Abstract Background Heart failure (HF) hospitalizations are associated with poor patient outcomes and generate a large burden on healthcare systems. Mitigation is a priority. Among HF patients with CIEDs, existing HF algorithm-based management solutions are manufacturer and device-specific, which imposes limitations. The SignalHF universal algorithm was developed to democratize access to HF management solutions by enabling prediction of HF decompensation based on CIED sensors data agnostic to manufacturer. Objective FORESEE-HF study aims to validate the SignalHF machine learning score and alerting system by matching alerts generated by SignalHF from CIED-related stored data in a universal platform (Implicity™) with HF events available in the French national health database containing medico-administrative data on patients’ characteristics, hospitalizations and deaths across 10 years. Methods The study cohort consisted of 19,013 patients (pacemaker (PM) 7,711, ICD 6,022, CRT-D 4,367 and CRT-P 913; age 70.0 ± 13.3, male 70.7%). Co-primary endpoints were sensitivity of HF hospitalizations detection with HF as primary diagnosis, the unexplained alert rate per patient-year defined as alert states not followed within 30 days by a hospitalization with HF as primary or secondary diagnosis (UAR PPY), and the lower quartile on alerting time defined as the lower bound on the time-to-event of 75% of true alert states (see Table 1). Results SignalHF for ICD/CRT-Ds demonstrated a sensitivity of 60%, a UAR PPY of 0.66 and a lower quartile on alerting time of 35 days. For PM/CRT-Ps, SignalHF demonstrated a sensitivity of 46%, a UAR PPY of 0.47 and a lower quartile on alerting time of 37 days. Conclusion FORESEE-HF demonstrates that SignalHF is an improvement over existing HF monitors for ICD/CRT-Ds, while also providing the same capability for PM/CRT-Ps. Being an algorithm that can be easily activated through a universal CIED remote monitoring platform, SignalHF widens the pool of patients that can benefit from HF monitoring solutions.Table 1:Study coprimary endpointsTable 2:Study results per device type