Halperin, Jonathan L.

Introduction: Heart failure (HF) is a leading cause of hospitalization, morbidity and mortality. Deep learning (DL) techniques appear to show promising results in risk stratification and prognosis in several conditions in medicine. However, few methods using DL exist to help quantitatively estimate prognosis of HF. We hypothesized that deep learning (DL) techniques could prognosis of HF using simple variables. We propose application of a custom-built deep-neural-network model to identify mortality in HF patients. Methods: Custom-built deep-neural-networks were assessed using survey data from 42,147 participants from the National Health and Nutrition Examination Survey 1999-2016 (NHANES). Variables were selected using clinical judgment and stepwise backward regressions to develop prediction models. We partitioned the data into training and testing sets and repetitive experiments. We then evaluated model performance based on discrimination and calibration including the area under the receiver-operator characteristics curve (C-statistics), balanced accuracy, probability calibration with sigmoid, and the Brier score, respectively. As sensitivity analyses, we examined results limited to cases with complete clinical information available. We validated models’ performance using Mount Sinai database. Results: Of 42,147 participants with 4,060 variables, 1,491 (3.5%) had HF and HF mortality was 51.8%. In validation cohort, of 26,333 HF patients, the mortality in HF patients was 405 (1.5%). Final model using only 20 variables (age, race, gender, BMI, smoking, alcohol consumption, HTN, COPD, SBP, DBP, HR, HDL, LDL, CRP, A1C, BUN, creatinine, hemoglobin, sodium level, on statin) was tested. A state-of-the-art deep learning models achieved high accuracy for predicting mortality in HF patients with an AUC of 0.96 (95% CI: 0.95-0.99) in the first cohort and AUC of 0.93 (95% CI: 0.91-0.96) in validation cohort. Conclusions: A deep neural network model has shown to have high predictive accuracy and discriminative and calibrative power for prediction of HF mortality. Further research can delineate the clinical implications of DL in predicting HF mortality.

Abstract Deep learning (DL) is a branch of machine learning (ML) showing increasing promise in medicine, to assist in data classification, novel disease phenotyping and complex decision making. Deep learning is a form of ML typically implemented via multi-layered neural networks. Deep learning has accelerated by recent advances in computer hardware and algorithms and is increasingly applied in e-commerce, finance, and voice and image recognition to learn and classify complex datasets. The current medical literature shows both strengths and limitations of DL. Strengths of DL include its ability to automate medical image interpretation, enhance clinical decision-making, identify novel phenotypes, and select better treatment pathways in complex diseases. Deep learning may be well-suited to cardiovascular medicine in which haemodynamic and electrophysiological indices are increasingly captured on a continuous basis by wearable devices as well as image segmentation in cardiac imaging. However, DL also has significant weaknesses including difficulties in interpreting its models (the ‘black-box’ criticism), its need for extensive adjudicated (‘labelled’) data in training, lack of standardization in design, lack of data-efficiency in training, limited applicability to clinical trials, and other factors. Thus, the optimal clinical application of DL requires careful formulation of solvable problems, selection of most appropriate DL algorithms and data, and balanced interpretation of results. This review synthesizes the current state of DL for cardiovascular clinicians and investigators, and provides technical context to appreciate the promise, pitfalls, near-term challenges, and opportunities for this exciting new area.

Introduction: Recent trials using insertable cardiac monitors (ICMs) show high rates of atrial fibrillation (AF) in those at risk of developing AF based on demographics. Further identifying subsets who could benefit most from ICMs is desirable. We evaluated if the recently-developed HAVOC risk score which has predicted AF in cryptogenic stroke patients also predicts AF detection by ICMs in those without a recent stroke. Methods: Participants from the prospective REVEAL AF study, which reported frequent clinically unrecognized AF in patients at risk for AF (CHADS 2 scores ≥3 or =2 with ≥1 additional risk factor), with ICM data and not on anti-arrhythmic drugs were included. Ischemic stroke <1 year ago excluded participation. HAVOC scores were calculated using points for hypertension (2), age ≥75 (2), valvular disease (2), vascular disease (1), obesity (1), congestive heart failure (4), and coronary artery disease (2). Prior stroke, an AF outcome, generally not a cause, is not included. Score classification is: low (0-4), intermediate (5-9), or high (10-14). Corresponding AF detection rates were compared via the log-rank test. Results: Among 391 participants, 95 (24%), 241 (62%), and 55 (14%) had low, intermediate, and high HAVOC scores, respectively. At 18 months, AF incidence was significantly less for those with low (19.5%) vs intermediate (32.1%) or high (34.2%) scores (p=0.045, Figure). At 18 months, AF incidence was similar among those with (n=79) vs. without (n=312) a history of remote stroke (27.1% vs. 29.9%; p>0.05, median time from stroke to ICM insertion=4 years). Conclusions: The HAVOC risk score identified a subset of high risk individuals at greatest risk of developing AF. AF incidence rates were similar among those with and without prior remote stroke. Remote strokes may not be a strong predictor of AF in this population. HAVOC scores could be a useful approach to identify those at high likelihood of manifesting AF, as best documented by long-term monitoring.

Background— The multicenter PROTECT AF study (Watchman Left Atrial Appendage System for Embolic Protection in Patients With Atrial Fibrillation) was conducted to determine whether percutaneous left atrial appendage closure with a filter device (Watchman) was noninferior to warfarin for stroke prevention in atrial fibrillation. Methods and Results— Patients (n=707) with nonvalvular atrial fibrillation and at least 1 risk factor (age >75 years, hypertension, heart failure, diabetes, or prior stroke/transient ischemic attack) were randomized to either the Watchman device (n=463) or continued warfarin (n=244) in a 2:1 ratio. After device implantation, warfarin was continued for ≈45 days, followed by clopidogrel for 4.5 months and lifelong aspirin. Study discontinuation rates were 15.3% (71/463) and 22.5% (55/244) for the Watchman and warfarin groups, respectively. The time in therapeutic range for the warfarin group was 66%. The composite primary efficacy end point included stroke, systemic embolism, and cardiovascular death, and the primary analysis was by intention to treat. After 1588 patient-years of follow-up (mean 2.3±1.1 years), the primary efficacy event rates were 3.0% and 4.3% (percent per 100 patient-years) in the Watchman and warfarin groups, respectively (relative risk, 0.71; 95% confidence interval, 0.44%–1.30% per year), which met the criteria for noninferiority (probability of noninferiority >0.999). There were more primary safety events in the Watchman group (5.5% per year; 95% confidence interval, 4.2%–7.1% per year) than in the control group (3.6% per year; 95% confidence interval, 2.2%–5.3% per year; relative risk, 1.53; 95% confidence interval, 0.95–2.70). Conclusions— The “local” strategy of left atrial appendage closure is noninferior to “systemic” anticoagulation with warfarin. PROTECT AF has, for the first time, implicated the left atrial appendage in the pathogenesis of stroke in atrial fibrillation. Clinical Trial Registration:— URL: http://www.clinicaltrials.gov . Unique identifier: NCT00129545.

Background and Purpose —Thoracic aortic plaque identified by transesophageal echocardiography heightens the risk of stroke associated with atrial fibrillation (AF). We sought to identify the prevalence, predictors, and implications of aortic plaque in patients with nonvalvular AF. Methods —Thoracic aortic plaque was prospectively sought in 770 persons with AF with the use of transesophageal echocardiography and classified as simple or complex on the basis of thickness ≥4 mm, ulceration, or mobility. Clinical and echocardiographic features of thromboembolism were correlated by multivariate analysis. Results —Aortic plaque was detected in 57% of the cohort, and complex plaque was detected in 25%. Both were found more frequently in the descending than in the proximal aorta. Potentially etiologic patient characteristics independently associated with complex plaque included advanced age, history of hypertension, diabetes, and past or present tobacco use. Comorbidities associated with aortic plaque were prior thromboembolism, increased pulse pressure, ischemic heart disease, stenosis or sclerosis of the aortic valve, mitral annular calcification (>10%), elevated serum creatinine concentration, spontaneous echo contrast in the left atrium or appendage, and left atrial appendage thrombus. The prevalence of complex plaque in patients aged <70 years with <10% mitral annular calcification, without ischemic heart disease, or without pulse pressure ≥65 mm Hg was 4% (95% CI, 1% to 6%). Conclusions —Aortic plaque is prevalent in patients with AF and is associated with atherosclerosis risk factors and with left atrial stasis or thrombosis, which are themselves independent stroke risk factors. Since the predominant location of complex plaque was in the descending aorta, the role of aortic plaque as a source of embolism in AF is uncertain.