Prediction of heart failure events based on physiologic sensor data in HINODE defibrillator patients

Abstract

Abstract Aims Hospitalizations are common in patients with heart failure and are associated with high mortality, readmission and economic burden. Detecting early signs of worsening heart failure may enable earlier intervention and reduce hospitalizations. The HeartLogic algorithm is designed to predict worsening heart failure using diagnostic data from multiple device sensors. The main objective of this analysis was to evaluate the sensitivity of the HeartLogic alert calculation in predicting worsening heart failure events (HFEs). We also evaluated the false positive alert rate (FPR) and compared the incidence of HFEs occurring in a HeartLogic alert state to those occurring out of an alert state. Methods The HINODE study enrolled 144 patients (81 ICD and 63 CRT‐D) with device sensor data transmitted via a remote monitoring system. HeartLogic alerts were then retrospectively simulated using relevant sensor data. Clinicians and patients were blinded to calculated alerts. Reported adverse events with HF symptoms were adjudicated and classified by an independent HFE committee. Sensitivity was defined as the ratio of the number of detected usable HFEs (true positives) to the total number of usable HFEs. A false positive alert was defined as an alert with no usable HFE between the alert onset date and the alert recovery date plus 30 days. The patient follow‐up period was categorized as in alert state or out of alert state. The event rate ratio was the HFE rate calculated in alert to out of alert. Results The patient cohort was 79% male and had an average age of 68 ± 12 years. This analysis yielded 244 years of follow‐up data with 73 HFEs from 37 patients. A total of 311 HeartLogic alerts at the nominal threshold (16) occurred across 106 patients providing an alert rate of 1.27 alerts per patient‐year. The HFE rate was 8.4 times greater while in alert compared with out of alert (1.09 vs. 0.13 events per patient‐year; P < 0.001). At the nominal alert threshold, 80.8% of HFEs were detected by a HeartLogic alert [95% confidence interval (CI): 69.9%–89.1%]. The median time from first true positive alert to an adjudicated clinical HFE was 53 days. The FPR was 1.16 (95% CI: 0.98–1.38) alerts per patient‐year. Conclusions Results suggest that signs of worsening HF can be detected successfully with remote patient follow‐up. The use of HeartLogic may predict periods of increased risk for HF or clinically significant events, allowing for early intervention and reduction of hospitalization in a vulnerable patient population.

Keywords

• HeartLogic
• heart failure
• remote monitoring
• ICD
• CRT
• hospitalization