WO2020257609A1 - Diagnostics and treatments for cardiac disease - Google Patents

Abstract

A method for predicting whether a patient suffers from an elevated pressure gradient of the cardiac outflow tract, e.g., obstructive hypertrophic cardiomyopathy (oHCM) and aortic stenosis (AS), includes receiving a plethysmography signal obtained from the patient. The plethysmography signal comprises a plurality of features related to measurements of left ventricle outflow tract obstruction (LVOTO), LV diastolic function and/or LV systolic function of the patient. The method further includes determining a score indicative of elevated pressure in the cardiac outflow tract and a likelihood that the patient is suffering from a disease associated with such event or is at risk at developing a disease associated with such event (including, but not limited to oHCM or AS), using a predictive model. The method also includes outputting the determined indicator score (e.g., oHCM score or AS score).

Description

Diagnostics and Treatments for Cardiac Disease

CROSS REFERENCE TO RELATED APPLICATIONS

[0001] This PCT application claims the benefit of United States provisional application nos. 62/864,910, filed June 21, 2019, and 62/876,484, filed July 19, 2019, the entire contents of both of which are hereby incorporated by reference.

TECHNICAL FIELD

[0002] This disclosure relates to diagnosing and treating a patient who knowingly or unknowingly has an elevated pressure gradient in his/her cardiac outflow tract. Diseases that are associated with an elevated pressure gradient in the cardiac outflow tract include obstructive hypertrophic cardiomyopathy (oHCM) and aortic stenosis (AS). In an example implementation, the likelihood of oHCM or AS can be predicted with a machine learning model.

BACKGROUND

[0003] Hypertrophic cardiomyopathy (HCM) is a disease of the heart muscle characterized by hypertrophy without a systemic etiology. Patients with HCM are at increased risk of heart failure, stroke and sudden cardiac death. Approximately one in three HCM patients have outflow tract obstruction (oHCM) at rest or with exertion, which is associated with worse clinical outcomes. Characteristic hemodynamic abnormalities in patients with outflow tract obstruction were noted in the earliest descriptions of HCM and have been observed using arterial pressure tracings, acoustic phonography and echocardiography.

[0004] Echocardiographic screening is a standard diagnostic for HCM. However, efforts to use electrocardiography (ECG) or echocardiography to screen asymptomatic individuals for HCM have been limited by test characteristics and cost. Further, the non specific symptoms of HCM (exercise intolerance, dyspnea, or fatigue) can delay referral of even symptomatic individuals for diagnostic cardiac imaging studies.

[0005] With increased life expectancy and aging of the population, aortic stenosis is now one of the most common valvular heart diseases. Early recognition and management of aortic stenosis are of paramount importance because untreated symptomatic severe disease is universally fatal. The progression of aortic valve disease and the need for eventual aortic valve replacement varies considerably.

[0006] In general, progression from aortic valve sclerosis to valve obstruction is seen in approximately 10% to 15% of patients over a period of 2 to 5 years; however, progression is universal and valve replacement is eventually required once mild valve obstruction is identified.

[0007] The progression of disease in patients with elevated pressure gradients in their cardiac outflow tract can be rapid or it can be slow. In either case, patients may not be seeing their physician at intervals that are optimal to identify or treat their diseases. Thus, there is a need for better approaches to properly direct potentially unrecognized patients for definitive diagnostic studies for disease status, to monitor“at risk” patients for disease progression, and/or to more timely and optimally treat such diseases in relation to disease progression.

SUMMARY

[0008] One aspect of the disclosure provides a method for training a predictive model for predicting whether a patient suffers from a cardiac disease characterized by an elevation of a pressure gradient in the patient’s cardiac outflow tract. The method includes receiving, at data processing hardware, a training dataset comprising a plurality of features extracted from training plethysmography signals. The training

plethysmography signals are obtained from a first group of subjects, each suffering from a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract, and from a second group of subjects who are healthy. The plurality of features are related to measurements of left ventricle outflow tract obstruction (LVOTO), diastolic function, and/or LV systolic function of the subjects. The method further includes training, by the data processing hardware, a predictive model of the cardiac disease with the training dataset. The predictive model of the cardiac disease is trained according to a multiple instance learning via embedded instance selection (MILES) approach with yet- another-radial-distance-based-similarity (YARDS) measure. [0009] Another aspect of the disclosure provides a method for predicting whether a patient is at risk of or suffers from a cardiac disease characterized by an elevation of a pressure gradient in the patient’s cardiac outflow tract. The method includes receiving, at data processing hardware, a plethysmography signal obtained from a patient. The plethysmography signal comprises a plurality of features related to measurements of LVOTO, LV diastolic function, and/or LV systolic function of the patient. The method further includes determining, by the data processing hardware, an indicator score indicating a likelihood of the patient is at risk of or suffering from the cardiac disease using a predictive model. The predictive model is configured to receive the plurality of features as inputs and is trained on a corpus of training plethysmography signals according to a multiple instance learning via embedded instance selection (MILES) approach with yet-another-radial-distance-based-similarity (YARDS) measure. Each training plethysmography signal comprises a corresponding plurality of features related to measurements of LVOTO, LV diastolic function, and/or LV systolic function. The method further includes outputting, by the data processing hardware, the determined indicator score.

[0010] Yet another aspect of the disclosure provides a method of diagnosing and treating a patient suffering from a cardiac disease characterized by an elevation of a pressure gradient in the patient’s cardiac outflow tract. The method includes diagnosing, by data processing hardware, the patient suffers from a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract of the patient based on an indicator score. The indicator score is determined by a predictive model. The predictive model is trained to predict the cardiac disease from a plurality of features related to measurements of left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function of the patient. The plurality of features are extracted from a plethysmography signal obtained from the patient. The method further includes instructing, by the data processing hardware, treatment of the diagnosed cardiac disease.

[0011] Still yet another aspect of the disclosure provides a method of determining whether a patient is at risk for a cardiac disease that is characterized by an elevation of a pressure gradient in the patient’s cardiac outflow tract. The method includes determining, by data processing hardware, a patient is at risk for the cardiac disease based on an indicator score determined by a predictive model trained to predict a risk of the cardiac disease from a plurality of features related to measurements of left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function of the patient extracted from a plethysmography signal obtained from the patient. The method further includes instructing, by the data processing hardware, the patient to seek guidance from a physician to address the determined risk for the cardiac disease.

[0012] Still another aspect of the disclosure provides a method for treating a patient suffering from a disease having an elevated pressure gradient of a cardiac outflow tract of the patient comprising: treating a patient with a drug to treat a disease causing an elevated pressure gradient of a cardiac outflow tract of the patient, wherein the patient has been diagnosed as having the elevated pressure gradient using a method of the present disclosure.

[0013] In some examples, the predictive model can be used to monitor a patient who is risk for a cardiac disease such as aortic stenosis and obstructive hypertrophic cardiomyopathy (oHCM), which are characterized by an elevation of a pressure gradient in the patient’s cardiac outflow tract. In one example, a patient to be monitored with a predictive aortic stenosis model of the present disclosure is an at risk patient for aortic stenosis such as a patient who has been diagnosed with a cardiac valve impairment (e.g., calicified valve, bicuspid valve, a unicuspid or quadricuspid aortic valve, aortic valve sclerosis) but has not yet developed or been diagnosed with aortic stenosis. In another example, a patient to be monitored with a predictive oHCM model of the present disclosure is an at risk patient for oHCM such as a patient who carries mutations associated with the development of oHCM but who has not developed or been diagnosed with oHCM. In another embodiment, a patient to be monitored with the predictive oHCM model has been diagnosed with New York Heart Association (NYHA) Class I heart failure.

[0014] The details of one or more implementations of the disclosure are set forth in the accompanying drawings and the description below. Other aspects, features, and advantages will be apparent from the description and drawings, and from the claims. DESCRIPTION OF DRAWINGS

[0015] FIG. l is a schematic view of an example system for predicting whether a patient is at risk for or suffering from a cardiac disease based on a plethysmography signal obtained from the patient.

[0016] FIG. 2A is a plot view of an example arterial pulse waveform.

[0017] FIG. 2B is a perspective view of an example plethysmography sensor that is worn around a patient’s fingertip.

[0018] FIG. 2C is a diagram illustrating the technique of photoplethysmography.

[0019] FIGS. 2D and 2E are graph views of example photoplethysmographic traces obtained from a healthy patient and a patient with obstructive hypertrophic

cardiomyopathy.

[0020] FIG. 2F is a perspective view of an example wearable medical device having a photoplethysmography sensor.

[0021] FIG. 3 is a schematic view of an example pulse engine for training a cardiac di sease predi ctive model .

[0022] FIG. 4 is a graph view of example photoplethysmographic traces, including beats with aberrant waveforms.

[0023] FIG. 5 is a schematic view of an example system for predicting a cardiac disease from a photoplethysmography signal obtained from a patient.

[0024] FIG. 6 is a flow chart of an example method for predicting whether a patient is at risk for or suffering from cardiac disease based on a plethysmography signal obtained from a patient.

[0025] FIG. 7 is a flow chart of an example method for diagnosing and treating a patient for a cardiac disease.

[0026] FIGS. 8A-8D are graph views illustrating data from an experimental protocol.

[0027] FIG. 9 is a schematic view of an example computing device that may be used to implement the systems and methods described herein.

[0028] Like reference symbols in the various drawings indicate like elements. DETAILED DESCRIPTION

[0029] The present invention relates to diagnosing and treating a patient who knowingly or unknowingly has an elevated pressure gradient in his/her cardiac outflow tract. Cardiac diseases that are associated with an elevated pressure gradient in the cardiac outflow tract include obstructive hypertrophic cardiomyopathy (oHCM) and aortic stenosis (AS). In one embodiment, the patient who is suffering from oHCM has an elevated pressure gradient that is greater than 50 millimeters of mercury (mmHg). In another embodiment, the patient having a valve impairment who is suffering from AS has a pressure gradient that is greater than 16 mmHg, greater than 36 mmHg, greater than 40 mmHg, or greater than 50 mmHg.

[0030] Hypertrophic cardiomyopathy (HCM) is a form of heart failure. The stage of a patient’ s heart failure can be classified under the New York Heart Association (NYHA) Functional Classification for heart failure, as provided in the following table.

[0031] HCM is a progressive disease that can result in shortness of breath, chest pain, an inability to participate in normal activities, disabling heart failure, stroke or sudden death. Obstructive HCM (oHCM) is a form of HCM characterized by a dynamic obstruction of the left ventricular outflow tract (LVOT) that results in abnormalities in arterial blood flow. The identification of obstruction in HCM is important for both prognostic and therapeutic reasons. The presence of obstruction portends a worse prognosis but also suggests potential established medical therapies (e.g., beta blocker, calcium channel blocker, disopyramide) and interventional therapies (e.g., septal reduction).

[0032] Stages of AS, similar to stages of heart failure, have been proposed by the current American Heart Association (AHA)/American College of Cardiology (ACC) guidelines to acknowledge the continuum of AS from those at risk for development of AS, as provided in the following table.

[0033] Existing methods for detection of ventricular and outflow tract obstruction include introduction of a pressure catheter into a patient’s left ventricle and

echocardiographic assessment of blood flow and pressure gradients in the left ventricle and outflow tract. Both of these techniques are limited, particularly for dynamic or induced obstruction (e.g. caused by dehydration, carrying out the Valsalva maneuver, ventricular premature beat or exercise), because they require trained operators, can only be performed episodically in a physician’s office, and may not reproduce the stimuli that provoke obstruction in an individual patient. This makes it difficult to correlate between study findings and a patient’s symptoms during daily activities.

[0034] FIG. 1 is a schematic view of an example diagnostic system 100 for predicting whether a patient 10 is at risk for or suffering from a cardiac disease that is characterized by an elevated pressure gradient in the patient’s cardiac outflow tract. The diagnostic system 100 includes a plethysmography sensor 200, which in the example shown is worn by the patient 10 around their wrist. The plethysmography sensor 200 generates a plethysmography signal 210. A computing device 110 collects the plethysmography signal 210 generated by the plethysmography sensor 200 and transmits the

plethysmography signal 210 over a network 114, for example, to a remote system 140. The remote system 140 may be a distributed system (e.g., a cloud or edge computing environment) having scalable/elastic computing resources 142 (e.g., data processing hardware) and/or storage resources 146 (e.g. memory hardware). In some

implementations, the remote system 140 executes a pulse engine 300 configured to receive the plethysmography signal 210 from the patient 10 and to output an indicator score 330 back to the patient 10. The indicator score 330 indicates a likelihood of the patient 10 is at risk for or is suffering from the cardiac disease.

[0035] The plethysmography sensor 200 obtains, in a non-invasive manner, information about how the volume of blood in a part of the patient’s body changes over time. The information gathered by the plethysmography sensor 200 can be represented, graphically, as an arterial pulse waveform (or simply waveform) as shown in FIG. 2A.

[0036] Referring to FIGS. 2B and 2C, an example of the plethysmography sensor 200 uses photoplethysmography and optically detects blood volume changes within the microvascular bed of the patient’s 10 fingertip. Light emitted by the plethysmography sensor 200 (e.g., by way of a LED) into the patient’s 10 fingertip is mostly absorbed by the underlying tissue. Some of the light, however, is reflected and is captured by the plethysmography sensor 200 (e.g., by way of a photodiode). The plethysmography sensor 200 samples reflected light many times within a second to construct a

plethysmography signal 210. The absorption of light varies with the changes in pulsatile arterial blood flow and generates a time-varying pulse waveform (also known as arterial pulse waves) corresponding to the plethysmography signal 210. By wearing the plethysmography sensor 200 around a part of the patient’s body, the plethysmography sensor 200 can perform blood volume measurements without penetrating the patient’s skin. Other examples of the plethysmography sensor 200 may use tonometry, radar spectroscopy, inductance plethysmography or impedance plethysmography to

detect/measure changes in a patient’s blood volume.

[0037] FIGS. 2D and 2E illustrate example photoplethysmographic traces obtained from a healthy individual (FIG. 2D) and from an individual with HCM and obstruction (FIG. 2E).

[0038] FIG. 2F shows another example of the plethysmography sensor 200. During measurement, the plethysmography sensor 200 is placed snugly on a patient’s arm above their wrist bone. The plethysmography sensor 200 comes with a removable sensor carriage 202 and a plastic band 204. The sensor carriage 202 contains one or more light emitting diodes 206 (LEDs) and an optical sensor 208 along with a battery (not shown) and an inductive charging coil (not shown). In a convenient configuration, the sensor carriage 202 contains two light emitting diodes (LEDs) 206 producing light of two wavelengths, e.g., green light, at about 520-530 nanometer (nm) and infrared light, at about 850-1000 nm. The LEDs 206 fire at a rate configurable between 20 and 95 Hz driven by a submillisecond resolution low-jitter external clock signal.

[0039] A fully integrated analog front end (not shown) receives and digitizes plethysmography 210. In some examples, the plethysmography sensor 200 digitizes plethysmography 210 with an amplitude resolution of 16 bits. Such resolution may be advantageous to predicting cardiac diseases, including oHCM and aortic stenosis, from plethysmography signals 210 according to the approaches described herein. In other examples, the plethysmography sensor 200 is also capable of collecting inertial motion data using a 3-axis accelerometer and 3-axis gyroscope built into the sensor carriage 202. The sampling rate and duty cycle of the light and motion sensors are configurable. For example, the plethysmography sensor 200 collects light sensor data at a time resolution greater than 30 Hz (e.g., 86 Hz) and collects motion data at 10 Hz.

[0040] Referring to FIG. 3, the pulse engine 300 includes a cardiac disease predictor 310 that uses a cardiac disease predictive model 320 to generate the indicator score 330. The cardiac disease predictive model 320 may be trained by a cardiac disease trainer 340 based on a training dataset 350, which may be obtained from a data store 360. In some examples, the cardiac disease predictor 310 uses a cardiac disease predictive model 320 that is configured to receive a plurality of features 212, 212a-n associated with the plethysmography signal 210 as feature inputs 212. The plurality of features 212a-n includes features 212 related to measurements of left ventricle (LV) diastolic function and/or LV systolic function of the patient 10. For example, the plurality of features 212a-n may include features 212 related to any one or any combination of following measurements: 1) a ratio of early filling (E) velocity to atrial filling (A) velocity; 2) a peak A wave velocity; 3) a peak E wave velocity; 4) an ejection fraction; 5) a LV global circumferential strain; 6) a LV global longitudinal strain; and 7) a LV outflow tract gradient. In some examples, the plurality of features 212a-n are wave features extracted from a plethysmography waveform corresponding to the plethysmography signal 210. [0041] In some implementations, the cardiac disease predictive model 320 is trained on the training dataset 350, which is obtained from a data store 360 residing on the storage resources 146 of the distributed system 140, or may reside at some other remote location in communication with the distributed system 140. The training dataset 350 includes a corpus of training plethysmography signals 210T collected from a first group of subjects who suffer from a cardiac disease, such as oHCM and aortic stenosis, and from a second group of subjects who are healthy (i.e., who are not suffering from a cardiac disease). Each training plethysmography signal 210T includes a corresponding plurality of features. For example, each training plethysmography signal 210T includes features related to measurements of left ventricle (LV) diastolic function and/or LV systolic function of each subject. The corresponding plurality of features may include features 212 related to any one or any combination of following subject measurements:

1) a ratio of early filling (E) velocity to atrial filling (A) velocity; 2) a peak A wave velocity; 3) a peak E wave velocity; 4) an ejection fraction; 5) a LV global

circumferential strain; 6) a LV global longitudinal strain; and 7) a LV outflow tract gradient.

[0042] In the example shown, the cardiac disease trainer 340 receives the training dataset 350 for training the cardiac disease predictive model 320. Based on the training dataset 350, the cardiac disease trainer 340 models cardiac disease score parameters 342 to train the cardiac disease predictive model 320. The cardiac disease predictor 310 uses the trained the cardiac disease predictive model 320 during inference for determining the indicator scores 330 for corresponding plethysmography signals 210. As such, the cardiac disease predictive model 320 is trained to determine indicator scores 330 using the training dataset 350 associated with the corpus of training plethysmography signals 21 Or_. each of which includes a corresponding plurality of features.

[0043] In some implementations, the cardiac disease predictive model 320 is trained according to a multiple instance learning via embedded instance selection (MILES) approach with yet-another-radial-distance-based-similarity (YARDS) measure. The MILES approach with YARDS measure is described in the article“A review of multi instance learning assumptions”, the entirety of which is incorporated by reference herein. (Foulds, J., & Frank, E. (2010). A review of multi-instance learning assumptions. The Knowledge Engineering Review, 25(1), 1-25.) Briefly, training the cardiac disease predictive model 320 under this approach includes: 1) for each of the training

plethysmography signals 210T, transforming feature vectors extracted from the training plethysmography signal 210T into a single vector for that training plethysmography signal and 2) fitting the resulting vectors with a support vector machine.

[0044] In some instances, the cardiac disease predictive model 320 is evaluated using Leave-One-Group-Out cross-validation (LOGO CV) with nested hyperparameter tuning. Tuning of the hyperparameters may be done using a k-fold (e.g., 68) cross-validation with random selection of training and testing cohorts (e.g., 70% testing and 30% training). In such an evaluation, given N number of subjects in the training dataset 350, for each subject, the cardiac disease predictive model 320 is trained and tuned using all plethysmography signals except for plethysmography signals for that subject. In this way, for N number of subjects, the LOGO CV runs N number of folds. In each fold, a validation sample corresponds to an excluded subject (i.e., the subject left out) and N-l number of test samples correspond to the N-l number of non-ex eluded subjects.

[0045] The hyperparameters in each fold of the N number of LOGO CV folds are selected by the k-fold random subject split cross-validation. That is, for each subject of the N number of subjects, the predictive model 320 trains on the plethysmography signals of the other N-l number of subjects and evaluates it subsequently on the excluded subject’s plethysmography signals. To ensure that the excluded subject (i.e., the validation sample) is entirely new to the predictive model 320, hyperparameter tuning excludes that subject. Thus, in each fold of the N number of LOGO CV folds, hyperparameter tuning by cross-validation is performed on a training dataset without the plethysmography signals from the excluded subject (viz., a N-l subject training set).

This nested cross-validation performs k number of folds (e.g., 68) of splitting of the N-l number of non-excluded subjects randomly into training (e.g., 70%) and testing (e.g., 30%) cohorts to select LOGO-fold-specific hyperparameters. The selected

hyperparameters are then used during training of the predictive model 320 on the N-l number of non-excluded subjects, which in turn are used to evaluate the LOGO CV fold on the excluded subject’s plethysmography signals. Because plethysmography signals of the excluded subject are evaluated by an instance of the predictive model 320 that has not “seen” the excluded subject, the plethysmography signals of all N number of subjects are included in a final confusion matrix describing the performance of the predictive model 320.

[0046] For some applications, the pulse engine 300 performs an initial processing step or steps on the plethysmography signal 210 prior to predicting a cardiac disease, including oHCM and aortic stenosis. In these applications, it can be said that the pulse engine 300 combines digital signal processing and machine learning to generate an indicator score from sensor signals. For example, the pulse engine 300 applies an eighth- order Butterworth low-pass filter (e.g., with a 3 decibel point (dB) at 18 hertz (Hz)) to the plethysmography signal 210. The low-pass filtering is designed to remove high frequency noise from the plethysmography signal 210, which may lead to more accurate prediction of a cardiac disease. Zero-phase filtering can also be implemented, which involves filtering the plethysmography signal 210 in both forward and backward directions, to eliminate phase distortion.

[0047] Another example of initial processing (or digital signal processing) by the pulse engine 300 includes generating a plethysmography waveform (arterial pulse waveform) from the plethysmography signal 210 and separating out the plethysmography waveform into individual beats. The plethysmography signal 210 can be segmented into beats by: 1) detecting peaks in the plethysmography signal 210 using, for example, a wavelet transform-based peak detection algorithm; 2) from the peaks detected, selecting peaks representing systolic onsets using, for example, time and frequency domain heuristics; and 3) identifying segments of the plethysmography signal 210 from one systolic onset to the next as beats. The location of the peaks in the plethysmography signal 210 can be refined using interpolation techniques to improve the signal time resolution.

[0048] FIG. 4 shows a plethysmography waveform 402 for patient 10, who is identified as“oHCM subject-1”. The plethysmography waveform 402 is generated from a plethysmography signal acquired from patient 10 and represents or corresponds to a set of beats collected over a period of time from patient 10. It may be beneficial, however, to determine the indicator score 330 for patient 10 based on a subset of those beats. For example, any beat that fails to satisfy a signal quality metric is determined to have an “aberrant” waveform and is filtered out or removed from the prediction making process. Beats having aberrant waveforms are labeled in the figure with Xs. Beats having a “normal” waveform, i.e., satisfying the signal quality metric, are labeled in the figure with Os. In the example shown, the plethysmography waveform 402 includes more beats with aberrant waveforms than beats with normal waveforms. In this case, it may be said that the quality of the plethysmography signal used to generate the plethysmography waveform 402 is low or“unsatisfactory”. The entire plethysmography signal may be not used to determine the indicator score 330 because of the lack of signal quality. In some instances, another plethysmography signal acquired from patient 10 is used instead. As such, the foregoing provides a quality control for the prediction making process. By way of comparison, a majority of beats of plethysmography waveform 404 for another patient identified as“oHCM subject-2” have normal waveforms. As such, it may be said that the quality of a plethysmography signal used to generate the plethysmography waveform 404 is high or“satisfactory”.

[0049] Whether a beat has an aberrant waveform may be determined, e.g., by the data processing hardware 144 of FIG. 1, using an artifact motion associated with the beat, a short-time Fourier transformation of the beat, a correlation of the beat with other beats from the plethysmography signal, other heuristic, or a combination of the

aforementioned. Motion artifacts caused by, for example, the plethysmography sensor 200 moving while the sensor 200 is acquiring data can lead to a plethysmography signal having beats that do not accurately represent the condition of the heart of the patient 10. To improve accuracy, the computing device 110 (FIG. 1) may discard beats that are acquired when a net acceleration amplitude is greater than a threshold motion (e.g., g/10, where g is the acceleration of gravity). In another example, a beat having an aberrant waveform is discarded when an average amplitude of the aberrant waveform over a threshold period of time (e.g., 60 seconds) is less than a threshold amplitude (e.g., 500 microvolts or 0.01% of a direct current signal component of a plethysmography signal corresponding to the beat).

[0050] A waveform shape or morphology of a beat can also be used to identify whether the beat has an aberrant waveform. For example, a beat without a clear systolic ramp up and diastolic ramp down with waves outside 0.3 Hz to 10Hz frequency domain is identified as having an aberrant waveform and is discarded. Comparing morphologies of a given beat to other beats in the same plethysmography signal can also be used to identify whether the given beat has an aberrant waveform. For example, a beat having frequency content over a waveform segment (e.g., two seconds) that is significantly different from frequency content of previous beats over previous waveform segments (e.g. having a threshold difference of 80% or more) is identified as having an aberrant waveform and is filtered out.

[0051] Some implantations of the pulse engine 300 assess a patient’s health with respect to a cardiac disease based on the indicator score 330 and provide the assessment to the patient 10 or to their healthcare provider. For example, the pulse engine 300 compares the indicator score 330 against a threshold value. If the indicator score 330 is greater than the threshold value, then the pulse engine 300 informs the patient/healthcare provider that the patient 10 is predicted to have or be at risk of having a cardiac disease, such as oHCM or aortic stenosis. Alternatively, the pulse engine 300 notifies the patient/healthcare provider the likelihood of the patient 10 having or risk of having the cardiac disease as a probability (e.g., 25%, 50%, or 75%) or a rating (e.g.,“likely” or“not likely”). Such assessment can then be used by the patient’s healthcare provider, for example, to order additional tests (e.g., echocardiogram or ECG) or to prescribe medication.

[0052] Other examples of the pulse engine 300 provide an action plan that is specific to a patient based on the indicator score 330 predicted for that patient. Providing such action plan may include determining and/or causing the execution of a suitable medication dosing regimen, such as the administration of an oHCM drug (e.g., mavacamten) or a cardiac myosin inhibitor. Developing the patient-specific medication dosing regimen as a function of plethysmography signals acquired from the patient (and from the information carried within those signals) may eliminate or reduce the trial-and- error aspect of conventional dosing regimen development. Additionally, the determined indicator score 330 can also be used to evaluate a patient’s response to a medication dosing regimen that is prescribed to them. Advantageously, the ability for these examples of the pulse engine 300 to accurately predict a cardiac disease or risk of a cardiac disease from plethysmography signals coupled with the ubiquitous and non- intrusive nature of wearable medical devices with plethysmography sensors, provide a broadly available and inexpensive solution for screening and treating cardiac disease, such oHCM and aortic stenosis.

[0053] FIG. 5 shows an example diagnostic system 500 for predicting a cardiac disease from a photoplethysmography (PPG) signal. The diagnostic system 500 obtains the PPG signal from a pulse oximetry sensor 502, which is worn around a patient’s finger 504, through a pulse oximeter front end 506. A processor 508 obtains plethysmographic information carried in the PPG signal from the pulse oximeter front end 506 and generates an arterial pulse waveform from that information based on changes detected in the arterial blood flow over time. In some example, the functionality of pulse oximeter front end 506 may be incorporated into the processor 508. While FIG. 5 shows the pulse oximetry sensor 502 communicating via a wired connection 505, in other examples the pulse oximetry sensor 502 can communicate with the processor 508 and the pulse oximeter front end 506 wirelessly (e.g., using Bluetooth® or WiFi® ). In some embodiments, other sensors are used to detect blood flow and/or blood flow parameters with which a possible LVOT obstruction can be diagnosed.

[0054] The processor 508 analyzes the arterial pulse waveform using the predicative approach described above to find characteristic shapes or signatures indicative of a cardiac obstruction. Such shapes or signatures may be symptoms of hypertrophic cardiomyopathy. The analysis may include, for example, analysis of pressure segment slope, uniformity, amplitude, frequency, and/or pulse width. The processor 508 may also compare the waveform generated from plethysmographic information with waveforms stored in memory 510 to identify shapes and/or signatures indicative of cardiac obstruction caused, for example, by hypertrophic cardiomyopathy. The generated waveform may be identified as an obstruction signature if an obstruction is observed simultaneously via another means, such as echocardiography, thereby calibrating the device. The example shown in FIG. 5 also includes an optional output 512, such as a display, for communicating the generated waveform and/or other results of the analysis to a patient or to the patient’s healthcare provider.

[0055] FIG. 6 is a flowchart of an example method 600 for providing an indicator score 330 to a patient. The method 600 may be described with reference to FIGS. 1 and 3. The method 600 includes, at operation 602, receiving, at data processing hardware 144, a plethysmography signal 210 acquired from patient 10. The plethysmography signal 210 is associated with a plurality of features 212a-n related to measurements of outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function of the patient 10.

[0056] The method 600, at operation 604, includes determining, by the data processing hardware 144, an indicator score 330 for the patient 10 using cardiac disease predictive model 320. The cardiac disease predictive model 320 is configured to receive the plurality of features 212a-n as feature inputs 212. The cardiac disease predictive model 320 is trained on a corpus of training plethysmography signals 210_T. Each of the training plethysmography signals 210_T includes a corresponding plurality of features.

The cardiac disease predictive model 320 may be trained according to the MILES approach with YARDS measure. The method 600, at operation 606, further includes outputting, by the data processing hardware 144, the determined indicator score 330 to the patient 10. The determined indicator score 330 indicates a likelihood that the patient 10 is at risk for or suffering from a cardiac disease like oHCM or aortic stenosis.

[0057] FIG. 7 is a flowchart of an example method 700 for treating a cardiac disease. The method 700 may be described with reference to FIG. 1 and FIG. 3. The method 700 includes, at operation 702, diagnosing, at data processing hardware 144, that a patient 10 suffers from a cardiac disease, such as oHCM or aortic stenosis, that is characterized by an elevation of a pressure gradient in the patient’s 10 cardiac outflow tract based on an indicator score 330. The indicator score 330 is determined from a plurality of features 212a-n using cardiac disease predictive model 320. The plurality of features 212a-n relate to measurements of the patient’s left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function. The plurality of features 212a-n are extracted from a plethysmography signal 210 obtained from the patient 10. In some examples of the method 700, the cardiac disease predictive model 320 is trained on a corpus of training plethysmography signals 210T. Each of the training plethysmography signals 210T includes a corresponding plurality of features. The cardiac disease predictive model 320 may be trained according to the MILES approach with YARDS measure. Finally, the method 700 includes, at operation 704, instructing, by the data processing hardware 144, treatment of the diagnosed cardiac disease.

[0058] When the diagnosed cardiac disease is oHCM, the method 700 provides instructions for administrating an oHCM drug, such as a myosin inhibitor, a MyBP-C inhibitor, a DNA methyltransferase inhibitor, a fatty acid beta-oxidation inhibitor, a MEF2 inhibitor, a mineralocorticoid receptor antagonist (MRA), a NPR-A/C agonist, a neprilysin inhibitor, mavacamten, CYK-274, CT-G20, Trimetazidine, Valsartan, Entresto, or Spironolactone. Other examples of the method 700 provide instructions for a surgical intervention to be performed, such as a septal myectomy.

[0059] When the diagnosed cardiac disease is aortic stenosis, the method 700 provides instructions for administrating an aortic stenosis drug, such as a beta-adrenergic receptor blocker, a cardiac glycoside, a diuretics, or an angiotensin-converting enzyme inhibitor. Other examples of the method 700 provide instructions for a surgical intervention to be performed, such as a valve replacement or a transaortic valve replacement.

[0060] FIGS. 8A-8D depicts data from an experimental protocol for predicting oHCM from a photoplethysmography (PPG) signal using an oHCM predicative model. Twenty-one patients ages 18 to 70 with HCM, New York Heart Association class II- III symptoms, and a resting LVOT gradient greater than 30 mmHg were enrolled in the study. Control data were obtained from sixty-four all healthy volunteers. These participants were identified as free of cardiovascular disease by history, physical examination, and echocardiography. [0061] Study subjects underwent resting echocardiography with standard two- dimensional, M-mode, and Doppler imaging by trained sonographers. PPG signals were collected for five minutes (one to five recordings per participant) at rest using a wristworn biosensor (Wavelet Health, Mountain View, CA). PPG signals (traces) from all patients were acquired by a single investigator, who underwent training on a documented procedure that minimizes the impact of differences in environmental factors, including ambient light and temperature. All devices ran identical firmware and signal processing methods to obtain high quality signals. Signals were transmitted by

Bluetooth® to an Apple iPad® and uploaded to a cloud database for analysis.

[0062] Of the twenty-one patients, two were excluded from the study because of sensor errors during data collection. The remaining nineteen enrolled oHCM patients were 22 to 70 years old and nine (47%) were women (see table below). Participants had left ventricular hypertrophy (interventricular septal thickness 1.64 +/- 0.20 cm) with severe resting LVOT obstruction (peak pressure gradient 70.1 +/- 42.8 mmHg). All were in sinus rhythm at the time of sensor recording. The sixty-four healthy volunteers comprised the control group (see table below). They were 18 to 49 years old (38% women), and none had evidence of left ventricular hypertrophy (interventricular septal thickness 0.83 +/- 0.13 cm), LVOT obstruction, or other cardiovascular disease. Out of the sixty-four healthy volunteers, the PPG recordings for one volunteer was identified to be technically inadequate and was excluded from the study. As such, only the PPG recordings collected from the remaining eighty-two patients were further analyzed.

[0063] Continuous PPG traces revealed differences in pulsewave patterns between the control subjects and the oHCM patients. In individual beats (FIG. 8 A), PPG traces from oHCM patients often had a steeper initial rate of rise and contained multiple peaks of variable intensity. When patterns were examined across multiple beats, PPG traces from the oHCM patients showed more frequent irregularly shaped beats and greater variability from beat to beat, including with respiration, than did healthy controls (FIG. 8B). A set of morphometric features were extracted from all PPG tracings (FIG. 8B), a subset of which differed significantly between the group of healthy volunteers and the group of oHCM patients (FIG. 8C), including measures of systolic ejection time, slope of the systolic upstroke and respiratory variation. These data suggest that, in aggregate, beats from oHCM patients are morphologically distinct from healthy volunteers.

[0064] PPG recordings from the remaining eighty-three subjects were segmented into beats and a multi-instance classifier was trained to assign each recording an oHCM score based on qualified beats (instances). A set of morphometric pulse features was extracted into a feature vector for each beat. The multiple-instance learning via embedded instance selection (MILES) method was used. The method consists of (i) transforming feature vectors from all beats in a recording into a single vector per recording and (ii) fitting the resulting vectors with a support vector machine.

[0065] To evaluate the MILES predictive model, Leave-One-Group-Out cross- validation (LOGO CV) was employed with nested hyperparameter tuning. Using this kind of evaluation, for each patient of the remaining eighty-two patients in the dataset, the predictive model was trained and tuned using all PPG recordings except for that patient’s PPG recordings. The hyperparameters in each of the eighty-two LOGO CV folds were selected by a sixty-eight-fold random patient split cross-validation. That is, for each of the eighty-two patients, the classifier trained on the other eighty-one patients’ recordings and evaluated it subsequently on the excluded patient’s recordings; thus, the LOGO CV ran eighty-two folds. [0066] To ensure that the excluded patient was entirely new to the classifier, hyperparameter tuning excluded that patient. Thus, in each LOGO CV fold,

hyperparameter tuning by cross-validation was performed on an eighty-one-patient training set. This nested cross-validation performed sixty-eight folds of splitting eighty- one patients randomly into training (70%) and testing (30%) cohorts solely to select LOGO-fold-specific hyperparameters. These hyperparameters were then used during training of the classifier on the eighty-one non-excluded patients, which in turn was used to evaluate the LOGO fold on the excluded patient’s recordings. Because all the patient’s recordings were evaluated by an instance of the classifier that never saw the patient, all recordings of the eighty-two patients were included in a final confusion matrix describing the performance of the classifier.

[0067] With additional reference to FIGS. 1 and 3, an oHCM predictive model 320 (e.g., an automated classifier) may distinguish recordings from oHCM patients and healthy volunteers. Although significant differences may exist in many morphometric pulse features, the substantial beat-to-beat variability within individual PPG traces (as illustrated in FIG. 8B) may limit the performance of predicative models based on averaging beats across a recording. To best account for this heterogeneity, the pulse engine 300 may employ a multi -instance classifier to calculate an oHCM score 330 for each recording. After training and cross-validation, the oHCM predictive model 320 may achieve a C-statistic for oHCM detection of 0.99 (95% Cl, 0.99-1.0). At an operating threshold that optimizes the sum of sensitivity (95%) and specificity (98%), the oHCM predictive model 320 may correctly classify 18/19 patients with oHCM and 62/63 healthy volunteers (98% accuracy) (FIG. 8D). The oHCM predictive model 320 thus achieves discrimination between patients with oHCM and healthy controls.

[0068] FIG. 9 is schematic view of an example computing device 900 that may be used to implement the systems and methods described in this document. The computing device 900 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The components shown here, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed in this document.

[0069] The computing device 900 includes a processor 910, memory 920, a storage device 930, a high-speed interface/controller 940 connecting to the memory 920 and high-speed expansion ports 950, and a low speed interface/controller 960 connecting to a low speed bus 970 and a storage device 930. Each of the components 910, 920, 930, 940, 950, and 960, are interconnected using various busses, and may be mounted on a common motherboard or in other manners as appropriate. The processor 910 can process instructions for execution within the computing device 900, including instructions stored in the memory 920 or on the storage device 930 to display graphical information, including, for example, indicator score 330 of FIG. 1 and FIG. 3 or the action plan described above, for a graphical user interface (GUI) on an external input/output device, such as display 980 coupled to high speed interface 940. In other implementations, multiple processors and/or multiple buses may be used, as appropriate, along with multiple memories and types of memory. Also, multiple computing devices 900 may be connected, with each device providing portions of the necessary operations (e.g., as a server bank, a group of blade servers, or a multi -processor system).

[0070] The memory 920 stores information non-transitorily within the computing device 900. The memory 920 may be a computer-readable medium, a volatile memory unit(s), or non-volatile memory unit(s). The non-transitory memory 920 may be physical devices used to store programs (e.g., sequences of instructions) or data (e.g., program state information) on a temporary or permanent basis for use by the computing device 900. Examples of non-volatile memory include, but are not limited to, flash memory and read-only memory (ROM) / programmable read-only memory (PROM) / erasable programmable read-only memory (EPROM) / electronically erasable programmable read only memory (EEPROM) (e.g., typically used for firmware, such as boot programs). Examples of volatile memory include, but are not limited to, random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), phase change memory (PCM) as well as disks or tapes. [0071] The storage device 930 is capable of providing mass storage for the computing device 900. In some implementations, the storage device 930 is a computer- readable medium. In various different implementations, the storage device 930 may be a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory or other similar solid state memory device, or an array of devices, including devices in a storage area network or other configurations. In additional implementations, a computer program product is tangibly embodied in an information carrier. The computer program product contains instructions that, when executed, perform one or more methods, such as those described above. The information carrier is a computer- or machine-readable medium, such as the memory 920, the storage device 930, or memory on processor 910.

[0072] The high speed controller 940 manages bandwidth-intensive operations for the computing device 900, while the low speed controller 960 manages lower bandwidth intensive operations. Such allocation of duties is exemplary only. In some

implementations, the high-speed controller 940 is coupled to the memory 920, the display 980 (e.g., through a graphics processor or accelerator), and to the high-speed expansion ports 950, which may accept various expansion cards (not shown). In some

implementations, the low-speed controller 960 is coupled to the storage device 930 and a low-speed expansion port 990. The low-speed expansion port 990, which may include various communication ports (e.g., USB, Bluetooth®, Ethernet, wireless Ethernet), may be coupled to one or more input/output devices, such as a keyboard, a pointing device, a scanner, or a networking device such as a switch or router, e.g., through a network adapter.

[0073] The computing device 900 may be implemented in a number of different forms, as shown in the figure. For example, it may be implemented as a standard server 900a or multiple times in a group of such servers 900a, as a laptop computer 900b, or as part of a rack server system 900c.

[0074] Various implementations of the systems and techniques described herein can be realized in digital electronic and/or optical circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and/or

interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0075] These computer programs (also known as programs, software, software applications or code) include machine instructions for a programmable processor, and can be implemented in a high-level procedural and/or object-oriented programming language, and/or in assembly/machine language. As used herein, the terms“machine-readable medium” and“computer-readable medium” refer to any computer program product, non- transitory computer readable medium, apparatus and/or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term “machine-readable signal” refers to any signal used to provide machine instructions and/or data to a programmable processor.

[0076] The processes and logic flows described in this specification can be performed by one or more programmable processors, also referred to as data processing hardware, executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer.

Generally, a processor will receive instructions and data from a read only memory or a random access memory or both. The essential elements of a computer are a processor for performing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices. Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0077] To provide for interaction with a user, one or more aspects of the disclosure can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube), LCD (liquid crystal display) monitor, or touch screen for displaying information to the user and optionally a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer. Other kinds of devices can be used to provide interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user’s client device in response to requests received from the web browser.

[0078] A number of implementations have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the disclosure. While the systems and methods disclosed herein are largely discussed in terms of HCM with obstruction, they can also be used to diagnose other conditions and hemodynamic abnormalities, such as HCM without obstruction, dilated cardiomyopathy, restrictive cardiomyopathy, valvular heart disease (such as aortic stenosis and aortic regurgitation), heart failure with reduced or preserved ejection fraction, etc. Accordingly, other implementations are within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method comprising:

receiving, at data processing hardware, a training dataset comprising a plurality of features extracted from training plethysmography signals obtained from:

a first group of subjects, each subject of the first group suffering from a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract; and

a second group of subjects, each subject of the second group being healthy,

wherein the plurality of features relate to measurements of at least one of left ventricle outflow tract obstruction (LVOTO), diastolic function, and LV systolic function of the subjects; and

training, by the data processing hardware, a predictive model of the cardiac disease with the training dataset, wherein the predictive model of the cardiac disease is trained according to a multiple instance learning via embedded instance selection (MILES) approach with yet-another-radial-distance-based-similarity (YARDS) measure.

2. The method of claim 1, wherein the cardiac disease is any one of obstructive hypertrophic cardiomyopathy and aortic stenosis.

3. The method of claim 1, wherein the plurality of features extracted from the training plethysmography signals relate to a ratio of early filling (E) velocity to atrial filling (A) velocity, a peak A wave velocity, a peak E wave velocity, or combinations thereof for each of the subjects.

4. The method of claim 1, wherein the plurality of features extracted from the training plethysmography signals relate to an ejection fraction, a LV global

circumferential strain, a LV global longitudinal strain, or combinations thereof for each of the subjects.

5. The method of claim 1, wherein the plurality of features extracted from the training plethysmography signals relate to a LV outflow tract gradient for each of the subjects.

6. The method of claim 1, wherein the training plethysmography signals are photoplethysmography signals obtained by a photoplethysmography sensor worn by each of the subjects.

7. The method of claim 6, wherein the photoplethysmography sensor is worn around a wrist of each subject and has a time resolution greater than 30 hertz and an amplitude resolution of 16 bits.

8. The method of claim 1, wherein the training plethysmography signals are obtained from the subjects using any technique selected from a group consisting of tonometry, radar spectroscopy, inductance plethysmography, and impedance

plethysmography.

9. The method of claim 1, further comprising:

receiving, at the data processing hardware, the training plethysmography signals; segmenting, by the data processing hardware, the training plethysmography signals into beats; and

filtering out, by the data processing hardware, any beat having an aberrant waveform failing to satisfy a signal quality metric.

10. The method of claim 9, wherein the aberrant waveform fails to satisfy the signal quality metric when a net acceleration amplitude of the aberrant waveform exceeds a threshold motion.

11. The method of claim 10, wherein the threshold motion is gl 10 meter/seconds^A2 and g is the acceleration of gravity.

12. The method of claim 9, wherein the aberrant waveform fails to satisfy the signal quality metric when an average amplitude of the aberrant waveform over a threshold period of time is less than a threshold amplitude.

13. The method of claim 12, wherein for beats of a respective training

plethysmography signal, the threshold period of time is 60 seconds and the threshold amplitude is 500 microvolts or 0.01% of a direct current signal component of the respective training plethysmography signal.

14. The method of claim 9, wherein for beats of a respective training

plethysmography signal, the aberrant waveform fails to satisfy the signal quality metric when frequency content of the aberrant waveform differs from frequency content of a previous beat in the respective training plethysmography signal by a threshold difference.

15. The method of claim 14, wherein the threshold difference is 80%.

16. A method comprising:

receiving, at data processing hardware, a plethysmography signal obtained from a patient, the plethysmography signal comprising a plurality of features related to measurements of at least one of left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and LV systolic function of the patient;

determining, by the data processing hardware, an indicator score indicating a likelihood of the patient is at risk of or suffering from a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract of the patient using a predictive model configured to receive the plurality of features as inputs, the predictive model trained on a corpus of training plethysmography signals according to a multiple instance learning via embedded instance selection (MILES) approach with yet-another- radial-distance-based-similarity (YARDS) measure, each training plethysmography signal comprising a corresponding plurality of features related to the measurements of the at least one of LVOTO, LV diastolic function, and LV systolic function; and

outputting, by the data processing hardware, the determined indicator score.

17. The method of claim 16, wherein the plurality of features relate to a ratio of early filling (E) velocity to atrial filling (A) velocity, a peak A wave velocity, a peak E wave velocity, or combinations thereof for the patient.

18. The method of claim 16, wherein the plurality of features relate to an ejection fraction, a LV global circumferential strain, a LV global longitudinal strain, or combinations thereof for the patient.

19. The method of claim 16, wherein the plurality of features relate to a LV outflow tract gradient for the patient.

20. The method of claim 16, wherein the plethysmography signal is a

photoplethysmography signal obtained from a photoplethysmography sensor worn by the patient.

21. The method of claim 20, wherein the photoplethysmography sensor is worn around a wrist of the patient and has a time resolution greater than 30 hertz and an amplitude resolution of 16 bits.

22. The method of claim 16, wherein the plethysmography signal is obtained from the patient using any technique selected from a group consisting of tonometry, radar spectroscopy, inductance plethysmography, and impedance plethysmography.

23. The method of claim 16, further comprising:

segmenting, by the data processing hardware, the plethysmography signal into beats; and filtering out, by the data processing hardware, any beat having an aberrant waveform failing to satisfy a signal quality metric.

24. The method of claim 23, wherein the aberrant waveform fails to satisfy the signal quality metric when a net acceleration amplitude of the aberrant waveform exceeds a threshold motion.

25. The method of claim 24, wherein the threshold motion is gl 10 meter/seconds^A2 and g is the acceleration of gravity.

26. The method of claim 23, wherein the aberrant waveform fails to satisfy the signal quality metric when an average amplitude of the aberrant waveform over a threshold period of time is less than a threshold amplitude.

27. The method of claim 26, wherein the threshold period of time is 60 seconds and the threshold amplitude is 500 microvolts or 0.01% of a direct current signal component of the plethysmography signal.

28. The method of claim 23, wherein the aberrant waveform fails to satisfy the signal quality metric when frequency content of the aberrant waveform differs from frequency content of a previous beat in the plethysmography signal by a threshold difference.

29. The method of claim 28, wherein the threshold difference is 80%.

30. The method of claim 16, further comprising:

assessing, by the data processing hardware, health of the patient with respect to the cardiac disease based on the indicator score; and

providing the assessment to the patient or a healthcare provider.

31. The method of claim 16, further comprising: determining, by the data processing hardware, an action plan for the patient based on the indicator score; and

causing, by the data processing hardware, administration of the action plan to the patient.

32. The method of claim 31, wherein causing administration of the action plan to the patient comprises displaying the action plan on a display in communication with the data processing hardware.

33. The method of claim 31, wherein causing administration of the action plan to the patient comprises instructing execution of a dosing regimen of a drug for the patient to treat the cardiac disease.

34. The method of claim 31, wherein causing administration of the action plan to the patient comprises changing a dosing regimen of a drug for the patient to treat the cardiac disease.

35. The method of claim 33, wherein the cardiac disease is obstructive hypertrophic cardiomyopathy (oHCM); and

wherein the drug is an oHCM drug selected from a group consisting of a myosin inhibitor, a MyBP-C inhibitor, a DNA methyltransferase inhibitor, a fatty acid beta- oxidation inhibitor, a MEF2 inhibitor, a mineralocorticoid receptor antagonist (MRA), a NPR-A/C agonist, a neprilysin inhibitor, mavacamten, CYK-274, CT-G20,

Trimetazidine, Valsartan, Entresto, and Spironolactone.

36. The method of claim 33, wherein the cardiac disease is aortic stenosis (AS); and wherein the drug is an oHCM drug selected from a group consisting of a beta- adrenergic receptor blocker, a cardiac glycoside, a diuretics, and an angiotensin converting enzyme inhibitor.

37. A method comprising:

diagnosing, by data processing hardware, a patient suffers from a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract of the patient based on an indicator score determined by a predictive model trained to predict the cardiac disease from a plurality of features related to measurements of left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function of the patient extracted from a plethysmography signal obtained from the patient; and

instructing, by the data processing hardware, treatment of the diagnosed cardiac disease.

38. The method of claim 37, wherein the diagnosed cardiac disease is obstructive hypertrophic cardiomyopathy (oHCM); and

wherein instructing treatment of the diagnosed cardiac disease, includes instructing, by the data processing hardware, administration of an oHCM drug selected from a group consisting of a myosin inhibitor, a MyBP-C inhibitor, a DNA

methyltransferase inhibitor, a fatty acid beta-oxidation inhibitor, a MEF2 inhibitor, a mineralocorticoid receptor antagonist (MRA), a NPR-A/C agonist, a neprilysin inhibitor, mavacamten, CYK-274, CT-G20, Trimetazidine, Valsartan, Entresto, and

Spironolactone.

39. The method of claim 37, wherein the diagnosed cardiac disease is obstructive hypertrophic cardiomyopathy (oHCM); and

wherein instructing treatment of the diagnosed cardiac disease, includes instructing, by the data processing hardware, that surgical intervention be performed, including septal myectomy.

40. The method of claim 37, wherein the diagnosed cardiac disease is aortic stenosis (AS); and

wherein instructing treatment of the diagnosed cardiac disease, includes instructing, by the data processing hardware, administration of an AS drug selected from a group consisting of a beta-adrenergic receptor blocker, a cardiac glycoside, a diuretics, and an angiotensin-converting enzyme inhibitor.

41. The method of claim 37, wherein the diagnosed cardiac disease is aortic stenosis (AS); and

wherein instructing treatment of the diagnosed cardiac disease, includes instructing, by the data processing hardware, a surgical intervention to be performed selected from a group consisting of valve replacement and transaortic valve replacement.

42. The method of claim 37, instructing treatment of the diagnosed cardiac disease, includes instructing a change to a drug dosing regimen.

43. The method of claim 37, wherein the predictive model is trained on a corpus of training plethysmography signals according to a multiple instance learning via embedded instance selection (MILES) approach with yet-another-radial-distance-based-similarity (YARDS) measure, each training plethysmography signal comprising a corresponding plurality of features related to measurements of LVOTO, LV diastolic function, and/or LV systolic function.

44. The method of claim 37, wherein the plurality of features extracted from the plethysmography signal relate to a ratio of early filling (E) velocity to atrial filling (A) velocity, a peak A wave velocity, a peak E wave velocity, or combinations thereof for the patient.

45. The method of claim 37, wherein the plurality of features extracted from the plethysmography signal relate to an ejection fraction, a LV global circumferential strain, a LV global longitudinal strain, or combinations thereof for the patient.

46. The method of claim 37, wherein the plurality of features extracted from the plethysmography signal relate to a LV outflow tract gradient for the patient.

47. The method of claim 37, wherein the plethysmography signal is a photoplethysmography signal from a photoplethysmography sensor worn by the patient.

48. The method of claim 47, wherein the photoplethysmography sensor is worn around a wrist of the patient and has a time resolution greater than 30 hertz and an amplitude resolution of 16 bits.

49. The method of claim 37, wherein the plethysmography signal obtained from the patient is obtained using any technique selected from a group consisting of tonometry, radar spectroscopy, inductance plethysmography, and impedance plethysmography.

50. A method comprising:

determining, by data processing hardware, a patient is at risk for a cardiac disease characterized by an elevation of a pressure gradient in a cardiac outflow tract of the patient based on an indicator score determined by a predictive model trained to predict a risk of the cardiac disease from a plurality of features related to measurements of left ventricle outflow tract obstruction (LVOTO), LV diastolic function, and/or LV systolic function of the patient extracted from a plethysmography signal obtained from the patient; and

instructing, by the data processing hardware, the patient to seek guidance from a physician to address the determined risk for the cardiac disease.

51. A method for treating a patient suffering from a disease having an elevated pressure gradient of a cardiac outflow tract of the patient comprising:

treating a patient with a drug to treat a disease causing an elevated pressure gradient of a cardiac outflow tract of the patient, wherein the patient has been diagnosed as having the elevated pressure gradient using the method according to any one of claims 16-30.

52. The method of claim 51, wherein the disease is selected from the group consisting of obstructive hypertrophic cardiomyopathy (oHCM) and aortic stenosis (AS).

53. The method of claim 52, wherein the drug to treat oHCM is selected from the group consisting of a beta adrenergic blocking agent, a myosin inhibitor, a MyBP-C inhibitor, a DNA methyltransferase inhibitor, a fatty acid beta-oxidation inhibitor, a MEF2 inhibitor, a mineralocorticoid receptor antagonist (MRA), a NPR-A/C agonist, a neprilysin inhibitor, mavacamten, CYK-274, CT-G20, Trimetazidine, Valsartan,

Entresto, and Spironolactone.

54. The method of claim 52, wherein the drug to treat AS is selected from the group consisting of a beta adrenergic blocking agent (e.g., beta-adrenergic receptor blocker), a cardiac glycoside, a diuretic, and an angiotensin-converting enzyme inhibitor.