StrainNet: Improved Myocardial Strain Analysis of Cine MRI by Deep Learning from DENSE.

Purpose: To develop a three-dimensional (two dimensions + time) convolutional neural network trained with displacement encoding with stimulated echoes (DENSE) data for displacement and strain analysis of cine MRI. Materials and Methods: In this retrospective multicenter study, a deep learning model (StrainNet) was developed to predict intramyocardial displacement from contour motion. Patients with various heart diseases and healthy controls underwent cardiac MRI examinations with DENSE between August 2008 and January 2022. Network training inputs were a time series of myocardial contours from DENSE magnitude images, and ground truth data were DENSE displacement measurements. Model performance was evaluated using pixelwise end-point error (EPE). For testing, StrainNet was applied to contour motion from cine MRI. Global and segmental circumferential strain (Ecc) derived from commercial feature tracking (FT), StrainNet, and DENSE (reference) were compared using intraclass correlation coefficients (ICCs), Pearson correlations, Bland-Altman analyses, paired t tests, and linear mixed-effects models. Results: The study included 161 patients (110 men; mean age, 61 years ± 14 [SD]), 99 healthy adults (44 men; mean age, 35 years ± 15), and 45 healthy children and adolescents (21 males; mean age, 12 years ± 3). StrainNet showed good agreement with DENSE for intramyocardial displacement, with an average EPE of 0.75 mm ± 0.35. The ICCs between StrainNet and DENSE and FT and DENSE were 0.87 and 0.72, respectively, for global Ecc and 0.75 and 0.48, respectively, for segmental Ecc. Bland-Altman analysis showed that StrainNet had better agreement than FT with DENSE for global and segmental Ecc. Conclusion: StrainNet outperformed FT for global and segmental Ecc analysis of cine MRI.Keywords: Image Postprocessing, MR Imaging, Cardiac, Heart, Pediatrics, Technical Aspects, Technology Assessment, Strain, Deep Learning, DENSE Supplemental material is available for this article. © RSNA, 2023.

An ECG-based machine learning model for predicting new-onset atrial fibrillation is superior to age and clinical features in identifying patients at high stroke risk.

BACKGROUND: Several large trials have employed age or clinical features to select patients for atrial fibrillation (AF) screening to reduce strokes. We hypothesized that a machine learning (ML) model trained to predict AF risk from 12­lead electrocardiogram (ECG) would be more efficient than criteria based on clinical variables in indicating a population for AF screening to potentially prevent AF-related stroke. METHODS: We retrospectively included all patients with clinical encounters in Geisinger without a prior history of AF. Incidence of AF within 1 year and AF-related strokes within 3 years of the encounter were identified. AF-related stroke was defined as a stroke where AF was diagnosed at the time of stroke or within a year after the stroke. The efficiency of five methods was evaluated for selecting a cohort for AF screening. The methods were selected from four clinical trials (mSToPS, GUARD-AF, SCREEN-AF and STROKESTOP) and the ECG-based ML model. We simulated patient selection for the five methods between the years 2011 and 2014 and evaluated outcomes for 1 year intervals between 2012 and 2015, resulting in a total of twenty 1-year periods. Patients were considered eligible if they met the criteria before the start of the given 1-year period or within that period. The primary outcomes were numbers needed to screen (NNS) for AF and AF-associated stroke. RESULTS: The clinical trial models indicated large proportions of the population with a prior ECG for AF screening (up to 31%), coinciding with NNS ranging from 14 to 18 for AF and 249-359 for AF-associated stroke. At comparable sensitivity, the ECG ML model indicated a modest number of patients for screening (14%) and had the highest efficiency in NNS for AF (7.3; up to 60% reduction) and AF-associated stroke (223; up to 38% reduction). CONCLUSIONS: An ECG-based ML risk prediction model is more efficient than contemporary AF-screening criteria based on age alone or age and clinical features at indicating a population for AF screening to potentially prevent AF-related strokes.

Endophenotype effect sizes support variant pathogenicity in monogenic disease susceptibility genes.

Accurate and efficient classification of variant pathogenicity is critical for research and clinical care. Using data from three large studies, we demonstrate that population-based associations between rare variants and quantitative endophenotypes for three monogenic diseases (low-density-lipoprotein cholesterol for familial hypercholesterolemia, electrocardiographic QTc interval for long QT syndrome, and glycosylated hemoglobin for maturity-onset diabetes of the young) provide evidence for variant pathogenicity. Effect sizes are associated with pathogenic ClinVar assertions (P < 0.001 for each trait) and discriminate pathogenic from non-pathogenic variants (area under the curve 0.82-0.84 across endophenotypes). An effect size threshold of ≥ 0.5 times the endophenotype standard deviation nominates up to 35% of rare variants of uncertain significance or not in ClinVar in disease susceptibility genes with pathogenic potential. We propose that variant associations with quantitative endophenotypes for monogenic diseases can provide evidence supporting pathogenicity.

Loss-of-Function FLNC Variants Are Associated With Arrhythmogenic Cardiomyopathy Phenotypes When Identified Through Exome Sequencing of a General Clinical Population.

BACKGROUND: The FLNC gene has recently garnered attention as a likely cause of arrhythmogenic cardiomyopathy, which is considered an actionable genetic condition. However, the association with disease in an unselected clinical population is unknown. We hypothesized that individuals with loss-of-function variants in FLNC (FLNCLOF) would have increased odds for arrhythmogenic cardiomyopathy-associated phenotypes versus variant-negative controls in the Geisinger MyCode cohort. METHODS: We identified rare, putative FLNCLOF among 171 948 individuals with exome sequencing linked to health records. Associations with arrhythmogenic cardiomyopathy phenotypes from available diagnoses and cardiac evaluations were investigated. RESULTS: Sixty individuals (0.03%; median age 58 years [47-70 interquartile range], 43% male) harbored 27 unique FLNCLOF. These individuals had significantly increased odds ratios for dilated cardiomyopathy (odds ratio, 4.9 [95% CI, 2.6-7.6]; P<0.001), supraventricular tachycardia (odds ratio, 3.2 [95% CI, 1.1-5.6]; P=0.048), and left-dominant arrhythmogenic cardiomyopathy (odds ratio, 4.2 [95% CI, 1.4-7.9]; P=0.03). Echocardiography revealed reduced left ventricular ejection fraction (52±13% versus 57±9%; P=0.001) associated with FLNCLOF. Overall, at least 9% of FLNCLOF patients demonstrated evidence of penetrant disease. CONCLUSIONS: FLNCLOF variants are associated with increased odds of ventricular arrhythmia and dysfunction in an unselected clinical population. These findings support genomic screening of FLNC for actionable secondary findings.

rECHOmmend: An ECG-Based Machine Learning Approach for Identifying Patients at Increased Risk of Undiagnosed Structural Heart Disease Detectable by Echocardiography.

BACKGROUND: Timely diagnosis of structural heart disease improves patient outcomes, yet many remain underdiagnosed. While population screening with echocardiography is impractical, ECG-based prediction models can help target high-risk patients. We developed a novel ECG-based machine learning approach to predict multiple structural heart conditions, hypothesizing that a composite model would yield higher prevalence and positive predictive values to facilitate meaningful recommendations for echocardiography. METHODS: Using 2 232 130 ECGs linked to electronic health records and echocardiography reports from 484 765 adults between 1984 to 2021, we trained machine learning models to predict the presence or absence of any of 7 echocardiography-confirmed diseases within 1 year. This composite label included the following: moderate or severe valvular disease (aortic/mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction <50%, or interventricular septal thickness >15 mm. We tested various combinations of input features (demographics, laboratory values, structured ECG data, ECG traces) and evaluated model performance using 5-fold cross-validation, multisite validation trained on 1 site and tested on 10 independent sites, and simulated retrospective deployment trained on pre-2010 data and deployed in 2010. RESULTS: Our composite rECHOmmend model used age, sex, and ECG traces and had a 0.91 area under the receiver operating characteristic curve and a 42% positive predictive value at 90% sensitivity, with a composite label prevalence of 17.9%. Individual disease models had area under the receiver operating characteristic curves from 0.86 to 0.93 and lower positive predictive values from 1% to 31%. Area under the receiver operating characteristic curves for models using different input features ranged from 0.80 to 0.93, increasing with additional features. Multisite validation showed similar results to cross-validation, with an aggregate area under the receiver operating characteristic curve of 0.91 across our independent test set of 10 clinical sites after training on a separate site. Our simulated retrospective deployment showed that for ECGs acquired in patients without preexisting structural heart disease in the year 2010, 11% were classified as high risk and 41% (4.5% of total patients) developed true echocardiography-confirmed disease within 1 year. CONCLUSIONS: An ECG-based machine learning model using a composite end point can identify a high-risk population for having undiagnosed, clinically significant structural heart disease while outperforming single-disease models and improving practical utility with higher positive predictive values. This approach can facilitate targeted screening with echocardiography to improve underdiagnosis of structural heart disease.

Monogenic and Polygenic Contributions to QTc Prolongation in the Population.

BACKGROUND: Rare sequence variation in genes underlying cardiac repolarization and common polygenic variation influence QT interval duration. However, current clinical genetic testing of individuals with unexplained QT prolongation is restricted to examination of monogenic rare variants. The recent emergence of large-scale biorepositories with sequence data enables examination of the joint contribution of rare and common variations to the QT interval in the population. METHODS: We performed a genome-wide association study of the QTc in 84 630 UK Biobank participants and created a polygenic risk score (PRS). Among 26 976 participants with whole-genome sequencing and ECG data in the TOPMed (Trans-Omics for Precision Medicine) program, we identified 160 carriers of putative pathogenic rare variants in 10 genes known to be associated with the QT interval. We examined QTc associations with the PRS and with rare variants in TOPMed. RESULTS: Fifty-four independent loci were identified by genome-wide association study in the UK Biobank. Twenty-one loci were novel, of which 12 were replicated in TOPMed. The PRS composed of 1 110 494 common variants was significantly associated with the QTc in TOPMed (ΔQTc/decile of PRS=1.4 ms [95% CI, 1.3 to 1.5]; P=1.1×10-196). Carriers of putative pathogenic rare variants had longer QTc than noncarriers (ΔQTc=10.9 ms [95% CI, 7.4 to 14.4]). Of individuals with QTc>480 ms, 23.7% carried either a monogenic rare variant or had a PRS in the top decile (3.4% monogenic, 21% top decile of PRS). CONCLUSIONS: QTc duration in the population is influenced by both rare variants in genes underlying cardiac repolarization and polygenic risk, with a sizeable contribution from polygenic risk. Comprehensive assessment of the genetic determinants of QTc prolongation includes incorporation of both polygenic and monogenic risk.

Generalizability and quality control of deep learning-based 2D echocardiography segmentation models in a large clinical dataset.

Use of machine learning (ML) for automated annotation of heart structures from echocardiographic videos is an active research area, but understanding of comparative, generalizable performance among models is lacking. This study aimed to (1) assess the generalizability of five state-of-the-art ML-based echocardiography segmentation models within a large Geisinger clinical dataset, and (2) test the hypothesis that a quality control (QC) method based on segmentation uncertainty can further improve segmentation results. Five models were applied to 47,431 echocardiography studies that were independent from any training samples. Chamber volume and mass from model segmentations were compared to clinically-reported values. The median absolute errors (MAE) in left ventricular (LV) volumes and ejection fraction exhibited by all five models were comparable to reported inter-observer errors (IOE). MAE for left atrial volume and LV mass were similarly favorable to respective IOE for models trained for those tasks. A single model consistently exhibited the lowest MAE in all five clinically-reported measures. We leveraged the tenfold cross-validation training scheme of this best-performing model to quantify segmentation uncertainty. We observed that removing segmentations with high uncertainty from 14 to 71% studies reduced volume/mass MAE by 6-10%. The addition of convexity filters improved specificity, efficiently removing < 10% studies with large MAE (16-40%). In conclusion, five previously published echocardiography segmentation models generalized to a large, independent clinical dataset-segmenting one or multiple cardiac structures with overall accuracy comparable to manual analyses-with variable performance. Convexity-reinforced uncertainty QC efficiently improved segmentation performance and may further facilitate the translation of such models.

Genomic Screening for Pathogenic Transthyretin Variants Finds Evidence of Underdiagnosed Amyloid Cardiomyopathy From Health Records.

BACKGROUND: New treatments for transthyretin amyloidosis improve survival, but diagnosis remains challenging. Pathogenic or likely pathogenic (P/LP) variants in the transthyretin (TTR) gene are one cause of transthyretin amyloidosis, and genomic screening has been proposed to identify at-risk individuals. However, data on disease features and penetrance are lacking to inform the utility of such population-based genomic screening for TTR. OBJECTIVES: This study characterized the prevalence of P/LP variants in TTR identified through exome sequencing and the burden of associated disease from electronic health records for individuals with these variants from a large (N = 134,753), primarily European-ancestry cohort. METHODS: We compared frequencies of common disease features and cardiac imaging findings between individuals with and without P/LP TTR variants. RESULTS: We identified 157 of 134,753 (0.12%) individuals with P/LP TTR variants (43% male, median age 52 [Q1-Q3: 37-61] years). Seven P/LP variants accounted for all observations, the majority being V122I (p.V142I; 113, 0.08%). Approximately 60% (n = 91) of individuals with P/LP TTR variants (all V122I) had African ancestry. Diagnoses of amyloidosis were limited (2 of 157 patients), although related heart disease diagnoses, including cardiomyopathy and heart failure, were significantly increased in individuals with P/LP TTR variants who were aged >60 years. Fourteen percent (7 of 49) of individuals aged ≥60 or older with a P/LP TTR variant had heart disease and ventricular septal thickness >1.2 cm, only one of whom was diagnosed with amyloidosis. CONCLUSIONS: Individuals with P/LP TTR variants identified by genomic screening have increased odds of heart disease after age 60 years, although amyloidosis is likely underdiagnosed without knowledge of the genetic variant.

Predictive Accuracy of a Clinical and Genetic Risk Model for Atrial Fibrillation.

BACKGROUND: Atrial fibrillation (AF) risk estimation using clinical factors with or without genetic information may identify AF screening candidates more accurately than the guideline-based age threshold of ≥65 years. METHODS: We analyzed 4 samples across the United States and Europe (derivation: UK Biobank; validation: FINRISK, Geisinger MyCode Initiative, and Framingham Heart Study). We estimated AF risk using the CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology AF) score and a combination of CHARGE-AF and a 1168-variant polygenic score (Predict-AF). We compared the utility of age, CHARGE-AF, and Predict-AF for predicting 5-year AF by quantifying discrimination and calibration. RESULTS: Among 543 093 individuals, 8940 developed AF within 5 years. In the validation sets, CHARGE-AF (C index range, 0.720-0.824) and Predict-AF (0.749-0.831) had largely comparable discrimination, both favorable to continuous age (0.675-0.801). Calibration was similar using CHARGE-AF (slope range, 0.67-0.87) and Predict-AF (0.65-0.83). Net reclassification improvement using Predict-AF versus CHARGE-AF was modest (net reclassification improvement range, 0.024-0.057) but more favorable among individuals aged <65 years (0.062-0.11). Using Predict-AF among 99 530 individuals aged ≥65 years across each sample, 70 849 had AF risk <5%, of whom 69 067 (97.5%) did not develop AF, whereas 28 681 had AF risk ≥5%, of whom 2264 (7.9%) developed AF. Of 11 379 individuals aged <65 years with AF risk ≥5%, 435 (3.8%) developed AF before age 65 years, with roughly half (46.9%) meeting anticoagulation criteria. CONCLUSIONS: AF risk estimation using clinical factors may prioritize individuals for AF screening more precisely than the age threshold endorsed in current guidelines. The additional value of genetic predisposition is modest but greatest among younger individuals.

3D-Encoded DENSE MRI with Zonal Excitation for Quantifying Biventricular Myocardial Strain During a Breath-Hold.

PURPOSE: Right ventricular (RV) function is increasingly recognized for its prognostic value in many disease states. As with the left ventricle (LV), strain-based measurements may have better prognostic value than typical chamber volumes or ejection fraction. Complete functional characterization of the RV requires high-resolution, 3D displacement tracking methods, which have been prohibitively challenging to implement. Zonal excitation during Displacement ENcoding with Stimulated Echoes (DENSE) magnetic resonance imaging (MRI) has helped reduce scan time for 2D LV strain quantification. We hypothesized that zonal excitation could alternatively be used to reproducibly acquire higher resolution, 3D-encoded DENSE images for quantification of bi-ventricular strain within a single breath-hold. METHODS: We modified sequence parameters for a 3D zonal excitation DENSE sequence to achieve in-plane resolution < 2 mm and acquired two sets of images in eight healthy adult male volunteers with median (IQR) age 32.5 (32.0-33.8) years. We assessed the inter-test reproducibility of this technique, and compared computed strains and torsion with previously published data. RESULTS: Data for one subject was excluded based on image artifacts. Reproducibility for LV (CoV: 6.1-9.0%) and RV normal strains (CoV: 6.3-8.2%) and LV torsion (CoV = 7.1%) were all very good. Reproducibility of RV torsion was lower (CoV = 16.7%), but still within acceptable limits. Computed global strains and torsion were within reasonable agreement with published data, but further studies in larger cohorts are needed to confirm. CONCLUSION: Reproducible acquisition of 3D-encoded biventricular myocardial strain data in a breath-hold is feasible using DENSE with zonal excitation.

Rare Coding Variants Associated With Electrocardiographic Intervals Identify Monogenic Arrhythmia Susceptibility Genes: A Multi-Ancestry Analysis.

BACKGROUND: Alterations in electrocardiographic (ECG) intervals are well-known markers for arrhythmia and sudden cardiac death (SCD) risk. While the genetics of arrhythmia syndromes have been studied, relations between electrocardiographic intervals and rare genetic variation at a population level are poorly understood. METHODS: Using a discovery sample of 29 000 individuals with whole-genome sequencing from Trans-Omics in Precision Medicine and replication in nearly 100 000 with whole-exome sequencing from the UK Biobank and MyCode, we examined associations between low-frequency and rare coding variants with 5 routinely measured electrocardiographic traits (RR, P-wave, PR, and QRS intervals and corrected QT interval). RESULTS: We found that rare variants associated with population-based electrocardiographic intervals identify established monogenic SCD genes (KCNQ1, KCNH2, and SCN5A), a controversial monogenic SCD gene (KCNE1), and novel genes (PAM and MFGE8) involved in cardiac conduction. Loss-of-function and pathogenic SCN5A variants, carried by 0.1% of individuals, were associated with a nearly 6-fold increased odds of the first-degree atrioventricular block (P=8.4×10-5). Similar variants in KCNQ1 and KCNH2 (0.2% of individuals) were associated with a 23-fold increased odds of marked corrected QT interval prolongation (P=4×10-25), a marker of SCD risk. Incomplete penetrance of such deleterious variation was common as over 70% of carriers had normal electrocardiographic intervals. CONCLUSIONS: Our findings indicate that large-scale high-depth sequence data and electrocardiographic analysis identifies monogenic arrhythmia susceptibility genes and rare variants with large effects. Known pathogenic variation in conventional arrhythmia and SCD genes exhibited incomplete penetrance and accounted for only a small fraction of marked electrocardiographic interval prolongation.

OASIS +: leveraging machine learning to improve the prognostic accuracy of OASIS severity score for predicting in-hospital mortality.

BACKGROUND: Severity scores assess the acuity of critical illness by penalizing for the deviation of physiologic measurements from normal and aggregating these penalties (also called "weights" or "subscores") into a final score (or probability) for quantifying the severity of critical illness (or the likelihood of in-hospital mortality). Although these simple additive models are human readable and interpretable, their predictive performance needs to be further improved. METHODS: We present OASIS +, a variant of the Oxford Acute Severity of Illness Score (OASIS) in which an ensemble of 200 decision trees is used to predict in-hospital mortality based on the 10 same clinical variables in OASIS. RESULTS: Using a test set of 9566 admissions extracted from the MIMIC-III database, we show that OASIS + outperforms nine previously developed severity scoring methods (including OASIS) in predicting in-hospital mortality. Furthermore, our results show that the supervised learning algorithms considered in our experiments demonstrated higher predictive performance when trained using the observed clinical variables as opposed to OASIS subscores. CONCLUSIONS: Our results suggest that there is room for improving the prognostic accuracy of the OASIS severity scores by replacing the simple linear additive scoring function with more sophisticated non-linear machine learning models such as RF and XGB.

Clinical Findings and Diagnostic Yield of Arrhythmogenic Cardiomyopathy Through Genomic Screening of Pathogenic or Likely Pathogenic Desmosome Gene Variants.

BACKGROUND: Genomic screening holds great promise for presymptomatic identification of hidden disease, and prevention of dramatic events, including sudden cardiac death associated with arrhythmogenic cardiomyopathy (ACM). Herein, we present findings from clinical follow-up of carriers of ACM-associated pathogenic/likely pathogenic desmosome variants ascertained through genomic screening. METHODS: Of 64 548 eligible participants in Geisinger MyCode Genomic Screening and Counseling program (2015-present), 92 individuals (0.14%) identified with pathogenic/likely pathogenic desmosome variants by clinical laboratory testing were referred for evaluation. We reviewed preresult medical history, patient-reported family history, and diagnostic testing results to assess both arrhythmogenic right ventricular cardiomyopathy and left-dominant ACM. RESULTS: One carrier had a prior diagnosis of dilated cardiomyopathy with arrhythmia; no other related diagnoses or diagnostic family history criteria were reported. Fifty-nine carriers (64%) had diagnostic testing in follow-up. Excluding the variant, 21/59 carriers satisfied at least one arrhythmogenic right ventricular cardiomyopathy task force criterion, 11 (52%) of whom harbored DSP variants, but only 5 exhibited multiple criteria. Six (10%) carriers demonstrated evidence of left-dominant ACM, including high rates of atypical late gadolinium enhancement by magnetic resonance imaging and nonsustained ventricular tachycardia. Two individuals received new cardiomyopathy diagnoses and received defibrillators for primary prevention. CONCLUSIONS: Genomic screening for pathogenic/likely pathogenic variants in desmosome genes can uncover both left- and right-dominant ACM. Findings of overt cardiomyopathy were limited but were most common in DSP-variant carriers and notably absent in PKP2-variant carriers. Consideration of the pathogenic/likely pathogenic variant as a major criterion for diagnosis is inappropriate in the setting of genomic screening.

Deep Neural Networks Can Predict New-Onset Atrial Fibrillation From the 12-Lead ECG and Help Identify Those at Risk of Atrial Fibrillation-Related Stroke.

BACKGROUND: Atrial fibrillation (AF) is associated with substantial morbidity, especially when it goes undetected. If new-onset AF could be predicted, targeted screening could be used to find it early. We hypothesized that a deep neural network could predict new-onset AF from the resting 12-lead ECG and that this prediction may help identify those at risk of AF-related stroke. METHODS: We used 1.6 M resting 12-lead digital ECG traces from 430 000 patients collected from 1984 to 2019. Deep neural networks were trained to predict new-onset AF (within 1 year) in patients without a history of AF. Performance was evaluated using areas under the receiver operating characteristic curve and precision-recall curve. We performed an incidence-free survival analysis for a period of 30 years following the ECG stratified by model predictions. To simulate real-world deployment, we trained a separate model using all ECGs before 2010 and evaluated model performance on a test set of ECGs from 2010 through 2014 that were linked to our stroke registry. We identified the patients at risk for AF-related stroke among those predicted to be high risk for AF by the model at different prediction thresholds. RESULTS: The area under the receiver operating characteristic curve and area under the precision-recall curve were 0.85 and 0.22, respectively, for predicting new-onset AF within 1 year of an ECG. The hazard ratio for the predicted high- versus low-risk groups over a 30-year span was 7.2 (95% CI, 6.9-7.6). In a simulated deployment scenario, the model predicted new-onset AF at 1 year with a sensitivity of 69% and specificity of 81%. The number needed to screen to find 1 new case of AF was 9. This model predicted patients at high risk for new-onset AF in 62% of all patients who experienced an AF-related stroke within 3 years of the index ECG. CONCLUSIONS: Deep learning can predict new-onset AF from the 12-lead ECG in patients with no previous history of AF. This prediction may help identify patients at risk for AF-related strokes.

Deep-learning-assisted analysis of echocardiographic videos improves predictions of all-cause mortality.

Machine learning promises to assist physicians with predictions of mortality and of other future clinical events by learning complex patterns from historical data, such as longitudinal electronic health records. Here we show that a convolutional neural network trained on raw pixel data in 812,278 echocardiographic videos from 34,362 individuals provides superior predictions of one-year all-cause mortality. The model's predictions outperformed the widely used pooled cohort equations, the Seattle Heart Failure score (measured in an independent dataset of 2,404 patients with heart failure who underwent 3,384 echocardiograms), and a machine learning model involving 58 human-derived variables from echocardiograms and 100 clinical variables derived from electronic health records. We also show that cardiologists assisted by the model substantially improved the sensitivity of their predictions of one-year all-cause mortality by 13% while maintaining prediction specificity. Large unstructured datasets may enable deep learning to improve a wide range of clinical prediction models.

Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network.

The electrocardiogram (ECG) is a widely used medical test, consisting of voltage versus time traces collected from surface recordings over the heart1. Here we hypothesized that a deep neural network (DNN) can predict an important future clinical event, 1-year all-cause mortality, from ECG voltage-time traces. By using ECGs collected over a 34-year period in a large regional health system, we trained a DNN with 1,169,662 12-lead resting ECGs obtained from 253,397 patients, in which 99,371 events occurred. The model achieved an area under the curve (AUC) of 0.88 on a held-out test set of 168,914 patients, in which 14,207 events occurred. Even within the large subset of patients (n = 45,285) with ECGs interpreted as 'normal' by a physician, the performance of the model in predicting 1-year mortality remained high (AUC = 0.85). A blinded survey of cardiologists demonstrated that many of the discriminating features of these normal ECGs were not apparent to expert reviewers. Finally, a Cox proportional-hazard model revealed a hazard ratio of 9.5 (P < 0.005) for the two predicted groups (dead versus alive 1 year after ECG) over a 25-year follow-up period. These results show that deep learning can add substantial prognostic information to the interpretation of 12-lead resting ECGs, even in cases that are interpreted as normal by physicians.

A Machine Learning Approach to Management of Heart Failure Populations.

BACKGROUND: Heart failure is a prevalent, costly disease for which new value-based payment models demand optimized population management strategies. OBJECTIVES: This study sought to generate a strategy for managing populations of patients with heart failure by leveraging large clinical datasets and machine learning. METHODS: Geisinger electronic health record data were used to train machine learning models to predict 1-year all-cause mortality in 26,971 patients with heart failure who underwent 276,819 clinical episodes. There were 26 clinical variables (demographics, laboratory test results, medications), 90 diagnostic codes, 41 electrocardiogram measurements and patterns, 44 echocardiographic measurements, and 8 evidence-based "care gaps": flu vaccine, blood pressure of <130/80 mm Hg, A1c of <8%, cardiac resynchronization therapy, and active medications (active angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker/angiotensin receptor-neprilysin inhibitor, aldosterone receptor antagonist, hydralazine, and evidence-based beta-blocker) were collected. Care gaps represented actionable variables for which associations with all-cause mortality were modeled from retrospective data and then used to predict the benefit of prospective interventions in 13,238 currently living patients. RESULTS: Machine learning models achieved areas under the receiver-operating characteristic curve (AUCs) of 0.74 to 0.77 in a split-by-year training/test scheme, with the nonlinear XGBoost model (AUC: 0.77) outperforming linear logistic regression (AUC: 0.74). Out of 13,238 currently living patients, 2,844 were predicted to die within a year, and closing all care gaps was predicted to save 231 of these lives. Prioritizing patients for intervention by using the predicted reduction in 1-year mortality risk outperformed all other priority rankings (e.g., random selection or Seattle Heart Failure risk score). CONCLUSIONS: Machine learning can be used to priority-rank patients most likely to benefit from interventions to optimize evidence-based therapies. This approach may prove useful for optimizing heart failure population health management teams within value-based payment models.