Deep Learning for Echo Analysis, Tracking, and Evaluation of Mitral Regurgitation (DELINEATE-MR).

BACKGROUND: Artificial intelligence, particularly deep learning (DL), has immense potential to improve the interpretation of transthoracic echocardiography (TTE). Mitral regurgitation (MR) is the most common valvular heart disease and presents unique challenges for DL, including the integration of multiple video-level assessments into a final study-level classification. METHODS: A novel DL system was developed to intake complete TTEs, identify color MR Doppler videos, and determine MR severity on a 4-step ordinal scale (none/trace, mild, moderate, and severe) using the reading cardiologist as a reference standard. This DL system was tested in internal and external test sets with performance assessed by agreement with the reading cardiologist, weighted κ, and area under the receiver-operating characteristic curve for binary classification of both moderate or greater and severe MR. In addition to the primary 4-step model, a 6-step MR assessment model was studied with the addition of the intermediate MR classes of mild-moderate and moderate-severe with performance assessed by both exact agreement and ±1 step agreement with the clinical MR interpretation. RESULTS: A total of 61 689 TTEs were split into train (n=43 811), validation (n=8891), and internal test (n=8987) sets with an additional external test set of 8208 TTEs. The model had high performance in MR classification in internal (exact accuracy, 82%; κ=0.84; area under the receiver-operating characteristic curve, 0.98 for moderate/severe MR) and external test sets (exact accuracy, 79%; κ=0.80; area under the receiver-operating characteristic curve, 0.98 for moderate or greater MR). Most (63% internal and 66% external) misclassification disagreements were between none/trace and mild MR. MR classification accuracy was slightly higher using multiple TTE views (accuracy, 82%) than with only apical 4-chamber views (accuracy, 80%). In subset analyses, the model was accurate in the classification of both primary and secondary MR with slightly lower performance in cases of eccentric MR. In the analysis of the 6-step classification system, the exact accuracy was 80% and 76% with a ±1 step agreement of 99% and 98% in the internal and external test set, respectively. CONCLUSIONS: This end-to-end DL system can intake entire echocardiogram studies to accurately classify MR severity and may be useful in helping clinicians refine MR assessments.

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.

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.

PheWAS and Beyond: The Landscape of Associations with Medical Diagnoses and Clinical Measures across 38,662 Individuals from Geisinger.

Most phenome-wide association studies (PheWASs) to date have used a small to moderate number of SNPs for association with phenotypic data. We performed a large-scale single-cohort PheWAS, using electronic health record (EHR)-derived case-control status for 541 diagnoses using International Classification of Disease version 9 (ICD-9) codes and 25 median clinical laboratory measures. We calculated associations between these diagnoses and traits with ∼630,000 common frequency SNPs with minor allele frequency > 0.01 for 38,662 individuals. In this landscape PheWAS, we explored results within diseases and traits, comparing results to those previously reported in genome-wide association studies (GWASs), as well as previously published PheWASs. We further leveraged the context of functional impact from protein-coding to regulatory regions, providing a deeper interpretation of these associations. The comprehensive nature of this PheWAS allows for novel hypothesis generation, the identification of phenotypes for further study for future phenotypic algorithm development, and identification of cross-phenotype associations.

Routinely reported ejection fraction and mortality in clinical practice: where does the nadir of risk lie?

AIMS: We investigated the relationship between clinically assessed left ventricular ejection fraction (LVEF) and survival in a large, heterogeneous clinical cohort. METHODS AND RESULTS: Physician-reported LVEF on 403 977 echocardiograms from 203 135 patients were linked to all-cause mortality using electronic health records (1998-2018) from US regional healthcare system. Cox proportional hazards regression was used for analyses while adjusting for many patient characteristics including age, sex, and relevant comorbidities. A dataset including 45 531 echocardiograms and 35 976 patients from New Zealand was used to provide independent validation of analyses. During follow-up of the US cohort, 46 258 (23%) patients who had undergone 108 578 (27%) echocardiograms died. Overall, adjusted hazard ratios (HR) for mortality showed a u-shaped relationship for LVEF with a nadir of risk at an LVEF of 60-65%, a HR of 1.71 [95% confidence interval (CI) 1.64-1.77] when ≥70% and a HR of 1.73 (95% CI 1.66-1.80) at LVEF of 35-40%. Similar relationships with a nadir at 60-65% were observed in the validation dataset as well as for each age group and both sexes. The results were similar after further adjustments for conditions associated with an elevated LVEF, including mitral regurgitation, increased wall thickness, and anaemia and when restricted to patients reported to have heart failure at the time of the echocardiogram. CONCLUSION: Deviation of LVEF from 60% to 65% is associated with poorer survival regardless of age, sex, or other relevant comorbidities such as heart failure. These results may herald the recognition of a new phenotype characterized by supra-normal LVEF.

Genomics-First Evaluation of Heart Disease Associated With Titin-Truncating Variants.

BACKGROUND: Truncating variants in the Titin gene (TTNtvs) are common in individuals with idiopathic dilated cardiomyopathy (DCM). However, a comprehensive genomics-first evaluation of the impact of TTNtvs in different clinical contexts, and the evaluation of modifiers such as genetic ancestry, has not been performed. METHODS: We reviewed whole exome sequence data for >71 000 individuals (61 040 from the Geisinger MyCode Community Health Initiative (2007 to present) and 10 273 from the PennMedicine BioBank (2013 to present) to identify anyone with TTNtvs. We further selected individuals with TTNtvs in exons highly expressed in the heart (proportion spliced in [PSI] >0.9). Using linked electronic health records, we evaluated associations of TTNtvs with diagnoses and quantitative echocardiographic measures, including subanalyses for individuals with and without DCM diagnoses. We also reviewed data from the Jackson Heart Study to validate specific analyses for individuals of African ancestry. RESULTS: Identified with a TTNtv in a highly expressed exon (hiPSI) were 1.2% individuals in PennMedicine BioBank and 0.6% at Geisinger. The presence of a hiPSI TTNtv was associated with increased odds of DCM in individuals of European ancestry (odds ratio [95% CI]: 18.7 [9.1-39.4] {PennMedicine BioBank} and 10.8 [7.0-16.0] {Geisinger}). hiPSI TTNtvs were not associated with DCM in individuals of African ancestry, despite a high DCM prevalence (odds ratio, 1.8 [0.2-13.7]; P=0.57). Among 244 individuals of European ancestry with DCM in PennMedicine BioBank, hiPSI TTNtv carriers had lower left ventricular ejection fraction (ß=-12%, P=3×10-7), and increased left ventricular diameter (ß=0.65 cm, P=9×10-3). In the Geisinger cohort, hiPSI TTNtv carriers without a cardiomyopathy diagnosis had more atrial fibrillation (odds ratio, 2.4 [1.6-3.6]) and heart failure (odds ratio, 3.8 [2.4-6.0]), and lower left ventricular ejection fraction (ß=-3.4%, P=1×10-7). CONCLUSIONS: Individuals of European ancestry with hiPSI TTNtv have an abnormal cardiac phenotype characterized by lower left ventricular ejection fraction, irrespective of the clinical manifestation of cardiomyopathy. Associations with arrhythmias, including atrial fibrillation, were observed even when controlling for cardiomyopathy diagnosis. In contrast, no association between hiPSI TTNtvs and DCM was discerned among individuals of African ancestry. Given these findings, clinical identification of hiPSI TTNtv carriers may alter clinical management strategies.

A genome-first approach to aggregating rare genetic variants in LMNA for association with electronic health record phenotypes.

PURPOSE: "Genome-first" approaches, in which genetic sequencing is agnostically linked to associated phenotypes, can enhance our understanding of rare variants' contributions to disease. Loss-of-function variants in LMNA cause a range of rare diseases, including cardiomyopathy. METHODS: We leveraged exome sequencing from 11,451 unselected individuals in the Penn Medicine Biobank to associate rare variants in LMNA with diverse electronic health record (EHR)-derived phenotypes. We used Rare Exome Variant Ensemble Learner (REVEL) to annotate rare missense variants, clustered predicted deleterious and loss-of-function variants into a "gene burden" (N = 72 individuals), and performed a phenome-wide association study (PheWAS). Major findings were replicated in DiscovEHR. RESULTS: The LMNA gene burden was significantly associated with primary cardiomyopathy (p = 1.78E-11) and cardiac conduction disorders (p = 5.27E-07). Most patients had not been clinically diagnosed with LMNA cardiomyopathy. We also noted an association with chronic kidney disease (p = 1.13E-06). Regression analyses on echocardiography and serum labs revealed that LMNA variant carriers had dilated cardiomyopathy and primary renal disease. CONCLUSION: Pathogenic LMNA variants are an underdiagnosed cause of cardiomyopathy. We also find that LMNA loss of function may be a primary cause of renal disease. Finally, we show the value of aggregating rare, annotated variants into a gene burden and using PheWAS to identify novel ontologies for pleiotropic human genes.

A genome-wide association study of polycystic ovary syndrome identified from electronic health records.

BACKGROUND: Polycystic ovary syndrome is the most common endocrine disorder affecting women of reproductive age. A number of criteria have been developed for clinical diagnosis of polycystic ovary syndrome, with the Rotterdam criteria being the most inclusive. Evidence suggests that polycystic ovary syndrome is significantly heritable, and previous studies have identified genetic variants associated with polycystic ovary syndrome diagnosed using different criteria. The widely adopted electronic health record system provides an opportunity to identify patients with polycystic ovary syndrome using the Rotterdam criteria for genetic studies. OBJECTIVE: To identify novel associated genetic variants under the same phenotype definition, we extracted polycystic ovary syndrome cases and unaffected controls based on the Rotterdam criteria from the electronic health records and performed a discovery-validation genome-wide association study. STUDY DESIGN: We developed a polycystic ovary syndrome phenotyping algorithm on the basis of the Rotterdam criteria and applied it to 3 electronic health record-linked biobanks to identify cases and controls for genetic study. In the discovery phase, we performed an individual genome-wide association study using the Geisinger MyCode and the Electronic Medical Records and Genomics cohorts, which were then meta-analyzed. We attempted validation of the significant association loci (P<1×10-6) in the BioVU cohort. All association analyses used logistic regression, assuming an additive genetic model, and adjusted for principal components to control for population stratification. An inverse-variance fixed-effect model was adopted for meta-analysis. In addition, we examined the top variants to evaluate their associations with each criterion in the phenotyping algorithm. We used the STRING database to characterize protein-protein interaction network. RESULTS: Using the same algorithm based on the Rotterdam criteria, we identified 2995 patients with polycystic ovary syndrome and 53,599 population controls in total (2742 cases and 51,438 controls from the discovery phase; 253 cases and 2161 controls in the validation phase). We identified 1 novel genome-wide significant variant rs17186366 (odds ratio [OR]=1.37 [1.23, 1.54], P=2.8×10-8) located near SOD2. In addition, 2 loci with suggestive association were also identified: rs113168128 (OR=1.72 [1.42, 2.10], P=5.2×10-8), an intronic variant of ERBB4 that is independent from the previously published variants, and rs144248326 (OR=2.13 [1.52, 2.86], P=8.45×10-7), a novel intronic variant in WWTR1. In the further association tests of the top 3 single-nucleotide polymorphisms with each criterion in the polycystic ovary syndrome algorithm, we found that rs17186366 (SOD2) was associated with polycystic ovaries and hyperandrogenism, whereas rs11316812 (ERBB4) and rs144248326 (WWTR1) were mainly associated with oligomenorrhea or infertility. We also validated the previously reported association with DENND1A1. Using the STRING database to characterize protein-protein interactions, we found both ERBB4 and WWTR1 can interact with YAP1, which has been previously associated with polycystic ovary syndrome. CONCLUSION: Through a discovery-validation genome-wide association study on polycystic ovary syndrome identified from electronic health records using an algorithm based on Rotterdam criteria, we identified and validated a novel genome-wide significant association with a variant near SOD2. We also identified a novel independent variant within ERBB4 and a suggestive association with WWTR1. With previously identified polycystic ovary syndrome gene YAP1, the ERBB4-YAP1-WWTR1 network suggests involvement of the epidermal growth factor receptor and the Hippo pathway in the multifactorial etiology of polycystic ovary syndrome.

ALG9 Mutation Carriers Develop Kidney and Liver Cysts.

BACKGROUND: Mutations in PKD1 or PKD2 cause typical autosomal dominant polycystic kidney disease (ADPKD), the most common monogenic kidney disease. Dominantly inherited polycystic kidney and liver diseases on the ADPKD spectrum are also caused by mutations in at least six other genes required for protein biogenesis in the endoplasmic reticulum, the loss of which results in defective production of the PKD1 gene product, the membrane protein polycystin-1 (PC1). METHODS: We used whole-exome sequencing in a cohort of 122 patients with genetically unresolved clinical diagnosis of ADPKD or polycystic liver disease to identify a candidate gene, ALG9, and in vitro cell-based assays of PC1 protein maturation to functionally validate it. For further validation, we identified carriers of ALG9 loss-of-function mutations and noncarrier matched controls in a large exome-sequenced population-based cohort and evaluated the occurrence of polycystic phenotypes in both groups. RESULTS: Two patients in the clinically defined cohort had rare loss-of-function variants in ALG9, which encodes a protein required for addition of specific mannose molecules to the assembling N-glycan precursors in the endoplasmic reticulum lumen. In vitro assays showed that inactivation of Alg9 results in impaired maturation and defective glycosylation of PC1. Seven of the eight (88%) cases selected from the population-based cohort based on ALG9 mutation carrier state who had abdominal imaging after age 50; seven (88%) had at least four kidney cysts, compared with none in matched controls without ALG9 mutations. CONCLUSIONS: ALG9 is a novel disease gene in the genetically heterogeneous ADPKD spectrum. This study supports the utility of phenotype characterization in genetically-defined cohorts to validate novel disease genes, and provide much-needed genotype-phenotype correlations.

Electronic health record phenotype in subjects with genetic variants associated with arrhythmogenic right ventricular cardiomyopathy: a study of 30,716 subjects with exome sequencing.

PurposeArrhythmogenic right ventricular cardiomyopathy (ARVC) is an inherited heart disease. Clinical follow-up of incidental findings in ARVC-associated genes is recommended. We aimed to determine the prevalence of disease thus ascertained.MethodsIndividuals (n = 30,716) underwent exome sequencing. Variants in PKP2, DSG2, DSC2, DSP, JUP, TMEM43, or TGFß3 that were database-listed as pathogenic or likely pathogenic were identified and evidence-reviewed. For subjects with putative loss-of-function (pLOF) variants or variants of uncertain significance (VUS), electronic health records (EHR) were reviewed for ARVC diagnosis, diagnostic criteria, and International Classification of Diseases (ICD-9) codes.ResultsEighteen subjects had pLOF variants; none of these had an EHR diagnosis of ARVC. Of 14 patients with an electrocardiogram, one had a minor diagnostic criterion; the rest were normal. A total of 184 subjects had VUS, none of whom had an ARVC diagnosis. The proportion of subjects with VUS with major (4%) or minor (13%) electrocardiogram diagnostic criteria did not differ from that of variant-negative controls. ICD-9 codes showed no difference in defibrillator use, electrophysiologic abnormalities or nonischemic cardiomyopathies in patients with pLOF or VUSs compared with controls.ConclusionpLOF variants in an unselected cohort were not associated with ARVC phenotypes based on EHR review. The negative predictive value of EHR review remains uncertain.

Association of Rare and Common Variation in the Lipoprotein Lipase Gene With Coronary Artery Disease.

Importance: The activity of lipoprotein lipase (LPL) is the rate-determining step in clearing triglyceride-rich lipoproteins from the circulation. Mutations that damage the LPL gene (LPL) lead to lifelong deficiency in enzymatic activity and can provide insight into the relationship of LPL to human disease. Objective: To determine whether rare and/or common variants in LPL are associated with early-onset coronary artery disease (CAD). Design, Setting, and Participants: In a cross-sectional study, LPL was sequenced in 10 CAD case-control cohorts of the multinational Myocardial Infarction Genetics Consortium and a nested CAD case-control cohort of the Geisinger Health System DiscovEHR cohort between 2010 and 2015. Common variants were genotyped in up to 305 699 individuals of the Global Lipids Genetics Consortium and up to 120 600 individuals of the CARDIoGRAM Exome Consortium between 2012 and 2014. Study-specific estimates were pooled via meta-analysis. Exposures: Rare damaging mutations in LPL included loss-of-function variants and missense variants annotated as pathogenic in a human genetics database or predicted to be damaging by computer prediction algorithms trained to identify mutations that impair protein function. Common variants in the LPL gene region included those independently associated with circulating triglyceride levels. Main Outcomes and Measures: Circulating lipid levels and CAD. Results: Among 46 891 individuals with LPL gene sequencing data available, the mean (SD) age was 50 (12.6) years and 51% were female. A total of 188 participants (0.40%; 95% CI, 0.35%-0.46%) carried a damaging mutation in LPL, including 105 of 32 646 control participants (0.32%) and 83 of 14 245 participants with early-onset CAD (0.58%). Compared with 46 703 noncarriers, the 188 heterozygous carriers of an LPL damaging mutation displayed higher plasma triglyceride levels (19.6 mg/dL; 95% CI, 4.6-34.6 mg/dL) and higher odds of CAD (odds ratio = 1.84; 95% CI, 1.35-2.51; P < .001). An analysis of 6 common LPL variants resulted in an odds ratio for CAD of 1.51 (95% CI, 1.39-1.64; P = 1.1 × 10-22) per 1-SD increase in triglycerides. Conclusions and Relevance: The presence of rare damaging mutations in LPL was significantly associated with higher triglyceride levels and presence of coronary artery disease. However, further research is needed to assess whether there are causal mechanisms by which heterozygous lipoprotein lipase deficiency could lead to coronary artery disease.

Impact of a Population Genomic Screening Program on Health Behaviors Related to Familial Hypercholesterolemia Risk Reduction.

BACKGROUND: Limited information is available regarding clinician and participant behaviors after disclosure of genomic risk variants for familial hypercholesterolemia (FH) from a population genomic screening program. METHODS: We conducted a retrospective cohort study of MyCode participants with an FH risk variant beginning 2 years before disclosure until January 16, 2019. We analyzed lipid-lowering prescriptions (clinician behavior), medication adherence (participant behavior), and LDL (low-density lipoprotein) cholesterol levels (health outcome impact) pre- and post-disclosure. Data were collected from electronic health records and claims. RESULTS: The cohort included 96 participants of mean age 57 (22-90) years with median follow-up of 14 (range, 3-39) months. Most (90%) had a hypercholesterolemia diagnosis but no specific FH diagnosis before disclosure; 29% had an FH diagnosis post-disclosure. After disclosure, clinicians made 36 prescription changes in 38% of participants, mostly in participants who did not achieve LDL cholesterol goals pre-disclosure (81%). However, clinicians wrote prescriptions for fewer participants post-disclosure (71/96, 74.0%) compared with pre-disclosure (81/96, 84.4%); side effects were documented for most discontinued prescriptions (23/25, 92%). Among the 16 participants with claims data, medication adherence improved (proportion of days covered pre-disclosure of 70% [SD, 24.7%] to post-disclosure of 79.1% [SD, 27.3%]; P=0.05). Among the 52 (54%) participants with LDL cholesterol values both before and after disclosure, average LDL cholesterol decreased from 147 to 132 mg/dL (P=0.003). CONCLUSIONS: Despite disclosure of an FH risk variant, nonprescribing and nonadherence to lipid-lowering therapy remained high. However, when clinicians intensified medication regimens and participants adhered to medications, lipid levels decreased.

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.

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.

Healthcare Utilization and Costs after Receiving a Positive BRCA1/2 Result from a Genomic Screening Program.

Population genomic screening has been demonstrated to detect at-risk individuals who would not be clinically identified otherwise. However, there are concerns about the increased utilization of unnecessary services and the associated increase in costs. The objectives of this study are twofold: (1) determine whether there is a difference in healthcare utilization and costs following disclosure of a pathogenic/likely pathogenic (P/LP) BRCA1/2 variant via a genomic screening program, and (2) measure the post-disclosure uptake of National Comprehensive Cancer Network (NCCN) guideline-recommended risk management. We retrospectively reviewed electronic health record (EHR) and billing data from a female population of BRCA1/2 P/LP variant carriers without a personal history of breast or ovarian cancer enrolled in Geisinger's MyCode genomic screening program with at least a one-year post-disclosure observation period. We identified 59 women for the study cohort out of 50,726 MyCode participants. We found no statistically significant differences in inpatient and outpatient utilization and average total costs between one-year pre- and one-year post-disclosure periods ($18,821 vs. $19,359, p = 0.76). During the first year post-disclosure, 49.2% of women had a genetic counseling visit, 45.8% had a mammography and 32.2% had an MRI. The uptake of mastectomy and oophorectomy was 3.5% and 11.8%, respectively, and 5% of patients received chemoprevention.

Electronic health record analysis identifies kidney disease as the leading risk factor for hospitalization in confirmed COVID-19 patients.

BACKGROUND: Empirical data on conditions that increase risk of coronavirus disease 2019 (COVID-19) progression are needed to identify high risk individuals. We performed a comprehensive quantitative assessment of pre-existing clinical phenotypes associated with COVID-19-related hospitalization. METHODS: Phenome-wide association study (PheWAS) of SARS-CoV-2-positive patients from an integrated health system (Geisinger) with system-level outpatient/inpatient COVID-19 testing capacity and retrospective electronic health record (EHR) data to assess pre-COVID-19 pandemic clinical phenotypes associated with hospital admission (hospitalization). RESULTS: Of 12,971 individuals tested for SARS-CoV-2 with sufficient pre-COVID-19 pandemic EHR data at Geisinger, 1604 were SARS-CoV-2 positive and 354 required hospitalization. We identified 21 clinical phenotypes in 5 disease categories meeting phenome-wide significance (P<1.60x10-4), including: six kidney phenotypes, e.g. end stage renal disease or stage 5 CKD (OR = 11.07, p = 1.96x10-8), six cardiovascular phenotypes, e.g. congestive heart failure (OR = 3.8, p = 3.24x10-5), five respiratory phenotypes, e.g. chronic airway obstruction (OR = 2.54, p = 3.71x10-5), and three metabolic phenotypes, e.g. type 2 diabetes (OR = 1.80, p = 7.51x10-5). Additional analyses defining CKD based on estimated glomerular filtration rate, confirmed high risk of hospitalization associated with pre-existing stage 4 CKD (OR 2.90, 95% CI: 1.47, 5.74), stage 5 CKD/dialysis (OR 8.83, 95% CI: 2.76, 28.27), and kidney transplant (OR 14.98, 95% CI: 2.77, 80.8) but not stage 3 CKD (OR 1.03, 95% CI: 0.71, 1.48). CONCLUSIONS: This study provides quantitative estimates of the contribution of pre-existing clinical phenotypes to COVID-19 hospitalization and highlights kidney disorders as the strongest factors associated with hospitalization in an integrated US healthcare system.

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.

GSTM1 Copy Number Is Not Associated With Risk of Kidney Failure in a Large Cohort.

Deletion of glutathione S-transferase µ1 (GSTM1) is common in populations and has been asserted to associate with chronic kidney disease progression in some research studies. The association needs to be validated. We estimated GSTM1 copy number using whole exome sequencing data in the DiscovEHR cohort. Kidney failure was defined as requiring dialysis or receiving kidney transplant using data from the electronic health record and linkage to the United States Renal Data System, or the most recent eGFR < 15 ml/min/1.73 m2. In a cohort of 46,983 unrelated participants, 28.8% of blacks and 52.1% of whites had 0 copies of GSTM1. Over a mean of 9.2 years follow-up, 645 kidney failure events were observed in 46,187 white participants, and 28 in 796 black participants. No significant association was observed between GSTM1 copy number and kidney failure in Cox regression adjusting for age, sex, BMI, smoking status, genetic principal components, or comorbid conditions (hypertension, diabetes, heart failure, coronary artery disease, and stroke), whether using a genotypic, dominant, or recessive model. In sensitivity analyses, GSTM1 copy number was not associated with kidney failure in participants that were 45 years or older at baseline, had baseline eGFR < 60 ml/min/1.73 m2, or with baseline year between 1996 and 2002. In conclusion, we found no association between GSTM1 copy number and kidney failure in a large cohort study.

Predicting Survival From Large Echocardiography and Electronic Health Record Datasets: Optimization With Machine Learning.

OBJECTIVES: The goal of this study was to use machine learning to more accurately predict survival after echocardiography. BACKGROUND: Predicting patient outcomes (e.g., survival) following echocardiography is primarily based on ejection fraction (EF) and comorbidities. However, there may be significant predictive information within additional echocardiography-derived measurements combined with clinical electronic health record data. METHODS: Mortality was studied in 171,510 unselected patients who underwent 331,317 echocardiograms in a large regional health system. The authors investigated the predictive performance of nonlinear machine learning models compared with that of linear logistic regression models using 3 different inputs: 1) clinical variables, including 90 cardiovascular-relevant International Classification of Diseases, Tenth Revision, codes, and age, sex, height, weight, heart rate, blood pressures, low-density lipoprotein, high-density lipoprotein, and smoking; 2) clinical variables plus physician-reported EF; and 3) clinical variables and EF, plus 57 additional echocardiographic measurements. Missing data were imputed with a multivariate imputation by using a chained equations algorithm (MICE). The authors compared models versus each other and baseline clinical scoring systems by using a mean area under the curve (AUC) over 10 cross-validation folds and across 10 survival durations (6 to 60 months). RESULTS: Machine learning models achieved significantly higher prediction accuracy (all AUC >0.82) over common clinical risk scores (AUC = 0.61 to 0.79), with the nonlinear random forest models outperforming logistic regression (p < 0.01). The random forest model including all echocardiographic measurements yielded the highest prediction accuracy (p < 0.01 across all models and survival durations). Only 10 variables were needed to achieve 96% of the maximum prediction accuracy, with 6 of these variables being derived from echocardiography. Tricuspid regurgitation velocity was more predictive of survival than LVEF. In a subset of studies with complete data for the top 10 variables, multivariate imputation by chained equations yielded slightly reduced predictive accuracies (difference in AUC of 0.003) compared with the original data. CONCLUSIONS: Machine learning can fully utilize large combinations of disparate input variables to predict survival after echocardiography with superior accuracy.