Long-term performance of the HeartLogic algorithm and the impact of device-detected sleep apnea in patients with ICDs and CRT-Ds
A Mazza, V Bianchi, G Savarese, G Vitulano, L Calo, R Calvanese, M Viscusi, C La Greca, V E Santobuono, A Santoro, G Arena, L Santini, M G Bendini, S Valsecchi, M MariniAbstract
Background
The HeartLogic algorithm integrates multiple implantable cardioverter-defibrillator (ICD) sensor signals to predict impending heart failure (HF) decompensation. In its validation study, HeartLogic demonstrated a sensitivity of 70%, an unexplained alert rate of 1.47 per patient-year, and a median lead time of 34 days before an HF event. Sleep apnea (SA) is known to exacerbate HF progression.
Purpose
This study evaluated the long-term performance of HeartLogic in a large cohort of patients implanted with ICDs and cardiac resynchronization therapy defibrillators (CRT-D). Additionally, we investigated whether the concomitant presence of a HeartLogic alert and a high respiratory disturbance index (RDI) could enhance the diagnostic performance of the system.
Methods
Data on HeartLogic Index and RDI were collected across the study cohort. The IN-alert HF state was defined as a multi-sensor HeartLogic Index >16. As these devices also measure respiratory patterns, a high RDI was defined as ≥30 episodes per hour (indicating severe sleep apnea)
Results
The HeartLogic feature was activated in 602 ICD recipients (393 with CRT-D; 78% male; mean age 68 ± 11 years; left ventricular ejection fraction 32 ± 9%) across 19 centers. The median follow-up was 45 months [IQR: 28–68]. During follow-up, 92 hospitalizations and 53 deaths occurred. Of these, 51 events were attributed to HF, and 374 outpatient visits involved intensification of HF therapy. A total of 2,384 HeartLogic alerts were recorded (1.09 per patient-year), with patients spending 13% of the total observation time in an alert state. The algorithm demonstrated a sensitivity of 80% [95% CI: 69–91] for detecting HF hospitalization or death, with a median lead time of 49 days [IQR: 21–85]. When including outpatient visits, sensitivity was 66% [95% CI: 62–71], with a median lead time of 20 days [IQR: 5–51]. The unexplained alert rate was 0.94 per patient-year [95% CI: 0.90–0.98]. The ICD-detected RDI was ≥30 episodes/h during 44% of the total observation period. Combining a HeartLogic alert with RDI ≥30 episodes/h (weekly median value at alert onset) yielded a sensitivity of 54% [95% CI: 38–71] for detecting HF hospitalization or death (p = 0.020 vs. HeartLogic alone) while significantly reducing the unexplained alert rate to 0.38 per patient-year [95% CI: 0.35–0.41] (p < 0.001).
Conclusions
This analysis confirms the high sensitivity and low unexplained alert rate of the HeartLogic algorithm for detecting HF-related clinical events. Concurrent verification of HeartLogic alerts and severe SA markedly decreases the rate of unexplained alerts but at the expense of reduced sensitivity.