Search

Despite the recent explosion of machine learning applied to medical data, very few studies have examined algorithmic bias in any meaningful manner, comparing across algorithms, databases, and assessment metrics. In this study, we compared the biases in sex, age, and race of 56 algorithms on over 130,000 electrocardiograms (ECGs) using several metrics and propose a machine learning model design to reduce bias. Participants of the 2021 PhysioNet Challenge designed and implemented working, open-source algorithms to identify clinical diagnosis from 2- lead ECG recordings. We grouped the data from the training, validation, and test datasets by sex (male vs female), age (binned by decade), and race (Asian, Black, White, and Other) whenever possible. We computed recording-wise accuracy, area under the receiver operating characteristic curve (AUROC), area under the precision recall curve (AUPRC), F-measure, and the Challenge Score for each of the 56 algorithms. The Mann-Whitney U and the Kruskal-Wallis tests assessed the performance differences of algorithms across these demographic groups. Group trends revealed similar values for the AUROC, AUPRC, and F-measure for both male and female groups across the training, validation, and test sets. However, recording-wise accuracies were 20% higher (p 0.01) and the Challenge Score 12% lower (p = 0.02) for female subjects on the test set. AUPRC, F-measure, and the Challenge Score increased with age, while recording-wise accuracy and AUROC decreased with age. The results were similar for the training and test sets, but only recording-wise accuracy (12% decrease per decade, p 0.01), Challenge Score (1% increase per decade, p 0.01), and AUROC (1% decrease per decade, p 0.01) were statistically different on the test set. We observed similar AUROC, AUPRC, Challenge Score, and F-measure values across the different race categories. But, recording-wise accuracies were significantly lower for Black subjects and higher for Asian subjects on the training (31% difference, p 0.01) and test (39% difference, p 0.01) sets. A top performing model was then retrained using an additional constraint which simultaneously minimized differences in performance across sex, race and age. This resulted in a modest reduction in performance, with a significant reduction in bias. This work provides a demonstration that biases manifest as a function of model architecture, population, cost function and optimization metric, all of which should be closely examined in any model.

Post-traumatic stress disorder (PTSD) is an independent risk factor for incident heart failure, but the underlying cardiac mechanisms remained elusive. Impedance cardiography (ICG), especially when measured during stress, can help understand the underlying psychophysiological pathways linking PTSD with heart failure. We investigated the association between PTSD and ICG-based contractility metrics (pre-ejection period (PEP) and Heather index (HI)) using a controlled twin study design with a laboratory-based traumatic reminder stressor. PTSD status was assessed using structured clinical interviews. We acquired synchronized electrocardiograms and ICG data while playing personalized-trauma scripts. Using linear mixed-effects models, we examined twins as individuals and within PTSD-discordant pairs. We studied 137 male veterans (48 pairs, 41 unpaired singles) from Vietnam War Era with a mean (standard deviation) age of 68.5(2.5) years. HI during trauma stress was lower in the PTSD vs. non-PTSD individuals (7.2 vs. 9.3 [ohm/s

Post-traumatic stress disorder (PTSD) has been associated with cardiovascular disease (CVD), but the mechanisms remain unclear. Autonomic dysfunction, associated with higher CVD risk, may be triggered by acute PTSD symptoms. We hypothesized that a laboratory-based trauma reminder challenge, which induces acute PTSD symptoms, provokes autonomic dysfunction in a cohort of veteran twins. We investigated PTSD-associated real-time physiologic changes with a simulation of traumatic experiences in which the twins listened to audio recordings of a one-minute neutral script followed by a one-minute trauma script. We examined two heart rate variability metrics: deceleration capacity (DC) and logarithmic low frequency (log-LF) power from beat-to-beat intervals extracted from ambulatory electrocardiograms. We assessed longitudinal PTSD status with a structured clinical interview and the severity with the PTSD Symptoms Scale. We used linear mixed-effects models to examine twin dyads and account for cardiovascular and behavioral risk factors. We examined 238 male Veteran twins (age 68 ± 3 years old, 4% black). PTSD status and acute PTSD symptom severity were not associated with DC or log-LF measured during the neutral session, but were significantly associated with lower DC and log-LF during the traumatic script listening session. Long-standing PTSD was associated with a 0.38 (95% confidence interval, −0.83,− 0.08) and 0.79 (−1.30, −0 .29) standardized unit lower DC and log-LF, respectively, compared to no history of PTSD. Traumatic reminders in patients with PTSD lead to real-time autonomic dysregulation and suggest a potential causal mechanism for increased CVD risk, based on the well-known relationships between autonomic dysfunction and CVD mortality.

Sleep disturbance is associated with autonomic dysregulation, but the temporal directionality of this relationship remains uncertain. The objective of this study was to evaluate the temporal relationships between objectively measured sleep disturbance and daytime or nighttime autonomic dysregulation in a co-twin control study.A total of 68 members (34 pairs) of the Vietnam Era Twin Registry were studied. Twins underwent 7-day in-home actigraphy to derive objective measures of sleep disturbance. Autonomic function indexed by heart rate variability (HRV) was obtained using 7-day ECG monitoring with a wearable patch. Multivariable vector autoregressive models with Granger causality tests were used to examine the temporal directionality of the association between daytime and nighttime HRV and sleep metrics, within twin pairs, using 7-day collected ECG data.Twins were all male, mostly white (96%), with mean (SD) age of 69 (2) years. Higher daytime HRV across multiple domains was bidirectionally associated with longer total sleep time and lower wake after sleep onset; these temporal dynamics were extended to a window of 48 h. In contrast, there was no association between nighttime HRV and sleep measures in subsequent nights, or between sleep measures from previous nights and subsequent nighttime HRV.Daytime, but not nighttime, autonomic function indexed by HRV has bidirectional associations with several sleep dimensions. Dysfunctions in autonomic regulation during wakefulness can lead to subsequent shorter sleep duration and worse sleep continuity, and vice versa, and their influence on each other may extend beyond 24 h.

Background Heart failure (HF) is a major cause of frequent hospitalization and death. Early detection of HF symptoms using smartphone-based monitoring may reduce adverse events in a low-cost, scalable way. Objective We examined the relationship of HF decompensation events with smartphone-based features derived from passively and actively acquired data. Methods This was a prospective cohort study in which we monitored HF participants’ social and movement activities using a smartphone app and followed them for clinical events via phone and chart review and classified the encounters as compensated or decompensated by reviewing the provider notes in detail. We extracted motion, location, and social interaction passive features and self-reported quality of life weekly (active) with the short Kansas City Cardiomyopathy Questionnaire (KCCQ-12) survey. We developed and validated an algorithm for classifying decompensated versus compensated clinical encounters (hospitalizations or clinic visits). We evaluated models based on single modality as well as early and late fusion approaches combining patient-reported outcomes and passive smartphone data. We used Shapley additive explanation values to quantify the contribution and impact of each feature to the model. Results We evaluated 28 participants with a mean age of 67 years (SD 8), among whom 11% (3/28) were female and 46% (13/28) were Black. We identified 62 compensated and 48 decompensated clinical events from 24 and 22 participants, respectively. The highest area under the precision-recall curve (AUCPr) for classifying decompensation was with a late fusion approach combining KCCQ-12, motion, and social contact features using leave-one-subject-out cross-validation for a 2-day prediction window. It had an AUCPr of 0.80, with an area under the receiver operator curve (AUC) of 0.83, a positive predictive value (PPV) of 0.73, a sensitivity of 0.77, and a specificity of 0.88 for a 2-day prediction window. Similarly, the 4-day window model had an AUC of 0.82, an AUCPr of 0.69, a PPV of 0.62, a sensitivity of 0.68, and a specificity of 0.87. Passive social data provided some of the most informative features, with fewer calls of longer duration associating with a higher probability of future HF decompensation. Conclusions Smartphone-based data that includes both passive monitoring and actively collected surveys may provide important behavioral and functional health information on HF status in advance of clinical visits. This proof-of-concept study, although small, offers important insight into the social and behavioral determinants of health and the feasibility of using smartphone-based monitoring in this population. Our strong results are comparable to those of more active and expensive monitoring approaches, and underscore the need for larger studies to understand the clinical significance of this monitoring method.

Objective. High morphological variability magnitude (MVM) and microvolt T wave alternans (TWA) within an electrocardiogram (ECG) signifies increased electrical instability and risk of sudden cardiac death. However, the influence of breathing rate (BR), heart rate (HR), and signal-to-noise ratio (SNR) is unknown and may inflate measured values. Approach. We synthesize ECGs with morphologies derived from the Physikalisch-Technische Bundesanstalt Database. We calculate MVM and TWA at varying BRs, HRs and SNRs. We compare the MVM and TWA of signal with versus without breathing at varying HRs and SNRs. We then quantify the percentage of MVM and TWA estimates affected by BR and HR in a healthy population and assess the effect of removing these affected estimates on a method for classifying individuals with and without post-traumatic stress disorder (PTSD). Main results. For signals with high SNR (>15 dB), MVM is significantly increased when BRs are > 9 respirations/minute (rpm) and HRs are < 100 beats/minute (bpm). Increased TWAs are detected for HR/BR pairs of 60/15, 60/30 and 120/30 bpm/rpm. For 18 healthy participants, 8.33% of TWA windows and 66.76% of MVM windows are affected by BR and HR. On average, the number of windows with TWA elevations > 47 μV decreases by 23% after excluding regions with significant BR and HR effect. Adding HR and BR to a morphological variability feature increases the classification performance by 6% for individuals with and without PTSD. Significance. Physiological BR and HR significantly increase MVM and TWA , indicating that BR and HR should be considered separately as confounders. The code for this work has been released as part of an open-source toolbox.

Post-Traumatic Stress Disorder (PTSD) is a psychiatric condition resulting from threatening or horrifying events. We hypothesized that circadian rhythm changes, measured by a wrist-worn research watch are predictive of post-trauma outcomes. Approach: 1618 post-trauma patients were enrolled after admission to emergency departments (ED). Three standardized questionnaires were administered at week eight to measure post-trauma outcomes related to PTSD, sleep disturbance, and pain interference with daily life. Pulse activity and movement data were captured from a research watch for eight weeks. Standard and novel movement and cardiovascular metrics that reflect circadian rhythms were derived using this data. These features were used to train different classifiers to predict the three outcomes derived from week-eight surveys. Clinical surveys administered at ED were also used as features in the baseline models. Results: The highest cross-validated performance of research watch-based features was achieved for classifying participants with pain interference by a logistic regression model, with an area under the receiver operating characteristic curve (AUC) of 0.70. The ED survey-based model achieved an AUC of 0.77, and the fusion of research watch and ED survey metrics improved the AUC to 0.79. Significance: This work represents the first attempt to predict and classify post-trauma symptoms from passive wearable data using machine learning approaches that leverage the circadian desynchrony in a potential PTSD population.

OBJECTIVES Heart rate variability is controlled by the autonomic nervous system. After brain death, this autonomic control stops, and heart rate variability is significantly decreased. However, it is unknown if early changes in heart rate variability are predictive of progression to brain death. We hypothesized that in brain-injured children, lower heart rate variability is an early indicator of autonomic system failure, and it predicts progression to brain death. We additionally explored the association between heart rate variability and markers of brain dysfunction such as electroencephalogram and neurologic examination between brain-injured children who progressed to brain death and those who survived. DESIGN Retrospective case-control study. SETTING PICU, single institution. PATIENTS Children up to 18 years with a Glasgow Coma Scale score of less than 8 admitted between August of 2016 and December of 2017, who had electrocardiographic data available for heart rate variability analysis, were included. EXCLUSION CRITERIA patients who died of causes other than brain death. Twenty-three patients met inclusion criteria: six progressed to brain death (cases), and 17 survived (controls). Five-minute electrocardiogram segments were used to estimate heart rate variability in the time domain (SD of normal-normal intervals, root mean square successive differences), frequency domain (low frequency, high frequency, low frequency/high frequency ratio), Poincare plots, and approximate entropy. INTERVENTIONS None. MEASUREMENTS AND MAIN RESULTS Patients who progressed to brain death exhibited significantly lower heart rate variability in the time domain, frequency domain, and Poincare plots (p < 0.01). The odds of death increased with decreasing low frequency (odds ratio, 4.0; 95% CI, 1.2-13.6) and high frequency (odds ratio, 2.5; 95% CI, 1.2-5.4) heart rate variability power (p < 0.03). Heart rate variability was significantly lower in those with discontinuous or attenuated/featureless electroencephalogram versus those with slow/disorganized background (p < 0.03). CONCLUSIONS These results support the concept of autonomic system failure as an early indicator of impending brain death in brain-injured children. Furthermore, decreased heart rate variability is associated with markers of CNS dysfunction such as electroencephalogram abnormalities.

We conducted a prospective clinical study (n=14; 29% female) to assess the accuracy of a three-dimensional (3D) photography-based method of torso geometry reconstruction and body surface electrodes localization. The position of 74 body surface electrocardiographic (ECG) electrodes (diameter 5 mm) was defined by two methods: 3D photography, and CT (marker diameter 2 mm) or MRI (marker size 10×20 mm) imaging. Bland-Altman analysis showed good agreement in X (bias −2.5 [95% limits of agreement (LoA) −19.5 to 14.3] mm), Y (bias −0.1 [95% LoA −14.1 to 13.9] mm), and Z coordinates (bias −0.8 [95% LoA −15.6 to 14.2] mm), as defined by the CT/MRI imaging, and 3D photography. The average Hausdorff distance between the two torso geometry reconstructions was 11.17 ± 3.05 mm. Thus, accurate torso geometry reconstruction using 3D photography is feasible. Body surface ECG electrodes coordinates as defined by the CT/MRI imaging, and 3D photography, are in good agreement.

Background The risk of sudden cardiac death (SCD) is known to be dynamic. However, the accuracy of a dynamic SCD prediction is unknown. We aimed to measure the dynamic predictive accuracy of ECG biomarkers of SCD and competing non-sudden cardiac death (non-SCD). Methods Atherosclerosis Risk In Community study participants with analyzable ECGs in sinus rhythm were included (n = 15,716; 55% female, 73% white, age 54.2 ± 5.8 y). ECGs of 5 follow-up visits were analyzed. Global electrical heterogeneity and traditional ECG metrics (heart rate, QRS, QTc) were measured. Adjudicated SCD was the primary outcome; non-SCD was the competing outcome. Time-dependent area under the receiver operating characteristic curve (ROC(t) AUC) analysis was performed to assess the prediction accuracy of a continuous biomarker in a period of 3,6,9 months, and 1,2,3,5,10, and 15 years using a survival analysis framework. Reclassification improvement as compared to clinical risk factors (age, sex, race, diabetes, hypertension, coronary heart disease, stroke) was measured. Results Over a median 24.4 y follow-up, there were 577 SCDs (incidence 1.76 (95%CI 1.63–1.91)/1000 person-years), and 829 non-SCDs [2.55 (95%CI 2.37–2.71)]. No ECG biomarkers predicted SCD within 3 months after ECG recording. Within 6 months, spatial ventricular gradient (SVG) elevation predicted SCD (AUC 0.706; 95%CI 0.526–0.886), but not a non-SCD (AUC 0.527; 95%CI 0.303–0.75). SVG elevation more accurately predicted SCD if the ECG was recorded 6 months before SCD (AUC 0.706; 95%CI 0.526–0.886) than 2 years before SCD (AUC 0.608; 95%CI 0.515–0.701). Within the first 3 months after ECG recording, only SVG azimuth improved reclassification of the risk beyond clinical risk factors: 18% of SCD events were reclassified from low or intermediate risk to a high-risk category. QRS-T angle was the strongest long-term predictor of SCD (AUC 0.710; 95%CI 0.668–0.753 for ECG recorded within 10 years before SCD). Conclusion Short-term and long-term predictive accuracy of ECG biomarkers of SCD differed, reflecting differences in transient vs. persistent SCD substrates. The dynamic predictive accuracy of ECG biomarkers should be considered for competing SCD risk scores. The distinction between markers predicting short-term and long-term events may represent the difference between markers heralding SCD (triggers or transient substrates) versus markers identifying persistent substrate.

Background Sex-based differences in sudden cardiac death (SCD) exist and screening methods for SCD are inadequate. Purpose To develop sex-specific lifetime risk prediction models using electrocardiographic (ECG) global electrical heterogeneity (GEH) and clinical characteristics. Methods Participants from the Atherosclerosis Risk in Communities study with analyzable ECGs (n=14,725; age, 54.2±5.8 yrs; 55% female, 74% white) were followed up for 24.4 years (median). Traditional ECG and GEH variables were measured on 12-lead ECGs. A Cox regression model was used to develop a prediction model. In women, the final model included race, age, coronary heart disease (CHD), stroke, hypertension, diabetes, smoking, high-density lipoprotein, albumin, uric acid, education level, heart rate, QTc, sum absolute QRST integral, spatial peak QRS-T angle. In men, the final prediction model included age, race, CHD, stroke, hypertension, diabetes, total cholesterol, physical activity, smoking, serum phosphorus, albumin, chronic kidney disease, spatial area QRS-T angle, area spatial ventricular gradient (SVG) elevation and magnitude, and peak SVG magnitude. Results There were a total of 530 SCDs. Our prediction models showed robust prediction of SCD in both sexes [(Harrell's C-statistic women 0.863 (95% CI 0.845–0.882), men 0.786 (95% CI 0.786–0.803)]. In women when ECG and GEH variables were added to clinical variables, the net reclassification improved by 9% (P=0.001) (Table). In men there was no significant reclassification improvement. Net reclassification Lifetime SCD Risk: Clinical + ECG + GEH Variables Women Men 15% Total 15% Total SCD Cases 15% 0 0 20 20 0 5 74 79 Lifetime SCD Risk: Total 89 73 30 192 115 137 86 338 Clinical Variables Only Non-Cases 15% 0 28 84 112 0 56 214 270 Total 7,136 668 128 7,932 4,621 1,379 262 6,262 Conclusions We were the first to develop sex-specific lifetime SCD prediction models. The addition of ECG GEH to clinical variables improved SCD risk reclassification in women, but not in men. Prediction of SCD was more accurate in women as compared to men.

Background In patients with end‐stage kidney disease, sudden cardiac death is more frequent after a long interdialytic interval, within 6 hours after the end of a hemodialysis session. We hypothesized that the occurrence of paroxysmal arrhythmias is associated with changes in heart rate and heart rate variability in different phases of hemodialysis. Methods and Results We conducted a prospective ancillary study of the Predictors of Arrhythmic and Cardiovascular Risk in End Stage Renal Disease cohort. Continuous ECG monitoring was performed using an ECG patch, and short‐term heart rate variability was measured for 3 minutes every hour (by root mean square of the successive normal‐to‐normal intervals, spectral analysis, Poincaré plot, and entropy), up to 300 hours. Out of enrolled participants (n=28; age 54±13 years; 57% men; 96% black; 33% with a history of cardiovascular disease; left ventricular ejection fraction 70±9%), arrhythmias were detected in 13 (46%). Nonsustained ventricular tachycardia occurred more frequently during/posthemodialysis than pre‐/between hemodialysis (63% versus 37%, P =0.015). In adjusted for cardiovascular disease time‐series analysis, nonsustained ventricular tachycardia was preceded by a sudden heart rate increase (by 11.2 [95% CI 10.1–12.3] beats per minute; P CI 0.9–11.2] ms; amplitude 1.5 [95% CI 1.0–3.1] ms; peak at 02:01 [95% CI 20:22–03:16] am ; P CI −1.67 to −1.15] ms/24 h; P Conclusions Sudden increase in heart rate during/posthemodialysis is associated with nonsustained ventricular tachycardia. Every‐other‐day hemodialysis preserves circadian rhythm, but a second day without dialysis is characterized by parasympathetic withdrawal.

BackgroundMechanisms of arrhythmogenicity in hypertrophic cardiomyopathy (HCM) are not well understood.ObjectiveTo characterize an electrophysiological substrate of HCM in comparison to ischemic cardiomyopathy (ICM), or healthy individuals.MethodsWe conducted a prospective case-control study. The study enrolled HCM patients at high risk for ventricular tachyarrhythmia (VT) (n=10; age 61±9 y; left ventricular ejection fraction (LVEF) 60±9%), and three comparison groups: healthy individuals (n=10; age 28±6 y; LVEF>70%), ICM patients with LV hypertrophy (LVH) and known VT (n=10; age 64±9 y; LVEF 31±15%), and ICM patients with LVH and no known VT (n=10; age 70±7y; LVEF 46±16%). All participants underwent 12-lead ECG, cardiac CT or MRI, and 128-electrode body surface mapping (BioSemi ActiveTwo, Netherlands). Non-invasive voltage and activation maps were reconstructed using the open-source SCIRun (University of Utah) inverse problem-solving environment.ResultsIn the epicardial basal anterior segment, HCM patients had the greatest ventricular activation dispersion [16.4±5.5 vs. 13.1±2.7 (ICM with VT) vs. 13.8±4.3 (ICM no VT) vs. 8.1±2.4 ms (Healthy); P=0.0007], the largest unipolar voltage [1094±211 vs. 934±189 (ICM with VT) vs. 898±358 (ICM no VT) vs. 842±90 µV (Healthy); P=0.023], and the greatest voltage dispersion [median(interquartile range) 215(161-281) vs. 189(143-208) (ICM with VT) vs. 158(109-236) (ICM no VT) vs. 110(106-168)µV (Healthy); P=0.041]. Differences were also observed in other endo-and epicardial basal and apical segments.ConclusionHCM is characterized by a greater activation dispersion in basal segments, a larger voltage, and a larger voltage dispersion through LV.

Background: Phthalates are used as plasticizers in the manufacturing of flexible, plastic medical products. Patients can be subjected to high phthalate exposure through contact with plastic medical devices. We aimed to investigate the cardiac safety and biocompatibility of mono-2-ethylhexyl phthalate (MEHP), a phthalate with documented exposure in intensive care patients. Methods: Optical mapping of transmembrane voltage and pacing studies were performed on isolated, Langendorff-perfused rat hearts to assess cardiac electrophysiology after MEHP exposure compared with controls. MEHP dose was chosen based on reported blood concentrations after an exchange transfusion procedure. Results: Thirty-minute exposure to MEHP increased the atrioventricular node (147 versus 107 ms) and ventricular (117 versus 77.5 ms) effective refractory periods, compared with controls. Optical mapping revealed prolonged action potential duration at slower pacing cycle lengths, akin to reverse use dependence. The plateau phase of the action potential duration restitution curve steepened and became monophasic in MEHP-exposed hearts (0.18 versus 0.06 slope). Action potential duration lengthening occurred during late-phase repolarization resulting in triangulation (70.3 versus 56.6 ms). MEHP exposure also slowed epicardial conduction velocity (35 versus 60 cm/s), which may be partly explained by inhibition of Na v 1.5 (874 and 231 µmol/L half-maximal inhibitory concentration, fast and late sodium current). Conclusions: This study highlights the impact of acute MEHP exposure, using a clinically relevant dose, on cardiac electrophysiology in the intact heart. Heightened clinical exposure to plasticized medical products may have cardiac safety implications—given that action potential triangulation and electrical restitution modifications are a risk factor for early after depolarizations and cardiac arrhythmias.

BackgroundIn end-stage kidney disease the dialytic cycle relates to the rate of sudden cardiac death. We hypothesized that circadian, dialytic cycles, paroxysmal arrhythmias, and cardiovascular risk factors are associated with periodic changes in heart rate and heart rate variability (HRV) in incident dialysis patients.MethodsWe conducted a prospective ancillary study of the Predictors of Arrhythmic and Cardiovascular Risk in End Stage Renal Disease cohort (n=28; age 54±13 y; 57% men; 96% black; 33% with a history of structural heart disease; left ventricular ejection fraction 70±9%). Continuous ECG monitoring was performed using an ECG patch (Zio Patch, iRhythm) and short-term HRV was measured for three minutes every hour. HRV was measured by root mean square of the successive normal-to-normal intervals (rMSSD), high and low frequency power, Poincaré plot, and sample and Renyi entropy.ResultsArrhythmias were detected in 46% (n=13). Non-sustained ventricular tachycardia (VT) was more frequent during dialysis or within 6 hours post-dialysis, as compared to pre-or between-dialysis (63% vs. 37%, P=0.015), whereas supraventricular tachycardia was more frequent pre-/ between-dialysis, as compared to during-/ post-dialysis (84% vs. 16%, P=0.015). In adjusted for cardiovascular disease and its risk factors autoregressive conditional heteroscedasticity panel (ARCH) model, VT events were associated with increased heart rate by 11.2 (95%CI 10.1-12.3) bpm (PConclusionCardiac arrhythmias associate with dialytic phase. Regular dialytic schedule preserves physiological circadian rhythm, but the second day without dialysis is characterized by parasympathetic withdrawal and a steady increase in sympathetic predominance.Subject TermsArrhythmias, Autonomic Nervous System, Electrocardiology (ECG), Treatment.

Objective: Vast 12-lead ECGs repositories provide opportunities to develop new machine learning approaches for creating accurate and automatic diagnostic systems for cardiac abnormalities. However, most 12-lead ECG classification studies are trained, tested, or developed in single, small, or relatively homogeneous datasets. In addition, most algorithms focus on identifying small numbers of cardiac arrhythmias that do not represent the complexity and difficulty of ECG interpretation. This work addresses these issues by providing a standard, multi-institutional database and a novel scoring metric through a public competition: the PhysioNet/Computing in Cardiology Challenge 2020. Approach: A total of 66 361 12-lead ECG recordings were sourced from six hospital systems from four countries across three continents; 43 101 recordings were posted publicly with a focus on 27 diagnoses. For the first time in a public competition, we required teams to publish open-source code for both training and testing their algorithms, ensuring full scientific reproducibility. Main results: A total of 217 teams submitted 1395 algorithms during the Challenge, representing a diversity of approaches for identifying cardiac abnormalities from both academia and industry. As with previous Challenges, high-performing algorithms exhibited significant drops ( ≲ 10%) in performance on the hidden test data. Significance: Data from diverse institutions allowed us to assess algorithmic generalizability. A novel evaluation metric considered different misclassification errors for different cardiac abnormalities, capturing the outcomes and risks of different diagnoses. Requiring both trained models and code for training models improved the generalizability of submissions, setting a new bar in reproducibility for public data science competitions.

BackgroundThe risk of sudden cardiac death (SCD) is known to be dynamic. However, an accuracy of a dynamic SCD prediction, and “expiration date” of ECG biomarkers is unknown. Our goal was to measure dynamic predictive accuracy of ECG biomarkers of SCD and competing outcomes.MethodsAtherosclerosis Risk In Community study participants with analyzable digital ECGs were included (n=15,768; 55% female, 73% white, age 54.2±5.8 y). ECGs of 5 follow-up visits were analyzed. Global electrical heterogeneity (GEH) and traditional ECG metrics were measured. Adjudicated SCD served as the primary outcome; non-sudden cardiac death served as competing outcome. Time-dependent area under the (receiver operating characteristic) curve (AUC) analysis was performed to assess prediction accuracy of a continuous biomarker in a period of 3,6,9 months, and 1,2,3,5,10, and 15 years, using survival analysis framework.ResultsOver a median 24.4 y follow-up, there were 581 SCDs (incidence 1.77 (95%CI 1.63-1.92)/1,000 person-years), and 838 nonSCDs [2.55 (95%CI 2.39-2.73)]. Resting heart rate was the strongest (AUC 0.930) short-term (3-month) non-specific SCD predictor, whereas spatial peak QRS-T angle predicted specifically SCD 15 years after ECG recording (AUC 0.719). QRS duration (AUC 0.885) and QTc (AUC 0.711) short-term predicted advanced structural heart disease better than SCD. “Expiration date” for most ECG biomarkers was two years after ECG recording. GEH significantly improved reclassification of SCD risk beyond age, sex, race, diabetes, hypertension, coronary heart disease and stroke.ConclusionShort-term predictors of SCD, nonSCD, and biomarkers of long-term SCD risk differed, reflecting differences in transient vs. persistent SCD substrates.

Extracardiac factors such as respiration, fluid overload and body habitus have important effects on the ECG voltage. Vectorcardiographic (VCG) Global Electrical Heterogeneity (GEH) is associated with sudden cardiac death (SCD). Risk of SCD is especially high in end-stage renal disease patients (ESRD) on dialysis. However, extracardiac factors challenge ECG interpretation in ESRD patients. The effects of extracardiac factors on GEH have not been fully studied. To1 assess effects of extracardiac factors on ECG, we conducted a multi-scale study. An experimental data of ESRD patients and a previously developed biophysically detailed heart-torso model were used to investigate the effects of respiration, fluid overload and body habitus on the VCG and GEH.