AbstractAimTo examine the ability of serum proteins in predicting future heart failure (HF) events, including HF with reduced or preserved ejection fraction (HFrEF or HFpEF), in relation to event time, and with or without considering established HF‐associated clinical variables.Methods and resultsIn the prospective population‐based Age, Gene/Environment Susceptibility Reykjavik Study (AGES‐RS), 440 individuals developed HF after their first visit with a median follow‐up of 5.45 years. Among them, 167 were diagnosed with HFrEF and 188 with HFpEF. A least absolute shrinkage and selection operator regression model with nonparametric bootstrap were used to select predictors from an analysis of 4782 serum proteins, and several pre‐established clinical parameters linked to HF. A subset of 8–10 distinct or overlapping serum proteins predicted different future HF outcomes, and C‐statistics were used to assess discrimination, revealing proteins combined with a C‐index of 0.80 for all incident HF, 0.78 and 0.80 for incident HFpEF or HFrEF, respectively. In the AGES‐RS, protein panels alone encompassed the risk contained in the clinical information and improved the performance characteristics of prediction models based on NT‐proBNP and clinical risk factors. Finally, the protein predictors performed particularly well close to the time of an HF event, an outcome that was replicated in the Cardiovascular Health Study (CHS).ConclusionA small number of circulating proteins accurately predicted future HF in the AGES‐RS cohort of older adults, and they alone encompass the risk information found in a collection of clinical data. Incident HF events were predicted up to eight years, with predictor performance significantly improving for events occurring less than one year ahead, a finding replicated in an external cohort study.This article is protected by copyright. All rights reserved.