Search

Supplemental Digital Content is available in the text., 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.

Introduction: Healthcare cost has increased drastically in the last decade, and over 50% of the cost can be attributed to a small portion (5-10%) of the population. Certain clinical programs, such as home-based care, aim to reduce this utilization but need methods to identify the most appropriate patients to enroll. We hypothesized that machine learning can predict patients with high future utilization with good accuracy. Methods: 683,160 cardiology patients (defined broadly as those with an ECG, echocardiogram or cardiology visit) with ~17 million clinical episodes since 2004 were identified from Geisinger’s electronic health records. Utilization was estimated as total cost of care for outpatient, inpatient and emergency department visits. Patients with the highest 10% utilization in a given year were defined as high utilizers. Machine learning models were used to predict high utilization over the next 3, 6 and 12 months. Input variables (n=191) included age, sex, smoking, 5 vital signs, 21 labs, 18 medications (current and past), 40 ECG and 44 echocardiographic measurements, 43 comorbidities, 7 time / cyclical features, 6 past utilization metrics and 4 social metrics (e.g. distance to healthcare facilities). Results: XGBoost achieved the best performance with areas under the ROC curve (AUC) of 0.82, 0.81 and 0.78 for 3, 6, 12-month models, and average precision scores (AP) of 0.31, 0.36 and 0.37, respectively, while the commonly used Charlson Comorbidity Index had poor performance with AUCs of 0.63 - 0.64 and APs of 0.1 - 0.17. Past utilization was the best predictor of future utilization. Targeting patients with the top 5 and 10% highest risk for utilization achieved sensitivities of 26 and 40% and positive predictive values of 50 and 38% (12-month model, Figure). Conclusions: Machine learning can be used to predict which patients will have high future healthcare utilization. This may help target populations for intervention programs aimed at reducing utilization.

Introduction: Managing heart failure patients relies on mortality prediction models such as the Seattle Heart Failure (SHF) model. We hypothesized that a deep learning model that leverages echocardiography video data could improve prediction performance. Methods: We trained deep neural networks (DNN) to predict 1-year all-cause mortality using echocardiography videos and 158 additional clinical variables (labs, vitals, diagnoses, etc.) acquired from 34,362 patients (812,278 videos). We tested both the DNN and the SHF model on a separate cohort of 2,404 patients with heart failure who underwent 3,384 echocardiograms. We computed the area under the receiver operating characteristic curve (AUC) and its bootstrapped 95% confidence interval (CI) in the test set. Results: The DNN model (AUC of 0.76, 95% CI [0.74, 0.77]) outperformed the SHF model (AUC of 0.70, 95% CI [0.68, 0.71]). This superior performance was observed for patients with both reduced (n=2,026, AUC of 0.76 [0.74, 0.78] vs 0.70 [0.67, 0.72]) and preserved left ventricular ejection fraction (n=1,356, AUC of 0.75 [0.72, 0.78] vs 0.69 [0.66, 0.72]). A Cox Proportional Hazard survival analysis showed that, despite the prediction being for 1-year all-cause mortality, the result held long-term predictive power over the next 9 years with a superior hazard ratio of 2.9 [2.6, 3.2] for the DNN model compared to 2.2 [2, 2.4] for the SHF model. At mid-range operating points, the DNN model also maintained a higher negative predictive value, predicting survival, compared to the SHF model (89% vs 83%, respectively), while maintaining the same positive predictive value of 40%. Conclusion: A deep neural network that automatically analyzes echocardiography videos can outperform traditional risk modeling approaches such as the Seattle Heart Failure Model. This provides additional evidence that deep learning can help improve our ability to make clinically relevant predictions by leveraging complex datasets.

Background: Atrial fibrillation (AF) is associated with stroke, especially when AF goes undetected. Deep neural networks (DNN) can predict incident AF from a 12-lead resting ECG. We hypothesize that use of a DNN to predict new onset AF from an ECG may identify patients at risk of sustaining a potentially preventable AF-related stroke. Methods: We trained a DNN model to predict new-onset AF using 382,604 ECGs prior to 2010. We then evaluated the model performance on a test set of ECGs from 2010 through 2014 linked to patients in an institutional stroke registry. There were 181,969 patients in the test set with at least one ECG and no prior history of AF. Of those patients 3,497 (1.9%) had a stroke following an ECG that did not show AF. Within the set of patients with stroke, 375 had the stroke within 3 years of the ECG and were diagnosed with new AF between -3 and 365 days of the stroke. We considered these potentially preventable AF-related strokes. We report the sensitivity and positive predictive value (PPV) of the model for appropriately risk stratifying these 375 patients who sustained a potentially preventable AF-related stroke. Results: We used F β scores to identify different risk prediction thresholds (operating points) for the model. Operating points chosen by F 0.5 , F 1 , and F 2 scores identified 4, 12, and 21% of the population as high risk for the development of AF within 1 year (Figure 1). Screening 1, 4, 12, and 21% of the overall population resulted in PPV of 28, 21, 15, and 12%, respectively, for identification of new onset AF in one year. Using those same thresholds yielded sensitivities of 4, 17, 45, and 62% for identifying potentially preventable AF-related strokes. The different risk prediction thresholds resulted in a low (120-162) number needed to screen to detect one potentially preventable AF-related stroke at 3 years. Conclusions: Use of a deep learning model to predict new onset AF may identify patients at high risk of sustaining a potentially preventable AF-related stroke.

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 (β=–12%, P =3×10 –7 ), and increased left ventricular diameter (β=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 (β=–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.

Objective Arrhythmogenic right ventricular cardiomyopathy (ARVC) is a rare myocardial disorder. Genetic etiology is ascribed to variants in 8 genes ( PKP2 , DSP , DSC2 , DSG2 , JUP , TMEM43 , TGFβ2 , RYR2 ). We aimed to establish the prevalence and clinical phenotype of individuals with pathogenic variants (PV) in these genes from an unselected cohort. Methods Whole-exome sequences for 31,036 patients in the MyCode™ Biorepository were evaluated for pathogenic or likely pathogenic (P/LP) ARVC variants reported in ClinVar or the ARVC database. Composite ICD-9 data from the electronic health records of PV individuals vs a non-PV control group were analyzed. Results 57 P/LP variants were identified in 240 individuals (0.77%), aged 61 ± 19 years. Characteristics of this PV group versus a non-PV control group are summarized in the table. Zero patients in the PV group had a documented diagnosis of ARVC. Use of implantable cardioverter defibrillators—the primary treatment for ARVC—was rare. Of individuals ≥55 years, 24% with PV had no history of cardiac disease (vs. 22% for non-PV). However, there was a modest increase in the prevalence of cardiac electrical abnormalities (e.g., tachycardia, fibrillation) in PV vs. non-PV. There was also a trend towards increased diagnoses of 'other primary cardiomyopathy'. Conclusions The prevalence of reported pathogenic ARVC variants (0.77%) greatly exceeds the disease prevalence (0.02 - 0.10%). In this cohort of 240 genotype-positive subjects, there was little evidence of overt ARVC disease, suggesting that the positive-predictive value for classic ARVC in individuals with an incidental positive genetic finding is low. The increased risk of certain clinical findings, not specific to ARVC, supports the notion that the spectrum of phenotypes associated with pathogenic variants in these genes is broader than classically described. Targeted “deep phenotyping” is needed to assess subclinical phenotypes in genomically screened populations.