Machine Learned Cellular Phenotypes in Cardiomyopathy Predict Sudden Death

Abstract

Rationale: Susceptibility to VT/VF (ventricular tachycardia/fibrillation) is difficult to predict in patients with ischemic cardiomyopathy either by clinical tools or by attempting to translate cellular mechanisms to the bedside.

Objective: To develop computational phenotypes of patients with ischemic cardiomyopathy, by training then interpreting machine learning of ventricular monophasic action potentials (MAPs) to reveal phenotypes that predict long-term outcomes.

Methods and results: We recorded 5706 ventricular MAPs in 42 patients with coronary artery disease and left ventricular ejection fraction ≤40% during steady-state pacing. Patients were randomly allocated to independent training and testing cohorts in a 70:30 ratio, repeated K=10-fold. Support vector machines and convolutional neural networks were trained to 2 end points: (1) sustained VT/VF or (2) mortality at 3 years. Support vector machines provided superior classification. For patient-level predictions, we computed personalized MAP scores as the proportion of MAP beats predicting each end point. Patient-level predictions in independent test cohorts yielded c-statistics of 0.90 for sustained VT/VF (95% CI, 0.76-1.00) and 0.91 for mortality (95% CI, 0.83-1.00) and were the most significant multivariate predictors. Interpreting trained support vector machine revealed MAP morphologies that, using in silico modeling, revealed higher L-type calcium current or sodium-calcium exchanger as predominant phenotypes for VT/VF.

Conclusions: Machine learning of action potential recordings in patients revealed novel phenotypes for long-term outcomes in ischemic cardiomyopathy. Such computational phenotypes provide an approach which may reveal cellular mechanisms for clinical outcomes and could be applied to other conditions.

Keywords: artificial intelligence; coronary disease; death, sudden, cardiac; heart failure; ion channels; systems biology.

Figure 1.

Ventricular monophasic action potentials (MAPs) in patients with ischemic LV dysfunction with (top row) or without (bottom row) events on long-term follow-up. Panels show (A) 79-year-old male, LVEF 29% with appropriate ICD therapy at 400 days; (B) 55-year-old male, LVEF 35% with ICD therapy at 598 days; (C) 70-year-old male, LVEF 25% who died from congestive heart failure at 95 days. Bottom panels show (D) 45-year-old female, LVEF 26%, (E) 63-year-old male, LVEF 40%, (F) 59-year-old male, LVEF 21%, each of whom had no event at > 3 years of follow-up. (G) Expanded MAP upstroke from patient in (F) (black) with dV/dt (red) illustrating signal fidelity.

Figure 2.

Data flow in study.

Figure 3.

Receiver operating characteristics of patient-level MAP scores for (A) sustained VT/VF and (B) all-cause mortality on 3-year follow-up.

Figure 4.. MAP Morphologies identified by Machine Learning to predict endpoints of (a) VT/VF and (b) overall mortality.

a. Average computed from all single beats that predicted VT/VF (red) or no VT/VF (blue). b. Average computed from all single beats that predicted mortality (red) or survival (blue). Means were computed from SVM results for all 10 folds in learn and test sets combined. Voltages are standardized within each curve and adjusted for small offsets at time zero.

Figure 5.. Cellular Biophysical Simulations Probe How Machine Learning of MAPs Predicted Clinical Outcomes.

A) Global sensitivity analysis of the contribution of I_kr (Kr), I_CaL (CaL), NCX, I_to (TO) and SERCA, which are altered in heart failure. The vertical scale is the normalized sensitivity (%) of action potential durations at 30% (APD30), 60% (APD60) and 90% (APD90) repolarization to each pathway (log scale). B) Ionic pathway densities for dataset 1, in which TO, Kr and CaL were each varied within 91 increments (91³ = 753,571 permutations) were fitted to clinically measured MAP durations predicting VT/VF or mortality. C) Ionic pathway densities for dataset 2, in which 753,571 permutations of TO, Kr and CaL were fitted to measured MAP durations predicting VT/VF or mortality. Action potentials from patients with VT/VF exhibited lower Kr, higher CaL or enhanced NCX (explaining higher phase II plateau) than those without events. Action potentials from patients who died exhibited elevated Kr or reductions in NCX compared to those who survived.