Abstract Background The HeartLogic algorithm integrates data from various implantable defibrillator (ICD) sensors to predict impending heart failure (HF) decompensation. The algorithm computes worsening in its sensors and weighs them based on a risk level. These sensors include accelerometer-based first (S1) and third (S3) heart sounds, intrathoracic impedance (TI), respiration rate (RR), the ratio of respiration rate to tidal volume (RSBI), and night heart rate (NHR). Objective This study assessed the relative worsening of ICD sensors at the onset of HeartLogic alerts, their association with patient characteristics and subsequent outcomes. Methods The HeartLogic feature was activated in 568 ICD patients (410 with CRT-D) across 26 centers, with a median follow-up of 26 months [25th–75th percentile: 16–37]. Results During the follow-up, 1200 HeartLogic alerts were recorded in 370 patients. The sensor with the highest worsening (SHW) at the alert onset was S3 (27% of alerts), followed by S3/S1 (25%), TI (16%), RR (15%), and NHR (11%), RSBI (6%). Patients with atrial fibrillation (AF) at implantation and those with chronic kidney disease (CKD) had higher alert prevalence (AF 84% vs. no-AF 58%, CKD 72% vs. no-CKD 59%; both p <0.05) and higher alert rates (AF 1.51/patient-year vs. no-AF 0.88/patient-year, CKD 1.30/patient-year vs. no-CKD 0.89/patient-year; both p <0.05). AF patients had alerts with every sensor as SHW, while CKD patients had alerts primarily with TI, RR, and RSBI as SHW (Figure). In 85% of cases among 247 patients with >1 alert, the SHW changed between successive alerts. Of the 88 (7%) alerts resulting in HF hospitalizations or deaths, a greater proportion featured RR or RSBI (11%) and NHR (11%) as SHW, followed by heart sounds (5%) (both p <0.05). Clinical events were more common with the first alert (12.6%) than subsequent alerts (5.2%, p <0.001). Conclusion HeartLogic alerts are predominantly associated with the highest worsening in heart sounds. Nevertheless, recurrent alerts often involve other sensors, suggesting varied HF progression mechanisms and potential divergent outcomes. The frequency of alerts and the pattern of worsening in ICD sensors are linked to patient characteristics.