The Use of the Multisensor HeartLogic Algorithm for Heart Failure Remote Monitoring in Patients With Left Ventricular Assist Devices.

Patients with left ventricular assist devices (LVADs) experience high rates of heart failure (HF), and all-cause, hospitalizations.1,2 Such hospitalizations are associated with increased costs and decreased survival.1 Thus, identifying strategies to reduce hospitalizations for LVAD patients is critical. HeartLogic is a multisensor device algorithm in implantable cardioverter defibrillator (ICD) devices by Boston Scientific (Marlborough, MA). Using thoracic impedance, the first and third heart sounds, respiration, heart rate, and activity level, the algorithm supplies clinicians with a single, unitless, daily value, that has been validated as a sensitive and timely predictor of HF decompensation in ambulatory cohorts.3 However, the use of HeartLogic for remote monitoring of LVAD patients is poorly described, despite the fact that 61–91% of LVAD patients have ICDs.4,5 Our objective was thus to characterize the use of HeartLogic monitoring among patients implanted with a durable LVAD at a single quaternary referral center. In this retrospective cohort study, all patients with a HeartLogic capable device who underwent HeartMate 3 LVAD (Abbott, North Chicago, IL) implantation at a quaternary center were identified using the electronic health record from 2018 to 2020. All HeartLogic daily values and sensor alerts were recorded. All hospitalizations were clinician adjudicated as a HF or non-HF–associated hospitalization by prespecified criteria. To qualify as a HF hospitalization, patients needed to report at least one symptom of HF, demonstrate at least one objective sign of decompensated HF (including physical examination, imaging, biomarker, or hemodynamic evidence of congestion), and have received an escalation in diuretic therapy within 48 hours of admission. Baseline characteristics of the cohort (n = 14) are shown (Table 1). HeartLogic data were available for a mean of 664.5 (±348.5) days (see Table 1, Supplemental Digital Content, https://links.lww.com/ASAIO/B9). The median HeartLogic daily composite index value was 7, (Q1–Q3: 1–15) (see Figure 1, Supplemental Digital Content, https://links.lww.com/ASAIO/B5). All 14 patients had at least one HeartLogic alert (index value ≥16) by 6 months of follow up and there were 2,221 total days in alert status (mean 159 ± 102 days). There were 29 total hospitalizations over the study period (1.1 per patient/year) (see Table 2, Supplemental Digital Content, https://links.lww.com/ASAIO/B10), of which 10 were classified as HF hospitalizations (0.4 per patient/year). Non-HF hospitalizations (n = 19) occurred due to arrhythmia (n = 7), infection (n = 5), gastrointestinal bleeding (n = 2), hypovolemia (n = 2), pain (n = 2), and orthostasis (n = 1). There were three deaths. Table 1. - Baseline Characteristics of the Study Cohort Characteristic All Patients (N = 14) Demographics Age (yrs) (median, IQR) 69 (66–72) Male sex 10/14 (71.4%) Comorbidities Chronic pulmonary disease 7/14 (50.0%) Atrial fibrillation/flutter 10/14 (71.4%) Hypertension 12/14 (85.7%) Coronary disease 14/14 (100.0%) Diabetes 7/14 (50.0%) BMI ≥30 kg/m2 8/14 (57.1%) Myocardial infarction* 4/14 (28.6%) Coronary artery bypass grafting* 0/14 (0.0%) CKD grade 3 or higher 10/14 (71.4%) Vitals LVEF (%) (n = 6) 17.5 (2.7) Systolic blood pressure (mm Hg) 115 (15.4) Diastolic blood pressure (mm Hg) 74.4 (13.4) BMI (kg/m2) 30.5 (9.2) Labs NT-proBNP (pg/ml) (n = 10) 2,627.9 (2,370) eGFR (ml/min/1.73 m2)† 61.2 (25.3) Sodium (mmol/L) 138 (2.7) Hemoglobin (g/dl) 11.8 (2.3) Concomitant medications MRA 7/14 (50.0%) Loop diuretics 13/14 (92.9%) Beta blockers 13/14 (92.9%) ACEi/ARB 11/14 (78.6%) Calcium channel blockers 3/14 (21.4%) Hydralazine 7/14 (50.0%) Nitrates 1/14 (7.1%) All listed as mean (SD) unless noted.*Occurring within a 6 month window around index date.†Use of the CKD-EPI equation for calculation of eGFR (includes race as a factor).ACEi, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; BMI, body mass index; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; MRA, mineralocorticoid receptor antagonist; NT-proBNP, N-terminal-pro-brain natriuretic peptide. In the 48 hours before admission, the mean HeartLogic daily index value was 12.0 for HF hospitalizations, and 18.6 for non-HF hospitalizations. In the 2 weeks preceding HF hospitalizations, thoracic impedance significantly decreased (indicative of increasing pulmonary edema) (see Figure 2, Supplemental Digital Content, https://links.lww.com/ASAIO/B6), however, the overall HeartLogic daily index value largely did not change (Figure 1). In the 2 weeks before non-HF hospitalizations, there was a substantial increase in HeartLogic index values (Figure 1), which appeared mainly driven by an increase in heart rate (see Figure 2, Supplemental Digital Content, https://links.lww.com/ASAIO/B6). Trends in additional HeartLogic components can be viewed in Figure 3, Supplemental Digital Content, https://links.lww.com/ASAIO/B7, Figure 4, Supplemental Digital Content, https://links.lww.com/ASAIO/B8.Figure 1.: Distribution of HeartLogic sensor data before heart failure versus non-heart failure associated hospitalizations. The red curve indicates the average.This is the largest retrospective cohort study to date on the use of the HeartLogic multisensor device algorithm for HF remote monitoring in LVAD patients. This has yielded several important observations. A majority of HF hospitalizations occurred with a HeartLogic value below alert status, and HeartLogic values remained largely unchanged in the 2 weeks before HF hospitalizations, despite decreasing thoracic impedance. Such results suggest unclear utility of the HeartLogic system to predict HF decompensation in LVAD patients. There are several possible explanations for this. First, the HeartLogic alert was initially validated in the MultiSENSE study of patients with chronic HF with reduced ejection fraction and an existing HeartLogic capable device.3 Using development and validation cohorts, the investigators identified a HeartLogic value of ≥16 as providing a sensitivity of 70% for predicting HF events. The difference in HeartLogic performance in our cohort versus MultiSENSE may be driven by the differences in patient populations. Although MultiSENSE contained an ambulatory HF cohort, our study is comprised of higher risk, LVAD patients with increased rates of comorbid conditions, and lower left ventricular ejection fraction. There are also several physiologic reasons for which HeartLogic may not perform equally in LVAD patients. In LVAD patients, right HF represents a disproportionate share of morbidity as compared to traditional heart failure with reduced ejection fraction populations.6 As such, HeartLogic algorithm inputs including thoracic impedance and changes in heart sounds may differ in a patient with a well-functioning LVAD and concomitant RV dysfunction, where pulmonary edema and elevated left-sided filling pressures may not feature as prominently. Additionally, it is unclear how accurately HeartLogic can differentiate heart sounds, which it achieves via the accelerometer detecting cardiac vibrations, amidst the mechanical hum and vibrations created by an LVAD. The mean HeartLogic index value increased significantly in the 2 weeks before non-HF hospitalizations, and was within alert status range (mean 18.6) 48 hours before such events. This appeared to be largely driven by increases in heart rate. These results suggest that HeartLogic may provide useful predictive information for non-HF hospitalizations in LVAD patients, which represent up to 81% of readmissions in contemporary LVAD cohorts.2 Given the contribution of heart rate to the HeartLogic score, it follows that arrhythmias may be well tracked by HeartLogic. Serious infections and bleeding, which may be accompanied by increased respiratory rate, and decreased activity, may also be well tracked. HeartLogic alerts could thus potentially enable LVAD clinicians to diagnose, and treat, such conditions more rapidly. However, further inquiries are needed. Though this is the largest report on the use of HeartLogic in LVAD patients to date, our findings must be interpreted in the context of the modest sample size. While it is possible that a different HeartLogic alert value cut-point could provide greater sensitivity to predict admissions in LVAD populations, our study is not suited to examine this question. Whether other remote monitoring devices, including CardioMEMS, are more effective in LVAD populations remains unclear.7 However, as such sensors are less prevalent in LVAD patients, they would require additional costs and invasive procedures to use, as compared with sensors within existent ICDs, such as HeartLogic. In conclusion, HeartLogic may hold promise for prediction of non-HF associated hospitalizations, with less certainty regarding HF-associated hospitalizations, in LVAD patients. Further investigation is warranted in larger, prospective cohorts.