Method and system for predicting an adverse patient clinical event

Abstract

The invention relates to a computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, comprising the steps of providing a first dataset (DS1) comprising medical data acquired by the implantable medical device (10), applying a machine learning algorithm (A) to the first dataset (DS1) for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event and outputting a second dataset (DS2) comprising a classification of the predicted and / or detected adverse patient clinical event. In addition, the invention relates to a computer-implemented method for providing a machine learning algorithm (A) and to a system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

Description

[0001] Method and system for predicting an adverse patient clinical event

[0002] The invention relates to a computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0003] Furthermore, the invention relates to a computer-implemented method for providing a machine learning algorithm for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0004] Moreover, the invention relates to a system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0005] Many diseases of the heart are accompanied by changes in the data acquired by an implantable medical device. Said changes in the data can be for medical reasons. There are currently no known technical solutions for a detection of an upcoming or acute clinical event based on medical data of IPG / ICD / ICM devices. Existing home monitoring solutions do not include services or data preparations / processing / visualizations to detect clinical events.

[0006] It is therefore an object of the present invention to provide a method and system for predicting an adverse patient clinical event, in particular to predict an adverse patient clinical event and / or detecting an acute adverse patient clinical event such that an adverse patient clinical event can be prevented.

[0007] The object is solved by a computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event having the features of claim 1. Furthermore, the object is solved by a computer-implemented method for providing a machine learning algorithm for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event having the features of claim 14.

[0008] Moreover, the object is solved by a system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event having the features of claim 15.

[0009] Further developments and advantageous embodiments are defined in the dependent claims.

[0010] The present invention provides a computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0011] The method comprises providing a first dataset comprising medical data acquired by an implantable medical device, applying a machine learning algorithm to the first dataset for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, and outputting a second dataset comprising a classification and / or a probability indication of the predicted and / or detected adverse patient clinical event.

[0012] Furthermore, the present invention provides a computer implemented method for providing a machine learning algorithm for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0013] The method comprises providing a first training dataset comprising medical data acquired by the implantable medical device.

[0014] In addition, the method comprises providing a second training dataset comprising a classification of a predicted and / or detected adverse patient clinical event, and training the machine learning algorithm by an optimization algorithm which calculates an extreme value of a loss function. Moreover, the present invention provides a system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event.

[0015] The system comprises an implantable medical device configured to provide a first dataset comprising medical data, a training computation unit configured to apply a machine learning algorithm to the first dataset to predict an upcoming adverse patient clinical event and / or to detect an acute adverse patient clinical event, and a data output unit configured to output a second dataset comprising a classification of the predicted and / or detected adverse patient clinical event.

[0016] An idea of the present invention is to enable an automated and regular analysis of home monitoring data from IPG / ICD / ICM devices to identify signals that could indicate an upcoming or acute patient clinical event. There are thus no further manual efforts e.g. for clinical staff. Due to the complexity of the home monitoring data, a powerful analysis approach using a machine learning algorithm is provided.

[0017] Machine learning algorithms are based on using statistical techniques to train a data processing system to perform a specific task without being explicitly programmed to do so. The goal of machine learning is to construct algorithms that can learn from data and make predictions. These algorithms create mathematical models that can be used, for example, to classify data or to solve regression type problems.

[0018] According to an aspect of the invention, the first dataset comprises a patient status, recorded episodes, therapy statistics and / or a device programming of the implantable medical device. By using a wide range of parameters, an upcoming adverse patient clinical event can be more likely predicted and / or an acute adverse patient clinical event effectively detected.

[0019] According to a further aspect of the invention, the patient status comprises a mean atrial heart rate, a mean ventricular heart rate, a mean PVC / h, an atrial burden, a fast non-sustained ventricular tachycardia (VT) per day and / or a number of atrial rate episodes per day, wherein the recorded episodes and / or IPGS / SECGs (e.g. ECG pictures, machine-readable ECG data) comprise an episode type, an episode duration, an episode start timestamp and / or an episode end timestamp, wherein the therapy statistics comprise an atrial pacing percentage, a biventricular (BiV) pacing percentage, an antitachycardia pacing (ATP) in VT zones started, an ATP in VT zones successful, an ATP one shot started and / or an ATP one shot successful, and wherein the device programming of the implantable medical device comprises an AT / AF ATP type, a mode switching mode and / or an AV delay. The method of the present invention thus takes into account a wide range of values to determine a patient clinical condition or event.

[0020] According to a further aspect of the invention, for predicting an upcoming adverse patient clinical event, the first dataset comprises medical data acquired by the implantable medical device during a first predetermined time period, preferably during the past 90 days, and wherein for detecting an acute adverse patient clinical event, the first dataset comprises medical data acquired by the implantable medical device during a second predetermined time period, preferably during the past 120 days. The medical data may be analyzed 90 days before a possible event and 30 days after. By analyzing not only a single dataset acquired at a particular point in time, e.g. at present, evolving patterns and / or changes in specific parameters can be effectively tracked.

[0021] According to a further aspect of the invention, if a ratio of a number of data transmissions comprising medical data sent by the implantable medical device during the first and / or second predetermined time period to a number of expected data transmissions comprising medical data sent by the implantable medical device is equal to or greater than a predetermined threshold value, the data transmissions are stored, and wherein if said ratio is less than the predetermined threshold value, the data transmissions are deleted. This way the dataset used as a basis for analysis fulfills the defined criteria for completeness.

[0022] According to a further aspect of the invention, medical parameters comprised by the first dataset which contain no data, in particular are null, or contain no variance during the first and / or second predetermined time period, in particular are static, are deleted. Hence, irrelevant data is deleted such that the analysis is performed only on useful data. According to a further aspect of the invention, medical parameters comprised by the first dataset are pre-processed, wherein invalid data classified by predetermined escape values are set to null, wherein if several data transmissions are received in a day, respective parameter values are averaged, in particular for numerical parameters, or randomly picked, in particular for non-numeric parameters, and wherein all parameters whose completeness is less than a predetermined threshold value, in particular less than 60%, are deleted. Said preprocessing thus provides a complete dataset free of invalid data.

[0023] According to a further aspect of the invention, any remaining data gaps of the medical parameters are eliminated using linear interpolation between a previous and a next value, adoption of a previous value and / or taking an average value of a predetermined time period. By eliminating said data gaps, the analysis is more accurate since invalid and / or missing data does not influence said analysis.

[0024] According to a further aspect of the invention, the first dataset comprises first data comprising, preferably pre-processed, raw medical data acquired by the implantable medical device, and wherein the first dataset comprises second data comprising medical parameters having a variability exceeding a predetermined threshold value, wherein for every medical parameter of said second data, a plurality of features are generated using a library for automated feature generation. Said raw sensor data provides the most accurate information regarding patient clinical issues since no data processing has been performed on this data. Furthermore, said data having a high variability can potentially assist in identifying patterns and / or changes in medical parameters that could be caused by patient clinical issues.

[0025] According to a further aspect of the invention, the first dataset comprises third data comprising medical parameters having a variability that is less than a predetermined threshold value, in particular static parameters such as programming settings, wherein for every medical parameter of said third data, a plurality of features are generated using the library for automated feature generation. Said data having a low variability can potentially assist in identifying parameters that deviate from an expected value due to potential patient clinical issues. According to a further aspect of the invention, the first dataset comprises fourth data comprising medical parameters, wherein the fourth data is a composition of the first data, the second data and the third data. Said fourth data thus combines the above-mentioned advantages of each of the other datasets.

[0026] According to a further aspect of the invention, the second dataset outputted by the machine learning algorithm comprises at least one class of an adverse patient clinical event and a probability for said class. Hence, an accurate determination of a likelihood of an adverse patient clinical event can be made.

[0027] Thus, not only a probability for a particular class is outputted but also a feature importance, i.e. which param eter(s) of the input data contributed most to the prediction or detection result.

[0028] According to a further aspect of the invention, the upcoming adverse patient clinical event and / or the acute adverse patient clinical event is an atrial fibrillation, a ventricular tachycardia, a ventricular fibrillation in combination with ventricular tachycardia, a pneumonia, a fall, a therapy change and / or a cardiac failure. The machine learning algorithm is thus capable of predicting and / or detecting a wide range of patient clinical conditions.

[0029] The herein described features of the computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event are also disclosed for the system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event and vice versa.

[0030] For a more complete understanding of the present invention and advantages thereof, reference is now made to the following description taken in conjunction with the accompanying drawings. The invention is explained in more detail below using exemplary embodiments, which are specified in the schematic figures of the drawings, in which: Fig. 1 shows a flowchart of a computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event according to a preferred embodiment of the invention;

[0031] Fig. 2 shows a flowchart of a computer-implemented method for providing a machine learning algorithm for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event according to the preferred embodiment of the invention; and

[0032] Fig. 3 shows a diagram of a system for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event according to the preferred embodiment of the invention.

[0033] The computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event shown in Fig. 1 comprises providing SI a first dataset DS1 comprising medical data acquired by an implantable medical device 10.

[0034] Moreover, the method comprises applying S2 a machine learning algorithm A to the first dataset DS1 for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, and outputting S3 a second dataset DS2 comprising a classification of the predicted and / or detected adverse patient clinical event.

[0035] The first dataset DS1 comprises a patient status, recorded episodes, therapy statistics and / or a device programming of the implantable medical device 10. Furthermore, the patient status comprises a mean atrial heart rate, a mean ventricular heart rate, a mean PVC / h, an atrial burden, a fast non-sustained VT per day and / or a number of atrial rate episodes per day, wherein the recorded episodes comprise an episode type, an episode duration, an episode start timestamp and / or an episode end timestamp, wherein the therapy statistics comprise an atrial pacing percentage, a BiV pacing percentage, an ATP in VT zones started, an ATP in VT zones successful, an ATP one shot started and / or an ATP one shot successful, and wherein the device programming of the implantable medical device 10 comprises an AT / AF ATP type, a mode switching mode and / or an AV delay.

[0036] For predicting an upcoming adverse patient clinical event, the first dataset DS1 comprises medical data acquired by the implantable medical device 10 during a first predetermined time period Tl, preferably during the past 90 days, and wherein for detecting an acute adverse patient clinical event, the first dataset DS1 comprises medical data acquired by the implantable medical device 10 during a second predetermined time period T2, preferably during the past 120 days.

[0037] If a ratio of a number of data transmissions comprising medical data sent by the implantable medical device 10 during the first and / or second predetermined time period Tl, T2 to a number of expected data transmissions comprising medical data sent by the implantable medical device 10 is equal to or greater than a predetermined threshold value, the data transmissions are stored, and wherein if said ratio is less than the predetermined threshold value, the data transmissions are deleted.

[0038] Furthermore, medical parameters comprised by the first dataset DS1 which contain no data, in particular are null, or contain no variance during the first and / or second predetermined time period Tl, T2, in particular are static, are deleted.

[0039] Medical parameters comprised by the first dataset DS1 are pre-processed, wherein invalid data classified by predetermined escape values are set to null, wherein if several data transmissions are received in a day, respective parameter values are averaged, in particular for numerical parameters, or randomly picked, in particular for non-numeric parameters, and wherein all parameters whose completeness is less than a predetermined threshold value, in particular less than 60%, are deleted.

[0040] Any remaining data gaps of the medical parameters are eliminated using linear interpolation between a previous and a next value, adoption of a previous value and / or taking an average value of a predetermined time period. The first dataset DS1 comprises first data DI comprising, preferably pre-processed, raw medical data acquired by the implantable medical device 10, and wherein the first dataset DS1 comprises second data D2 comprising medical parameters having a variability exceeding a predetermined threshold value, wherein for every medical parameter of said second data D2, a plurality of features are generated using a library for automated feature generation.

[0041] Moreover, the first dataset DS1 comprises third data D3 comprising medical parameters having a variability that is less than a predetermined threshold value, in particular static parameters such as programming settings, wherein for every medical parameter of said third data D3, a plurality of features are generated using the library for automated feature generation.

[0042] In addition, the first dataset DS1 comprises fourth data D4 comprising medical parameters, wherein the fourth data D4 is a composition of the first data DI, the second data D2 and the third data D3. The second dataset DS2 outputted by the machine learning algorithm A comprises at least one class of an adverse patient clinical event and a probability for said class.

[0043] The upcoming adverse patient clinical event and / or the acute adverse patient clinical event is an atrial fibrillation, a ventricular tachycardia, a ventricular fibrillation in combination with ventricular tachycardia, a pneumonia, a fall, a therapy change and / or a cardiac failure.

[0044] Fig. 2 shows a flowchart of a computer-implemented method for providing a machine learning algorithm A for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event according to the preferred embodiment of the invention.

[0045] The method comprises providing SI’ a first training dataset TD1 comprising medical data acquired by the implantable medical device 10. In addition, the method comprises providing S2’ a second training dataset TD2 comprising a classification of a predicted and / or detected adverse patient clinical event, and training S3’ the machine learning algorithm A by an optimization algorithm which calculates an extreme value of a loss function.

[0046] Fig. 3 shows a diagram of a system 1 for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event according to the preferred embodiment of the invention. The system 1 comprises an implantable medical device 10 configured to provide a first dataset DS1 comprising medical data.

[0047] Furthermore, the system 1 comprises a training computation unit 12 configured to apply a machine learning algorithm A to the first dataset DS1 to predict an upcoming adverse patient clinical event and / or to detect an acute adverse patient clinical event, and a data output unit 14 configured to output a second dataset DS2 comprising a classification of the predicted and / or detected adverse patient clinical event.

[0048] Reference Signs

[0049] 1 system

[0050] 10 implantable medical device

[0051] 12 training computation unit

[0052] 14 data output unit

[0053] A machine learning algorithm

[0054] DI first data

[0055] D2 second data

[0056] D3 third data

[0057] D4 fourth data

[0058] DS1 first dataset

[0059] DS2 second dataset

[0060] TD1 first training dataset

[0061] TD2 second training dataset

[0062] SI -S3 method steps

[0063] S 1’ -S3 ’ training method steps

[0064] T1 first predetermined time period

[0065] T2 second predetermined time period

Claims

Claims1. Computer-implemented method for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, comprising the steps of: providing (SI) a first dataset (DS1) comprising medical data acquired by an implantable medical device (10); applying (S2) a machine learning algorithm (A) to the first dataset (DS1) for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event; and outputting (S3) a second dataset (DS2) comprising a classification of the predicted and / or detected adverse patient clinical event.

2. Computer-implemented method of claim 1, wherein the first dataset (DS1) comprises a patient status, recorded episodes, therapy statistics and / or a device programming of the implantable medical device (10).

3. Computer-implemented method of claim 2, wherein the patient status comprises a mean atrial heart rate, a mean ventricular heart rate, a mean PVC / h, an atrial burden, a fast non-sustained VT per day and / or a number of atrial rate episodes per day, wherein the recorded episodes comprise an episode type, an episode duration, an episode start timestamp and / or an episode end timestamp, wherein the therapy statistics comprise an atrial pacing percentage, a BiV pacing percentage, an ATP in VT zones started, an ATP in VT zones successful, an ATP one shot started and / or an ATP one shot successful, and wherein the device programming of the implantable medical device (10) comprises an AT / AF ATP type, a mode switching mode and / or an AV delay.

4. Computer-implemented method of claims 1 to 3, wherein for predicting an upcoming adverse patient clinical event, the first dataset (DS1) comprises medical data acquired by the implantable medical device (10) during a first predetermined time period (Tl), preferably during the past 90 days, and wherein for detecting an acute adverse patient clinical event, the first dataset (DS1) comprises medical data acquired by theimplantable medical device (10) during a second predetermined time period (T2), preferably during the past 120 days.

5. Computer-implemented method of claim 4, wherein if a ratio of a number of data transmissions comprising medical data sent by the implantable medical device (10) during the first and / or second predetermined time period (Tl, T2) to a number of expected data transmissions comprising medical data sent by the implantable medical device (10) is equal to or greater than a predetermined threshold value, the data transmissions are stored, and wherein if said ratio is less than the predetermined threshold value, the data transmissions are deleted.

6. Computer-implemented method of claim 4 or 5, wherein medical parameters comprised by the first dataset (DS1) which contain no data, in particular are null, or contain no variance during the first and / or second predetermined time period (Tl, T2), in particular are static, are deleted.

7. Computer-implemented method of claim 4 or 5, wherein medical parameters comprised by the first dataset (DS1) are pre-processed, wherein invalid data classified by predetermined escape values are set to null, wherein if several data transmissions are received in a day, respective parameter values are averaged, in particular for numerical parameters, or randomly picked, in particular for non-numeric parameters, and wherein all parameters whose completeness is less than a predetermined threshold value, in particular less than 60%, are deleted.

8. Computer-implemented method of claim 7, wherein any remaining data gaps of the medical parameters are eliminated using linear interpolation between a previous and a next value, adoption of a previous value and / or taking an average value of a predetermined time period.

9. Computer-implemented method of any one of the preceding claims, wherein the first dataset (DS1) comprises first data (DI) comprising, preferably pre-processed, raw medical data acquired by the implantable medical device (10), and wherein the firstdataset (DS1) comprises second data (D2) comprising medical parameters having a variability exceeding a predetermined threshold value, wherein for every medical parameter of said second data (D2), a plurality of features are generated using a library for automated feature generation.

10. Computer-implemented method of claim 9, wherein the first dataset (DS1) comprises third data (D3) comprising medical parameters having a variability that is less than a predetermined threshold value, in particular static parameters such as programming settings, wherein for every medical parameter of said third data (D3), a plurality of features are generated using the library for automated feature generation.

11. Computer-implemented method of claim 9 or 10, wherein the first dataset (DS1) comprises fourth data (D4) comprising medical parameters, wherein the fourth data (D4) is a composition of the first data (DI), the second data (D2) and the third data (D3).

12. Computer-implemented method of any one of the preceding claims, wherein the second dataset (DS2) outputted by the machine learning algorithm (A) comprises at least one class of an adverse patient clinical event and a probability for said class.

13. Computer-implemented method of any one of the preceding claims, wherein the upcoming adverse patient clinical event and / or the acute adverse patient clinical event is an atrial fibrillation, a ventricular tachycardia, a ventricular fibrillation in combination with ventricular tachycardia, a pneumonia, a fall, a therapy change and / or a cardiac failure.

14. Computer-implemented method for providing a machine learning algorithm (A) for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, comprising the steps of: providing (ST) a first training dataset (TD1) comprising medical data acquired by the implantable medical device (10);providing (S2’) a second training dataset (TD2) comprising a classification of a predicted and / or detected adverse patient clinical event; and training (S3’) the machine learning algorithm (A) by an optimization algorithm which calculates an extreme value of a loss function.

15. System (1) for predicting an upcoming adverse patient clinical event and / or for detecting an acute adverse patient clinical event, comprising: an implantable medical device (10) configured to provide a first dataset (DS1) comprising medical data; a training computation unit (12) configured to apply a machine learning algorithm (A) to the first dataset (DS1) to predict an upcoming adverse patient clinical event and / or to detect an acute adverse patient clinical event; and a data output unit (14) configured to output a second dataset (DS2) comprising a classification of the predicted and / or detected adverse patient clinical event.

Images