Evaluating an alert-based multiparametric algorithm for predicting heart failure hospitalisations in patients with implantable cardioverter-defibrillators: a meta-cohort study.

Abstract

Background: The alert-based HeartInsight algorithm predicts risk of worsening heart failure hospitalisations (WHFHs) by evaluating temporal trends of seven physiologic parameters obtained through automatic daily remote monitoring of implantable cardioverter-defibrillators. The aim of the present study was to evaluate the predictive performance of HeartInsight in a larger and more heterogeneous meta-cohort of patients, incorporating newer device generations and including patients managed with the most recent guideline-directed medical therapy (GDMT).

Methods: The HeartInsight algorithm was retrospectively applied to data from four clinical trials in which WHFH events were adjudicated by independent external boards and remote monitoring was activated to provide relevant parameter trends. The analysis comprised 1352 patients with New York Heart Association (NYHA) class II/III, and no long-standing atrial fibrillation.

Results: During a median follow-up of 599 days, 110 patients (median age 68 years (IQR, 61-75), 75.7% male) had a total of 165 WHFHs. The estimated sensitivity of WHFH prediction, as determined by generalised estimating equations, was 51.5% (95% CI 43.0% to 59.9%). The false alert rate was 0.85 per patient-year, the median alerting time was 34 days (IQR, 16-78) and the specificity was 81.4% (95% CI 80.4 to 82.4%). The results were verified in the multivariable analysis with two adjusting covariates (newer/older device generation and quadruple/other GDMT) and in the univariable analysis of prespecified patient subgroups according to NYHA class, aetiology and sex, showing no significant differences.

Conclusions: Study results underscore the robustness of the predictive algorithm in a heterogeneous and contemporarily managed heart failure population.