Apparatus, systems and methods for predicting, screening and monitoring of mortality and other conditions

Abstract

To provide an apparatus, systems and methods that relate to predicting, screening, and monitoring for mortality and other negative patient outcomes.SOLUTION: Systems and methods may include receiving one or more signals from one or more sensing devices; processing the one or more signals to extract one or more features from the one or more signals; analyzing the one or more features to determine one or more values for each of the one or more features; comparing at least one of the one or more values or a measure based on at least one of the one or more values to a threshold; determining a presence, absence, or likelihood of the subsequent mortality, falls or extended hospital stays for a patient based on the comparison; and outputting an indication of the presence, absence, or likelihood of the subsequent development of poor outcomes or death for the patient.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] cross reference

[0001] This application is a joint application of "Apparatus, Systems" filed on April 4, 2019. This application claims priority to U.S. Provisional Patent Application No. 62 / 829,411, entitled "Devices, Systems and Methods for Predicting, Screening and Monitoring of Mortality and Other Conditions," which is incorporated herein by reference in its entirety.

[0002] government support This invention is supported by the United States Government under No. 1,664,364 awarded by the National Science Foundation. The United States Government has certain rights in this invention.

[0003]

[0003] In this specification, there is provided a method for the treatment of rhodium and rhodium-containing ... Various devices, systems and methods are described. [Background technology]

[0004]

[0004] Delirium is characterized by inattention, cognitive impairment, psychomotor disturbances, and a waxing and waning course. Delirium is an acute confusional state that can be triggered by a variety of factors. Delirium is particularly common among older hospitalized adults and affects a significant number of patients on general medical floors, post-operative care units, including those receiving electroconvulsive therapy, and intensive care units.

[0005] Delirium in hospitalized elderly patients is common, dangerous, and costly. It is also significantly undertreated. It is often diagnosed and therefore undertreated. It is estimated that there are at least 2 to 3 million cases of delirium per year in the United States alone. Delirium is a strong predictor of poor patient outcomes. Delirium increases mortality, complications, length of hospital stay, and post-discharge institutionalization. These patients, even if they survive, are at high risk for long-term cognitive impairment. If undetected, delirium can increase healthcare costs by thousands of dollars per patient per year, thereby generating billions of dollars in additional healthcare costs.

[0006]

[0006] Delirium is common and dangerous, yet it is underdetected and undertreated. Screening questionnaires are subjective and poorly implemented within busy hospital workflows. Electroencephalography (EEG) can objectively detect the diffuse slow-wave pattern characteristic of delirium, but the size, cost, and expertise required for lead placement and interpretation make it unsuitable for high-throughput screening.

[0007]

[0007] Because dementia is one of the risk factors for delirium, the relationship between delirium and dementia is unclear. Furthermore, delirium is known to accelerate the progression of dementia. In addition, delirium and dementia are related to patient outcomes, including death. If a patient has both delirium and dementia, the patient's mortality rate increases.

[0008]

[0008] The art provides efficient and reliable methods for predicting and screening for mortality. Therefore, there is a need for a better device, system and method. Summary of the Invention [Means for solving the problem]

[0009]

[0009] The present invention provides a method for detecting, identifying, or otherwise treating death and / or other conditions in a patient. Various related devices, systems, and methods for predicting otherwise are described. In various implementations, a device is used to detect diffuse slowing, a characteristic of the above condition.

[0010]

[0010] Embodiments of the present disclosure are directed to predicting, screening for, and monitoring mortality or other conditions. The present disclosure relates to systems and methods for monitoring, and more particularly to systems and methods for determining the presence or likelihood of death or other subsequent conditions in a patient through signal analysis. The output data includes providing indicators of risk for poor outcomes, including death, prolonged hospitalization, institutionalization after discharge, and likelihood of falls in the hospital. In various implementations, the output is a continuous score, with higher scores indicating a higher likelihood that the patient will have a poor outcome. In further implementations, the systems, methods, and devices of the present disclosure include implementing an intervention or treatment to prevent an undesirable outcome.

[0011]

[0011] Prognosis, including mortality, prolonged hospitalization, institutionalization after discharge, and likelihood of falls in the hospital Systems and methods are described that use various means and procedures for predicting, screening, and monitoring maladies. In certain embodiments, the means and procedures disclosed herein may be used with one or more additional means and / or procedures for predicting, screening, and monitoring mortality. The examples described herein relate to prediction, screening, and monitoring for illustrative purposes only. In the case of a multi-step process or method, the steps may be performed by one or more different parties, servers, processors, etc.

[0012]

[0012] A system consisting of one or more computers is a system that, when in operation, A computer program may be configured to perform a particular operation or action by having software, hardware, or a combination thereof installed on a system that causes the system to perform the operation. One or more computer programs may be configured to perform a particular operation or action by containing instructions that, when executed by a data processing device, cause the device to perform the action.

[0013] In Example 1, a method for patient screening for prognostic risk is provided. The device records raw BSEEG values ​​and calculates NBSEEG. This involves normalizing the raw BSEEG values ​​for the purpose of predicting the outcome and outputting a prognostic BSEEG score.

[0014] Example 2 relates to the method of Example 1, wherein the NBSEEG is a BSEEG of raw BSEEG. It is calculated by comparing the BSEEG population mean and dividing the population result by the BSEEG population standard deviation.

[0015] Example 3 relates to the method of Example 1, and the prognostic NBSEEG score is NBSEEG positive. score or NBSEEG negative score.

[0016] Example 4 relates to the method of Example 1, wherein the prognostic NBSEEG score is continuous.

[0016]

[0017] Example 5 relates to the method of example 1, where the recording is performed at the primary point of care.

[0018] Example 6 relates to the method of Example 1, and the prognosis NBSEEG is the length of hospital stay (LOS "), discharge disposition, and / or mortality risk correlates with at least one

[0017]

[0019] In Example 7, a handheld system for screening patients for mortality risk is provided. The system includes at least two sensors configured to record one or more brain frequencies, a processor, and at least one module configured to record raw BSEEG values, normalize the raw BSEEG values ​​to calculate a NBSEEG, and output a prognostic NBSEEG score.

[0018]

[0020] Example 8 relates to the system of Example 7, and the prognosis NBSEEG is based on the hospital LOS, discharge correlates with at least one of the following: time trend, risk of death, and / or mortality risk.

[0021] Example 9 relates to the system of example 7, further comprising outputting threshold data. .

[0019]

[0022] Example 10 relates to the system of Example 7, and compares the prognostic NBSEEG score with a threshold. The method further includes:

[0023] Example 11 relates to the system of example 7, further comprising a signal processing device.

[0020]

[0024] In Example 12, a method for screening a subject for risk of death involves collecting unreported data from the subject. It involves recording processed BSEEG values ​​with a handheld device, normalizing the raw BSEEG values ​​to calculate NBSEEG, and outputting a prognostic NBSEEG score.

[0021]

[0025] Example 13 relates to the method of Example 12, comparing the prognostic NBSEEG score to a threshold value. It further includes:

[0026] Example 14 relates to the method of example 12, wherein the raw BSEEG values ​​are obtained from a handheld device. The signal is processed by a signal processing module or feature analysis module within the device.

[0022]

[0027] Example 15 relates to the method of Example 12, wherein the prognostic NBSEEG score is one or Patients are classified as low, intermediate, or high risk by comparison with multiple thresholds.

[0028] Example 16 relates to the method of example 12, further comprising maintaining the BSEEG population norm. Further includes:

[0023]

[0029] Example 17 relates to the method of Example 16, wherein the NBSEEG is a raw BSEEG. It is calculated by comparing to the mean of the SEEG population norm and dividing the population result by the BSEEG population standard deviation.

[0024]

[0030] Example 18 relates to the method of Example 17, further comprising recording subject outcome. .

[0031] Example 19 relates to the method of Example 18, wherein the BSEEG population norm is It is updated to include EEG values ​​and subject outcomes.

[0025]

[0032] Example 20 relates to the method of Example 19, and is a method for determining whether the prognosis NBSEEG is related to the length of hospital stay (" correlate with at least one of the following: LOS) and / or discharge disposition.

[0033] In certain implementations, embodiments of the present disclosure include a ratio of high frequency components to low frequency components. In a particular implementation, the EEG signal is recorded by a point-of-care portable EEG device with a limited number of electrodes. In a particular implementation, the raw EEG signal is processed by spectral density analysis and then by an algorithm to combine high-frequency and low-frequency power, such as a ratio of two or more powers, to generate a raw BSEEG value.

[0026]

[0034] Various embodiments may use the mean of the raw BSEEG scores and the BSEEG score population norm. Assigning a prognostic NBSEEG score ("NBSEEG score") to the normalized BSEEG by dividing the difference from the mean by the standard deviation of the BSEEG population norm. In implementations, raw BSEEG values ​​are evaluated relative to a population BSEEG score distribution in terms of mean and standard deviation. The population mean can be defined by a specific patient group or a healthy population group. In certain implementations, the NBSEEG score is calculated as (raw BSEEG value - population norm BSEEG mean) divided by the standard deviation of the BSEEG from the population norm.

[0027]

[0035] Various embodiments may use the resulting prognostic NBSEEG score to determine two outcomes, such as mortality prognosis. , and outputting it as one of three or more different levels. The NBSEEG score can be used as a continuous value as a new vital sign, just like body temperature, blood pressure, and heart rate. The risk threshold for death is thresholded by the current risk score defined by epidemiological studies. Data presented in this disclosure showed that a high NBSEEG score may lead to higher mortality rates, while a low NBSEEG score may be associated with less risk. When the scores were divided into three groups, the scores showed a dose-dependent relationship with the risk of death.

[0028]

[0036] One general aspect is a device comprising a housing and a device for recording one or more brain signals and includes a system for patient screening, including a handheld screening device including at least two sensors configured to generate a plurality of values, a processor, and at least one module configured to perform a spectral density analysis of one or more values ​​and output data providing an indication of the presence or likelihood of subsequent mortality. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the above-described methods.

[0029]

[0037] Implementations may include one or more of the following features: a system in which the module is configured to compare one or more values ​​from one or more brain signals with a threshold; a system in which the threshold is a ratio including the number of occurrences of high frequencies to the number of occurrences of low frequencies; a system in which the one or more brain signals are electroencephalogram (EEG) signals; a system in which there are two sensors; a system in which the housing includes a display; a system in which the processor is disposed within the housing; a system in which the one or more values ​​are selected from a group including high frequencies, low frequencies, and combinations thereof; a system in which the one or more values ​​are a numerical representation of the number of occurrences of each of one or more features over a period of time; a system in which the threshold is predetermined; a system in which the threshold is set based on a machine learning model; a system further including a handheld housing including a display, at least two sensors in electrical communication with the housing, a processor disposed within the housing, the display configured to display output data; a system further including a validation module configured to evaluate the brain signals, the processor converting one or more brain frequencies into signal data, and the validation module discarding signal data that exceeds at least one predetermined signal quality threshold; a system in which the signal data is partitioned into windows of equal duration; a device further including a signal processing module; a device further including a validation module. A device further comprising a threshold module.Implementations of the described technologies may include hardware, a method or process, or computer software on a computer-accessible medium.

[0030]

[0038] One general aspect is a system for assessing the presence of mortality risk, comprising one or Other embodiments of this aspect include a system including at least two sensors configured to record a plurality of brain frequencies, a processor, and at least one module configured to compare the EEG frequencies over time, perform a spectral density analysis of the EEG frequencies to establish a ratio, compare the ratio to a set threshold, and output data providing an indication of the presence or likelihood of subsequent death. Other embodiments of this aspect include a system including at least two sensors configured to record a plurality of brain frequencies, a processor, and at least one module configured to compare the EEG frequencies over time, perform a spectral density analysis of the EEG frequencies to establish a ratio, compare the ratio to a set threshold, and output data providing an indication of the presence or likelihood of subsequent death. It also includes corresponding computer systems, apparatus, and computer programs stored on one or more computer storage devices.

[0031]

[0039] Implementations may include one or more of the following features: a threshold value a system in which the threshold is predetermined; a system in which the threshold is set based on a machine learning model; a system further including a handheld housing including a display, at least two sensors in electrical communication with the housing, a processor disposed within the housing, and the display configured to display output data; a system further including a validation module configured to evaluate a signal brain, the processor converting one or more brain frequencies into signal data, and the validation module discarding signal data that exceeds at least one predetermined signal quality threshold; a system in which the signal data is partitioned into windows of equal duration; a device further including a signal processing module; a device further including a validation module; a device further including a threshold module. Implementations of the described technology may include hardware, methods or processes, or computer software on a computer-accessible medium.

[0032]

[0040] One common aspect is the use of handheld devices to assess whether or not subsequent death of the patient has occurred. and a handheld device including a housing, at least one sensor configured to generate at least one electroencephalogram signal, at least one processor, at least one system memory, at least one program module configured to perform a spectral density analysis of the at least one electroencephalogram signal and generate patient output data, and a display configured to display the patient output data. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the above-described methods.

[0033]

[0041] Implementations may include one or more of the following features. A device further comprising a processing module. A device further comprising a verification module. A device further comprising a threshold module. A device further comprising a feature analysis module. A device further comprising a signal processing module. Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.

[0034]

[0042] One or more computing devices may be represented in computer readable form. The required functionality may be provided by accessing software instructions stored in a computer. When software is used, any suitable programming, scripting, or other type of language or combination of languages ​​may be used to implement the techniques described herein. However, software need not be used exclusively, or even at all. For example, some embodiments of the methods and systems described herein may be implemented by hardwired logic or other circuitry, including, but not limited to, application-specific circuitry. A combination of computer-executed software and hardwired logic or other circuitry may also be suitable.

[0035]

[0043] Although multiple embodiments are disclosed, those skilled in the art will recognize that the disclosed apparatus, systems, and methods Still other embodiments of the present disclosure will become apparent from the following detailed description, which shows and describes illustrative embodiments of the present disclosure. It should be understood that the apparatus, systems, and methods of the present disclosure are capable of modification in various obvious aspects, all without departing from the spirit and scope of the present disclosure. Accordingly, the drawings and detailed description are to be regarded as illustrative in nature and not restrictive.

[0036]

[0044] It is included to provide a better understanding of the invention and The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate preferred embodiments of the invention and, together with the detailed description, serve to explain the principles of the invention. [Brief explanation of the drawings]

[0037] [Figure 1A]

[0045] FIG. 1 shows an overview of an exemplary screening system in use on a patient. [Figure 2]

[0046] FIG. 2A is a diagram showing an example of diffuse slowing in electroencephalograms.

[0047] FIG. 2B shows an example of a normal state of electroencephalogram brain waves. [Figure 3]

[0048] FIG. 3A is a diagram showing an example of widespread slowing of electroencephalograms.

[0049] FIG. 3B shows an example of two-channel widespread slowing of the electroencephalogram.

[0050] FIG. 3C is a diagram showing an example of two-channel normal state electroencephalograms appearing on an electroencephalogram. [Figure 4A]

[0051] FIG. 1 illustrates an exemplary system for mortality risk prediction, screening, and monitoring. [Figure 4B]

[0052] FIG. 1 illustrates an exemplary system for computational aspects of mortality risk prediction, screening and monitoring. [Figure 5A]

[0053] FIG. 1 illustrates an exemplary system for predicting, screening, and monitoring high-risk brain signals for poor prognosis, including death. [Figure 5B]

[0054] FIG. 1 illustrates an exemplary system for predicting, screening, and monitoring high-risk brain signals for poor prognosis, including death. [Figure 5C]

[0055] FIG. 1 illustrates an exemplary system for predicting, screening, and monitoring high-risk brain signals for poor prognosis, including death. [Figure 5D]

[0056] FIG. 1 illustrates another exemplary system for predicting, screening, and monitoring high risk brain signals for poor prognosis, including death. [Figure 6]

[0057] FIG. 1 illustrates program modules and program data of a screening device, according to one exemplary embodiment. [Figure 7]

[0058] FIG. 1 illustrates an overview of a method for predicting, screening, and monitoring high risk brain signals for poor prognosis, including death, according to an exemplary embodiment. [Figure 8-1]

[0059] FIG. 1 illustrates an overview of a method for predicting, screening, and monitoring high risk brain signals for poor prognosis, including death, according to an exemplary embodiment. [Figure 8-2]

[0059] Figure 1 shows an overview of a method for predicting, screening and monitoring high risk brain signals for poor prognosis, including death, according to an exemplary embodiment. [Figure 9]

[0060] FIG. 9A shows an example of a raw electroencephalogram signal.

[0061] FIG. 9B shows an example of a raw electroencephalogram signal.

[0062] FIG. 9C is a diagram showing an example of a spectral density analysis.

[0063] FIG. 9D is a diagram showing an example of a spectral density analysis. [Figure 10]

[0064] FIG. 10A is a diagram illustrating an exemplary screening system in use on a patient.

[0065] FIG. 10B illustrates a model flow chart for predicting a prognostic BSEEG score, according to one embodiment. [Figure 11]

[0066] Figure 1 shows the distribution of normalized BSEEG (NBSEEG) scores. Scores are based on a total of 2938 recordings from 428 patients across all age groups. The NBSEEG score is defined as the number of standard deviations from the mean, where the mean is a score of 0. [Figure 12]

[0067] Figure 1 shows the patient enrollment flow chart. The target population was classified in two ways: clinical delirium status and BSEEG score. [Figure 13]

[0068] Figure 360-day survival curves based on clinical delirium status (left panel) and NBSEEG score (right panel). Patients with clinical delirium had an increased mortality rate compared with patients without clinical delirium (P = 0.0038). Regardless of clinical delirium status, NBSEEG-positive patients had a higher mortality rate than NBSEEG-negative patients (P = 0.0032). [Figure 14]

[0069] Figure 1 shows 360-day survival curves based on the three NBSEEG categories. Mortality is directly proportional to the NBSEEG score, with higher NBSEEG score groups being associated with higher mortality (P=0.005). [Figure 15]

[0070] Subgroup analysis of mortality based on both clinical delirium status and NBSEEG category. Patients who were both clinically delirious and NBSEEG-positive had the highest mortality rate (purple line). Patients with clinical delirium but negative NBSEEG had a lower mortality rate (blue line), nearly as low as patients who were both clinically non-delirious and NBSEEG-negative (orange line). In contrast, patients with clinically non-delirium but positive NBSEEG had a higher mortality rate (green line), indicating that NBSEEG score is a better predictor of mortality than clinical delirium status. [Figure 16]

[0071] This figure shows a predictive model for EEG feature-based diagnosis using random forests (RF). The predictive power of individual genera (EEG features) for clinical conditions was evaluated using the Boruta feature selection algorithm. The three horizontal lines of the box represent the first, second (median), and third quartiles, respectively, with whiskers extending to the interquartile range (IQR) of 1.5. The blue box plot represents shuffled actual genera introduced into the RF classifier, corresponding to the minimum, mean, and maximum Z-scores of the shadow genera, which serve as a benchmark for detecting true predicted genera. Red, yellow, and green represent rejected, suggested, and confirmed genera by Boruta selection. Here, importance is defined as the average decrease in classification accuracy. [Figure 17]

[0072] Figure 17A shows 180-day survival curves based on NBSEEG categories for 2 NBSEEG categories in 228 subjects (replication cohort). Figure 17B shows 180-day survival curves based on NBSEEG categories for 3 NBSEEG categories in 228 subjects (replication cohort). Figure 17C shows 180-day survival curves based on NBSEEG categories for 2 NBSEEG categories in 502 subjects (discovery cohort and replication cohort). Figure 17D shows 180-day survival curves based on NBSEEG categories for 3 NBSEEG categories in 502 subjects (discovery cohort and replication cohort). Abbreviations: NBSEEG: bispectral electroencephalography, B(-): NBSEEG negative, B(+): NBSEEG positive. [Figure 18]

[0073] Figure 18. 180-day survival curves based on dementia and NBSEEG category. Abbreviations: NBSEEG: normalized bispectral electroencephalography, B(-): NBSEEG negative, B(+): NBSEEG positive, D(-): dementia negative, D(+): dementia positive. [Figure 19]

[0074] Figure 19A shows 30-day short-term mortality rates based on NBSEEG category in 502 subjects (discovery and replication cohorts). Figure 19B shows 60-day short-term mortality rates based on NBSEEG category in 502 subjects (discovery and replication cohorts). Figure 19C shows 90-day short-term mortality rates based on NBSEEG category in 502 subjects (discovery and replication cohorts). Note: *Relative risk was significantly higher than the NBSEEG-negative group. Abbreviations: NBSEEG: normalized bispectral electroencephalography. [Figure 20]

[0075] Figure 1 shows short-term mortality according to dementia and NBSEEG category in 502 subjects (discovery and replication cohorts). *Relative risk was significantly higher than in the NBSEEG-negative group. Abbreviations: NBSEEG: normalized bispectral electroencephalography. DETAILED DESCRIPTION OF THE INVENTION

[0038]

[0076] Various embodiments disclosed or contemplated herein may be used to assess the risk of death using objective clinical measures. The present invention relates to systems, methods, and devices that can provide a constant value for detecting the presence of diffuse slowing in a patient's electroencephalogram. Implementations described herein involve performing a spectral density analysis of electroencephalograms recorded from a small number of discrete locations on the patient's head. Diffuse slowing can be detected by electroencephalography (EEG), thereby enabling easier bedside diagnosis, such as with a handheld device. That is, various implementations can record EEG via two or more leads placed on the patient's head, evaluate the ratio of recorded low-frequency to high-frequency waves, and run an algorithm to compare that ratio to a determined threshold to identify risk of death. In further embodiments, these implementations use machine learning and additional data, such as data from medical records, to improve diagnostic accuracy.

[0039]

[0077] The presently disclosed normalized bispectral electroencephalography ("NBSEEG") method, system and The device can predict patient outcomes, including hospital length of stay, discharge disposition, and mortality from NBSEEG scores obtained on the patient's first day of hospitalization. Brain signals are acquired from the patient's frontal region, and a novel algorithm is used to calculate raw BSEEG data, which is then compared to mass data from approximately 3,000 raw BSEEG recordings from the patient to determine a normalized BSEEG (NBSEEG) score. Higher scores are associated with longer hospital stays, a higher likelihood of discharge away from home, and higher mortality. The described implementations can be used to screen large numbers of patients and provide an objective score for predicting patient outcomes, thereby enabling earlier intervention to improve patient outcomes.

[0040]

[0078] The systems, devices, and methods of the present disclosure are different from those found in the prior art. This disclosure relates to noninvasive point-of-care diagnostics using fewer leads than 24-lead EEG. For example, as generally shown in FIG. 1 , certain implementations employ a two-lead BSEEG screening system 1 that can be performed using a handheld screening device 10 by applying two leads 12A, 12B to the forehead of a patient 30 for less than 10 minutes. While these implementations use two leads or channels in BSEEG for illustrative purposes, it should be understood that multiple leads or channels are contemplated herein. In various implementations, the device 10 can display graphical and / or numerical representations 7 of useful information, such as recent measurements 4, trends 6, and signal quality 8, for use in predicting mortality, brain dysfunction, and / or prolonged hospitalization and the associated risks. It should be understood that these representations 7 can be the result of well-known graphical user interface techniques on a display 16.

[0041]

[0079] Brain waves can have various frequencies and / or frequency bands. "Diffuse slowing" is It is a strong predictor of death. Figures 2A to 3D show several EEG readings from patients experiencing symptoms of death2 compared to normal controls3. As will be apparent to those skilled in the art, in various circumstances, the EEG in the death state is characterized by "pervasive slowing"2, meaning that slowed waveforms can be observed in each of the observation channels. As can be seen from Figures 2A and 3A, this slowing is pervasive rather than localized, so that the slowing (shown in Figures 2A and 3A) is observed in most, and typically all, of the various electrodes of the EEG.

[0042]

[0080] As shown in Figure 3A, the appearance of wave 2, which is slow compared to the number of higher frequency waves, Patient death or other negative medical outcome as described herein This may be a sign that you are facing or are likely to face a

[0043]

[0081] As shown in Figure 3B, diffuse slowing usually occurs in all or almost all E Because they are observed with EEG electrodes, fewer EEG channels than the standard 16 to 24 can be used to identify diffuse slowing and therefore predict risk of mortality. In these implementations, the two leads used in the BSEEG implementations described herein may be sufficient to detect diffuse slowing, and require appropriate signal processing and user input. This interface allows for easy handheld screening without requiring specialized expertise for placement or interpretation. To quantify risk of death, raw BSEEG values ​​are compared to a set population norm distribution to accurately calculate a normalized BSEEG (NBSEEG) score, which can be expressed as NBSEEG positive (BSEEG(+)) or NBSEEG negative (BSEEG(-)).

[0044]

[0082] One implementation of the screening device 10 is shown in FIG. Therefore, the systems and methods for mortality prediction, screening, and monitoring disclosed herein may use such handheld or other portable screening devices 10. In these implementations, the screening device 10 is configured to receive signals from, for example, one or more sensors 12A, 12B. Because diffuse slowing is readily identifiable throughout a patient's brain, these devices 10 may use far fewer than 20 sensors 12, such as two, three, four, five, or more. In certain implementations, between six and 20 or more sensors are used. Thus, because fewer than 20 sensors 12 are used, the devices and systems of the present disclosure are easily portable and can be used on a patient 30 without the need for a typical prior art EEG cap.

[0045]

[0083] In the implementation of FIG. 4A, one or more sensors 12A, 12B are mounted on the patient. The signal may be a brain sensor, such as, but not limited to, an electrode placed on the patient's brain. In certain embodiments, the signal may be an electroencephalography ("EEG") signal from one or more electrodes measuring the patient's brain activity. The signal may be processed to extract one or more features of the signal. The one or more features may be analyzed to determine one or more values ​​for each of the one or more features. These values, or a measure based on one or more of the values, may be compared to a threshold to determine the presence or likelihood of subsequent death if the patient is not currently exhibiting clinical signs or symptoms of disease.

[0046]

[0084] With continued reference to FIGS. 4A and 4B, the screening device 10 includes a housing. The housing 14 may include a display 16 and an interface 18, such as buttons or a touchscreen, as will be appreciated by those skilled in the art. In various implementations, the sensors 12A, 12B connect to the device 10 via ports 22A, 22B and wires 24A, 24B, while other implementations use a wireless interface such as Bluetooth®. As shown in FIG. 4A , a ground lead 13 may also be included, which may be bundled with one of the wires 24A, 24B for simplified application. In various implementations, a transmission cord 26 or other connection may be used to place the device 10 in electrical communication with a server or other computing device, as described with respect to FIGS. 5A and 5B .

[0047]

[0085] As shown in FIG. 4B, in various implementations, the display 16 may include one or more The display 16 may display a graphical and / or numerical representation 7 of information useful for use in diagnosing death and / or delirium, including at least one of a plurality of EEG signals 2A, 2B, a most recent measurement 4, a trend 6, and signal quality 8. In an exemplary implementation, a graphical representation of a threshold step, described below, comparing spectral density to a set threshold, is shown as the most recent measurement 4. It should be understood that these representations 7 may include any program data 67 and may be presented to a healthcare provider as a result of well-known graphical user interface techniques on the display 16. In these embodiments, the EEG signals 2A, 2B are derived from a sensor 12 placed on the patient and may be used to diagnose and / or predict signs of death in a clinical or other setting. In various implementations, the housing The device housing 14 also includes a power source and computing components such as a microprocessor, memory, etc. In further implementations, calculations and other processing may occur within the device housing 14, such as by program modules (described in connection with FIGS. 6 and 7), while additional processing may occur elsewhere as described herein. In each of these implementations, the screening device 10 and / or system 1 includes a processor configured to assess the patient's diffuse slowing by spectral density analysis performed on fewer than 20 channels to identify diffuse slowing.

[0048]

[0086] FIG. 5A shows a diagram of mortality prediction, screening, and monitoring according to one implementation. 4A illustrates an exemplary system 1 for screening a patient 30. In this implementation, the system 1 may include one or more screening devices 10 (e.g., screening device 1, screening device 2, ..., screening device n), which in a particular implementation is device 10 of FIG. 1. The screening device 10 in this implementation is operably coupled to one or more sensors 12-1 to 12-n. The one or more sensors 12 may be individual sensors, an array of sensors, a medical device, or other computing devices, remote access devices, etc. In particular embodiments, the one or more sensors 12 may be one or more electrodes and / or other brain function monitoring devices. The one or more sensors 12 may be directly or indirectly coupled to a patient 30 to monitor the patient's biosignals. In particular embodiments, a particular sensor 12 of the one or more sensors 12 may be directly coupled to and / or integrated with the screening device 10 via a port 22 (best shown in FIG. 4A ).

[0049]

[0087] As shown in FIG. 5A, the signals received from the sensors 12-1, 12-2, 12-3, and 12-n are Various processing and analysis of the detected signals can be performed by the screening device 10 as described in connection with Figures 6 and 8. In certain other implementations, such as those shown in Figures 5B-5D, the screening device 10 can be used in conjunction with other computing devices.

[0050]

[0088] As shown in the implementation of FIG. 5B, the screening device is a bedside device. The bedside device 11 may be connected to or otherwise interfaced with a patient monitor or other integrated bedside monitoring device 11. In various implementations, the bedside device 11 may be used by a healthcare provider to observe and / or perform certain analysis or observation steps.

[0051]

[0089] As shown in FIGS. 5C and 5D, one or more screening devices 1 System 1 may be operatively connected directly and / or indirectly, such as via a network, to one or more server / computing devices 42, system databases 36 (e.g., Database 1, Database 2, ..., Database n). Various devices may be connected to the system, including, but not limited to, medical devices, medical monitoring systems, client computing devices, consumer computing devices, healthcare provider computing devices, remote access devices, etc. System 1 may receive one or more inputs 38 and / or one or more outputs 40 from various sensors, medical devices, computing devices, servers, databases, etc.

[0052]

[0090] As shown in FIG. 5D, in certain embodiments, one or more screening devices The device 10 may be directly coupled to and / or integrated with one or more server / computing devices 42 via connection 32 and / or may be connected to one or more servers / computing devices 42. may be coupled to one or more server / computing devices 42 via multiple network interface connections 32. The screening device 10 and / or one or more server / computing devices 42 may be operatively connected directly and / or indirectly, such as via a network, to one or more third-party servers / databases 34 (e.g., Database 1, Database 2, ..., Database n). The one or more server / computing devices 42 may correspond to any one or more of the following: a server, a personal computer (PC), a laptop, a smartphone, a tablet, etc.

[0053]

[0091] In various implementations, the connection 32 may be, for example, a hardwired connection, a wireless connection, an internet connection, or the like. The network may correspond to any combination of local area networks such as the Internet, an intranet, a wide area network, a cellular network, and / or a Wi-Fi network, etc. One or more sensors 12 may themselves include at least one processor and / or memory and may correspond to any set of sensors, medical devices, or other computing devices running applications, each transmitting data input to and / or receiving data output from one or more screening devices 10 or server / computing devices 42. Such server / computing devices 42 may include, for example, one or more desktop computers, laptops, mobile computing devices (e.g., tablets, smartphones, wearable devices), server computers, etc. In certain implementations, input data may include analog and / or digital signals, such as from an EEG system, for processing by one or more server / computing devices 10, such as other electroencephalogram measurements.

[0054]

[0092] In various implementations, the data output 40 may include, for example, medical instructions, recommendations, notifications, etc. The data may include information, alerts, data, and / or images. Some embodiments of the present disclosure may also be used for collaborative projects with multiple users logging in from various locations to perform various operations on the data project. Certain embodiments may be computer-based, web-based, smartphone-based, tablet-based, and / or human-wearable device-based.

[0055]

[0093] In another exemplary implementation, the screening device 10 (and / or server 6, the server / computing device 42 may include at least one processor 44 coupled to a system memory 46. The system memory 46 may include computer program modules 48 and program data 50. As noted above, operations associated with each computer program instruction in the program modules 48 may be distributed across multiple computing devices.

[0056]

[0094] Leads spectral density analysis from various electronic and computing mechanisms 6, program modules 48 are provided. These program modules may include a signal processing module 52, a feature analysis module 54, a validation module 55, a diffuse slowing or death threshold module 56, an output module 59, and / or other program modules 58, such as an operating system, device drivers, etc. In various implementations, each program module 52-58 may include a respective set of computer program instructions executable by the processor 44.

[0057]

[0095] FIG. 6 shows an example of a set of program modules. Other numbers and configurations of modules are contemplated depending on any particular design and / or architecture of server / computing device 10 and / or system 1. In one exemplary implementation, program data 50 may include signal data 60, feature data 62, verification data 63, a diffuse slowing or death determination module 64, output data 67, and other program data 66, such as data inputs, third party data, and / or other data configured to perform the various steps described in FIGS.

[0058]

[0096] In various implementations, the program data 50 may include the various program modules described above. 12 , or any other computing, storage, or communication device in electrical or physical communication with the output module 67. It should be understood that the output data 67 provides characteristics of the patient's electroencephalogram to a point-of-care provider for diagnosis or otherwise identification of signs of delirium or death. For example, certain implementations may provide the provider with the most recent measurement readings of spectral density and qualitative measurements of signal quality and trends, as described elsewhere herein.

[0059]

[0097] Figure 7 shows the mortality prediction, screening, and monitoring and the NBSEEG prediction. 1 shows an overview of an exemplary method 5 according to one embodiment for establishing a prognostic NBSEEG score. To perform mortality prediction, screening and monitoring, the system and method may perform several optional steps.

[0060]

[0098] One optional step is recording step 70. In this step, one Or input data, such as a plurality of raw BSEEG value input signals, may be received and / or recorded by one or all of the sensors 12, device 10, processor 40 and / or system memory 46, for example, as shown in Figures 4A, 4B, 5A-5D, and 6.

[0061]

[0099] Continuing with reference to FIG. 7, in optional process step 71, one or more signals The signals may be processed, such as by program data 50 and / or signal processing module 52 shown in FIG. 6. These signals may also be processed to partition the signals into windows, as also shown in FIG. 8. The signals may also be processed to extract one or more values ​​from one or more signals and / or windows, as described in more detail in connection with FIG. 8A. These values ​​may include characteristics of raw BSEEG signal values ​​from the EEG in certain implementations. In various implementations, system 1 may trend these values ​​to improve the accuracy of the system, as described in more detail in connection with FIGS. 11A and 11B.

[0062]

[0100] 7, an optional analysis step 72 may be performed in which one or more values ​​may be analyzed to determine particular characteristics of the signal or window. In certain implementations, this analysis step may include performing a Fast Fourier Transform 100 (as shown at 64 in FIG. 6) to create or otherwise compare characteristic data. .

[0063]

[0101] In certain implementations, the signal processing module 52 and / or feature analysis module 54 are configured to normalize the raw BSEEG values, such as by differencing the recorded BSEEG values ​​from a population mean or threshold and then dividing by the standard deviation of the BSEEG population norm or threshold, to establish a normalized BSEEG (NBSEEG) score in an analyzing step 72. When a threshold is used, the NBSEEG can be classified as a NBSEEG positive (NBSEEG(+)) score or a NBSEEG negative (NBSEEG(-)) score, as described in the examples below.

[0064]

[0102] In another optional step, a validation step 73 can be performed. In such an implementation, as shown in FIG. 8, the spectral density 102 and / or other raw BSEEG signal values ​​104A, 104B, or other features, can be used to compare individual readings from segmented signal windows (shown schematically at 84A, 84B) for inclusion or exclusion for use in subsequent steps. In various implementations, this step can be performed using validation module 55 and validation data 63. In various implementations, validation step 73 can be performed iteratively, in parallel with other steps described herein, or in other manners, until the resulting signal data is ready for subsequent analysis, and error correction algorithms can be used. For example, as detailed in FIG. 8, validation step 73 can be used to ensure that all readings above or below a certain determined predetermined error value have been excluded from further processing.

[0065]

[0103] With continued reference to FIG. 7 , a diffuse slowing or death determination step or threshold step 74 may be performed. In this step, at least one of the one or more values ​​or characteristics, or a measure based on at least one of the one or more values, may be compared to a set diffuse slowing threshold (shown at 65 in FIG. 8 ), which may be implemented by diffuse slowing threshold data (such as shown at 64 in FIG. 6 ). In certain embodiments, this threshold comparison may be used to determine the presence or likelihood of a negative outcome, such as prolonged hospitalization, discharge away from home, or death. For example, as described in the Examples below, using the average of the 3 Hz / 10 Hz readings as threshold data 64 and a threshold value 65 of 1.44 (positive >= 1.44 and negative < 1.44), the screening device 10 may graphically display positive and / or negative readings with or without also displaying the threshold value 65. It should be understood that the threshold value 65 may be modified over time as a result of system improvements and enhancements, and based on additional data 66 and other factors.

[0066]

[0104] In an exemplary implementation, a graphical representation of the threshold steps is displayed on the screening device 10 or other monitoring system, showing a comparison of the spectral density to the set threshold as the most recent measurement 4. Each of these optional steps is described in more detail below in connection with examples of the present disclosure.

[0067]

[0105] As shown in FIG. 8 , in one exemplary implementation of system 1, a patient 30 is monitored by sensors 12A, 12B, such as using screening device 10 of FIG. 1 . In this implementation, sensor 12 generates one or more signals 80A, 80B from patient 30. One or more signals 80A, 80B may be analog and / or digital signals. One or more sensors 12A, 12B may be separate from a system receiving one or more signals, or may be incorporated into a system receiving one or more signals. While system 1 in this description relates to collecting EEG signals, in another implementation, one or more sensors 12A, 12B may also collect physiological signals such as: It is to be understood that one or more of the following conditions may be measured, detected, determined and / or monitored: heart rate, pulse rate, EKG, cardiac variability, respiratory rate, skin temperature, exercise parameters, blood pressure, oxygen level, core body temperature, heat flow from the body, galvanic skin response (GSR), electromyogram (EMG), electro-oculogram (EOG), body fat, hydration level, activity level, oxygen consumption, glucose or blood sugar level, body position, muscle or bone compression, and ultraviolet (UV) absorption.

[0068]

[0106] In various implementations, as described in connection with FIGS. 4A-6 , system 1 includes one or more signal processing devices, which may be located within screening device 10 or elsewhere. The processing may determine one or more characteristics by analyzing one or more signals to look for signal information corresponding to particular signal features. In certain embodiments, one or more signals may be processed to determine the presence of one or more high frequencies and / or one or more low frequencies. The one or more characteristics may be types of waves, such as, but not limited to, high frequencies and / or low frequencies.

[0069]

[0107] In the implementation of Figure 8, signals 80A, 80B are received via a first channel 82A and a second channel 82B for processing as described above. Other implementations are possible. The method may be incorporated into an 8-bit or 16-bit embedded device environment or a more rugged environment. The method may also be incorporated into an existing hospital patient workflow.

[0070]

[0108] Also, as shown in FIG. 8 , in certain implementations, signals 80A, 80B can be processed to window the signal into equal, contiguous intervals (shown schematically at 84A and 84B). In the illustrated implementation, a 10-minute signal duration is shown, and the signal is divided into 10 separate 1-minute windows (also shown schematically at 84A and 84B), although it should be understood that other signal durations and numbers of windows can be used in alternative implementations. In various implementations, the signal may be less than one second long or more than one hour long, or any number of minutes long. Similarly, there can be any number of windows, and the windows can range from a fraction of a second to many fractions or more of a time.

[0071]

[0109] In the implementation of Figure 8, after signals 80A, 80B are segmented, several optional processing and analysis steps can be performed, as described in connection with Figure 7. In various implementations, values ​​86A, 86B are extracted, and various optional steps can be performed on the values ​​86A, 86B to identify raw data features 104A, 104B, such that minimum / maximum amplitudes 88A, 88B, mean amplitudes 90A, 90B, interquartile range (IQR) amplitudes 92A, 92B, mean deviations of amplitudes 94A, 94B, and / or signal entropy 96A, 96B can be established for each signal 80A, 80B. As will be appreciated by those skilled in the art, other raw signal features 104A, 104B can also be identified.

[0072]

[0110] The analysis step can use spectral density analysis 102 to identify diffuse slowing. Also shown in FIG. 8 , in certain embodiments, a fast Fourier transform 100 can be performed on one or more signals 80A, 80B to identify additional features 104A, 104B. The fast Fourier transform 100 can be used alone or in combination with other analytical means to evaluate spectral density. Spectral density analysis 102 can be performed on one or more signals to determine one or more spectral density features 104A, 104B, such as low frequency density 105A, high frequency density 105B, and a ratio of low frequency density to high frequency density 105C. These features 104A, 104B from the spectral density analysis 102 can be used alone or derived from raw signals to predict and / or forecast mortality using various optional steps and combinations. The values ​​104A, 104B may be used in conjunction with various other values.

[0073]

[0111] In certain implementations, spectral density analysis 102A, 102B can be performed on each of one or more signals 80A, 80B, such as the EEG signals shown, to distinguish between different patient conditions. In certain embodiments, the spectral density analysis 102A, 102B can provide values ​​104A, 104B including a ratio 105C of high-frequency to low-frequency brain electrical activity. For example, to confirm widespread slowing, the ratio of approximately 10 Hz signals to approximately 2 Hz, 3 Hz, or 4 Hz signals can be compared. One or more bands or windows within one or more signals 80A, 80B can be identified for use in the systems and methods described herein.

[0074]

[0112] In certain implementations, a validation step 73 (shown in box 106) can be performed. In such implementations, the spectral density 102 and / or other raw signal values ​​104A, 104B can be used to compare individual readings from segmented signal windows (shown schematically at 84A, 84B) for inclusion or exclusion in the analysis. For example, in certain implementations, a correction algorithm 108 can be performed. In one implementation, the error correction algorithm 108 performs several optional steps. In one optional step, various values ​​104A, 104B within a window that are above or below a certain predetermined error threshold are discarded 110. In another optional step, if the IQR and / or density ratio 105C are within a certain predetermined proximity 112, these window signals are retained for aggregation and recombination 116, as described below. Other optional steps are also possible.

[0075]

[0113] In these implementations, an optional additional recombination step 116 can be performed in which windows 84A, 84B that were not removed as a result of the verification step 73 (box 106) can be combined such that the values ​​104, 104B, 104A, 104B from those windows are aggregated as the widespread slowing threshold data 64.

[0076]

[0114] The diffuse slowing threshold data (shown at 64 in FIG. 6 ) is used to perform a diffuse slowing or death thresholding step 74, as further shown in FIGS. 9A-9D . In these implementations, the aggregated threshold data 64 can be compared to a set threshold 65 using a threshold module 56. In various implementations, data from other sources 66, such as electronic medical records, can also be compared to the threshold 65 to set an output 67 that can be presented graphically in any of the other devices and systems described herein, such as in connection with FIGS. 5A-5D . One such example is a display of output data 67, which can include the latest measurement 4, trend 6, quality 8, power 75, and other graphical representations, such as those shown in FIGS. 9A-9D .

[0077]

[0115] As shown in Figures 2A through 3D, the EEG of a potentially dying patient may be characterized by "pervasive slowing," i.e., electrodes from the EEG, preferably all electrodes, exhibiting slowed waveforms. As discussed above, EEGs may have various frequencies and / or frequency bands. Pervasive slowing may mean that slowed waves, lower frequency waves, are seen across most, if not all, electrodes on the EEG. The appearance of slow waves relative to the number of higher frequency waves may be a sign that the patient is dead or likely to die. Two leads (and a ground) may be sufficient to detect pervasive slowing, and with appropriate signal processing and user interface, may not require specialized expertise for placement or interpretation.

[0078]

[0116] In certain embodiments, even the number of high frequencies in one or more signals and / or the number of low frequencies in one or more signals may be A spectral density analysis 102A, 102B is performed on one or more values ​​104A, 104B. As shown in Figure 9A, each channel 82A, 82B can be analyzed by the steps described in connection with Figure 8A to determine the frequency and other characteristics of the "high" and "low" frequencies as output data 67. In various implementations, these representations can be used in this process as any of the various forms of data described above and displayed on the device display 16 (shown in Figure 4B).

[0079]

[0117] In certain embodiments, values ​​104A, 104B may be calculated over the entirety of one or more signals 80A, 80B as part of any of the steps described above. In certain embodiments, values ​​104A, 104B may be calculated over a subset of one or more signals or a subset of time periods for one or more signals. For example, if one or more signals have a duration of five minutes, values ​​104A, 104B may be calculated over a time period less than five minutes, such as four minutes, three minutes, two minutes, one minute, or 30 seconds. Thus, one or more values ​​104A, 104B may be features 104A, 104B and / or values ​​104A, 104B over a predetermined time period. In certain embodiments, one or more values ​​104A, 104B may be the number of high frequencies over a certain period of time and / or the number of low frequencies over a certain period of time. In certain embodiments, one or more values ​​104A, 104B may be the ratio of the number of high frequencies to the number of low frequencies. In certain embodiments, one or more values ​​104A, 104B may be a ratio of the number of high frequencies over a period of time to the number of low frequencies over a period of time.

[0080]

[0118] As shown in FIGS. 8A, 8B, and 9A-9D, in various implementations of device 10, system 1, and method 5, the diffuse slowing or death determination step (shown at 74 in FIG. 7) can plot and compare the ratio of high to low frequencies for each channel. In these implementations, at least one of one or more values ​​or features, or a measure based on at least one of one or more values, included in threshold data 64 may be compared to a diffuse slowing threshold 65 established by module 56. In certain embodiments, this comparison may be used to determine the presence or likelihood of subsequent patient death. In various implementations, the frequencies for each signal are compared over time, and the results are analyzed. As described in more detail below, FIGS. 9A and 9B show raw EEG channel signals over time for each channel. FIGS. 9C and 9D show a spectral density analysis 102 of the results for each channel.

[0081]

[0119] 10A and 10B, in various implementations, system 1 and method 5 are configured to execute a series of optional steps by a server / computing device 42 and / or processor 44, such as by various program modules 48 as described in connection with FIGS. 6-8. In the implementations of FIGS. 10A and 10B, system 1 is configured such that a screening device 10 is used to record (box 200) raw BSEEG scores from a subject. In these and other implementations, the output normalized BSEEG (NBSEEG) score can be quantified as NBSEEG positive (NBSEEG(+)) or NBSEEG negative (NBSEEG(-)), which can be used to predict death or other negative outcomes as described herein.

[0082]

[0120] In various implementations, to output a prognostic NBSEEG score (box 204) by NBSEEG, NBSEEG (box 202) is calculated by (the difference between the recorded raw BSEEG and the BSEEG population mean) / (the standard deviation of the BSEEG population). The prognostic NBSEEG can be used as a classification such as NBSEEG positive (NBSEEG(+)) or NBSEEG negative (NBSEEG(-)).

[0083]

[0121] These prognostic NBSEEG scores can also be used as a continuous number, as is the case for other vital signs such as blood pressure or temperature: NBSEEG positive (NBSEEG(+)) or NBSEEG negative (NBSEEG(-)).

[0084]

[0122] Experimental results are presented and conclusions are drawn in the accompanying examples. [Example]

[0085] Screening Device Evaluation

[0123] In this example, the initial training set of the dataset included a total of 186 patient EEG samples that correlated with clinical evidence of delirium or CAM evidence. These samples represented 5 positive cases, 179 negative cases, and 2 negative cases for which the data quality was insufficient for analysis and were therefore excluded from further consideration.

[0086]

[0124] In this example, a 15 Hz low pass filter was initially used, but preliminary results showed uneven attenuation of FFT frequency information between positive and negative cases, so this low pass filter was removed.

[0087]

[0125] During processing of the processed samples, a 4 second window was observed to be sufficient to show good results, and in this example, a threshold of 500 μV was used to exclude windows containing high amplitude peaks, as shown in Figures 9A and 9B.

[0088]

[0126] Figures 9C and 9D show the spectral densities of the channels, where intensity (in W / Hz) can be compared with each frequency (in Hz) to set the low:high frequency ratio. In this training set, it was observed that using a 3 Hz / 10 Hz ratio produced better prediction results than a 2 Hz / 10 Hz or 4 Hz / 10 Hz ratio.

[0089]

[0127] Values, Features, and Thresholds. As used herein, the terms "value" and "feature" may be interchangeable and may refer to raw and analyzed data, whether numerical, time-scale, graphical, or other data. In various implementations, such as those described herein, a value, such as a high frequency count, may be compared to a threshold. Alternatively, or in addition, a ratio of two or more values ​​may be compared to a threshold. The threshold may be a predetermined value. The threshold may be based on statistical information regarding the presence or likelihood of delirium sequelae, such as information from a population of individuals. In certain embodiments, the threshold may be predetermined for one or more patients. In certain embodiments, the threshold may be a consistent value for all patients. In certain embodiments, the threshold may be specific to one or more characteristics of the patient, such as current health status, age, sex, race, medical history, or other medical conditions. In certain embodiments, the threshold may be adjusted based on pathological data in the patient's electronic medical record (EMR).

[0090]

[0128] In certain embodiments, the threshold may be the ratio of high frequency to low frequency. In certain embodiments, the threshold may be the ratio of high frequency over a period of time to low frequency over a period of time. Throughout this disclosure, this ratio will be referred to as the ratio of high frequency to low frequency, but it should be understood that this ratio may also be the ratio of low frequency to high frequency, as long as the format of the ratio is consistent throughout the process. For example, the comparison may be a comparison of the ratio of high frequency to low frequency, or vice versa, i.e., low frequency / high frequency.

[0091]

[0129] One or more characteristics or values ​​may be predetermined. For example, a range of waves that are high frequency may be predetermined as being above a set value. Similarly, a range of waves that are low frequency may be predetermined as being above a set value. A range of frequencies may be predetermined as being below a set value, which may be the same for all patients or may vary depending on the characteristics of a particular patient.

[0092]

[0130] Other features or values ​​of one or more signals may be extracted. For example, signal-to-noise ratio may also be determined in other applications. Data quality may be assessed by looking for non-physiological frequencies of electrical activity. Data quality may be limited to stop data collection and / or interpretation if it falls below an acceptable level.

[0093]

[0131] Device Characteristics and Signal Collection. The systems and methods described herein can provide a dedicated screening device 10, system 1, and method 5 that are simple, convenient, and easy to use. In certain embodiments, the systems and methods may utilize electroencephalogram (EEG) technology simplified for end users. The systems and methods may automatically interpret the data and provide guidance to healthcare professionals regarding patient delirium monitoring, screening, or sequelae. Traditionally, EEG data is visually inspected by an experienced neurologist, and this process is not automated. In certain embodiments described herein, there may be options for interfacing with standard monitoring equipment, mobile devices, cloud technology, etc. to create an automated process.

[0094]

[0132] Certain embodiments described herein may be useful in various medical areas, such as, but not limited to, intensive care, pre- and post-operative care, geriatrics, nursing homes, emergency rooms, and trauma care. Monitoring, screening, or prediction can improve patient care while in a hospital or other medical setting. Patients may also use personal healthcare devices and monitoring to enable remote monitoring of their condition when not in a medical setting. For example, personal healthcare devices may monitor patients at home or other locations outside of a medical setting and provide delirium monitoring, screening, or prediction. Remote sensing and / or analysis systems may interface with systems used by medical personnel.

[0095]

[0133] As shown in various implementations, one or more sensors 12 may be in communication with the patient 30. In certain embodiments, the one or more sensors 12 may be one or more brain sensors, such as, but not limited to, an EEG device, such as one or more EEG leads / electrodes. In this disclosure, the terms "lead" and "electrode" are used interchangeably. The one or more signals may be EEG signals. The EEG signals may include voltage fluctuations caused by ionic currents in neurons of the patient's brain. In certain embodiments, there may be multiple sensors. In certain embodiments, there may be two sensors, such as two EEG electrodes. The use of an EEG system with fewer electrodes than traditional 16- or 24-electrode systems can reduce the cost and complexity of delirium prediction, screening, or monitoring. In various implementations, two or more leads or sensors are used. In certain implementations, 2, 3, 4, 5, 6, 7, 8, 9, or 10 sensors are used. In further implementations, 11, 12, 13, 14, or 15 sensors are used. In still other implementations, more than 15 sensors are used. In various implementations, a minimal number of EEG leads that are easier to deploy than those shown in the prior art may be used, thereby eliminating and / or reducing the need for skilled EEG technicians and / or specialized neurologists. In various implementations, at least one ground lead is used, while in other implementations, multiple ground leads are used, such as 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 or more ground leads.

[0096]

[0134] In certain embodiments, one or more sensors 12 may be non-invasive. In certain embodiments, non-invasive electrodes may be placed on the patient's skin. In certain embodiments, there may be a minimum of two skin contact points: a sensor and a ground sensor. The sensor may be passive in that no external current flows through it. In certain embodiments, electrically active sensors may be used. In certain embodiments, one or more sensors may be an adhesive patch with a printed circuit or a non-adhesive headset that couples the sensor to the skin. In certain embodiments, one or more sensors may be placed on the patient's head, such as on the forehead and / or behind one or more of the patient's ears. In certain embodiments, a minimum separation distance may be provided between one or more sensors so that the one or more sensors do not contact each other.

[0097]

[0135] In certain embodiments, minimally invasive or invasive sensors may be used that may provide one or more signals as an indication of a physiological condition, such as brain activity.

[0098]

[0136] One or more signals may be converted from analog to digital signals if necessary. The conversion may occur in the processing device before the one or more signals are received by the processing device, or in a separate device. If one or more signals are generated or received as digital signals, conversion may not be necessary.

[0099]

[0137] The one or more signals can be indicative of one or more brain functions of the patient. In certain embodiments, the one or more signals can provide information about the patient's brainwave activity. EEG can be measured on the patient. In certain embodiments, the EEG can be performed by EEG, which can be a recording of the brain's electrical activity from the scalp. The recorded waveforms can reflect electrocortical activity. In certain embodiments, the EEG signal strength can be small and measured in microvolts. Traditionally, there are several frequencies and / or frequency bands that can be detected using EEG. The definition of high and low frequencies can vary greatly depending on various factors, including, but not limited to, the patient population. In certain embodiments, the definition of high and low frequencies can be consistent across the entire patient population. In certain embodiments, low frequencies can be waves below about 7.5 Hz, below about 7.0 Hz, below about 6.5 Hz, below about 5.5 Hz, below about 5 Hz, below about 4.5 Hz, below about 4.0 Hz, below about 3.5 Hz, or below about 3.0 Hz. In certain embodiments, the high frequency may be greater than about 7.5 Hz, greater than about 8.0 Hz, greater than about 8.5 Hz, greater than about 9.0 Hz, greater than about 9.5 Hz, greater than about 10.0 Hz, greater than about 10.5 Hz, greater than about 11.0 Hz, greater than about 11.5 Hz, greater than about 12.0 Hz, greater than about 12.5 Hz, greater than about 13.0 Hz, or greater than about 14.0 Hz.

[0100]

[0138] In certain embodiments, the one or more signals may be a real-time or near-real-time stream of data, hi certain embodiments, the one or more signals may be measured and / or stored for a period of time prior to processing and / or analysis.

[0101]

[0139] Although not required, the systems and methods are described in the general context of software and / or computer program instructions executed by one or more computing devices, which may take the form of traditional servers / desktops / laptops, mobile devices such as smartphones or tablets, wearable devices, medical devices, other healthcare systems, etc. The computing devices may include one or more processors coupled to data storage for computer program modules and data. Key technologies include multi-industry standards for Microsoft and Linux / Unix-based operating systems, databases such as SQL Server, Oracle, NOSQL and DB2, business analytics / intelligence tools such as SPSS, Cognos, SAS, etc. , Java, development tools such as .NET Framework (e.g., VB.NET, ASP.NET, AJAX.NET), and other e-commerce products, computer languages, and development tools. Such program modules may include computer program instructions, such as routines, programs, objects, components, etc., for execution by one or more processors to perform particular tasks, use data, data structures, and / or implement particular abstract data types. While systems, methods, and apparatuses are described in the above context, the acts and operations described below may also be implemented in hardware. [Example]

[0102] NBSEEG, delirium and death

[0140] Methods: This was a prospective study to measure bispectral EEG ("BSEEG") from elderly patients to assess their prognosis. Normalized BSEEG ("NBSEEG") scores were defined based on the distribution of 2938 BSEEG recordings from 428 subjects evaluated for delirium. The primary outcomes measured were hospital length of stay ("LOS"), discharge disposition, and mortality.

[0103]

[0141] Results: 274 patients had NBSEEG score data available for analysis. Delirium was significantly associated with NBSEEG score (P<0.001). Higher NBSEEG scores were significantly associated with LOS (P<0.001) and discharge away from home (P<0.01). The hazard ratio for survival control for age, sex, Charlson comorbidity index, and delirium status was 1.35 (95% confidence interval = 1.04 to 1.76, P=0.025).

[0104]

[0142] Described herein is an efficient and reliable device that provides an objective measure of brain functional status. NBSEEG scores are significantly associated with clinical outcomes for patient mortality, hospital length of stay, and discharge disposition. NBSEEG scores actually better predict mortality than clinical delirium status. Use of the devices and systems described herein allows for the identification of a previously unrecognized subpopulation of patients without clinical features of delirium who are at increased risk of mortality.

[0105]

[0143] The electrophysiological signal characteristics of delirium are often reported as "diffuse slowing." This term refers to the fact that most channels of EEG exhibit reduced frequency. The appearance of low-frequency waves indicates the possibility of delirium. The fact that all channels can detect the same frequency reduction suggests that only a few channels are sufficient to obtain relevant data. BSEEG uses only two channels and, when combined with appropriate signal analysis algorithms, can be easily applied by non-experts, making it extremely easy to use as a screening tool. Due to its objectivity, inter-rater reliability is unaffected by BSEEG and it can be more strongly correlated with patient prognosis. The following example demonstrates whether BSEEG values ​​and NBSEEG scores can predict patient outcomes, including mortality.

[0106]

[0144] method

[0145] Study design and management: This study tested the utility of the BSEEG approach to patient care and examined the association between the BSEEG score from this algorithm and patient outcomes.

[0107]

[0146] Variables and data sources. The CAM-ICU, DRS-R-98, and Delirium Observation Screening Score (DOSS) were used to measure clinical symptoms of delirium. The Montreal Cognitive Assessment (MoCA) was used to assess cognitive function. The CAM-ICU and DRS-R-98 were administered to each subject twice daily unless the subject declined the assessment request. DOSS was collected by clinical nursing staff during routine nursing visits and obtained through medical record review. Delirium was defined based on any positive questionnaire screening: a positive CAM-ICU, a DRS-R-98 score ≥ 18, a DOSS score > 2, or clinical documentation of altered mental status or confusion consistent with delirium in the medical record. Each case was discussed at a weekly study meeting led by a board-certified consultant liaison psychiatrist.

[0108]

[0147] BSEEG Data Collection. A handheld two-channel EEG device was used for EEG recording. Raw BSEEG values ​​were collected twice daily unless the subject declined the assessment request. As shown in Figure 10, to obtain 10-minute raw BSEEG values, one electrode in the center of the forehead was used as a ground, two electrodes were placed on the left and right sides of the forehead, and two electrodes were placed on either side of the earlobes as references. The 10-minute duration was selected as a length of time that would allow for the collection of a sufficient amount of EEG data without sacrificing efficiency and throughput, which are essential features of a screening test. Recordings were obtained while the patient was at their highest level of consciousness, as awake and alert as possible given their clinical condition. Patients were instructed to keep their eyes closed, their jaws relaxed, and to remain as quiet and still as possible. The acquired raw BSEEG data was converted into a spectral density plot, and a signal processing algorithm was used to generate one or more raw BSEEG values.

[0109]

[0148] Spectral Density Analysis and NBSEEG Score. Raw EEG signals from each channel were subjected to power spectral density analysis to determine the relative presence of "high" and "low" frequency components. An iterative approach was used to generate a score reflecting the relative presence of high- and low-frequency activity. Means and standard deviations (SDs), shown in Figure 11, were calculated from 2938 recordings of raw BSEEG values ​​from all 428 study patients. The NBSEEG score was defined as the number of SDs from the mean of the study population.

[0110]

[0149] Prognostic Measures. Three patient outcomes were tracked and measured: 1) hospital LOS; 2) discharge away from home, including in-hospital death; and 3) death at study outcome. LOS, discharge outcome, and mortality status were obtained from each subject's hospitalization record. Mortality was also assessed by follow-up telephone interviews and death records.

[0111]

[0150] Statistical Methods and Analysis. Regression analysis was used to determine how the proposed NBSEEG score correlated with clinical delirium and patient outcomes, including hospital LOS, discharge away from home, and death. Specifically, logistic regression was performed with delirium and discharge away from home as binary response variables, and linear regression was used to assess the relationship between hospital LOS and NBSEEG score. Furthermore, Cox proportional hazards regression analysis was used to calculate the hazard ratio for death. Age, sex, and severity of illness were controlled for in the regression analysis. The association between death and NBSEEG score was further clarified by comparing two nonparametric survival functions for the NBSEEG-positive and NBSEEG-negative groups. The survival function is a series of Kaplan-Meier estimates calculated from the number of deaths and the total number of individuals at risk at that time. A log-rank test was performed to determine whether the two survival functions differed. A two-sided P value of 0.05 or less was considered statistically significant. All analyses were performed using R software, version 3.4.3.

[0112]

[0151] result

[0152] Participants, descriptive and prognostic data. 428 patients were enrolled in this study. Of the 428, 337 patients were aged 55 years or older, and 274 of the 337 were included in the analysis. The available NBSEEG scores were used. Based on questionnaire screening or clinical descriptions, 37.2% of patients in the 55+ age group were classified as delirious. Furthermore, as shown in Figure 12, based on thresholds for distinguishing patient outcomes as described in the following section, the study population was independently divided into two groups: an NBSEEG-positive group with higher NBSEEG scores indicating more low-frequency components in the EEG, and an NBSEEG-negative group with lower NBSEEG scores indicating fewer low-frequency components in the EEG. Otherwise, these cohorts were balanced with respect to the overall baseline characteristics shown in Table 1.

[0113]

[0153] Association between NBSEEG scores and clinical delirium. Data from 274 subjects were analyzed to establish the association between NBSEEG scores and clinical delirium. Logistic regression showed a significant association between delirium category and NBSEEG scores (P = 6.39 × 10 -6 , unadjusted; P=1.22×10 -5 , adjusted for age, sex, and CCI).

[0114]

[0154] NBSEEG score and patient outcomes: To test the utility of the NBSEEG score in predicting patient outcomes, we used prognosis data from 274 subjects aged 55 years or older to examine the association between the NBSEEG score obtained at study entry and patient outcomes commonly affected by delirium. Specifically, we evaluated hospital LOS, hospital discharge trends, and mortality.

[0115]

[0155] First, there was a significant association between LOS and NBSEEG score (P = 0.00099, unadjusted; P = 0.0014, adjusted for age, sex, and CCI), with higher NBSEEG scores corresponding to increased patient LOS.

[0116]

[0156] Second, we compared discharge outcomes with NBSEEG scores. When comparing the NBSEEG scores of those discharged home with those discharged to a location other than home, including those who died during hospitalization, higher NBSEEG scores were significantly associated with discharge to a location other than home (P = 0.0038, adjusted for age, sex, and CCI).

[0117]

[0157] Third, when the mortality rate of subjects was analyzed, controlling for age, sex, and CCI, the hazard ratio based on a 1 SD change in NBSEEG score was 1.44 (1.12 to 1.84, P = 0.004). Even after controlling for age, sex, and CCI as well as clinical delirium status, the HR based on NBSEEG score remained significant at 1.35 (95% CI = 1.04 to 1.76, P = 0.025).

[0118]

[0158] In addition to mathematical correlations, we analyzed NBSEEG scores as a potentially useful measure for assessing patients' risk of poor outcomes. As described above, we divided the study population into NBSEEG-positive and NBSEEG-negative groups. Next, we evaluated the data to determine whether there was a correlation between the groups based on NBSEEG scores and all-cause mortality at the end of the study period for patients in our dataset. First, we assessed the overall survival of study participants to determine whether the clinical classification of delirium was sufficiently valid to replicate the established association between delirium and high mortality. Results showed a difference in mortality between patients with and without clinical delirium (P = 0.0038) (Figure 13). Second, we examined group differences based on the NBSEEG cutoff score and confirmed that the NBSEEG-positive group exhibited worse survival compared to the NBSEEG-negative group (P = 0.0032) (Figure 13). This classification also significantly differentiated other outcomes, including LOS and hospital discharge trends (Table 2).

[0119]

[0159] The NBSEEG score not only measures the presence of delirium but also represents the severity of delirium. Based on the NBSEEG score, subjects were classified as NBSEEG high, NBSEEG medium, and The survival curves showed a "dose-dependent" relationship between increasing mortality and increasing NBSEEG scores (Figure 14, P = 0.005), suggesting a strong relationship between NBSEEG scores and mortality.

[0120]

[0160] The cohort was divided into four groups based on clinical delirium diagnosis and BSEEG. Clinical delirium subjects with positive NBSEEG scores had the highest mortality rate. In contrast, patients classified as having clinical delirium but with negative NBSEEG scores had a lower mortality rate similar to that of non-delirium subjects with negative NBSEEG scores. Furthermore, subjects who were considered non-delirium based on the results of clinical assessment but with positive NBSEEG scores had a higher mortality rate compared with patients with clinical delirium but with negative NBSEEG scores (Figure 15).

[0121]

[0161] Discussion - Key Results and Interpretation: NBSEEG scores were significantly associated with the clinical presence of delirium, even after controlling for age, sex, and CCI. More importantly, NBSEEG scores were strongly correlated with patient outcomes, including hospital LOS, discharge disposition, and mortality. Importantly, this association was based on NBSEEG scores obtained at enrollment, often within 24 hours of admission. These results suggest that a single NBSEEG score obtained at the onset of hospitalization can predict patient outcomes. The results indicate that among patients who cannot be clinically identified as having delirium, a subset distinguishable by differences in EEG activity detected by NBSEEG is at increased risk for death. This condition can be classified as silent cerebral dysfunction (SBF). Therefore, identifying this population using NBSEEG methods may result in earlier intervention and potentially improved survival.

[0122]

[0162] The data disclosed herein demonstrate the utility of the NBSEEG score in distinguishing delirium cases from non-delirium patients and predicting patient outcomes, such as hospital LOS, discharge disposition, and mortality, among elderly hospitalized patients.

[0123]

[0163] Such NBSEEG-based biomarkers may enable early intervention and improve current drug and surgical treatment practices for patients at risk for delirium. For example, NBSEEG analysis may be an important factor in the decision to perform elective surgery or can be used for close postoperative monitoring. If high-risk patients are identified through NBSEEG analysis, hospital resources may be allocated more efficiently and effectively compared with the current standard of care.

[0124]

[0164] NBSEEG monitoring may be applicable in additional settings, such as primary care clinics, emergency departments, and nursing homes or home health care settings. Delirium is particularly dangerous when patients face it outside of hospitals, due to a lack of awareness and resources to manage it. The simple and noninvasive nature of this test makes it ideal for routine screening. The NBSEEG test described herein can also be used as a monitoring tool to assess mortality risk in appropriate populations. For example, with the rapidly growing aging population, NBSEEG could be implemented as an efficient modality for mortality risk screening.

[0125]

[0165] Limitations: Forehead electrode placement was used as a convenient screening method. Of course, other lead placement configurations may be possible.

[0166] In some implementations, the devices and methods described herein can be used to investigate and evaluate the effectiveness of various therapies, such as ramelteon and suvorexant, to determine the impact of specific therapies / drugs on NBSEEG scores and prognosis. Thus, the devices and methods of the present disclosure can provide and investigate better treatments for delirium. This can be used to improve patient outcomes, ultimately leading to improved outcomes.

[0126]

[0167] Implications for clinical practice: Noninvasive point-of-care EEG acquisition combined with NBSEEG scoring can predict poor patient outcomes, including death. Importantly, it can identify specific patient populations at high risk of death who are unable to be identified by current clinical assessment. [Example]

[0127]

[0168] This example evaluates the use of two-channel frontal EEG activity to quantitatively characterize delirium and predict outcomes, including fall risk and mortality.

[0169] method

[0170] Frontal EEG activity (Fp1 and Fp2 EEG locations) was collected from patients after admission or emergency room visit. Subjects were assessed for the clinical presence of delirium, and the primary outcomes measured were delirium diagnosis, disposition at discharge, mortality, and fall history. EEG features (different combinations of band power and low- and high-frequency activity) were calculated for both channels and averaged. k-nearest neighbor, logistic regression, support vector machine (SVM), kernelized SVM, and neural network methods were used to evaluate the performance of EEG features in predicting delirium status, survival, and falls using 5-fold cross-validation.

[0128]

[0171] result

[0172] EEG features and prognosis data for 274 patients were available for analysis. Random forests were used to select the top nine predictive features from EEG. Of all these classification methods, kernelized SVM demonstrated the highest predictive accuracy of 69%, 81%, and 89% for delirium status, death, and falls, respectively. Frontal EEG can be used to objectively measure delirium, given the variability in clinical causes, and to predict associated clinical outcomes, including fall risk and mortality.

[0129]

[0173] Placing only two channels on the head—BSEEG—allows non-experts to apply the device, thereby eliminating the need for specialized neurologists and technicians and enabling widespread adoption of this technology. In this example, the disclosed BSEEG method was used in conjunction with point-of-care technology to predict patient outcomes (delirium diagnosis, mortality, fall risk, and discharge status). Specifically, power spectral density analysis from limited frontal EEG leads can predict patient outcomes.

[0130]

[0174] method

[0175] Questionnaire and Definition of Delirium. The CAM-ICU, DRS-R-98, and DOSS were used to measure clinical symptoms of delirium. The MoCA was used to assess baseline cognitive function. Delirium was defined based on a positive questionnaire screening, such as a positive CAM-ICU, DRS-R-98 > 18, or DOSS > 2, or clinical documentation of altered mental status or confusion consistent with delirium in the medical record. Each case was discussed at a weekly research conference led by a board-certified psychosomatic psychiatrist.

[0131]

[0176] BSEEG data collection and processing procedures. A handheld two-channel EEG device was used for EEG recording. To obtain a 10-minute BSEEG signal, one electrode was placed at the center of the forehead as a ground, two electrodes were placed on the left and right sides of the forehead, and two electrodes were placed on either side of the earlobes as references. The acquired BSEEG data were converted into spectral density plots, and signal processing algorithms were run to extract EEG features.

[0132]

[0177] EEG signal processing, analysis, and interpretation. The recorded EEG data were examined in European data format for further analysis. Each channel of the EEG data was extracted, followed by a 4-second window. The BSEEG signals were divided into four sections, which were then filtered for excess noise. Interfering sectioning windows were excluded from further analysis. The power spectral density (PSD) of the remaining windows was determined by fast Fourier transformation and aggregated as the median of all remaining windows. BSEEG features (different combinations of band power and low-to-high frequency activity) for both channels were calculated and averaged. Various PSD ratios (PSDRs) of low-to-high frequency activity were used to determine features for further analysis at baseline admission.

[0133]

[0178] Outcomes. Patient outcomes were measured as follows: 1) delirium diagnosis, 2) survival, and 3) discharge to home, including in-hospital death. All outcomes were obtained from patient admission records. Mortality was also assessed by follow-up telephone interviews and death records. Endpoint classification was determined by study members blinded to BSEEG features.

[0134]

[0179] Predictive model using random forest: We evaluated the predictive power of BSEEG features using random forest (RF) along with the Voluta algorithm, which can predict pathology based on an ensemble of decision trees. RF was used to create a predictive model based on the EEG profile using all EEG features as input. The relative importance of each EEG feature in the predictive model was evaluated using the mean drop in accuracy and the Gini coefficient.

[0135]

[0180] Classification Algorithms: After establishing the association between delirium state and EEG features, classification analysis was used to examine whether the onset of poor outcomes, such as risk of death and falls, was associated with the EEG signal. Various machine learning (ML) algorithms were applied to the EEG dataset for different types of classification tasks. The resulting classification models were then used directly to predict patient outcomes, such as delirium, death, falls, and hospital discharge. In the machine learning hierarchy, classification tasks fall under supervised learning tasks, which means that, unlike unsupervised learning tasks, feedback is available to the learning system. This feedback is also known as gold standard, training data, example data, or labeled data.

[0136]

[0181] k-Nearest Neighbor Algorithm. The k-Nearest Neighbor algorithm (k-NN) stores training data. When a new data point arrives, the k-NN finds the closest, or nearest, point in the training dataset to the new data point, where k is the number of nearest neighbors considered. The k-NN can then make predictions using majority voting among the k nearest neighbors. The k-NN algorithm takes only one input parameter for training: k. In our model, we adjusted the value of k from 1 to 10 to obtain the k that showed the best prediction accuracy on the test dataset. The k-NN algorithm is quite easy to understand and often provides reasonable performance without significant tuning. However, since it must calculate the distance between the new data point and every data point in the training dataset in real time, predictions are slow for large training datasets. It also performs poorly on datasets with many features or sparse datasets. For these reasons, the k-NN algorithm is not often used in practice; instead, it serves as a good baseline method to try before considering more advanced techniques.

[0137]

[0182] Logistic regression. The logistic regression algorithm is based on the linear regression algorithm, which uses a linear function of the input feature variables to make predictions of the objective function. The difference between the two algorithms is that the linear regression algorithm makes predictions on continuous values, while the logistic regression algorithm makes predictions on predefined class labels, and can therefore be used for classification tasks. To make predictions on class labels, the entire linear function is put into another function called a sigmoid function, which ranges between 0 and 1. If the sigmoid function is greater than 0.5, it predicts the class as +1, and if it is less than 0.5, it predicts the class as -1. The logistic regression algorithm is a binary classification algorithm that returns +1 and -1. Binary classification algorithms can be used for classification tasks that involve multiple classes. Several techniques, such as one-vs-all or one-vs-other, have been proposed to extend linear models to multi-class classification algorithms. Linear models are fast to train and predict. These techniques scale to extremely large datasets and work well with sparse data. They are relatively straightforward in how predictions are made using linear functions. On the other hand, linear models, as the name suggests, are based on the strong assumption that the target variable can be predicted by a linear combination of feature variables, which may be too weak to apply to real-world problems.

[0138]

[0183] Support Vector Machine. The support vector machine algorithm, or SVM, is based on the intuition of large margins. In other words, it seeks to find a maximum-margin strict line, surface, or hyperplane that represents the maximum separation, or margin, between two classes. Typically, only a subset of data points—specifically, data points on the boundary between classes, called support vectors—are important in determining the decision boundary. To make a prediction for a new data point, the distance to each support vector is measured, and a classification decision is made based on the distance to the support vector and the support vector weights learned during training. SVM algorithms work well with high-dimensional data, which means they can draw complex decision boundaries. In this case, the distance between data points can be measured using a Gaussian kernel. SVM algorithms that use a Gaussian kernel function are called kernelized SVM algorithms. The inventors tuned two parameters for training: a penalty parameter for the error term for regularization and a kernel coefficient, or gamma. The SVM algorithm performs extremely well on a variety of datasets, which is why it is known as one of the most commonly used classification algorithms. It allows for complex decision boundaries, as described above, even when the dataset has few features. It also works well for low-dimensional data with few features and high-dimensional data with many features. However, it is extremely sensitive to data scaling and parameter settings. In some cases, it can be difficult to understand why the algorithm made a particular decision.

[0139]

[0184] Neural Networks. Neural network algorithms are inspired by the actual biological neural networks that make up animal brains. These algorithms are essentially a generalization of linear models, using multiple stages of processing to reach a decision. For example, a logistic regression model can be represented as a two-layer neural network consisting of an input layer with input feature nodes and an output layer with a goal node. A new hidden layer with several hidden nodes can then be added between the input and output layers to increase the model's complexity. Various hidden layers and hidden nodes can be added to increase the model's complexity. All initial weights are randomly set, and therefore, this random initial setting can affect the trained model. The inventors adjusted the random state values ​​for the random weight initial setting between 0 and 10 to obtain the random state that provided the best predictive accuracy for the test dataset. Neural network algorithms can capture information, including large amounts of data, and can build highly complex models. Given sufficient computation time, data, and careful parameter tuning, they often outperform other machine learning algorithms. However, training often takes a long time. They also require careful data preprocessing and parameter tuning.

[0140]

[0185] result

[0186] The Voluta algorithm was used to select significant BSEEG features and nine genera were identified for their feature importance in prediction, shown in Figure 16. The prediction model identified certain BSEEG features, specifically the 3 to 10 Hz PSD ratio, theta to alpha power ratio, and the 5 to 10 Hz PSD ratio, as predictive of delirium state.

[0141]

[0187] KNN, logistic regression, SVM, kernelized SVM, and neural network algorithms were used to predict delirium diagnosis, survival, falls, and outcomes of discharge to a non-home location, including in-hospital mortality. Results of all analyses are shown in Tables 3 to 6.

[0142]

[0188] Consideration

[0189] Our results demonstrated the utility of simplified, portable, automated EEG using bispectral density analysis (BSEEG) for predicting patient outcomes. Compared with traditional EEG, which requires >20 leads to be placed across the patient's head by a skilled EEG technician, this system requires only a few leads placed on the forehead and therefore minimal training. Screening can be performed in minimal time (i.e., a few minutes), and extended monitoring can also be performed dynamically through longer recording durations. This is a significant advantage compared to traditional EEG readings that require expert interpretation, which introduce significant delays. BSEEG is also an improvement over many screening methods currently used in practice, such as question-based methods, which are prone to subjective variability among examiners, and mental status tests, which require extensive training and a long time to administer.

[0143]

[0190] Continuous monitoring with BSEEG can classify patients into three or more different mortality levels. Conventional EEGs read by neurologists can only classify patients into two categories: diffuse slow-waves or normal. In fact, BSEEG can predict mortality as well as conventional EEG.

[0144]

[0191] BSEEG is an improvement in that it requires fewer electrodes and does not require interpretation by a specialized neurologist. In various implementations, BSEEG can provide continuous measurements, thereby providing more comprehensive and additional information regarding risk of death than can be obtained from a traditional EEG and neurologist.

[0145]

[0192] The BSEEG devices and methods described herein can also predict patient outcomes related to delirium. In clinical practice, this method can be used as an additional biomarker for predicting patient outcomes. Electrodes may be placed on the forehead, although other lead placements are possible. In various implementations described herein, EEG features were as important as traditional clinical endpoints, such as hospital length of stay, in predicting delirium ( FIG. 16 ).

[0146]

[0193] BSEEG may also be useful for distinguishing delirium cases from normal subjects and for predicting patient outcomes, including mortality, fall risk, and discharge prognosis, in elderly hospitalized patients. BSEEG can be used not only in patients with overt mental status alterations, but also in a broader patient cohort.

[0147]

[0194] BSEEG allows for early intervention and prevention of delirium-related outcomes and can improve current medical practices for patients at risk for delirium. For example, if high-risk patients are identified through BSEEG analysis, hospital resources can be allocated more efficiently and effectively compared to current standard care. BSEEG monitoring may also be applicable in additional settings, such as primary care clinics, emergency departments, and nursing homes or home care settings. Delirium is particularly dangerous when patients experience it outside of hospitals, as there is no on-site medical attention available. The devices, systems, and methods described herein may be used for routine screening and monitoring. [Example]

[0148]

[0195] Prediction of 1-month all-cause mortality by NBSEEG. The NBSEEG score can detect delirium and independently predict early mortality in elderly patients, as early as 30 days or less. Yes, the NBSEEG score can predict mortality in patients with dementia.

[0149]

[0196] Results: In both the replication cohort (N = 228) and the combined cohort (N = 502), the 180-day mortality rate was higher in the NBSEEG-positive group than in the NBSEEG-negative group. Mortality rates showed a dose-dependent increase in both cohorts. The 30-day mortality rate in the NBSEEG-positive group was significantly higher than in the NBSEEG-negative group (relative risk = 3.65%; 95% CI, 1.73 to 7.69; P < 0.001). In patients with dementia who tested positive for NBSEEG, their mortality rates were significantly higher at both 60 days (relative risk = 3.00; 95% CI, 1.17 to 7.70; P = 0.025) and 90 days (relative risk = 3.80; 95% CI, 1.52 to 9.48; P = 0.002) than in patients with dementia who tested negative for NBSEEG.

[0150]

[0197] Delirium was screened using the following questionnaires: the CAM-ICU (Intensive Care Unit), the Delirium Rating Scale-Revised 98 (DRS-R-98), and the Delirium Observation Screening Scale (DOSS). Delirium status was defined according to the following screening results: a positive CAM-ICU, a DRS-R-98 score ≥ 19, or a DOSS score ≥ 3. Baseline cognitive function was measured using the Montreal Cognitive Assessment (MoCA). Dementia was recorded based on medical record review. Delirium and dementia status were finally adjudicated by a board-certified consultant liaison psychiatrist using the results of the measurements and detailed medical record review.

[0151]

[0198] NBSEEG data were collected using a portable EEG device such as the one shown in Figure 10A.

[0199] All statistical analyses were performed using R. T-tests were performed to compare continuous data between cases and controls for delirium and dementia, and for positive and negative NBSEEG scores. A log-rank test was performed to compare two survival time functions at 180 days. To examine how early NBSEEG can differentiate mortality risk, we also compared the mortality rates of both the NBSEEG-positive and -negative groups at 30 days. Furthermore, we calculated the relative risk of death at 30 days between the NBSEEG-positive and -negative groups. Cox proportional hazards regression analysis was performed to calculate hazard ratios adjusting for age, sex, and the Charlson Comorbidity Index (CCI). A p-value of less than 0.05 was determined as statistically significant.

[0152]

[0200] Results—Replication of the Utility of NBSEEG in Predicting Mortality. Analysis of data from 228 subjects (replication cohort) confirmed the utility of NBSEEG in predicting mortality. Demographic characteristics of the subjects are shown in Table 7. Age and CCI were significantly higher in patients with delirium, dementia, and the NBSEEG-positive group compared with their respective control groups (Table 7). The proportion of women was significantly higher in patients with dementia compared with the control group (Table 7). The unadjusted 180-day mortality rate in the NBSEEG-positive group was higher than that in the NBSEEG-negative group (Figure 17A). When patients were divided into three categories based on their NBSEEG scores—NBSEEG low, intermediate, and high—with approximately equal sample sizes, mortality showed a dose-dependent increase based on the NBSEEG category (Figure 17B). The results of a Cox proportional hazards model adjusted for age, sex, CCI, and delirium showed that NBSEEG was a significant predictor of 180-day mortality (95% CI, 1.33 to 6.00; P = 0.007) (Table 8). Furthermore, age and CCI were significant predictors of death (Table 8).

[0153]

[0201] The demographic characteristics of the 502 analyzed subjects are shown in Table 9. Age and CCI were significantly higher in patients in the delirium, dementia, and NBSEEG-positive groups compared with the respective control groups (Table 9). The overall mortality rate was higher in the NBSEEG-negative group than in the NBSEEG-negative group (Figure 17C). Furthermore, when patients were divided into three categories based on NBSEEG score—low, intermediate, and high—to achieve approximately equal sample sizes, the mortality rate showed a score-dependent increase based on the NBSEEG category (Figure 17D). The results of a Cox proportional hazards model adjusted for age, sex, CCI, and delirium indicated that NBSEEG was a significant predictor of 180-day mortality (95% CI, 1.55 to 3.82; P < 0.001) (Table 10). Age, delirium status, and CCI were significant predictors of death (Table 10).

[0154]

[0202] Usefulness of BSEEG in predicting mortality in patients with and without dementia: To examine the usefulness of NBSEEG for predicting mortality in patients with dementia, 502 subjects were analyzed. Patients with dementia who were NBSEEG-positive had a higher mortality rate than patients with dementia who were NBSEEG-negative (Figure 18). When dementia was added as a covariate in the Cox proportional hazards model, BSEEG remained a significant predictor of 180-day mortality (95% CI, 1.55 to 3.82; P < 0.001) (Table 11).

[0155]

[0203] Usefulness of NBSEEG in predicting short-term mortality. To examine how early NBSEEG can differentiate mortality risk among a total of 502 subjects (discovery and replication cohorts), we compared the 30-, 60-, and 90-day mortality rates. The 30-day mortality rate in the NBSEEG-positive group was significantly higher than that in the NBSEEG-negative group (relative risk = 3.65; 95% CI, 1.73 to 7.69; P < 0.001) (Figure 19A). Similarly, the 60-day mortality rate in the NBSEEG-positive group was significantly higher than that in the NBSEEG-negative group (relative risk = 2.96; 95% CI, 1.74 to 5.03; P < 0.001) (Figure 19B). The 90-day mortality rate in the NBSEEG-positive group was also significantly higher than that in the NBSEEG-negative group (relative risk = 2.86; 95% CI, 1.78 to 4.60; P < 0.001) (Figure 19C).

[0156]

[0204] Among a total of 502 subjects (discovery and replication cohorts), short-term mortality was analyzed to demonstrate differences between patients with and without dementia. The 60-day mortality rate for the NBSEEG-positive group with dementia was significantly higher than that for the NBSEEG-negative group (relative risk = 3.00; 95% CI, 1.17 to 7.70; p = 0.025) and significantly higher than that for those without dementia (relative risk = 2.78; 95% CI, 1.46 to 5.28; p < 0.001) (Figure 20). Similarly, the 90-day mortality rate for the NBSEEG-positive group with dementia was significantly higher than that for the NBSEEG-negative group (relative risk = 3.80; 95% CI, 1.52 to 9.48; p = 0.002) and significantly higher than that for those without dementia (relative risk = 2.39; 95% CI, 1.35 to 4.22; p = 0.003) (Figure 20).

[0157]

[0205] This study demonstrated the usefulness of NBSEEG in predicting mortality in an independent cohort through a replication study. Furthermore, the mortality rate of patients with dementia who had a high NBSEEG score was higher than that of patients with dementia who had a negative NBSEEG score. This result was consistent with our hypothesis that NBSEEG scores can predict mortality in patients with dementia. To the best of our knowledge, this is the first study to demonstrate the usefulness of NBSEEG scores in predicting mortality in patients with dementia.

[0158]

[0206] NBSEEG was shown to be useful in predicting mortality in an independent cohort and a cohort with an increased sample size. Furthermore, the score-dependent increase in mortality according to the NBSEEG score was replicated as shown in the previous cohort. Because it is important to assess the risk of outcomes, including mortality, in elderly patients in order to optimize interventions and care plans, the following As shown below, many metrics have been developed to assess the risk of death. For example, the CCI is used to predict death by assessing comorbidities. Similarly, various metrics, such as the Multidimensional Prognostic Index (MPI), the Elixhauser Comorbidity System, and the Single General Self-Rated Health Survey (GSRH), have been used to predict death. However, the above metrics have the limitation of lacking a biological basis. In addition to the above metrics, the NBSEEG score has the potential to be used as an electrophysiological biomarker for predicting death.

[0159]

[0207] NBSEEG has also been shown to be useful for predicting mortality in patients with dementia. These results suggest that using NBSEEG scores, rather than simply relying on a clinical diagnosis of delirium, may be able to predict mortality in patients with dementia. Appropriate intervention can improve the prognosis of patients with delirium, but detecting delirium in patients with dementia is notoriously difficult. Therefore, prompt intervention following the detection of patients with a positive NBSEEG score may improve the prognosis of patients with or without dementia.

[0160]

[0208] Importantly, there was a significant difference in mortality between the NBSEEG-positive and -negative groups, even at 30 days. Approximately 1 in 8 NBSEEG-positive patients died at 30 days, compared with 1 in 32 NBSEEG-negative patients. Predicting short-term prognosis in elderly patients is important because their prognosis may be directly related to mortality. NBSEEG may be useful for predicting both short-term and long-term mortality in elderly patients. Furthermore, the short-term mortality rate and relative risk of NBSEEG-positive patients were higher in patients with dementia compared with patients without dementia. Approximately 1 in 3 NBSEEG-positive patients with dementia died at 90 days, compared with 1 in 6 NBSEEG-positive patients without dementia. These results suggest that NBSEEG may be useful for predicting short-term mortality in patients with dementia.

[0161]

[0209] The NBSEEG score can predict mortality in elderly patients in general, and in patients with dementia, as early as 30 days after hospitalization.

[0210] Although the present disclosure has been described with reference to preferred embodiments, workers skilled in the art will recognize that changes may be made in form and detail without departing from the spirit and scope of the disclosed devices, systems and methods. [Table 1] [Table 2] [Table 3] [Table 4] [Table 5] [Table 6] [Table 7] Table 8 Table 9 Table 10 Table 11

Claims

1. 1. A method of screening a patient for prognostic risk, comprising: recording raw BSEEG values ​​with a handheld device; normalizing the raw BSEEG values ​​to calculate NBSEEG; and outputting a prognostic NBSEEG score.

2. The NBSEEG is comparing the raw BSEEG to a BSEEG population mean; and dividing the result by the BSEEG population standard deviation.

3. 2. The method of claim 1, wherein the prognostic NBSEEG score comprises an NBSEEG positive score or an NBSEEG negative score.

4. The method of claim 1 , wherein the prognostic NBSEEG score is continuous.

5. The method of claim 1 , wherein the recording step occurs at the primary point of care.

6. 10. The method of claim 1, wherein the prognostic BSEEG correlates with at least one of hospital length of stay ("LOS"), disposition at discharge, and / or risk of mortality.

7. 1. A handheld system for patient screening for mortality risk, comprising: a. at least two sensors configured to record one or more brain frequencies; b. a processor; c. at least one module, i. Record the raw BSEEG values; ii. normalizing the raw BSEEG values ​​to calculate NBSEEG; iii. Output the prognostic NBSEEG score and at least one module configured to:

8. 8. The system of claim 7, wherein the prognostic NBSEEG correlates with at least one of hospital LOS, discharge disposition, and / or mortality risk.

9. The system of claim 7 further comprising outputting the threshold data.

10. The system of claim 7 , further comprising comparing the prognostic NBSEEG score to a threshold value.

11. The system of claim 7 further comprising a signal processing device.

12. 1. A method for screening a subject for risk of mortality, comprising: recording raw BSEEG values ​​from the subject with a handheld device; normalizing the raw BSEEG values ​​to calculate NBSEEG; and outputting a prognostic NBSEEG score.

13. 13. The method of claim 12, further comprising comparing the prognostic NSBF score to a threshold value. method.

14. The method of claim 12 , wherein the raw BSEEG values ​​are processed by a signal processing module or a feature analysis module within the handheld device.

15. 13. The method of claim 12, wherein the prognostic NBSEEG score is classified as low risk, intermediate risk, or high risk by comparison to one or more thresholds.

16. The method of claim 12 further comprising maintaining a BSEEG population norm.

17. The NBSEEG is comparing the raw BSEEG to the mean of the BSEEG population norm; and dividing the result by the BSEEG population standard deviation.

18. 20. The method of claim 17, further comprising recording subject outcome.

19. 20. The method of claim 18, wherein the BSEEG population norm is updated to include the raw BSEEG values ​​and subject outcome.

20. 20. The method of claim 19, wherein the prognostic NBSEEG correlates with at least one of hospital length of stay ("LOS") and / or hospital discharge disposition.

Images