metrics are determined based on the current patient and / or metrics from a new patient are determined, the machine learning classifier can be retrained for the increased number of metrics or for new, different metrics to provide an updated classifier model. The training population can be updated as new metrics for the current patient and / or metrics for a new patient are determined, or after a patient progresses through the end of device use.

[0336] Reference is made to the example model referred to as "Model 80". Figure 8A , with the x-axis of the graph showing scores for patients who received an appropriate shock. The y-axis of the graph does not necessarily represent any values, and for ease of viewing, the plotted values ​​are separated along the y-axis. As shown, for this example model, a threshold value can be set so that an estimated risk of event score of 0.425 or above can be considered to classify the patient as having an increased risk of receiving a shock compared to other patients (e.g., patients with an estimated risk of event score below 0.425). For example, a patient with an estimated risk of event score that classifies the patient as having an increased risk can have a relatively higher likelihood of experiencing a medical event than a patient who is not classified as having an increased risk.

[0337] For this example model, another threshold can be set so that an estimated event risk score greater than 0.52 can be considered to classify the patient as being at risk of receiving an immediate shock, compared to other patients (those with an estimated event risk score less than 0.52). For example, a patient with an estimated event risk score that classifies the patient as being at risk of an immediate shock may have a relatively higher likelihood of experiencing a medical event compared to patients not classified as being at risk of an immediate shock, and thus may require immediate attention to the patient requiring action. In response, one or more actions can be taken and / or the patient can be instructed to perform one or more actions intended to reduce the patient's risk of experiencing a medical event. For example, as described in more detail below, one or more operating parameters of a medical device and / or one or more thresholds for determining whether a medical event has occurred can be adjusted to accommodate patients at risk of an immediate shock. In this way, patients classified as being at risk of an immediate shock can have their risk reduced, while patients not classified as being at risk of an immediate shock may not be required to take measures to reduce their risk (e.g., measures that may be inconvenient and therefore counterproductive for the patient). Although one or more actions can be taken for patients classified as being at risk of an immediate shock, these actions do not necessarily need to be taken immediately.

[0338] It should be understood that although the above example implementations describe two thresholds, there may be more (three or more) thresholds or only one threshold. Each threshold may correspond to a patient population for which different actions may be taken. For example, in some implementations, instead of having both elevated and immediate risk thresholds, there may be only one threshold, such as an immediate risk threshold. Patients with risk scores that place them above this immediate risk threshold may be more likely to take one or more immediate actions than patients who are not above this threshold.

[0339] For example, patients who are classified as having a high risk but for whom no immediate action may be taken (e.g., in the above example, patients with a risk score of 0.425 or greater but less than 0.52) may have their physiological parameters monitored more regularly compared to patients who are not classified as being at high risk. For example, a physician and / or technician may cause a medical device (e.g., a wearable defibrillator) to stream the patient's ECG data in substantially real time for analysis. Alternatively or in addition, the physician and / or technician may request additional monitoring of the patient for at least the period of time that the patient remains at high risk. Furthermore, for patients for whom immediate action is deemed necessary (e.g., patients with a risk score of 0.52 or greater), the physician may deem it necessary to instruct the patient to perform the following: more frequent device checks (e.g., ensuring battery charge is at an optimal level), minimize certain activities, minimize the time that the medical device is not actively monitoring the patient, etc.

[0340] In some examples, the scores from the various models can be combined in a predetermined manner to generate a single event estimated risk score for the patient. In some examples, clustering techniques can be implemented during validation to select the most predictive set of classifier models. For example, the most predictive valid set of classifiers can be identified based on sensitivity values ​​occurring within a predetermined range (e.g., exceeding 0.25 or any other predefined threshold). In some examples, a single model, such as the example model 80 shown above, can be selected and used to determine the event estimated risk score for a given patient. Any other implementation can be used to select an appropriate classifier model and / or combine the scores from the various models.

[0341] More than one summary measure can be considered (e.g., more than one estimated risk of an event or more than one combination of multiple estimated risk scores for an event from multiple models), allowing investigation of different aspects of a time series risk score. Different summary measures can be combined into a single result to characterize the overall status of a patient, such as whether the patient's risk profile is getting better or worse. For example, the time series risk score can be analyzed using the techniques described in "Analysis of serial measurements in medical research," by JNS Matthews, Douglas G Altman, MJ Campbell, and Patrick Royston, the contents of which are incorporated herein by reference in their entirety.

[0342] In some examples, the above-described process determines an event-estimated risk score to characterize the patient's risk. Based on the above-described validation process, appropriate thresholds are selected to achieve a predefined level of specificity. In cases where time series measurements are included, it can be observed that a patient's score may increase or decrease over a certain period of time. Consequently, observing changes in a patient's score over a certain period of time until the end of device use enables the determination of various information as described below.

[0343] Multiple techniques can be used to evaluate the time series data of each patient to assess whether the patient's risk is increasing or decreasing over time. For example, the time series values ​​can be plotted and the area under the curve can be calculated as a weighted average measurement. In some examples, the mean of the observations can be used. If the intervals between observations remain the same (e.g., hourly or weekly), the mean can be closely related to the weighted average. In implementation, the mean or weighted average can be compared to respective thresholds. In some examples, the amount by which the mean / average exceeds the corresponding threshold can be considered when judging how the subject's condition is changing.

[0344] In some examples, the number of times a risk score exceeds a threshold can be noted, and a trend in this exceeding can be extracted from this data. For a gradient boosting model, the range of the predicted score trend thus obtained is given by a logit transformation of the underlying probability score (e.g., typically falling within the range of -4.5 to 4.5, where a logit value of 0 corresponds to a probability value of 0.5).

[0345] For example, the following metrics have been shown to be highly predictive of final outcome (eg, appropriate shock vs. no shock). Hypothetical values ​​are shown below for the metrics to represent how a patient might be classified as high risk or immediate risk.

[0346] o Gender: Male (1) - more likely than Female (0)

[0347] o Age 56 – generally higher risk range

[0348] o Acute Myocardial Infarction (MI) diagnosis. (Optionally, added patient metrics and / or time series measurements can improve predictions for patients with congestive heart failure (MI in the presence of heart failure, etc.)

[0349] o Ejection fraction (EF) less than, for example, 35% (for example, the actual EF value is 25%)

[0350] o Other morbidities such as diabetes, renal failure, tobacco use, etc.

[0351] For a baseline ECG, for example 60 seconds, fed into a machine learning classifier model, the following morphological parameters can predict the patient's final outcome: QRS morphology and variation characteristics, QRS width, QRS height, single-lead QRS morphology, and dual-lead QRS morphology. For example, for a classifier previously trained as described above, at the end of the analysis, a model (e.g., a model similar to model 80 described above) may produce a score of 0.6 for the example hypothetical patient described above. Over time, the patient's score may change to, for example, 0.65, 0.5, 0.55, 0.45, and 0.6. As shown here, for example, for Figure 11 and 12 As mentioned above, the time series analysis described above can be used to recommend the nature of the action to be taken.

[0352] In various implementations, the parameters or metrics used in the calculation of the event estimated risk score can be grouped into three main categories, such as A) heart rate, heart rate variability, and heart rate irregularity; B) QRS morphology and variability; and C) ST segment and T wave morphology and variability. The main categories can be organized using subcategories of the metrics in the table or tables. The ECG can be obtained within a time period of 45 to 60 seconds; however, the example embodiments are not limited to this, and the length of the ECG can be longer or shorter. The following lists example metrics that can be determined based on the ECG; however, the example embodiments are not limited to these metrics, and various other metrics can be used to calculate and classify the event estimated risk score.

[0353] In an example based on a wearable monitoring and / or therapeutic device, each metric is evaluated independently on the lateral (SS) channel and the front-back (FB) channel. For example, the SS channel and the FB channel represent the position of the corresponding ECG sensor in relation to the patient's body during use of the wearable monitoring and / or therapeutic device. In various examples, an equal number of metrics for the SS channel and the FB channel can be used. For example, in some implementations, heart rate-related measurements can be similar between SS and FB. Conversely, in some implementations, morphological measurements can be different between SS and FB. In the example, the measurement from the SS channel is used to make the prediction, and in other schemes such as the scheme discussed below, both measurements (SS and FB) can be used.

[0354] Metrics of heart rate, heart rate variability, and heart rate irregularity include metrics derived from estimates of RR (R-wave-R-wave) intervals, as detected using a QRS detector. In some implementations, evaluation of these metrics includes distinguishing between normal beats (N) and ectopic beats (e.g., PVCs). For example, a PVC detector can be used to identify PVCs. The interval between two normal beats is referred to as an NN interval. The metrics table or subcategories of metrics included in this category may include:

[0355] 1) Heart rate measurement;

[0356] 2) Heart rate variability (time domain) measurement;

[0357] 3) non-sustained VT episode count measure; and

[0358] 4)PVC counting measurement.

[0359] QRS morphology and variability metrics include metrics that describe the shape and changes over time of the QRS complex as detected by a QRS detector. In some implementations, evaluation of these metrics includes distinguishing between normal beats (N) and ectopic beats (e.g., PVCs). For example, a PVC detector can be used to identify PVCs. The metrics table and subcategories included in this category may include:

[0360] 1) QRS width measurement;

[0361] 2) QRS height measurement;

[0362] 3) single-lead QRS morphology metrics (e.g., similarity score relative to a normal template); and

[0363] 4) Two-lead QRS morphology metrics (e.g., descriptors describing the vectorcardiogram).

[0364] Metrics of ST segment and T wave morphology and variability include metrics that describe the shape and changes of the ST segment and T wave over time as depicted by a T wave detector. In some implementations, evaluation of these metrics includes distinguishing between normal beats and ectopic beats. For example, a PVC detector may be used to identify PVCs. Thus, a metric may include an indication of the presence or absence of an ectopic beat in one or more of the underlying metrics. For example, a T wave metric may be evaluated on a normal beat, and because the T wave cannot be clearly identified or defined on an ectopic beat, the ectopic beat is excluded from the T wave analysis. The metrics table or subcategories included in this category may include:

[0365] 1) QT variability measurement;

[0366] 2) ST depression / elevation and slope measurement;

[0367] 3) T wave alternans measurement;

[0368] 4) T wave variability measures; and

[0369] 5) Two-lead T-wave morphology (eg, cosine RT) measurement.

[0370] In an example, single-lead QRS morphology metrics may include similarity scores for the SS and FB channels. For example, these similarity scores may include a metric representing the average of the scores over time for the SS and FB channels. In an implementation, similarity scores are obtained by comparing the QRS complex to a normal template (e.g., one for the SS and one for the FB). For example, such similarity score metrics may or may not include ectopic beats.

[0371] For example, Figure 8B As shown, the master normal template is compared with the QRS complex of the ECG signal (arrows in the plot). The obtained similarity scores are averaged to obtain a unique morphology score. In this example, ectopic beats can be excluded from the calculation. The score can describe the variability of the shape compared to the normal beat. For example, if one of the compared QRS morphologies deviates from the normal template, the similarity score can take a smaller value. A lower average score indicates a higher variability of the QRS morphology in the normal ECG rhythm.

[0372] In another example, a heart rate variability (HRV) metric may include an SDNN, where the SDNN represents the standard deviation of the NN intervals over time. In some implementations, the SDNN calculation may not include ectopic beats. Figure 8C is a graph of an example ECG signal used to determine a heart rate variability metric. Figure 8CAs shown, five normal intervals 801a to 801e are used to calculate the SDNN. The intervals before and after the PVC can be excluded from the SDNN evaluation. The SDNN is calculated as the standard deviation of (NN1, NN2, NN3, NN4, NN5).

[0373] In an example, the QRS width metric can be based on any of a standard deviation, a mean, or other average value of the respective estimated widths of the QRS complexes over time. In some implementations, the QRS width metric can exclude ectopic beats such as PVCs. In some implementations, the QRS width metric can include an indication of the presence or absence of ectopic beats. Figure 8D is a graph showing an example ECG signal illustrating an example of QRS width.

[0374] In an example, the QRS height metric can be based on any of a standard deviation, a mean, or other average value of the estimated height of the QRS complex over time. In some implementations, the QRS height metric can exclude ectopic beats such as PVCs. In some implementations, the QRS height metric can include an indication of the presence or absence of ectopic beats. Figure 8E is a graph showing an example ECG signal illustrating an example of QRS height.

[0375] The external medical device can monitor and collect physiological information of the subject, such as cardiac electrophysiological information, and classify the patient into risk categories based on the above-mentioned machine learning classifier model, together with known risk factors associated with sudden cardiac death (such as, but not limited to, left ventricular ejection fraction). The device may include an external defibrillator for applying treatment such as defibrillation to the subject when necessary (e.g., after detecting sustained ventricular arrhythmia or ventricular fibrillation).

[0376] The technical principles of the risk classification method involve developing an algorithm using a machine learning application as described above. Machine learning algorithms include, but are not limited to, random forests, gradient boosting, neural networks, and classification and regression trees. The classification method uses any combination of inputs from demographic, medical, and electrophysiological information obtained from subjects whose outcomes of sudden cardiac death are known to train and test the classification algorithm. Validation of the risk prediction algorithm can be achieved through prospective studies of independent groups of subjects; however, retrospective studies of registries of subjects can also be used. Cardiac electrophysiological information in the form of an electrocardiogram recording can be obtained once for each subject to develop a baseline risk score and / or time interval, thereby developing a time series that can monitor changes in the risk score. Electrophysiological information can be collected at specified predefined event intervals or in response to external events such as physician requests, and the electrophysiological information can also be used to determine risk scores and classifications.

[0377] The electrophysiological data may include electrocardiogram (ECG) records of variable length. For example, a recording of 45 to 60 seconds in length may be obtained for each subject and / or at time intervals. Recordings of shorter or longer lengths may also be used, and for example, these records may be within the time range of 1 minute to 2 weeks of continuous cardiac electrophysiological monitoring. Using the ECG data may include applying software algorithms to process the cardiac signals into cardiac-related metrics such as, but not limited to, heart rate, heart rate variability, ST segment elevation, premature ventricular contractions, rhythm morphology, early repolarization, and the like.

[0378] Non-electrophysiological data may include subject demographic information and medical history information. Subject demographic information may include, but is not limited to, gender, age, body mass index, family history of heart disease, cardiac diagnosis, comorbidities, and left ventricular ejection fraction.

[0379] An estimated risk score for an event for an individual subject can be obtained by applying a machine learning algorithm to electrophysiological data or to both electrophysiological and non-electrophysiological data. The interpretation of the risk score relies on threshold settings, enabling the risk score to be classified using a clinically meaningful risk stratification. A threshold is determined that is specific to each model algorithm and can be continuously refined using the available registration of subject data.

[0380] In some examples, a threshold value for a classifier model can be determined by applying a previously developed machine learning algorithm to historical data available in a registry of subject data whose results are known for sudden cardiac death (including cardiac arrest and cardiac asystole). Subjects can be selected according to any number of criteria (such as continuous input to the registry, cardiac diagnosis, gender, etc.). The determination of the threshold value can be based on predetermined performance criteria, such as 95% specificity or other diagnostic test performance criteria. For example, in the case of 95% specificity, the risk scores of a group of subjects with negative results from the test criteria can be sorted by value, and a cutoff value for the 95th percentile can be determined using standard methods. In addition, threshold values ​​for other performance criteria such as sensitivity, positive predictive value, the number required for diagnosis, etc. can be predetermined in a similar manner.

[0381] Thresholds can be selected to achieve any clinically meaningful level of specificity, sensitivity, positive predictive value, number required for diagnosis, or other diagnostic performance characteristics. In the example, the threshold for the interpretation of the event estimation risk score is predefined to achieve performance at 95% specificity. This decision determines the level of other performance characteristics. For example, in a sample of 10,000 subjects and a prevalence of cardiac arrest of 1.4%, the corresponding threshold classifies 493 subjects as false positives and 9,367 subjects as true negatives. If the sensitivity performance of the model is 20%, 28 subjects are classified as true positives and 112 subjects are classified as false negatives. In a similar manner, additional performance characteristics such as positive predictive value, negative predictive value, number required for diagnosis, etc. can be pre-specified to achieve individual performance measurements of predetermined clinical significance.

[0382] In some examples, an event estimated risk score can be evaluated by including electrophysiological information collected at different times. Time series data can provide updated risk scores at intervals ranging from seconds or minutes to days. The use of multiple risk scores can be optimized to follow any number of methods to determine the subject's overall risk for sudden cardiac death. For example, any risk score above a threshold that occurs can be used to classify the subject as presenting a high risk. In addition, other variables can be used to evaluate risk, such as, but not limited to, changes in risk values ​​(increasing or decreasing), the length of time the subject remains above a risk threshold, or methods based on the area under the curve.

[0383] Identification of estimated risk scores for events associated with true positive subjects can yield actionable results that impact the subject's health. The performance of the risk score can be effective for timeframes spanning seconds or minutes to days, resulting in different recommended actions. In the event that a subject's risk score is greater than a threshold, possible actions that can be taken include, but are not limited to, a shorter time interval between detection of a tachyarrhythmia and defibrillation, or disabling of a response button on a medical device, thereby ensuring timely delivery of treatment.

[0384] In some examples, a subject's estimated risk score for an event greater than a threshold can result in direct notification to the subject, the subject's medical team, or a responsible third party such as a close family member. For example, notification of the subject's high-risk condition can take the form of a text message, email, or telephone contact. Additional subject notification schemes can include using the wearable medical device's audio and vibration alert systems to prompt the patient to increase his or her compliance with wearing the device. High-risk notifications can also be sent to the subject's medical team, including a primary care physician and a cardiologist, with a recommendation that both contact the patient to evaluate the subject's compliance with optimized medication therapy.

[0385] Figure 8F 806 is a flow chart of a process for determining an event estimated risk score and a corresponding classification according to some embodiments. In stage 802, input data from ECG data sources and non-ECG data sources are collected and cleaned to provide a set of weighted metrics for each subject. The metrics may include, but are not limited to, measurements of heart rate, counts of premature ventricular contractions, gender, and age. In stage 804, a model such as a predictive machine learning algorithm or classifier is applied, and an event estimated risk score is generated. In stage 806, the event estimated risk score of the subject is interpreted by a level of risk using thresholds (e.g., predetermined thresholds) based on performance criteria such as sensitivity and specificity as described herein to classify the risk score.

[0386] Figure 8G8000 is a block diagram of a risk prediction system for determining, classifying, and informing an estimated risk score for an event, according to some embodiments. In an example, electrocardiogram (ECG) data 8005 of a patient is received by the risk prediction system 8000 (e.g., using an external medical device). The input of data across system boundaries (e.g., from a device monitoring a subject's physiological parameters) enables the development of ECG-based metrics 8010, such as, but not limited to, heart rate, heart rate variability, ST-segment elevation, premature ventricular contractions, rhythm morphology, and the like. Where available, the ECG metrics can be augmented with demographic data, medical history, and previously acquired ECG metrics 8020. The complete dataset of metrics is passed to a classifier 8030, which is used to calculate a risk score for the patient and return a risk classification 8040. If the patient's risk classification is elevated above a typical patient's risk classification, e.g., above a threshold set for the classifier model (e.g., at a threshold set for high risk or elevated and imminent risk, as described above), the system can prompt for specific actions. For example, an action 8050 can be indicated to an external medical device (e.g., a wearable defibrillator), such as, but not limited to, triggering a change in a controller that reduces the time from detection of cardiac arrest to treatment. In addition, detection of an increased risk category can prompt the system to notify 8060 the patient, physician, responsible third party, medical team, and / or technical support personnel of the change in risk. In this way, patients with increased risk categories can be marked for more frequent future observation. For example, in some implementations, based on the risk classification, a caregiver and / or technical support personnel can cause the external medical device to perform specific additional actions.

[0387] In some implementations, if the patient is classified as being at high risk (e.g., above 0.425), including a higher immediate risk classification, the caregiver and / or technical support personnel can cause the external medical device to begin streaming the patient's ECG data 8005 (and / or other physiological data) to a remote server (e.g., arrow 8065) ​​in real time or substantially real time for additional monitoring and analysis. In some examples, the time period for reporting and / or metric measurement and analysis for such patients can be increased (e.g., from once every several hours or more to once every hour). In some examples, additional monitoring and / or reporting of the patient's condition and various physiological parameters can be initiated during the time period that the patient remains in the high risk classification.

[0388] In some implementations, if the patient is classified as being at high risk but no immediate action is deemed necessary (e.g., above 0.425 and less than 0.52 in the above example), the physician or technical support personnel can initiate more frequent reporting of the patient's physiological condition (e.g., a medical report of the patient's condition) or begin streaming ECG data. Furthermore, if the patient is placed in an immediate risk classification (e.g., above 0.52), the physician or technical support personnel can take additional actions, including, for example, instructing more frequent checks of external medical devices and / or ensuring that downtime for continuous monitoring is kept to a minimum.

[0389] In the event of a prediction of cardiac arrest, actions that may be taken include, but are not limited to, notifying the patient and responsible third parties to seek medical attention and notifying the responsible medical team of the patient's risk status. Figure 11 and 12 and here with Figure 11 and 12 Other actions that may be taken are discussed in detail.

[0390] In some embodiments, the control unit 120 can calculate an event estimation risk score by monitoring one or more physiological measurements or parameters of the subject to detect sudden changes in one or more states of the subject. These changes can indicate instability of the subject's underlying state and / or statistical changes in a series of data related to the subject. The state change can occur in days, hours, minutes, or seconds before a medical event including cardiac arrest, respiratory arrest, cardiac arrest, non-sudden death, or other medical events. Figure 8H In stage 750, a plurality of physiological measurements or parameters may be extracted from the subject's ECG signal (e.g., by control unit 120). Control unit 120 may store the extracted physiological measurements and use them to analyze the subject's ECG in real time or through post-analysis. As described in more detail below, control unit 120 may combine all or a portion of the extracted measurements or parameters into a multivariate parameter signal and process or analyze the multivariate parameter signal.

[0391] The one or more physiological measurements extracted from the ECG signal may include heart rate, heart rate variability, PVC burden or count, activity, noise quantization, atrial fibrillation, transient pauses, heart rate oscillations, QRS height, QRS width, changes in morphology size or shape, cosine RT, artificial pacing, corrected QT interval, QT variability, T wave width, T wave alternans, T wave variability, ST segment changes, early repolarization, late potentials, fragmented QRS / HF content, and fragmented T wave / HF content.

[0392] The control unit 120 can detect reference points in the ECG signal (e.g., points corresponding to the P wave, Q wave, R wave, S wave, and T wave) to extract individual measurement values ​​(e.g., QRS, PVC, etc.) from the physiological parameter data. For example, the QT interval can provide a measure of heart failure in the subject, and the distance between the Q point and the T point can be determined and extracted based on the physiological parameter signal.

[0393] Figure 8I 1 shows a dual-lead ECG signal in which the control unit 120 identifies PVC parameters in the ECG signal for extraction. For example, the control unit 120 can apply a QRS detector and a PVC detector to the ECG signal to extract the QRS measurement value and the PVC measurement value. The extracted QRS measurement value and the PVC measurement value and / or the reference points determined using the QRS detector and the PVC detector can be used by the control unit 120 to identify and extract various other measurement values ​​or parameters (e.g., HRV, RR interval, etc.) in the ECG signal. For example, the output from the QRS detector and the PVC detector can be used by an AFIB detector, a pause detector, a pace detector, a morphology detector, a T wave detector, and / or any other detector in the control unit 120 for extracting measurement values ​​or parameters from the ECG signal. QRS detectors, PVC detectors, and other detectors are well known in the art, as described, for example, in (Kohler, Hennig et al.) The Principles of Software QRS Detection, IEEE ENGINEERING IN MEDICINE AND BIOLOGY, (January / February 2002), which is incorporated herein by reference in its entirety.

[0394] An accelerometer data stream can be received from an accelerometer located at the wearable medical device 100 or the medical support device 305 or other parts of the subject's body to detect whether the subject is vertical or horizontal, or other activities and / or movements of the subject. For example, data related to the subject's movement due to chest compressions or other movements of the subject can be used to remove ECG signal artifacts caused by the subject's movement from the subject's ECG signal. The accelerometer can be a 3-axis accelerometer, which can be used to correlate posture (e.g., the degree of tilt of a supine subject) with changes in the subject's ECG.

[0395] In some implementations, one or more actions may occur for a patient identified as belonging to an increased risk category (e.g., a patient classified as being at "immediate risk" or "high risk"). For example, a medical device may adjust one or more operating parameters in response to the patient being identified as belonging to an increased risk category. In some implementations, a medical device may adjust one or more thresholds used to determine whether a medical event has occurred or is likely to occur. For example, if a patient is susceptible to a particular cardiac condition, the medical device may relax one or more threshold requirements for physiological parameters used to diagnose the particular cardiac condition (e.g., to minimize false negatives).

[0396] In some implementations, in response to a patient being identified as belonging to an increased risk category, one or more non-essential functions of the medical device can be disabled, or the frequency of execution of non-essential functions can be minimized. For example, the medical device can enter a power-saving mode in which only essential medical operations continue to run. In some implementations, the medical device can avoid checking for and installing automatic updates (e.g., for software and / or firmware). In some implementations, the medical device can reduce the timeout of display devices and other visual devices that may consume relatively large amounts of battery power. In some implementations, the medical device can avoid or reduce the frequency of performing self-checks.

[0397] Reference again Figure 8HIn stage 752, the control unit 120 may apply a change point (CP) analysis to the extracted measurements or parameters, which may be represented as a multivariate parameter signal, to determine one or more change points. Mathematical techniques may be used to analyze the data stream of the multivariate parameter signal to determine whether a change has occurred, the nature of the change, the level of significance of the change, and / or the confidence that a change has occurred. Each of these factors may be used to determine what action is appropriate in response to the detected change (if any). In general, a method for detecting a change in a parameter or characteristic of a subject may include determining initial statistical characteristics of a signal and analyzing the signal for any significant persistent changes. For example, known techniques for change point analysis are described by (M. Basseville & I. Nikiforov, (1993). Detection of Abrupt Changes: Theory and Application. Prentice Hall, Englewood Cliffs, NJ.) and (Pettitt AN, (1980). A simple cumulative sum type statistic for the change point problem with zero-one observations. Biometrika, 67: 79-84), the entire contents of which are incorporated herein by reference. Other known methods for change point analysis, such as Shewhart control charts, can be used to detect changes in characteristics and assess whether the detected changes are of sufficient magnitude.

[0398] The control unit 120 may determine the baseline of the multivariate parameter signal as intervals from the start of the signal until the first or any subsequent change point, or as intervals from the start of the signal until a desired time point. The control unit 120 may apply distance analysis to the extracted parameters of the multivariate parameter signal at various time intervals for the baseline or any other time period in stage 754. The time intervals may be determined using the detected change points or other arbitrary time points. For example, these time intervals may be located near corresponding change points in the multivariate parameter signal. Figure 8JAs shown, the extracted example parameter signals (e.g., QRS height, maxNN, sdNN, and detrended heart rate) constituting the multivariate parameter signal can be divided into time intervals. For example, the time interval R2 (e.g., event time (eventTime)) can be associated with the change point of the multivariate parameter signal. The control unit 120 can calculate the variance, mean, median, standard deviation or other statistical values ​​of the multivariate parameter signal for each time interval, and determine whether the distance of the statistical value for each time interval from the baseline meets one or more thresholds. For example, a given time interval R associated with a change point of the multivariate parameter signal can be i The calculated mean is compared with a baseline R determined based on the time interval from the start of the multivariate parameter signal until the first change point detected in the multivariate parameter signal. B Compare to determine the distance between the two scores, i.e. D i =R i -R B , where: R B is the baseline score, R i is the score for interval i, and D i is the distance between two scores for interval i.

[0399] The determined distance can be used in combination with an anomaly score output by an anomaly detector (e.g., a neural network, a density-based technique, a support vector machine, a cluster analysis, an ensemble technique, or any other known process for anomaly detection) and can be applied to the multivariate parameter signal by a classifier in the control unit 120 to determine an estimated event risk score for the subject. The control unit 120 can determine the anomaly score continuously or at set intervals (e.g., at set time intervals or at time intervals corresponding to each change point). Figure 8K shows a multivariate summation of example abnormality scores for treated subjects. Figure 8HIn parallel with the change point analysis in stage 752 and the distance analysis in stage 754, in stage 756, the control unit 120 may apply anomaly detection to the multivariate parameter signal. For example, the control unit 120 may train a neural network on a baseline time period of the multivariate parameter signal and apply the trained neural network to the multivariate parameter signal to determine an anomaly score for the subject. Thus, a metric or parameter for determining an event estimated risk score is generated by both the change point analysis and the neural network. The control unit 120 may perform three separate analyses of the multivariate parameter signal, wherein the three separate analyses include a baseline parameter difference measure, a combined anomaly detection magnitude value, and the original single parameter output from the anomaly detection algorithm. Each of these three analyses, which may have multiple parameters as outputs, may be fed into a classifier or other decision logic that may take the form of logistic regression, fuzzy logic, traditional "if then else" or "threshold" logic to determine the event estimated risk score. Figure 8L State change results including percent sensitivity and percent specificity for predicting medical precursor events from an example test set of 200 patients are shown.

[0400] In stage 758, a classifier in the control unit 120 receives the distances and anomaly scores corresponding to each time interval and determines an estimated risk score for the event based on these distances and anomaly scores. The classifier in the control unit 120 uses one or more machine learning processes to classify the distances and corresponding anomaly scores for the time intervals.

[0401] Machine learning can be used to train the decision threshold of a classifier. For example, the classifier can be a random forest classifier using multiple decision trees to improve the classification rate. Decision tree learning is a method commonly used in data mining to create a model that predicts the value of a target variable based on multiple input variables. Assuming that the values ​​of the input variables are represented by a path from the root to the leaves, each internal node in the decision tree corresponds to one of the input variables, and each leaf represents the value of the target variable. Decision trees thus provide a simple representation for classifying examples and are one of the most successful techniques used for supervised classification learning.

[0402] Various models can be used to define Figure 8H The process described for calculating an estimated risk score for an event. Figure 8M A multivariate feature selection process is shown for determining which of multiple metrics extracted from a multivariate parameter signal to use in a model to determine an estimated risk score for an event. The multivariate feature selection process can evaluate the model for all possible combinations of the extracted parameters to determine the best parameters to use for the model.

[0403] Definition Figure 8HThe processed real-time predictive model can optimize the area under the multivariate parameter curve for the optimized parameters to determine the receiver operating curve.

[0404] Including additional clinical history variables can also improve Figure 8H The real-time model presented. The medical history variables may include one or more of age, BMI, gender, etc. In some embodiments, patients may be clustered based on the medical history variables. Different real-time models may be developed for each cluster generated based on the medical history analysis.

[0405] Definition Figure 8H The processed baseline model may evaluate only the first time interval for the baseline period. For example, the baseline model may evaluate only the first 12 hours of physiological parameter data of the subject, such as comparing the first 6 hours of baseline data with the first time interval of the next 6 hours to predict risk in a future time period (e.g., the next week or the next two weeks).

[0406] An estimated risk score for an event may be calculated using a scoring model or a mathematical model, such as a model based on logistic regression, etc. Exemplary logistic regression models that can be used to calculate an estimated risk score for an event include univariate analysis and multivariate nonlinear regression.

[0407] The control unit 120 can apply a logistic regression mathematical model to data from samples of cardiovascular subjects and non-cardiovascular subjects. The logistic regression mathematical model can be fitted separately using a combination of demographic parameters (e.g., age), vital signs, and other ECG parameters of the cardiovascular subjects and non-cardiovascular subjects. The estimated risk can be analyzed by receiver operating characteristic (ROC) analysis, sensitivity analysis, specificity analysis, positive predictive value (PPV) analysis, and / or negative predictive value (NPV) analysis. Based on the logistic model, an event estimated risk score that can include a score of 0 to 100% is generated periodically (e.g., approximately every 10 seconds). Optionally, the interval for calculating the event estimated risk score can be a function of the perceived risk. For example, the frequency of calculating the event estimated risk score can increase as the estimated risk score of the previous interval increases (e.g., when it is determined that the event estimated risk score of the previous interval meets a threshold level for increasing the frequency of calculation).

[0408] The log-logistic distribution, also known as the Fisk distribution, is a general function that provides a parametric model for survival analysis. Unlike the more commonly used Weibull distribution, the log-logistic distribution can have a non-monotonic hazard function: when the shape parameter β>1, the hazard function is unimodal (when the shape parameter β≤1, the hazard function is monotonically decreasing). The cumulative distribution function can be written in closed form, which is particularly useful for analyzing survival data with censoring. The log-logistic distribution can be used as the basis for accelerated failure time models by varying the scale parameter α, which also serves as the median of the distribution, between groups, or more generally by modeling log(α) as a linear function of the covariate to introduce covariates that affect the scale parameter α but not the shape parameter β.

[0409] The survival function is expressed by the following equation (4):

[0410] S(t)=1-F(t)=[1+(t / α) β ] -1 (4)

[0411] Where: S(t) is the survivability of the subject as a function of time, e.g., S(t) can be the property of an arbitrary random variable that maps a set of events typically associated with the death or failure of some system (e.g., medical precursor events) onto time, and the survivability function S(t) captures the probability that the system (e.g., the subject) will survive (e.g., not experience a medical precursor event) beyond a specified time t; F(t) is the failure distribution as a function of time t, which is the cumulative distribution function used to describe the probability of failure (at least) up to time t (including time t); t is time; α is a scale parameter; and β is a shape parameter.

[0412] Therefore, the risk function is expressed by the following equation (5):

[0413]

[0414] Where: h(t) is the hazard rate or hazard function as a function of time, which becomes the instantaneous failure rate as Δt approaches zero; f(t) is the density function as a function of time t, which is the density of a continuous random variable, that is, a function used to describe the relative likelihood of a random variable taking a given value; S(t) is the survival function of equation (4); t is time; α is the scale parameter; and β is the shape parameter. Confidence intervals and confidence bands can be calculated for the survival function S(t) and / or the hazard function h(t).

[0415] Individual event estimated risk scores are calculated for different time periods. The time intervals can be more finely spaced, such as by minutes, seconds, or milliseconds, to provide, for example, Figure 6BThe wearable medical device 100 can display the estimated risk scores for an event as a list, can select one or more scores from the list based on user input, and / or can display the risk as a curve on the display. The wearable medical device 100 can thus perform a time-dependent calculation and evolution of the estimated risk of an event and display the results to the subject or a medical professional.

[0416] As described in U.S. Provisional Patent Application No. 61 / 945,424, the contents of which are incorporated herein by reference, the control unit 120 may perform VCG vector loop bifurcation. Figure 9 , shows an example method 770 for determining a cardiac risk indication based on identification of T wave bifurcations. In one embodiment, the method 770 is implemented by the wearable medical device 100 or the medical support device 305. However, other embodiments are also possible.

[0417] In stage 772, the patient is monitored by recording physiological data including one or more types of ECG signals. The ECG signal can be received from any appropriate source of the patient's ECG data. For example, ECG data can be received in real time from two or more or three or more ECG electrodes installed to the patient, or previously recorded data can be received from a storage device. ECG data can have any appropriate type. ECG data can be recorded from multiple lead positions on the surface of the patient's body. In some implementations, a standard 12-lead ECG record (e.g., Leads I, II, III, aVR, aVF, aVL, V1, V2, V3, V4, V5, and V6) can be obtained based on the signal obtained using 10 ECG electrodes. Any appropriate number of ECG electrodes installed to appropriate body parts can be used. The example of other ECG lead systems includes a "Frank" electrode lead system (e.g., six electrodes), a McFee-Parungao lead system, a SVEC III lead system, a Fischmann Barber-Weiss lead system, and a Nelson lead system. Other examples include right precordial leads, posterior leads, and leads placed in the upper or lower intercostal spaces.

[0418] In some implementations, information related to the source of the ECG data can be provided to the wearable medical device 100 or the medical support device 305. For example, the wearable medical device 100 or the medical support device 305 can change the configuration of the display and / or analysis tools based on the source of the ECG data (such as the position of the ECG leads relative to the heart, body, or other leads). In some implementations, the patient monitoring device can perform real-time ECG signal preprocessing. Real-time ECG signal preprocessing can include removing DC components using a high-pass filter, amplifying the ECG signal, limiting the signal bandwidth using a low-pass filter, and digitally sampling the ECG signal. In some implementations, the ECG signal is received along with additional patient data, where the additional patient data includes patient demographics, other physiological data records, medical history, physical examination findings, and other medical information that the user may request. The patient data can be used in conjunction with patient-specific ECG data used for data processing and display, or the patient data can be used to correlate information extracted from the ECG data. In stage 774, a VCG signal is determined (e.g., generated) based on the received ECG signal. ECG data provides a time-dependent voltage that is processed as a dipole with an origin located at the center of the patient's heart to describe the heart's electrical activity. Multiple ECG lead sites provide different time-dependent voltage waveforms that reflect the overall cardiac electrical activity. A time-dependent cardiac vector representing the magnitude and position of the time-varying electric dipole can be calculated by approximating the cardiac electrical activity.

[0419] Additionally, in stage 774, three or more ECG leads may be used to generate a vectorcardiogram, typically using the X, Y, and Z quadrature components of a VCG vector. A conversion matrix may be used to convert a particular set of leads into the Vx, Vy, and Vz quadrature components of a VCG vector. In some implementations, an inverse transform (e.g., an inverse Dower matrix, a Levkov matrix) may be used to determine the VCG cardiac vector from the ECG. Any conversion method may be used to generate the VCG cardiac vector based on the ECG data.

[0420] In stage 776, a cyclic trajectory comprising a portion of an ECG cycle is determined. The portion of the ECG cycle can be a P, QRS, or T wave. Determining the portion of the ECG cycle can include detecting a starting point of the portion of the ECG cycle and an end of the portion of the ECG cycle, isolating the portion of the ECG cycle based on detecting the starting point of the portion of the ECG cycle and the end of the portion of the ECG cycle, and filtering the isolated portion of the ECG cycle. The portion of the ECG cycle can be determined for each cardiac cycle of the received ECG.

[0421] In some implementations, characterization values ​​for the plurality of cardiac cycles are calculated using information related to the plurality of cardiac cycles to generate a control cycle trajectory and the control cycle trajectory is stored. In some implementations, the characterization value may include a spline estimate of the cycle trajectory corresponding to the plurality of cardiac cycles.

[0422] Statistical shape analysis can be used to characterize cycles or groups of cycles. For example, a control cycle trace group can be automatically generated at the beginning of a monitoring session. There may be user input to the wearable medical device 100 or the medical support device 305 to enable the user to manually initiate a new acquisition of a control cycle trace group. The control group can include two or more ECG cycles, although more typically it will include 30 seconds to 5 minutes of ECG data. This time period can be configured in the non-volatile memory of the patient monitoring device.

[0423] In stage 778, the control cycle group is characterized. The cycle trajectories for different cardiac cycles are analyzed (in some examples, statistically analyzed). Thus, the shape of the control cycle represents a statistical representation, often in the form of a mean, of the shape of the cycle trajectories observed during the control window.

[0424] In some examples, only odd or even beats are used to generate control or test cycle traces. This way, the analysis can be tailored to the effect of T wave alternans. Thus, the control group can be every odd element in a time period, and the test group can be every even element in the same time period, or vice versa. This will provide a more accurate measurement of T wave alternans. In some implementations, every kth cycle L of the group under analysis can be used. k To determine the shape characteristics of the control and / or test groups. For example, for k equal to 3, the control group will include the 1st, 4th, 7th, 10th, etc. of the original control group. The test group will include the 2nd, 5th, 8th, etc. after the control group acquisition period ends. The test group can be updated for each cycle (e.g., the 3rd, 6th, 9th, etc., then the 4th, 7th, 10th, etc.).

[0425] In some implementations, for any particular value of k, multiple subgroups of control time intervals can be created. For example, for k=4, the subgroups can start at cycles 1, 2, 3, or 4, thereby creating four different groups of cycles for shape characterization using this method as a statistical shape analysis. One of these subgroups can be selected as a control group by analyzing the statistical variance of the shapes of the subgroups and selecting the subgroup with the lowest variance in shape. Thus, for each of up to 8 or 10 intervals, or as many intervals as are practically calculated using a microprocessor in the state of the art, a control group can be created for each value of k, even as quickly as with each new cycle, or periodically. It is also possible to decompose the test group into k subgroups and compare the degree of trajectory bifurcation to the control subgroups.

[0426] In some implementations, the characterization value may be based on an average or median value of cyclic traces corresponding to a plurality of cardiac cycles.In stages 780-786, a set of test cyclic traces is generated.

[0427] In stage 780, analysis of a new test cycle group is started. This analysis of the new test cycle group can be based on a time threshold (e.g., analyzing a new set of 15 cycles every 10 minutes or every 30 minutes), or can be based on a physiological trigger such as an increase in risk score or another factor representing a change in the patient's condition. In step 782, a VCG cycle trajectory is determined for the specific beats to be included in the test cycle group. For example, two or more ECG leads can be used to generate a vectorcardiogram (VCG) typically by utilizing the X, Y, Z orthogonal components of the representation of the VCG vector. In step 784, the system determines whether the length of the test group has been reached. For example, the size of the test group can be based on a threshold number of cycles and / or a time-based threshold. If the length has not been reached, the system continues to determine whether to add to the test cycle trajectory group. If the length has been reached, then in stage 786, the system characterizes the test cycle group.

[0428] The cyclic trajectory determined in stages 776 (control loop) and 782 (test loop) can be at least three-dimensional. For example, a first dimension can include a first spatial component of the cyclic trajectory, a second dimension can include a second spatial component of the cyclic trajectory that is perpendicular to the first dimension, and a third dimension can include a third spatial component of the cyclic trajectory that is perpendicular to the first and second dimensions. In some implementations, the cyclic trajectory can include more than three dimensions. For example, a fourth dimension can include additional physiological signals that are registered with the ECG signal.

[0429] In stage 788, the system compares the control cycle group with the test cycle group. In stage 790, a trajectory bifurcation is identified by comparing the test cycle trajectory with the control cycle trajectory. In some implementations, trajectory bifurcation can be identified based on statistical analysis. For example, a change in the cycle trajectory relative to the control cycle trajectory that occurs for a portion of the ECG signal and exceeds the standard deviation of the control cycle trajectory can be identified as a trajectory bifurcation. In some examples, the statistical analysis used to identify trajectory bifurcation can be a non-Gaussian statistical analysis.

[0430] In some implementations, the act of identifying a trajectory bifurcation may include: calculating an area of ​​the loop trajectory; and subtracting the area of ​​the loop trajectory from the area of ​​the control loop trajectory. In some implementations, identifying a trajectory bifurcation results in automatically generating a bifurcation.

[0431] In some implementations, stage 790 is repeated at least three times to compare trajectory bifurcations for at least three consecutive cardiac cycles of the plurality of cardiac cycles, thereby determining a trajectory bifurcation trend and defining a scenario trajectory bifurcation based on the trend. The steps for identifying trajectory bifurcations can be repeated periodically over the plurality of cardiac cycles. For example, after determining the control cycle trajectory, trajectory bifurcations can be identified for each recorded cardiac cycle.

[0432] In stage 792, an event estimated risk score may be generated based on the identification of trajectory bifurcations. In some implementations, the event estimated risk score may include a cardiac risk score (eg, a quantitative risk estimate) based on the identification of trajectory bifurcations.

[0433] Before the system is used on a subject in conjunction with real-time data, the vector cycle bifurcation measurements can be combined with other subject health and demographic information to create a database of retrospective subject health and demographic information and the associated time and severity of medical deterioration, such as cardiac arrest, etc. Parameters used in the survival function and associated confidence intervals, such as the log-logistic function, also known as the Fisk distribution, can be estimated using modeling techniques well known to those skilled in the art, such as linear or nonlinear regression analysis.

[0434] In some implementations, such as described in U.S. Patent 8,668,644, the contents of which are incorporated herein by reference in their entirety, a heart rate variability measure may be used instead of or in addition to a vector loop bifurcation measure combined with subject demographics and other information, wherein: demographic parameters of age are used; vital sign parameters of temperature, pulse rate, respiratory rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), SpO2, and pain score; and HRV parameters of average RR interval, STD, average heart rate (avHR), standard deviation heart rate (sdHR), RMSS, nn50 (counts), pnn50 (%), RR interval triangular index, TINN (ms), LS-VLF power (ms2), LS-LF power (ms2), LS-HF power (ms), LS total power (ms2), LS-LF power (nu), LS-HF power (nu), and LS-LF / HF ratio.

[0435] In some embodiments, the control unit 120 applies a class of medical precursor event estimation and detection algorithms, which may be referred to as "reverse maze" methods, to calculate an estimated event risk score. The general concept is that a maze is sometimes easier to solve by traversing back from the maze's destination to the maze's entrance. The reverse maze method can use less severe or less critical medical precursor events to predict more severe or more critical medical precursor events.

[0436] Medical precursor events can be ranked in terms of intensity (or criticality) levels from less severe to more severe as follows: 1) T wave abnormality; 2) ectopic beats; 3) a series of ectopic beats (scaled by the amount of ectopic beats); 4) a series of ventricular tachycardias (scaled by the duration of the series of tachycardias and / or the number of ventricular beats); 5) ventricular fibrillation. Similar scaling for medical precursor event criticality can be used for bradycardia events and ranked as follows: 1) increased QRS duration; 2) decreased heart rate (scaled by % decrease in heart rate); 3) intermittent heart block (scaled by the duration of the block); 4) complete heart block and ventricular escape (scaled by the duration of the block); 5) cardiac arrest.

[0437] In contrast to parameters for discretely occurring events such as ectopic beats, there are some parameters (such as T-wave measurements, heart rate, QRS width, pulse oximetry, or other physiological parameter measurements) that are continuous. Therefore, in some embodiments, parameters measured on a more continuous basis can be used to increase the accuracy of estimating risk from less critical medical precursor events to more critical medical precursor events. The multidimensional parameter space can be an M-dimensional parameter space, where M = J + K + L, J is the number of elements of the subject's history and demographic data; K is the number of physiological parameters such as vital sign measurements, ECG, SpO2, respiratory rate, etc.; and L is the number of different event types. A 1×M-dimensional vector tracks trajectories in the space, where the update rate of each parameter can be: 1) regular, such as sensor inputs sampled by a machine, such as an ECG; 2) semi-regular, such as vital sign parameters manually measured and entered by a nurse or other caregiver; or 3) irregular events.

[0438] In some embodiments, while working with a database of collected subject data and associated results, thresholds for detecting "tangible," "intermediate level," or "high order" types of medical events can be defined during a training period of time during which the reverse maze algorithm is developed. The thresholds can be set empirically to maximize detection performance, or alternatively, thresholds can be set automatically or semi-automatically using well-known statistical signal processing methods to optimize detection performance and maximize performance measures such as receiver operating characteristics (ROCs). The training database can include a substantially equal number of subjects who have experienced a medical condition exacerbation (cardiac arrest in the case of a cardiac event) and subjects who have not experienced a medical condition exacerbation.

[0439] The reverse maze method works backward from the deterioration of the medical condition in the case of a deteriorating subject, or works backward from the end of the recorded file for a subject in which there is no deterioration of the medical condition. As described in this paragraph for the immediate precursor events of cardiac arrest, threshold levels are set so that the sensitivity for detecting these immediate precursor events is maximized, and specificity is maximized by not having these immediate precursor events appear anywhere in the file for subjects in which there is no cardiac arrest. Preferably, the file for subjects in which there is no cardiac arrest should be at least the length of the time period desired for risk estimation. In files in which there is an immediate precursor (IPP) event and there is also cardiac arrest (IPPCA), the probability and / or error band of risk and time can be calculated. For IPPCA, a threshold can be set for the medical precursor event immediately preceding the IPP via the above method. The process continues in this manner back and thresholds are used to detect all types of events during real-time application to the subject.

[0440] During the training period of algorithm development, after the thresholds for event detection were initially established, the database was divided into separate sets of subjects who had a medical deterioration (cardiac arrest in the case of a cardiac event) and those who did not.

[0441] Using an M-dimensional parameter space, it is possible to use tracking algorithms (e.g., Kalman filters or particle filters, etc.) used in applications such as missile and satellite tracking to estimate and predict vector loop paths. The location of the most recent event and the direction and path of the trajectory are used to estimate the risk and degree of certainty (e.g., confidence bands) of a higher intensity medical precursor event occurring at some point in the future. As more events of various types occur, the trajectory path and the estimation of the future trajectory path can be better estimated, and when estimating risk, the confidence band of the estimate can be narrowed for greater certainty.

[0442] The Kalman filter estimates the process by using a form of feedback control. The Kalman filter estimates the state of the process at a certain time and then obtains feedback in the form of (noisy) measurements. As such, the equations used by the Kalman filter are divided into two groups: time update equations and measurement update equations. The time update equations are responsible for advancing the current state and error covariance estimates (in time) to obtain a priori estimates for the next time stage. The measurement update equations are responsible for feedback, that is, for incorporating new measurements into the priori estimates to obtain improved posterior estimates. The time update equations can also be considered prediction equations, while the measurement update equations can be considered correction equations. In fact, the final estimation algorithm is similar to the prediction-correction algorithm used to solve numerical problems.

[0443] In order to use a Kalman filter to estimate the internal state of a process assuming a sequence of observations containing only noise, the process must be modeled according to the framework of the Kalman filter. This means specifying the following metric for each time step k: The state transition model F k ; Observation model H k ; Covariance Q of the processing noise k ; Covariance R of observation noise k ; and sometimes the control input model B k The Kalman filter model assumes that the true state at time k evolves from the state at time (k-1) according to equation (6):

[0444] x k =F k x k-1 +B k u k +w k (6)

[0445] Among them: F k is applied to the previous state x k-1 State transition model of B k is applied to the control vector u k The control input model of w k It is assumed that the covariance Q k The processing noise induced by the zero-mean multivariate normal distribution is w k ~N(0,Q k ).

[0446] At time k, the true state x is calculated according to equation (7) k The observation (or measurement) z k :

[0447] z k =H k x k +v k (7)

[0448] Among them: H k is the observation model that maps the true state space to the observation space, and v k is assumed to have covariance R k The observation noise of zero-mean Gaussian white noise, that is, v k ~N(0,R k )v k Assume that the initial state and the noise vectors of each step {x0,w1,...,w k ,v1...v k}All independent of each other.

[0449] The Kalman filter is a recursive estimator. This means that only the estimated state from the previous time step and the current measurement are needed to compute the estimate of the current state. In contrast to batch estimation techniques, no history of observations and / or estimates is required. In the following, the notation Denotes the estimated value x at time n taking into account the observations up to and including time m, m≤n.

[0450] The state of the filter is represented by two variables: the posterior state estimate at time k given the observations up to and including time k The posterior error covariance matrix (a measure of the accuracy of the state estimate) P k|k .

[0451] The Kalman filter can be written as a single equation, but is most often conceptualized as two distinct phases: "prediction" and "update." The prediction phase uses the state estimate from the previous time step to produce an estimate of the state at the current time step. This predicted state estimate is also known as the a priori state estimate because, although it is an estimate of the state at the current time step, it does not include observations from the current time step onward. In the update phase, the current a priori prediction is combined with the current observations to refine the state estimate. This improved estimate is called the a posteriori state estimate.

[0452] Typically, two phases are alternated: prediction, which advances the state until the next scheduled observation; and update, which incorporates the observation. However, this is not essential; if an observation is unavailable for some reason, the update can be skipped and multiple prediction steps performed. Similarly, if multiple independent observations are available simultaneously (typically using different observation matrices H), then k ) can perform multiple update steps.

[0453] The predicted (a priori) state estimate can be expressed as follows:

[0454]

[0455] The predicted (a priori) state covariance can be expressed by the following equation (9):

[0456]

[0457] The innovation or measurement residual can be expressed by the following equation (10):

[0458]

[0459] The innovation (residual) covariance can be expressed by the following equation (11):

[0460]

[0461] The optimal Kalman gain can be expressed by the following equation (12):

[0462]

[0463] The updated (a posteriori) state estimate can be expressed as follows:

[0464]

[0465] The updated (a posteriori) estimated covariance can be expressed as follows:

[0466] Pk|k =(IK k H k )P k|k-1 (14)

[0467] The initial task during the measurement update is to calculate the Kalman gain K k , and the next stage is the actual measurement processing to achieve this by incorporating the measured value z k To obtain and generate the posterior state estimate. The final stage is to obtain the posterior error covariance estimate P k After each and measurement update pair, the process is repeated using the previously used a posteriori estimate to project or predict a new a priori estimate. This recursive property is one of the very attractive features of the Kalman filter and makes practical implementation more feasible than, for example, a Wiener filter that is designed to operate directly on all data used for each estimate. Instead, the Kalman filter recursively adjusts the current estimate for all past measurements. Thus, equation (13) is called a prediction equation.

[0468] A major limitation of the Kalman filter is that the Kalman filter models linear systems with Gaussian distributions that are not often encountered in physiological settings. An example of an algorithm for solving the problem of non-Gaussian nonlinear filtering is the extended Kalman filter (EKF). The EKF is based on the principle of linearizing the measurement and evolution model using a Taylor series expansion. However, the series approximation in the EKF algorithm may cause the performance of the nonlinear function and probability distribution of interest to deteriorate. As a result, the EKF may diverge. Based on the assumption that it is easier to approximate a Gaussian distribution than an arbitrary nonlinear function, other researchers have developed a filter known as a lossless Kalman filter (UKF). It has been shown that the UKF obtains more accurate results than the EKF, and the UKF generates a better estimate of the covariance of the states (which the EKF often seems to underestimate). However, there is the following limitation in the UKF: the UKF is not applied to general non-Gaussian distributions, which is often the case for ECG spectrum distributions. The Sequential Monte Carlo method, also known as a particle filter, overcomes this limitation and enables a complete representation of the posterior distribution of these states, making it easy to calculate arbitrary statistical estimates such as the mean, mode, kurtosis, and variance. Therefore, particle filters can handle arbitrary nonlinearities or distributions. Particle filters rely on importance sampling, which requires the design of a distribution of proposals that can approximate the posterior distribution very reasonably. Generally, it is difficult to design these proposals. The most common strategy is to sample from a probabilistic model of the state evolution (before the transition). However, this strategy may fail if new measurements appear in the tail of the prior or if the peak of the likelihood is too large compared to the prior.

[0469] Some other implementations use an estimation / prediction trajectory tracking technique known as an unscented particle filter (UPF), as disclosed in "The Unscented Kalman Filter for Nonlinear Estimation" (2000), by Wan, Eric A. and van der Merwe, Rudolph, the contents of which are incorporated herein by reference in their entirety. For example, Table 2 below provides an unscented Kalman filter (UKF) equation that can be used to calculate an event estimated risk score.

[0470] Table 2

[0471]

[0472]

[0473] For example, for initialization t=0 and for i=1,...N, the state (particle) x0 is derived from the prior p(x0) according to the following equations (15), (16), (17) and (18): (i) And set it up:

[0474]

[0475]

[0476]

[0477]

[0478] For t = 1, 2, ..., importance sampling step: For i = 1, ..., N: Update the particles using UKF according to the following equations (19), (20), (21), (22), (23), (24), (25), (26), (27), (28), and (29):

[0479] Calculate Sigma Points:

[0480]

[0481] Predicting future particles (event updates)

[0482]

[0483]

[0484]

[0485] Incorporate new observations (measurement updates)

[0486]

[0487]

[0488]

[0489]

[0490] -sampling

[0491] -set up and

[0492] For i=1,…N, evaluate the importance weights until a normalization constant is reached:

[0493]

[0494] For i=1,…N, the importance weights are normalized.

[0495] The selection step is performed according to the following equations (30), (31), (32) and (33):

[0496] Use high / low importance weights respectively

[0497]

[0498] To multiply / suppress particles

[0499]

[0500] To obtain N random particles.

[0501] a) Output: The output of the algorithm is a set of samples that can be used to approximate the posterior distribution as follows:

[0502]

[0503] For some attention function g t (e.g., marginal conditional means or marginal conditional covariances or other moments), yielding the following estimates:

[0504]

[0505] After a medical precursor event of lower intensity occurs, multiple potential larger and smaller medical precursor events "lurk" in the future multi-parameter space. Based on the vector trajectory estimated using one or more of the methods described herein, the likelihood or probability of any of these future medical precursor events occurring can be calculated. Medical precursor events can be viewed as sequentially oriented "bursts" or text, and decoding methods such as Viterbi or hidden Markov model (HMM) methods can also be used to improve estimation and detection accuracy. One or more refinement methods (e.g., ensemble prediction) can be used to refine the uncertainty of the vector trajectory estimate and the medical precursor event estimate.

[0506] Ensemble forecasting is a numerical forecasting method that attempts to generate a representative sample of possible future states of a dynamical system. Ensemble forecasting is a form of Monte Carlo analysis: multiple numerical forecasts are made using slightly different initial conditions that are all plausible given the past and current set of observations or measurements. Sometimes, ensemble forecasting can use different forecast models for different members or different formulations of the forecast model. Multiple simulations are performed to account for two common sources of uncertainty in the forecast model: (1) errors introduced by using imperfect initial conditions that are amplified by the chaotic nature of the evolution equations of the dynamical system (often referred to as sensitive dependence on the initial conditions); and (2) errors introduced due to imperfections in the model formulation (such as the approximate mathematical methods used to solve the equations). Ideally, the verified future dynamical system states should fall within the spread of the forecast ensemble, and the amount of spread should be related to the uncertainty (error) of the forecast.

[0507] External defibrillators such as AEDs, manual defibrillators with advisory algorithms, and wearable defibrillators classify shock-related ECG data into two groups: "shockable" rhythms and "non-shockable" rhythms. For certain time periods (e.g., time periods of about 1 hour or less), especially for wearable defibrillators, at least a third category can be defined. This third category can be described as a situation where the subject should be unconscious but is surprisingly still standing. More technically, this third category can be described as a physiological condition disorder or simply a "condition disorder" (SD). Reference Figure 10, the upper ECG waveform 1002 shows a wide and complex ventricular spontaneous beat rhythm at a rate of approximately 120 BPM for the example subject, as recorded using the front-to-back (FB) electrode configuration of the example subject, and the lower ECG waveform 1004 shows a wide and complex ventricular spontaneous beat rhythm at a rate of approximately 120 BPM for the example subject, as recorded using the lateral (SS) electrode configuration of the example subject. In many cases, the subject exhibiting the rhythm shown in the lower ECG waveform 1004 is unconscious. However, it is known from the recorded files of the example subject that the subject was conscious for at least 30 minutes after the arrhythmia. There is a significant delay period from the time of the wide and complex ventricular spontaneous beat rhythm until the wearable defibrillator initiates charging of the high-voltage capacitor and performs the subject notification sequence. Because the detection threshold in conventional wearable defibrillators is set to produce a high level of specificity, so that no false shocks are delivered to a conscious subject and the subject is not burdened or unnecessarily concerned about false notifications and / or defibrillator charging, the wearable defibrillator does not initiate charging and proceed with the subject notification sequence. The higher level of specificity comes at the expense of lower sensitivity for detecting potentially lethal arrhythmias; thus, there is a trade-off between sensitivity and specificity in the detection algorithm.

[0508] In some embodiments, to achieve short-term (up to 4 hours) early warning, the control unit 120 can implement a second shockable rhythm detection algorithm that has been tuned for higher sensitivity of the ECG data (e.g., perhaps at the expense of specificity). For example, the second algorithm can be set to trigger detection of a conditional disorder (SD) for wide and complex ventricular tachycardias (VTACH) with a rate exceeding about 115 BPM, which is different from existing algorithms that typically have a cutoff rate for VTACH of about 150 BPM or about 180 BPM. The QRS width cutoff criteria for VTACH detection can be set for a narrower (less stringent) width than the width typically used to detect SD. In the event that the control unit detects SD using the second shockable rhythm detection algorithm, action can be taken using the wearable medical device 100 (such as Figure 12 actions shown, etc.) to respond to SD.

[0509] Alternatively, the same algorithm used for shockable rhythm detection can have at least two different sets of detection criteria thresholds. The control unit 120 can determine detection based on various thresholds that are more or less sensitive, and use the two or more determined responses for different purposes, for example, one response for early warning and another response for defibrillation. Alternatively, the control unit 120 can analyze the ECG or other physiological waveform data by applying different algorithms to different ECG or other physiological waveform time periods (e.g., alternating three-second time periods, alternating two-second time periods, etc.). If the control unit 120 employs more than two thresholds or more than two detection algorithms, the algorithms or decision thresholds can be used on a sequentially alternating basis (e.g., for algorithms "A," "B," and "C"), and the algorithms can be run on the ECG in the sequence ABCABCABC.

[0510] Alternatively, in addition to ECG waveform analysis, the wearable medical device can detect SD by using a more sensitive shockable rhythm detection on the ECG data and detecting SD via one or more sensors. The second or third sensor may include a pulse oximeter clipped to the concha of the subject's ear to detect the subject's pulse, an accelerometer located on the wearable defibrillator or other part of the subject's body to detect whether the subject is vertical or horizontal or whether the subject is moving, an acoustic sensor located on the subject's body near the subject's heart to detect heartbeat sounds, lung sounds, etc. In the event that the more sensitive ECG rhythm detector detects a "shockable" rhythm, but the auxiliary sensor detects viability such as a pulse, heart sounds, subject movement, or vertical position, the medical device or system can determine that the subject is in the SD stage and perform functions appropriate to that SD stage. If the auxiliary sensor indicates no viability, the wearable medical device 100 can enter the true shockable stage indicated by the more sensitive shockable rhythm detection.

[0511] As described herein, event estimation risk scores can be used to determine a response or action plan, or to provide feedback to a subject and / or healthcare provider based on the likelihood or probability that the subject will experience a medical event (e.g., an adverse cardiac event or other worsening of a medical condition). One category of event estimation risk scores can be based on the likelihood that the subject will experience an adverse cardiac event within a relatively short time period from the current time, such as within about 5 minutes to about 30 minutes, etc. (e.g., within about 5 minutes, within about 10 minutes, within about 15 minutes, within about 20 minutes, and within a time period of about 30 minutes or more). If the subject and / or healthcare professional sets a window of about 5 to 30 minutes of advance notice of a possible adverse cardiac event, various actions can be taken to mitigate the impact of such an event. These actions range from providing an alert or information to the subject and / or caregiver, alerting the healthcare professional, and / or arming or triggering a defibrillator. These time periods may be used to drive system, clinician, user, and device activity via communications to the subject or clinician remote or local to the subject, or in the form of therapeutic interventions such as pacing, defibrillation, neurostimulation, etc. Some embodiments may use any number of time periods, time periods of varying duration, and / or employ methods such as Figure 6B More continuous risk measures are shown.

[0512] In some embodiments, if the event estimated risk score meets an acceptable warning threshold, an early warning can be provided to the subject. For example, the wearable medical device 100 can provide an audio and / or visual alert to the subject to reduce the possibility of secondary injuries (e.g., falls, car accidents, etc.). The early warning system can enable the subject to plan recent activities and / or avoid activities in which the subject is unprotected, such as activities in which the defibrillator device is removed (e.g., showering, changing clothes, holding a child), etc. For example, the early warning can include a "quick" screen containing audio and / or video output, wherein the audio and / or video output indicates to the subject wearing the device that the subject should not take off the wearable defibrillator to shower or should not change his or her clothes. The early warning can enable the subject to reattach the defibrillation pads before an adverse cardiac event occurs. For example, if the subject is in the shower and receives an early warning signal based on a high cardiac event estimated risk score, the subject can reattach the defibrillation pads to be prepared in the event of an adverse cardiac event.

[0513] The type and manner of warning to be provided to the subject based on the estimated risk score of a cardiac event can be changed based on the criticality measurement and confidence measurement of the estimated risk score of the event. For example, if the confidence measurement does not meet the confidence threshold or the criticality measurement does not meet the criticality threshold, the subject may be advised not to take off the defibrillator and stop any current activities; however, if the predicted criticality measurement and confidence measurement meet their respective thresholds, the warning may be more sudden and warn the subject that a shock may be about to occur, thereby giving the subject time to prepare for the event. In some implementations, if the confidence measurement does not meet the confidence threshold, the device may avoid taking an action (e.g., delivering a shock) or take an action different from the action taken if the confidence measurement meets the confidence threshold. In some implementations, if the confidence measurement does not meet the confidence threshold, the device may take measures intended to improve the confidence measurement. For example, the device may perform additional measurements on the subject, increase the frequency of measurements, etc. Thus, the system provides a systematic and diverse response based on the criticality of the event and the probability that the event will occur.

[0514] If the estimated time for a potential event is relatively short (e.g., less than about 15 minutes), the wearable medical device 100 can be used to instruct the subject to begin the Valsalva maneuver. The subject can be instructed by a physician who receives an alert on a cell phone or computer via a wired and wireless communication network, or by a voice command or other written or audio instruction from the wearable medical device 100, or by a voice prompt or instruction from a computing device 550 with which the wearable device communicates, to perform the Valsalva maneuver. The Valsalva maneuver involves the subject taking a deep breath, pinching his or her nose for a few seconds, and releasing the breath. The Valsalva maneuver simulates the parasympathetic nervous system, which may prevent fatal ventricular arrhythmias. The subject can also be instructed to perform actions such as bending over, beating the chest with a thump in front of the heart, etc. Alternatively, the wearable medical device 100 can, in response to a high estimated event risk score, electrically pace the vagus nerve using a transcutaneous electrical stimulation (TENS) waveform using electrodes placed in a wearable device shoulder pad near the subject's clavicle. The stimulation can be magnetic.

[0515] One type of feedback provided to the subject can be designed to limit the stress exhibited by the subject in response to a warning. For example, a warning that advises the subject that they are exhibiting an increased likelihood of an adverse cardiac event may cause other stress-related adverse health effects. Therefore, in addition to or in lieu of more other warnings, a more moderate warning such as "Proceeding advanced diagnostics; keep wearing the vest for 10 minutes" or "Check electrodes" may be transmitted to the subject using the wearable medical device 100. The more moderate warning or other similar warnings may be generated randomly or at preset time intervals to acclimate the subject to the presence of these warnings so that the subject does not associate these warnings with an increased likelihood of an adverse cardiac event that may cause other stress-related adverse health effects.

[0516] The wearable medical device 100 can selectively charge the capacitor of the defibrillator device based on the estimated cardiac event risk score. Energy in the defibrillator device for applying an electric shock to the subject needs to be accumulated, such as by charging the capacitor, and this energy accumulation may require a considerable length of time. Since the estimated cardiac event risk score provides an early indication of a possible adverse cardiac event, the system can start charging the capacitor when the estimated cardiac event risk score meets the charging threshold. For example, the system can analyze the needs of the subject in advance and start fully charging the capacitor or other appropriate energy delivery mechanism before estimating the need for a shock so that the shock can be delivered as soon as a shock is needed (for example, when a shockable rhythm is detected). Thus, the estimated cardiac event risk score can be used to start charging a defibrillator such as an AED. Advanced charging according to some embodiments enables easy-to-use defibrillation and can extend the battery life of the defibrillator. For example, an example method for charging a defibrillator is described in U.S. Patent Application Publication No. 2011 / 0202101, entitled “DEFIBRILLATOR CHARGING,” filed on February 11, 2011, the contents of which are incorporated herein by reference in their entirety. Providing early defibrillator charging based on a calculated estimated cardiac event risk score can minimize the size of a battery unit in a wearable medical device because the capacitor can be charged over a longer period of time due to the early warning provided by the estimated event risk score.

[0517] If the wearable medical device 100 comprises a LifeVest®, such as that available from ZOLL Medical, Inc., located in Chelmsford, Massachusetts, TM If a wearable defibrillator such as a wearable cardioverter-defibrillator is used, the cardiac event estimation risk score can enable a reduction in the amount of hardware included in the Holter-based wearable medical device. For example, the wearable medical device may include multiple therapy electrodes (e.g., Figure 2The wearable medical device 100 may be a device that is electrically connected to a control unit and capable of delivering one or more therapeutic defibrillation shocks to the subject's body. To reduce hardware, the electrodes provided in the wearable medical device may be dry electrodes that do not require gel to be dispersed on the subject's skin before use. For example, the dry electrodes may include barbs configured to attach to the subject's skin when delivering the shock. The barb electrodes may provide defibrillation capabilities for unexpected events. However, using wet electrodes to provide defibrillation shocks may be beneficial, so in the event that the estimated risk score for a cardiac event meets a threshold indicating that defibrillation may be required, the wearable medical device 100 may suggest that the subject change the electrodes connected to the device from the dry electrodes that are typically initially provided to a pair of wet electrodes.

[0518] The cardiac event estimation risk score can be used to warn medical professionals of potential adverse cardiac events. For example, if the subject is in a hospital environment, the wearable medical device 100 can trigger an alarm based on the event estimation risk score meeting the alarm threshold to speed up the emergency response, such as sending an alarm to a code group to respond to the subject. The alarm can be sent to a member of the Medical Emergency Response Team (MERT) or Medical Emergency Team (MET) via the communication network 580, which can be a hospital communication network. The MERT or MET team can respond to the subject before the arrest occurs to prevent the arrest.

[0519] As described above, since some of the algorithms used for medical precursor event estimation and detection can be computationally intensive, some or all of the processing can be performed on server hardware that is separate from the wearable medical device worn by the subject or near the subject. For example, in a hospital, data from the subject can be streamed from the wearable medical device 100 in real time to a server in the hospital network for processing. If the subject is not in a hospital environment, the medical device can send the subject's physiological data (e.g., ECG data) and / or a response to the physiological parameters determined by the wearable medical device 100 to the subject's doctor, and / or send a message to the emergency room of a nearby hospital. Messages can be sent to other caregivers to warn them of the possibility of an adverse cardiac event.

[0520] In some implementations, a portion of the processing of the algorithm used for medical precursor event estimation and detection is performed on the medical device, and a portion of the processing is performed on the server. For example, the algorithm processing can initially be performed by the medical device. If the processing becomes too computationally intensive for the medical device, the server can provide assistance. In this way, the algorithm processing can be performed solely by the medical device when possible, and the server is only called upon for auxiliary processing when necessary.

[0521] The estimated cardiac event risk score can be used to initiate clinical treatment to reduce the likelihood of a cardiac event and / or mitigate the adverse effects of a potential cardiac event. For example, the wearable medical device 100 or another medical device can provide automatic stimulation to the subject's heart to prevent the onset of an adverse cardiac event, such as through pacing. If the estimated cardiac event risk score meets a medication treatment threshold for administering medication, the wearable medical device can alert the user to administer an oral, injectable, or inhalable medication. For example, an inhaled device can be used to administer an inhaled medication such as nitric oxide or sublingual nitroglycerin, which acts as a vasodilator. Medication can be delivered to the subject using an autoinjector, such as a medical device for delivering measured doses of medication. The autoinjector can take the form of a pen-type device in which a spring-loaded needle is ejected from the front end of the device and penetrates the subject's skin to deliver the medication via subcutaneous injection. Exemplary medications that can be included in the autoinjector include aspirin, vitamin E, anti-inflammatory drugs, beta-blockers, and / or benzodiazepines such as diazepam.

[0522] If the estimated time to medical emergency exceeds approximately 30 minutes, the wearable medical device 100 or a physician responding to a message from the wearable medical device 100 may instruct the subject to take a pill containing a drug that can protect the subject's heart and brain even in the event of cardiac arrest. As described by O'Rourke, Brian, in "Evidence for Mitochondrial K+ Channels and Their Role in Cardioprotection," Circ Res. 2004;94:420-432 (the contents of which are incorporated herein by reference in their entirety), one class of such drugs is KATP channel openers.

[0523] like Figure 11As shown, the wearable medical device 100 can select various actions based on a combination of a criticality (risk) measurement and a confidence measurement for an event estimated risk score for a medical precursor event. For example, in the event that the criticality measurement and the confidence measurement of the event estimated risk score for a particular medical event each meet a criticality threshold and a confidence threshold, respectively, within an associated time period, the wearable medical device can perform or recommend one or more actions. If one of the criticality measurement or the confidence measurement does not meet the corresponding threshold, the action may not be recommended or a different action may be recommended. The thresholds may be different based on the type of action to be recommended and the associated time period. For example, the criticality threshold and confidence threshold required to perform an action of notifying the wearer that the medical device is performing advanced diagnostics (box 902) may be lower than the criticality threshold and confidence threshold required to perform an action of warning that an electric shock may be about to be administered (box 904) or preparing the device for treatment (e.g., charging) (box 912). As Figure 11 As shown, even if the criticality measurement value used for the medical precursor event meets the criticality threshold for alerting the medical professional (box 906), if the confidence measurement value used for the medical precursor event does not meet the confidence threshold for alerting the medical professional within the associated time period, the wearable medical device 100 can also be used to select a different action or response that meets the confidence threshold, such as opposing the removal of the device (box 908) or suggesting that the subject make behavioral modifications (box 910). For actions that are more invasive to the subject, the criticality threshold and confidence threshold may be higher. By setting different thresholds for different actions, the wearable medical device 100 can provide a graded response to the subject's condition. Similar sets of criticality thresholds and confidence thresholds can be associated with the time periods described herein.

[0524] In some embodiments, based on the different degrees of sensitivity and / or specificity of the event-estimated risk score, the subject may be advised or an early warning system (e.g., Figure 5For example, a response determined based on an event estimation risk score having a first sensitivity and a first specificity may be different from a response determined based on an event estimation risk score having a different second sensitivity and / or a different second specificity. For example, a relatively high level of specificity may be required to charge a shock mechanism or deliver treatment so that no pseudo shock or treatment is delivered to the subject and the subject is not burdened or worried about pseudo notifications and / or defibrillator charging. A relatively low level of specificity may be required for responses such as instructions for behavior modification. A relatively high sensitivity can be used to ensure that medical precursor events are detected; however, a higher level of sensitivity may result in a lower level of specificity and the detection of more pseudo events. Therefore, the response determined for the event estimation risk score can change based on the sensitivity and specificity of the event estimation risk score.

[0525] The early warning system 10 can be modified based on the risk of the subject 104 experiencing a medical event. For example, if the risk of the subject 104 experiencing a medical event is relatively high, such as a relatively high risk of cardiac arrest, the sensitivity of the algorithm used to determine the estimated risk score for the event can be modified to be more sensitive to the detection of the event. The sensitivity and specificity can be set for each patient. When a high-risk event is detected, a notification can be provided to a central monitoring station (e.g., a technical support center for the medical device manufacturer). The subject 104 can be instructed to charge the battery of the wearable medical device 100 at a higher frequency than the typical charging frequency when the risk of the medical event is relatively high (e.g., above a risk threshold) to ensure that the device 100 is ready to respond to the medical event.

[0526] exist Figure 12 Various example time periods and associated actions that may be taken based on corresponding criticality thresholds and confidence thresholds are shown in and described below.

[0527] About 1 hour before the incident

[0528] The event estimation risk score can be used to alert medical professionals of potential adverse cardiac or medical events that may occur within approximately 1 hour. For example, if the subject is in a hospital setting, the wearable medical device 100 can trigger an alarm to notify a cardiologist or other medical professional to examine the subject. If the subject is not in a hospital setting, the wearable medical device 100 can send the subject's ECG to the cardiologist or other medical professional so that the medical professional can follow up with the subject. Messages can be sent to other caregivers to warn them of the possibility of an adverse cardiac event.

[0529] The estimated event risk score can be used to initiate clinical treatment to reduce the likelihood of a cardiac event and / or mitigate the adverse effects of a potential cardiac event. If the estimated event risk score meets a threshold for administering medication, the wearable medical device can alert the user to administer an oral, injectable, or inhalable medication, such as those described herein. The type of medication that can be administered 1 hour before a predicted event may be different from the type of medication that is sufficiently fast-acting to be administered within 10 minutes of the predicted onset of an adverse cardiac event.

[0530] Within 3 hours before the incident

[0531] The estimated event risk score can be used to warn the subject and / or medical professional of a potential adverse cardiac medical event that is likely to occur within approximately 3 hours. If the estimated event risk score indicates that an adverse cardiac event is likely to occur within approximately 3 hours, the wearable medical device 100 can take actions to ensure that the battery has sufficient charge to provide defibrillation if medically necessary. For example, if the estimated cardiac event risk score meets a charging threshold, the wearable medical device can check the battery status and, if the battery charge is less than a predetermined percentage, recommend that the subject replace the battery. However, if there is a likelihood that an adverse cardiac event will occur in the near future, the subject is not recommended to replace the battery. Therefore, if the estimated cardiac event risk score for a first time period does not meet a first threshold (e.g., the likelihood of an adverse cardiac event in that time period is low), and the estimated cardiac event risk score for a second time period at a later time period meets a second threshold (e.g., the likelihood of an adverse cardiac event at a later time period is medium or high), the wearable medical device can recommend replacing the battery. By comparing the estimated cardiac event risk scores from these two time periods with the threshold, the subject reduces the risk of replacing the battery within a time period when an adverse cardiac event is likely to occur.

[0532] In the event that the estimated risk score of the event meets the threshold, the wearable medical device 100 can notify the family members or caregivers of the subject by sending a message via the communication network 580. The family members to be notified can be programmed in the medical device 100 and / or the base unit 530. The medical device 100 and / or the base unit 530 can be used to automatically notify a call center (e.g., with General Motors' "On-Star"). TM The call center can contact the subject via a telephone or other communication method (such as Skype or a pager). The call center can contact the subject's physician to triage the incident to the physician. The call center personnel can instruct the subject to keep the device on and, if appropriate, call the EMS system (e.g., 911).

[0533] In some embodiments, one or more of the warnings discussed above for less than 1 hour until an event may also be included.

[0534] About 1 day to 3 days before the event

[0535] A cardiac event estimation risk score can be used to warn medical professionals of the likelihood of a potential adverse cardiac event occurring within about 1 to about 3 days. The medical event can include an epileptic seizure, and particularly for anesthetized epileptic subjects, hospital personnel can be warned of the seizure in advance. If an event is predicted using the wearable medical device 100 and the subject is utilizing a wearable defibrillator with reduced or no treatment options, a three-day warning can be used to trigger automatic delivery of a platform with more robust treatment features. Alternatively, the wearable medical device 100 can include a conventional defibrillator or other physiological monitoring device and can be used to triage and assess the subject's temporal risk profile upon admission to the hospital, and as a result of the test, a tiered treatment response can be provided. For example, if the risk is greater than 50% with a certainty of more than + / - 10% of potential cardiac arrest within a time period of hospitalization (e.g., about 1 to 3 days), the subject may be required to wear a wearable defibrillator (such as one manufactured by ZOLL Medical) during his or her hospital stay. Type wearable defibrillator, etc.). If the risk is in the range of about 15% to about 50%, the subject may be required to wear a wearable monitor, wherein the wearable monitor does not include a defibrillator for treating the subject in the event that the subject experiences cardiac arrest. If the risk is less than 15%, the subject may be monitored periodically (e.g., at 1 hour intervals) by medical staff only. For less severe cases of the wearable monitor and intermittent monitoring by medical staff, medical precursor event estimation and detection can be performed continuously, and if the level of risk and certainty of the subject increases during his or her hospitalization, the subject can be switched from the wearable monitor to a combined wearable monitoring defibrillator, or from nurse monitoring to one of the wearable devices. Based on the subject's triage time risk profile, the subject can be sent to a ward with a more appropriate acuity level, or the subject can be moved to a lower acuity ward, or even be notified to be discharged (e.g., while protected by a wearable defibrillator). If the subject is not in a hospital setting, the wearable medical device 100 may send the subject's ECG via the communication network 580 to the subject's cardiologist or other medical professional so that the medical professional can follow up with the subject if necessary.

[0536] The subject may wear a wearable medical device 100 with reduced or no treatment options, such as a device that does not include defibrillation capabilities. Based on the estimated risk score of a cardiac event for 1 to 3 days, the wearable medical device 100 may notify the subject of an increased likelihood of an adverse event and instruct the subject to switch to using a wearable medical device 100 that includes treatment options. For example, if the estimated risk score for the event meets a switching threshold, a warning may be issued to the subject to switch from a medical device with reduced or no treatment to a wearable medical device that includes treatment options. If the estimated risk score for the event meets the switching threshold and the subject does not own a device with treatment options, a wearable medical device that includes treatment options may be automatically shipped to the subject if the estimated risk score for the event indicates an increased likelihood of an adverse event. For example, the wearable medical device 100 may automatically send an order for a wearable medical device that includes treatment options via the communication interface 580 to a hospital or medical device provider that ships the medical device to the subject.

[0537] Event estimation risk scores over a 1-day or 3-day time period can be used to assist medical professionals in determining and planning the discharge of a subject from a hospital setting. In some embodiments, three threshold levels can be used to facilitate discharge decisions. A first threshold can indicate a low likelihood of an adverse event and be associated with a recommendation to discharge the subject. A second, higher threshold can indicate a moderate likelihood of an adverse event and be associated with a recommendation to discharge the subject with a wearable medical device that includes therapeutic / defibrillation capabilities. A third threshold that is higher than both the first and second thresholds can indicate an increased or higher likelihood of an adverse cardiac event and be associated with a recommendation to keep the subject in a hospital setting for further observation. Thus, event estimation risk scores can provide a structured, systematic approach to planning discharge from a hospital setting.

[0538] In some embodiments, one or more of the warnings discussed above for less than 1 hour and / or less than 3 hours to an event may also be included.

[0539] About 1 week to 1 month before the incident

[0540] The estimated risk score for an event over a longer period of time can be used to increase compliance with the use of the wearable medical device. For example, even in cases where the specificity is low or the estimated risk score for an event is low, the wearable medical device 100 can provide feedback to the subject regarding the likelihood of an adverse cardiac event to motivate the subject to improve his or her compliance with wearing the wearable medical device 100 including therapeutic capabilities.

[0541] The estimated risk score for an event over a longer time period can be used to determine extended care for the subject. For example, if the estimated risk score for an event meets a threshold, the medical professional can be prompted to extend the recommended length of use of the wearable medical device. For example, if the subject is nearing the end of the time they are required to wear the wearable medical device, the physician or medical professional can extend the length of time the subject is required to wear the wearable medical device if the criticality and likelihood of an adverse cardiac event meet associated thresholds.

[0542] For subjects who have had a myocardial infarction but whose ejection fraction is below the criteria for receiving a wearable defibrillator during the waiting period prior to implantation of an implantable defibrillator, an elevated estimated risk score for an event can be used to justify providing the subject with a wearable defibrillator. For subjects who have a syncope in the emergency room and whose risk score is estimated to extend beyond the expected duration of the hospital stay, an elevated estimated risk score for an event can be used to prescribe that the subject be received a wearable defibrillator and be discharged with the wearable defibrillator as protection.

[0543] The estimated risk score for an event over a longer period of time can enable the subject to remain aware of potential warning signs that indicate the onset of an event. For example, if the estimated risk score for an event meets a threshold, the wearable medical device can ask the subject if he or she has noticed any warning signs, and if so, prompt the subject to speak with a medical professional.

[0544] In some embodiments, one or more of the warnings discussed above for less than 1 hour, less than 3 hours, less than 1 day, and / or less than 3 days to an event may also be included.

[0545] More than 3 months before the incident

[0546] Event estimated risk scores for longer time periods (such as scores for predicting events that are likely to occur within a 1- to 3-month timeframe or a 6-month to 1-year timeframe) can be used to guide long-term treatment. For example, event estimated risk scores can be used to guide long-term device planning. Event estimated risk scores can be used to guide treatment based on risk trends (such as continuously increasing event estimated risk scores). If the long-term event estimated risk score continues to rise or increase, the wearable medical device 100 can recommend rehabilitation efforts to reduce risk, such as the administration of medication, additional invasive testing, or extended use.

[0547] As described herein, an event estimated risk score can include a calculated criticality measure and a calculated confidence measure, and the combination of criticality and confidence can be used to determine an appropriate response to a medical precursor event within an associated time period. For example, some treatments, such as defibrillation, may require a higher confidence value before implementation. Other treatments, such as instructing the subject to sit down, may require a lower confidence value before implementation.

[0548] In some implementations, the system can be configured to generate a configurable ... Figure 5 The early warning system described herein may be at least partially implemented and / or controlled by executing software on a computing device 550. For example, an application (e.g., a mobile application) configured to run on a mobile phone and / or tablet computer may facilitate the operation of the early warning system.

[0549] Figure 14A An example of a parameter input screen 1402 for an example cardiac arrest prediction (CAP) application (e.g., a mobile application) is shown. The parameter input screen 1402 may be presented when a clinician is using the application for the first time. The patient, a medical professional, or both may enter patient information into the application. Some of the information entered into the application, such as ECG data, may be provided by a medical device (e.g., a defibrillator, a cardiac monitor, etc.). For example, some information may be provided via a wireless connection. The computing device may then use the entered patient information (and, for example, data received by the medical device) to determine an estimated risk score for an event and / or an increased estimated risk score for an event. Information that may be entered via the application's user interface may include the patient's age, Glasgow Coma Scale (GCS) score, temperature, heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure, SpO2 measurement, and pain score (e.g., on a scale of 1 to 10). As described below (e.g., while the medical device is transmitting ECG data), one or more physiological parameters included in the patient information may be collected by the medical device and provided to the application. In some implementations, the parameter input screen 1402 can display the values ​​of physiological parameters automatically measured and transmitted by the medical device. The medical professional can edit (e.g., overwrite) the automatically measured values ​​by manually entering the values ​​of the physiological parameters. If one or more physiological parameters are missing values, the medical professional can be prompted to manually enter these values ​​before any patient score is calculated. In some implementations, the data entered by the user and / or the clinician is not overwritten by the data automatically measured by the medical device.

[0550] The application and / or medical device may check one or more of the physiological parameters to ensure that the physiological parameters are within acceptable ranges. For example, the age parameter may require a value greater than or equal to 18, the temperature parameter may require a value greater than or equal to 90°F and less than or equal to 110°F, the heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure, and SpO2 parameters may require values ​​greater than 0, and the systolic blood pressure may require a value greater than the diastolic blood pressure value.

[0551] Before or after providing initial patient information, the medical device may monitor the patient by measuring and / or recording one or more types of physiological data (including receiving signals containing ECG data). Figure 14B As shown in the screenshot of , upon receiving ECG data, the application displays a data collection indication 1404. In some implementations, five minutes of ECG samples are measured from the patient at 250 Hz. In some implementations, the five minutes of ECG data can be continuous or non-continuous. In some implementations, the five minutes of ECG data must have fewer than a predetermined number of interruptions to be considered an acceptable data set. In some implementations, the ECG data must have an amount of noise that falls below a tolerable threshold. An initial CAP score can be determined for the patient based on the received ECG data and the input vital signs data. The CAP score is sometimes referred to as an event estimated risk score or simply an event risk score.

[0552] In some implementations, a modified early warning (MEW) score is also determined for the patient. A relatively high MEW score may indicate that the patient may be experiencing a worsening condition and / or acute illness. The MEW score may be determined separately and / or independently of the CAP score. In some implementations, the MEW score may be based on at least some information different from that used to determine the CAP score. Alarm, Voice, Pain, Unresponsiveness (AVPU) parameters may be required to calculate the MEW score. The AVPU parameters are indicative of the patient's responsiveness and / or level of consciousness. The AVPU parameters may be a simplification of the GCS score. The AVPU parameters may have an "alarm" value, indicating that the patient is fully awake (although not necessarily aware). A patient with an "alarm" parameter will spontaneously open their eyes, will respond to voice (although perhaps confused), and will have bodily motor function. The AVPU parameters may have a "voice" value, indicating that the patient can respond in some way when spoken to. This response may take the form of any of the three component measurements: eyes, voice, or movement. For example, a patient who opens their eyes in response to being asked, "How are you?" may be assigned the AVPU parameter "voice." The response may be as small as a grunt, groan, or slight movement of a limb when prompted by a voice prompt from the rescuer. The AVPU parameter may have a "pain" value, where the "pain" value indicates that the patient responded in the form of any of the three component measures of eyes, voice, or movement in response to the application of a painful stimulus (such as a central pain stimulus like rubbing the sternum or a peripheral stimulus such as squeezing the patient's fingers). A patient with some degree of consciousness (for example, a fully conscious patient will not need a painful stimulus) may respond by using his voice, moving his eyes, or moving a part of his body (for example, making an unusual posture). The AVPU parameter may have a "no response" value indicating that the patient did not give any eye, voice, or movement response to the voice or pain stimulus. The appropriate parameter can be selected in the AVPU selection menu 1408. Reference Figure 14C , an initial CAP score 1408 and a MEW score 1410 are determined for the patient based on the received ECG data and the selected AVPU parameters 1406. Figure 15 The process for calculating the CAP score 1408 and the MEW score 1410 is described in more detail.

[0553] Figure 14D An example of a connection screen 1412 for creating a communication association between a mobile application and a medical device is shown. The patient and / or caregiver can enter the IP address of the medical device to be associated with the application, thereby enabling information to be downloaded from the medical device to the application (and, for example, the computing device on which the application is running). Figure 14AAfter receiving the patient information, the mobile application presents a connection screen 1412. After establishing a connection between the application and the medical device, the mobile application can begin receiving data and / or measuring data and / or recording physiological data as described above. In some implementations, a medical device identifier other than an IP address can be provided instead of or in addition to the IP address. In some implementations, the application can be configured to automatically associate with the medical device without the assistance of the connection screen 1412.

[0554] Once a connection is established between the application and the medical device, the computing device on which the application is running can be configured to provide information related to the patient to another device, such as a device of an external entity and / or a remote server (e.g., a remote host server). For example, patient data can be streamed from the computing device (e.g., in real time) to a server in a hospital network for processing. In some implementations, the patient data can be streamed to a cloud-hosted server for data storage and more detailed data analysis. If the patient is not in a hospital environment, the computing device can send information related to the patient (e.g., the patient's physiological parameters, ECG data, responses to physiological parameters determined by the medical device, information related to CAP scores, information related to MEW scores, etc.) to the patient's doctor and / or send a message to the emergency room of a nearby hospital. Messages can also be sent to other caregivers to warn them of the possibility of an adverse cardiac event.

[0555] As mentioned above, data processing can be done in the early warning system ( Figure 5 10) occurs at various locations (e.g., on various devices). For example, at least a portion of the data (e.g., ECG data received by the medical device, data input via an application, etc.) can be processed on any combination of a computing device (e.g., a mobile device), a medical device, a base unit, a server, the Internet, a cloud-hosted server, and / or a second medical device (e.g., a medical device such as a defibrillator or a heart monitor that can communicate with the medical device). In some implementations, the patient data (e.g., ECG data) is sent to a mobile device, and at least a portion of the processing of the patient data is performed on the mobile device. In some examples, because some algorithms used for medical precursor event estimation and detection can be computationally intensive, at least a portion of the processing for determining an event estimated risk score and / or actions in response thereto can be performed on server hardware separate from the mobile device and the medical device.

[0556] Figure 15Examples of various physiological parameters that may be considered when calculating the MEW score are shown. The MEW score is the sum of the individual physiological parameter scores. For example, a patient with a systolic blood pressure of less than 70 mmHg may contribute a MEW score of 3; a patient with a systolic blood pressure of 71-80 mmHg, a heart rate of less than 40 beats / minute, a respiratory rate of less than 9 breaths / minute, a body temperature of less than 35°C, and an AVPU parameter of "speech" may have a MEW score of 9; a patient with a systolic blood pressure of 81-100 mmHg, a heart rate of 41-50 beats / minute, a respiratory rate of 9-14 breaths / minute, a body temperature of 35°C to 38.4°C (e.g., including both end points), and an AVPU parameter of "alarm" may have a MEW score of 2; a patient with a systolic blood pressure of 101-199 mmHg, a heart rate of 51-100 beats / minute, a respiratory rate of 9-14 breaths / minute, a body temperature of 35°C to 38.4°C (e.g., including both end points), and an AVPU parameter of "alarm" may have a MEW score of 4. The EW score can be 0; the MEW score of a patient with a systolic blood pressure of 101-199 mmHg, a heart rate of 101-110 beats / minute, a respiratory rate of 15-20 breaths / minute, a body temperature of 35°C-38.4°C (for example, including both end points) and an AVPU parameter of "speech" can be 3; the MEW score of a patient with a systolic blood pressure greater than 200 mmHg, a heart rate of 111-129 beats / minute, a respiratory rate of 21-29 breaths / minute, a body temperature greater than or equal to 38.5°C and an AVPU parameter of "pain" can be 10; and the MEW score of a patient with a systolic blood pressure greater than 200 mmHg, a heart rate greater than or equal to 130 beats / minute, a respiratory rate greater than or equal to 30 breaths / minute, a body temperature greater than or equal to 38.5°C and an AVPU parameter of "no response" can be 13.

[0557] In some implementations, patients with relatively high MEW scores (e.g., MEW scores of 4 or 5) should be observed more frequently by a medical professional. For example, depending on the particular MEW score, it may be appropriate to observe the patient at least once every 12 hours, once every 6 hours, once every 3 hours, once every 1 hour, once every 30 minutes, or even continuously. In some implementations, patients with a MEW score of 1 may also require more frequent observation. In some implementations, patients with a MEW score of 1 may be observed more frequently if a subsequent MEW score is also 1 or higher.

[0558] Figure 16An example of various patient characteristics that may be considered when calculating the Glasgow Coma Scale (GCS) score is shown. The GCS score includes three tests based on the patient's eye function, language function, and motor function. The patient is assigned a separate value for each of these functions, and the composite value is the GCS score. A patient may receive an eye function value of 1 (does not open eyes) to 4 (opens eyes spontaneously); a patient may receive a language function value of 1 (does not make sounds) to 5 (oriented, talks normally); and a patient may receive a motor function value of 1 (no movement) to 6 (obeys commands). The lowest possible GCS score is 3, which may indicate that the patient is in a deep coma or dead, and the highest possible GCS score is 15, which may indicate that the patient is fully awake and responsive.

[0559] Figure 17 17. A process for receiving additional data to calculate a modified early warning score (MEWS) (sometimes referred to as an enhanced event estimation risk score) is shown. In stage 1702, the wearable medical device 100 monitors a subject by measuring and / or recording one or more types of physiological data, including receiving a signal containing ECG data. The ECG data may be received from any suitable source of ECG data of the subject. For example, the ECG data may be received from ECG electrodes attached to the subject (e.g., Figure 2 The signal including the ECG data may be received in real time from 112 of the present invention or from previously recorded data received from a storage device. The signal including the ECG data may be received along with additional subject data, wherein the additional subject data includes subject statistics, other physiological data records, medical history, physical examination findings, and other medical information that the user may request. The additional subject data may be used in conjunction with the subject-specific ECG data for data processing and display.

[0560] In stage 1704, the control unit 120 calculates event estimated risk scores for a plurality of time-to-event time periods. Each event estimated risk measure may include a calculated crit...

Claims

1. A method for estimating a medical precursor event, comprising: Acquire, by one or more processors, a first set of physiological information of the subject received in a first time period and at least partially based on a first ECG signal of the subject, and a second set of physiological information of the subject received in a second time period; calculating, by the one or more processors, a first risk score and a second risk score associated with estimating a risk of a potential arrhythmic event for the subject based on applying the first set of physiological information and the second set of physiological information to one or more machine learning classifier models trained with training metrics, wherein the training metrics include at least one of: i) a plurality of cardiac electrophysiological metrics of the first subject; and ii) at least one of a demographic metric and a medical history metric of the plurality of first subjects, wherein the one or more machine learning classifier models are validated with validation metrics of a plurality of second subjects, and one or more thresholds of the one or more machine learning classifier models are set based on the validation; providing, by the one or more processors, at least the first risk score and the second risk score associated with the potential arrhythmic event as a time-varying series of risk scores; and The first risk score and the second risk score associated with estimating the risk of the potential arrhythmic event of the subject are classified, by the one or more processors, based on the one or more thresholds.

2. The method according to claim 1, wherein: The one or more thresholds include at least a high risk threshold and an immediate risk threshold, and the classification includes: For each of the first risk score and the second risk score, Classifying the risk of the potential arrhythmic event of the subject as high risk based on the first risk score or the second risk score exceeding the high risk threshold; and The risk of the potential arrhythmia event of the subject is classified as an immediate occurrence risk based on the first risk score or the second risk score exceeding the immediate occurrence risk threshold.

3. The method according to claim 1, wherein: The classification includes a time-varying classification of the subject's risk of the potential arrhythmic event based on the time-varying series of risk scores.

4. The method according to claim 1, wherein: The classifying includes adjusting a base specificity of the one or more machine learning classifier models to reduce Type 1 errors in a base classification of the subject's risk of the potential arrhythmic event.

5. The method according to claim 4, wherein: The Type 1 error is a false positive.

6. The method according to claim 1, wherein: The validation metrics include one or more of cardiac electrophysiological metrics, demographic metrics, and medical history metrics of the plurality of second subjects.

7. The method according to claim 1, wherein: The method further comprises: updating, by the one or more processors, the validation metric by at least one of: 1) adjusting one or more of the validation metrics; and 2) expanding the validation metric based on adding additional one or more subjects to the plurality of second subjects; and The one or more thresholds are refined, by the one or more processors, based on the updated validation metric.

8. The method according to claim 1, wherein: The method further comprises: updating, by the one or more processors, the training metric by at least one of: 1) adjusting one or more of the training metrics; and 2) expanding the training metric based on adding additional one or more subjects to the plurality of first subjects; and Retraining, by the one or more processors, the one or more machine learning classifier models based on the updated training metrics.

9. The method according to claim 1, wherein: The validation metrics of the plurality of second subjects are independent of the training metrics of the plurality of first subjects.

10. An externally worn defibrillator for monitoring a patient's physiological condition, the defibrillator comprising: a plurality of ECG sensing electrodes configured to be disposed relative to the patient's skin and to receive ECG signals; one or more auxiliary sensors disposed on the patient; a plurality of therapy electrodes configured to be disposed relative to the patient's skin and to provide an electrical shock to the patient; A control unit configured to: receiving the ECG signals from the plurality of ECG sensing electrodes and an output signal from the auxiliary sensor; analyzing the ECG signal to determine whether the patient is experiencing a shockable arrhythmia, and initiating a therapy sequence if the patient is experiencing a shockable arrhythmia; as well as The ECG signal and the output signal from the auxiliary sensor are analyzed to determine whether the patient is experiencing a condition disorder state, and if the patient is experiencing a condition disorder state, an early warning regarding the condition disorder state is provided to the patient and / or the patient's caregiver.

Images