BackgroundTo predict worsening heart failure hospitalizations (WHFHs), the HeartInsight multiparametric algorithm calculates a heart failure (HF) score based on temporal trends of physiologic parameters obtained through automatic daily remote monitoring of implantable cardioverter-defibrillators (ICDs). ObjectiveWe studied the association of the baseline HF score, determined at algorithm activation, with long-term patient outcomes. MethodsData from 9 clinical trials were pooled, including 1841 ICD patients with a preimplantation ejection fraction ≤35%, New York Heart Association class II/III, and no long-standing atrial fibrillation. The primary end point was a composite of death or WHFH. ResultsAfter a median follow-up of 631 days (interquartile range, 385–865 days), there were 243 WHFHs in 173 patients (9.4%) and 122 deaths (6.6%), 52 of which (42.6%) were cardiovascular. The primary end point occurred in 265 patients (14.4%). A multivariable time-to-first-event analysis showed that a high baseline HF score (>23, as determined by a time-dependent receiver operating characteristics curve analysis) was significantly associated with the occurrence of the primary end point (adjusted hazard ratio [HR], 2.05; 95% confidence interval [CI], 1.54–2.71; P < .0001), all-cause death (HR, 2.37; CI, 1.56–3.58; P < .0001), cardiovascular death (HR, 2.19; CI, 1.14–4.22; P = .019), and WHFH (HR, 1.91; CI, 1.35–2.71; P = .0003). In a hierarchical event analysis of all-cause death as the outcome with highest priority and WHFHs as repeated event outcomes, the win ratio was 2.47 (CI, 1.89–3.24; P < .0001). ConclusionBased on a retrospective analysis of clinical trial data with adjudicated events, baseline HF score derived from device-monitored variables was able to stratify patients at higher long-term risk of death or WHFH.