ABSTRACT
Background
Patients with heart failure (HF) can be monitored remotely using cardiovascular implantable electronic devices (CIEDs). In Japan, the HeartInsight algorithm has been implemented in selected implantable cardioverter defibrillators and cardiac resynchronization therapy devices. This retrospective study investigated whether elevations above, and daily fluctuations in, the HeartInsight score threshold aid prediction of hospitalization for HF.
Methods
This study included 61 patients. Data were extracted from the BIOTRONIK Home Monitoring system; the number of days of HeartInsight score calculation, coefficient of variation, standard deviation, maximum value during the observation period, and mean values were analyzed. The primary endpoint was the first hospitalization for HF after CIED implantation. Patients were divided into two groups (HF hospitalization and non‐HF hospitalization), and each parameter was compared between the groups.
Results
The first hospitalization for HF after CIED implantation occurred in 10 of 61 patients (16%). Although the HF hospitalization group was younger than the non‐HF hospitalization group, the two groups exhibited no differences in coronary risk factors, pharmacological therapies, or threshold elevation in HeartInsight scores. However, the HF hospitalization group demonstrated a tendency toward higher values, with significant differences observed in the maximum value and the difference between the maximum and mean values. Although few patients exhibited threshold elevation, greater daily variability was associated with a higher tendency toward HF hospitalization.
Conclusions
Although predicting HF hospitalization based solely on threshold elevation is challenging, monitoring daily fluctuations may aid prediction. Evaluating score variability is important, even among patients without threshold elevation.
Keywords: heart failure, hospitalization, implantable defibrillators, remote patient monitoring, risk assessment
HeartInsight threshold elevation alone may be insufficient to predict heart failure hospitalization. Monitoring daily variability in HeartInsight score may improve prediction of heart failure exacerbation, even in patients without threshold elevation or alert activation.
1. Introduction
In recent years, heart failure (HF) management has undergone major transformations owing to advances in pharmacological and nonpharmacological therapies. However, the prevention of acute HF onset through routine outpatient care alone is challenging, necessitating advancements to improve survival and maintain and improve quality of life. Therefore, patient education, disease management, and shared decision‐making have become important components of care.
In the field of nonpharmacological therapy, cardiovascular implantable electronic devices (CIEDs) play a major role in HF management. Implantable cardioverter defibrillators (ICDs) contribute to the prevention of sudden cardiac death, and cardiac resynchronization therapy (CRT) reduces mortality among patients with HF. Furthermore, remote monitoring (RM) offers advantages in the management of patients with CIEDs—including the early detection of clinical events, reduction in the frequency of routine outpatient visits, and decreased mortality—and has been established as a standard management strategy [1, 2, 3, 4].
Moreover, RM enables the confirmation of device battery status, lead impedance, sensing amplitude, pacing threshold, arrhythmic events, and other parameters with an accuracy comparable to that of in‐person consultations.
Recently, HF management functions using CIEDs have expanded, and manufacturers have developed proprietary algorithms with unique features for HF management [5, 6, 7].
HeartInsight was introduced by BIOTRONIK (Berlin, Germany) in November 2023. HeartInsight calculates a “HeartInsight score,” which quantifies HF risk based on physiological data accumulated in the BIOTRONIK Home Monitoring Service Center.
The HeartInsight score has been reported to be useful in predicting hospitalization for HF, and threshold elevation indicates changes in patient status, potentially resulting in early therapeutic interventions [8, 9]. However, a limitation of the measure is that its criteria remain unclear compared with those of conventional biomarkers and echocardiographic parameters [10].
In this study, to predict hospitalization owing to worsening HF and enable early therapeutic intervention, we retrospectively examined the usefulness of not only the threshold elevation of the HeartInsight score but also its temporal changes in predicting HF deterioration.
2. Methods
2.1. Study Design and Participants
This retrospective observational analysis, single‐center interventional study was conducted at the Department of Cardiovascular Medicine, Faculty of Medical Sciences, University of Fukui, Japan. Written informed consent was obtained from all patients. This study was approved by the institutional ethics committee (UMIN000023847) and conducted in accordance with the Declaration of Helsinki for research involving human participants.
The implementation of the HF hospitalization prediction algorithm HeartInsight (BIOTRONIK, Berlin, Germany) using RM was initiated in November 2023. The study population comprised patients who underwent implantation or generator replacement with compatible devices (Acticor 7 ICD ProMRI, Acticor 7 CRT‐D ProMRI, and Rivacor 7 CRT‐D ProMRI) and consented to undergo RM. Among these patients, those in whom RM data were continuously received for at least 30 days and in whom the HeartInsight score could be calculated were included in the analysis. Figure 1 depicts the participant selection process.
FIGURE 1.
Flow diagram of patient selection for HeartInsight analysis. Among 119 patients who underwent implantation of HeartInsight‐compatible ICD or CRT devices, patients with atrial fibrillation (n = 13), death before initiation of HeartInsight operation (n = 15), home monitoring transmission failure (n = 5), home monitoring reception period of ≤ 30 consecutive days (n = 8), and transfer to another hospital (n = 17) were excluded. A total of 61 patients were eligible for the HeartInsight analysis.
The exclusion criteria were persistent or permanent atrial fibrillation, death before the initiation of HeartInsight implementation, RM transmission failure, RM data reception for < 30 consecutive days, and transfer to another hospital (Figure 1).
The target observation period for this study was approximately 1 year after the initiation of HeartInsight score monitoring.
After CIED implantation, the patients were divided into two groups—namely, the HF hospitalization and non‐HF hospitalization groups. The primary endpoint was the usefulness of predicting hospitalization for HF based on the variability in the HeartInsight score. Additionally, the patients were classified into four groups based on the HeartInsight score threshold and standard deviation cutoff value. The incidence of hospitalization owing to HF was compared among the four groups.
2.2. Remote Monitoring
At our institution, cardiologists and clinical engineers provided explanations to patients when introducing RM using written materials and pamphlets based on the “Statement on Remote Monitoring of Cardiac Implantable Electronic Devices,” revised by the Japanese Heart Rhythm Society (JHRS) in 2022 [2]. In addition to RM, face‐to‐face clinical visits were conducted at least once a year, during which patient guidance was provided based on the RM transmission status.
With BIOTRONIK RM, data transmission and reception are performed daily, commencing from the day after registration, and healthcare professionals receive information regarding device function, arrhythmias, and therapy episodes. Alerts allow physicians to be promptly informed of the initial detection of atrial fibrillation, anti‐tachycardia pacing, or defibrillation therapy for ventricular tachycardia/ventricular fibrillation, battery depletion, and other events. When the non‐transmission status persists, telephone support is provided, and a new transmitter is prepared to maintain RM continuity.
2.3. HeartInsight
The HeartInsight score is calculated using physiological data stored in the BIOTRONIK Home Monitoring Service Center, including mean ventricular rate, mean resting ventricular rate, heart rate variability, activity, AT/AF burden, mean number of premature ventricular contractions per hour, and thoracic impedance [9, 11]. Calculation of the score requires activation of the program's thoracic impedance measurement. Although the algorithm used to calculate the score has not been publicly disclosed, each parameter's contribution is presented (Figure 2).
FIGURE 2.
Representative time course of the HeartInsight score (A) and its contributing parameters during the observation period (B) in patients with a HeartInsight‐compatible ICD or CRT device. The HeartInsight score was calculated from multiple physiological parameters, including mean ventricular rate, resting mean ventricular rate, heart rate variability, activity, AT/AF burden, premature ventricular contractions, and thoracic impedance.
HeartInsight was originally developed based on the SELENE HF study, in which baseline patient risk assessed by the SHFM was integrated with longitudinal remote monitoring parameters to improve prediction of heart failure hospitalization. An alert is generated when the score exceeds 45 for more than 3 days, and the recovery threshold is set at 35, with the score decreasing after therapeutic intervention [8, 9]. An alert is generated when the score exceeds 45 for more than 3 days, and the recovery threshold is set at 35, with the score decreasing after therapeutic intervention.
2.4. Seattle Heart Failure Model Calculator
HeartInsight was originally developed based on the SELENE HF study, in which baseline patient risk assessed by the SHFM was integrated with longitudinal remote monitoring parameters to improve prediction of heart failure hospitalization.
The SHFM was developed to predict prognosis based on data from patients with chronic HF in Western countries. In this study, the Japanese version of the SHFM reported by Shiraishi et al. [12] was used to predict hospitalization for HF after CIED implantation.
Blood test data obtained immediately before device implantation and medication information at discharge after CIED implantation were used for the analysis.
2.5. Statistical Analysis
Baseline characteristics of the study population (age, blood pressure, heart rate, and left ventricular ejection fraction) are presented as mean ± standard deviation, and the prevalences of hypertension, chronic kidney disease, diabetes mellitus, and medication are described as percentages.
For comparisons between the two groups, HeartInsight score–related variables are also presented as mean ± standard deviation. For statistical analysis, the t‐test was used for continuous variables, while the chi‐square test was used for categorical variables. EZR (Easy R) was used for the statistical analysis.
Additional multivariable logistic regression analysis adjusted for clinically important factors was performed to evaluate the association between HeartInsight score variability and HF hospitalization.
EZR is a statistical software program that extends the functionality of R and R Commander. A p value of < 0.05 was considered statistically significant.
3. Results
3.1. Patient Characteristics
Table 1 shows that among the 119 patients who underwent ICD or CRT device implantation at our institution, 61 who consented to undergo RM were included in the analysis. Among these patients, 10 (16.4%) reported hospitalization for HF after CIED implantation. Among the 51 patients who did not experience hospitalization for HF, 27 (52.9%) showed alerts because of threshold elevation; however, no hospitalization for HF occurred during the observation period. The overall mean age was 69.0 ± 11.1 years, and the group comparison indicated that patients in the HF hospitalization group were significantly younger than those in the non‐hospitalization group (62.4 ± 12.6 vs. 70.9 ± 9.9 years, p = 0.037).
TABLE 1.
Patient characteristics.
| N = 61 | HF hospitalization (N = 10) | HF hospitalization group (N = 10) | Non‐HF hospitalization group (N = 51) | p |
|---|---|---|---|---|
| Age, y.o. | 69.0 ± 11.1 | 62.4 ± 12.6 | 70.2 ± 9.9 | 0.037 |
| Gender, M/F | 45/16 | 9/1 | 36/15 | 0.208 |
| BMI, M/F | 22.8/22.6 | 24.6/26.2 | 22.3/22.4 | 0.192 |
| NYHA class, I/II/III/IV | 0/57/4/0 | 0/9/1/0 | 0/48/3/0 | N/A |
| ICM/non‐ICM | 30/31 | 5/5 | 25/26 | 0.955 |
| CIED, n (%) | ||||
| ICD/CRT‐D | 24 (39.3)/37 (60.7) | 4 (40.0)/6 (60.0) | 20 (39.2)/31 (60.8) | N/A |
| Blood pressure | ||||
| Systolic, mmHg | 107.2 ± 16.2 | 101.8 ± 9.5 | 108.3 ± 17.1 | 0.255 |
| Diastolic, mmHg | 65.0 ± 10.8 | 58.9 ± 7.9 | 66.2 ± 10.9 | 0.051 |
| Pulse rate, bpm | 71.1 ± 9.5 | 71.7 ± 5.8 | 71.2 ± 10.0 | 0.890 |
| LVEF, % | 34.8 ± 12.6 | 34.6 ± 11.1 | 34.8 ± 12.9 | 0.964 |
| pEF/mrEF/rFF, n (%) | 10 (16.4)/7 (11.5)/44 (72.1) | 1 (10.0)/2 (20.0)/7 (70.0) | 9 (17.6)/5 (9.8)/37 (72.5) | N/A |
| Hypertension, n (%) | 32 (52.5) | 3 (30.0) | 29 (56.8) | 0.124 |
| CKD, n (%) | 21 (34.4) | 3 (30.0) | 18 (35.2) | 0.752 |
| Diabetes mellitus, n (%) | 30 (49.2) | 5 (50.0) | 25 (49.0) | 0.956 |
| Medications, n (%) | ||||
| Diuretics, n (%) | 33 (54.1) | 6 (60.0) | 27 (52.9) | 0.688 |
| β‐blocker, n (%) | 52 (85.2) | 8 (80.0) | 44 (86.3) | 0.616 |
| ACEI/ARB, n (%) | 26 (42.6) | 3 (30.0) | 23 (45.1) | 0.386 |
| ARNI, n (%) | 4 (6.6) | 0 (0) | 4 (7.8) | N/A |
| MRA, n (%) | 26 (42.6) | 6 (23.1) | 21 (80.7) | 0.281 |
| Statin, n (%) | 33 (54.1) | 5 (50.0) | 28 (54.9) | 0.781 |
| Antihyperuricemics, n (%) | 6 (9.8) | 1 (10.0) | 5 (9.8) | 0.985 |
| SGLT2 inhibitors, n (%) | 20 (32.8) | 4 (40.9) | 16 (31.4) | 0.602 |
Abbreviations: ACEI, angiotensin‐converting enzyme inhibitor; ARBs, angiotensin II receptor blocker; ARNI, electronic device; CKD, chronic kidney disease; CRT‐D, cardiac resynchronization therapy with defibrillator; HFmrEF, heart failure with mildly reduced EF; HFpEF, heart failure with preserved EF; HFrEF, heart failure with reduced EF; ICD, implantable cardioverter‐defibrillator; LVEF, left ventricular ejection fraction; MRA, mineralocorticoid receptor antagonist; NYHA, New York Heart Association; SGLT2, sodium‐glucose cotransporter 2.
In addition, in the multivariable analysis, HeartInsight score variability showed a tendency toward association with HF hospitalization; however, statistical significance was not reached after adjustment for CKD and LVEF (OR 1.03, 95% confidence interval [CI] 0.98–1.09, p = 0.287).
No significant differences were observed in the proportion of men, body mass index by sex, prevalence of underlying diseases, vital signs, or medication use between the two groups.
3.2. HeartInsight Score
As presented in Table 2, the two groups exhibited no significant difference in the RM observation period.
TABLE 2.
Results of the analysis of the HeartInsight score and the Seattle Heart Failure Model.
| HF hospitalization (N = 10) | Non‐HF hospitalization (N = 51) | p | |
|---|---|---|---|
| Observation period, days | 379.5 ± 119.5 | 333.6 ± 138.7 | 0.341 |
| HeartInsight score | |||
| Max | 108.0 ± 57.6 | 78.4 ± 49.9 | 0.051 |
| Minimum | 9.6 ± 13.8 | 9.8 ± 9.2 | 0.946 |
| Mean | 37.2 ± 13.9 | 31.3 ± 18.1 | 0.344 |
| Max–mean | 70.8 ± 56.5 | 47.0 ± 38.5 | 0.043 |
| Coefficient variation | 0.5 ± 0.3 | 0.4 ± 0.2 | 0.300 |
| SHFM | |||
| SHFM score | 0.193 ± 0.6 | 0.086 ± 0.4 | 0.612 |
| One‐year survival rate | 93.3 ± 4.7 | 93.1 ± 2.4 | 0.866 |
Abbreviations: HF, heart failure; SHFM, Seattle Heart Failure Model.
The mean follow‐up period was 337.9 ± 136.8 days, and the median observation period was 358.5 days (interquartile range: 259.8–477.5 days; range: 38.0–493.0 days).
In the HF hospitalization group, the median observation period was 442 days (IQR: 275.5–467.3 days; range: 116–489 days). In the non‐HF hospitalization group, the median observation period was 340 days (IQR: 269–479 days; range: 38–493 days).
The maximum and mean values of the HeartInsight score during the observation period were calculated (Figure 3). The group comparison revealed significant differences in the maximum value and the difference between the maximum and mean values (Table 2).
FIGURE 3.
HeartInsight score variability during the observation period. This figure illustrates the variability in the HeartInsight score during the observation period. The mean value, standard deviation, coefficient of variation, and maximum and minimum values of the HeartInsight score were calculated for each patient and used as indices of score variability.
Regarding the coefficient of variation and mean value during the observation period, although the daily fluctuations were large, no threshold elevation was observed, and no significant differences were identified. No significant differences were observed between the HF hospitalization and non‐HF hospitalization groups in either SHFM score (0.193 ± 0.6 vs. 0.086 ± 0.4, p = 0.612) or 1‐year survival rate predicted by the SHFM (93.3 ± 4.7 vs. 93.1 ± 2.4, p = 0.866).
As illustrated in Figure 4, the mean HeartInsight score during the observation period exceeded the score threshold of 45 in 13 patients, whereas more than half of the patients remained below the threshold throughout the observation period.
FIGURE 4.
HeartInsight score variability and hospitalization for heart failure. Scatter plots show the relationship between the HeartInsight score and standard deviation (SD) of the HeartInsight score for each patient. The vertical axis represents the HeartInsight score, and the horizontal axis represents the SD of the HeartInsight score. Score variability was assessed using the median SD of the entire study population, and the median value (14.7) was used as the cutoff value. Red stars indicate patients who experienced heart failure hospitalization during the observation period. Dashed lines indicate the predefined HeartInsight score threshold (45) and SD cutoff value (14.7).
Regarding the standard deviation, zone classification was performed using an overall median value of 14.7 as the cutoff.
Among patients without threshold elevation and with low daily score variability, only three progressed to hospitalization for HF, indicating a relatively low incidence. By contrast, among patients who experienced hospitalization for HF, although no threshold elevation was observed, many progressed to HF when the standard deviation was large.
As depicted in Figure 5, zone classification was performed, and the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and false‐positive rate were evaluated using the HeartInsight score and standard deviation. In Group I (HeartInsight score < 45, SD < 14.7), hospitalization for HF occurred in 3 of 34 patients (8%), with a sensitivity of 0.30, specificity of 0.392, PPV of 8.8%, and NPV of 74.1%, indicating low sensitivity and a high false‐positive rate. Contrastingly, in Group II (HeartInsight score < 45, SD > 14.7), hospitalization for HF occurred in 5 of 14 patients (36%), with a sensitivity of 0.50, specificity of 0.824, PPV of 35.7%, and NPV of 89.4%. In Group III (HeartInsight score > 45, SD < 14.7), the hospitalization rate for HF was 0 of 3 patients (0%). In Group IV (HeartInsight score > 45, SD > 14.7), the HF hospitalization rate was 2 of 8 patients (20%), with a sensitivity of 0.20, specificity of 0.843, PPV of 20.0%, and NPV of 84.3%. Among Groups I–IV, Group II displayed the highest sensitivity, specificity, PPV, and NPV for predicting hospitalization for HF. Although significant differences were observed among Groups I to IV in these factors, the baseline characteristics of the patients classified into these groups are shown in Table 3, and no significant differences were found in any of the variables.
FIGURE 5.
Distribution of hospitalization for heart failure based on the threshold and variability of the HeartInsight score. The patients were classified into four groups based on the HeartInsight score threshold (45) and standard deviation cutoff value (14.7). The incidence of hospitalization due to heart failure was compared among the four groups. Sensitivity, specificity, positive predictive value, and negative predictive value were evaluated to assess the predictive performance of the combined use of HeartInsight score threshold and variability.
TABLE 3.
Baseline patient characteristics according to Groups I–IV.
| N = 61 | Group I (N = 34) | Group II (N = 14) | Group III (N = 3) | Group IV (N = 10) |
|---|---|---|---|---|
| Age, y.o. | 69.9 ± 9.5 | 63.7 ± 12.1 | 67.3 ± 9.2 | 74 ± 10.9 |
| Gender, M/F | 24/10 | 11/13 | 3/0 | 7/3 |
| BMI, M/F | 21.8/23.2 | 25.3/21.8 | 23.6/0 | 21.6/21.7 |
| NYHA class, I/II/III/IV | 0/32/2/0 | 1/12/1/0 | 0/3/0/0 | 0/9/1/0 |
| ICM/non‐ICM | 30/31 | 5/9 | 2/0 | 7/3 |
| CIED, n (%) | ||||
| ICD/CRT‐D | 16 (47.1)/18 (52.9) | 5 (35.7)/9 (64.3) | 1 (33.3)/2 (66.7) | 2 (20.0) /8 (80.0) |
| Blood pressure | ||||
| Systolic, mmHg | 109.7 ± 17.5 | 104.7 ± 13.9 | 102.7 ± 9.8 | 101.9 ± 13.2 |
| Diastolic, mmHg | 66.2 ± 10.8 | 62.5 ± 11.4 | 67.3 ± 3.3 | 63.7 ± 10.6 |
| Pulse rate, bpm | 71.4 ± 8.9 | 69.3 ± 10.5 | 68.3 ± 10.1 | 68.9 ± 8.8 |
| LVEF, % | 36.3 ± 14.3 | 34.1 ± 8.7 | 33.9 ± 11.5 | 30.7 ± 10.1 |
| pEF/mrEF/rFF, n (%) | 6 (17.6)/4 (11.8)/24 (70.6) | 1 (7.1)/2 (14.3) /11 (78.6) | 1 (33.3)/0 (0) /2 (66.7) | 1 (10.0)/1 (10.0)/8 (80.0) |
| Hypertension, n (%) | 20 (64.5) | 4 (28.6) | 2 (66.7) | 6 (60.0) |
| CKD, n (%) | 11 (32.5) | 8 (57.1) | 0 (0) | 2 (20.0) |
| Diabetes mellitus, n (%) | 13 (38.2) | 8 (57.1) | 1 (33.3) | 8 (80.0) |
| Medications, n (%) | ||||
| Diuretics, n (%) | 19 (55.8) | 7 (50.0) | 1 (33.3) | 6 (60.0) |
| β‐blocker, n (%) | 27 (79.4) | 12 (85.7) | 3 (100) | 10 (100) |
| ACEIs/ARBs, n (%) | 16 (47.1) | 7 (50.0) | 1 (33.3) | 1 (10.0) |
| ARNI, n (%) | 4 (11.7) | 1 (7.1) | 0 (0) | 3 (30.0) |
| MRAs, n (%) | 12 (35.3) | 9 (64.2) | 0 (0) | 6 (60.0) |
| Statins, n (%) | 15 (44.1) | 10 (71.4) | 2 (66.7) | 6 (60.0) |
| Antihyperuricemics, n (%) | 3 (8.8) | 3 (21.4) | 0 (0) | 0 (0) |
| SGLT2 inhibitors, n (%) | 9 (26.5) | 5 (35.7) | 1 (33.3) | 5 (50.0) |
Abbreviations: ACEI, angiotensin‐converting enzyme inhibitor; ARBs, angiotensin II receptor blocker; ARNI, electronic device; CKD, chronic kidney disease; CRT‐D, cardiac resynchronization therapy with defibrillator; HFmrEF, heart failure with mildly reduced EF; HFpEF, heart failure with preserved EF; HFrEF, heart failure with reduced EF; ICD, implantable cardioverter‐defibrillator; LVEF, left ventricular ejection fraction; MRA, mineralocorticoid receptor antagonist; NYHA, New York Heart Association; SGLT2, sodium‐glucose cotransporter 2.
In addition, to clarify the incremental predictive value of HeartInsight score variability beyond conventional prediction of HF hospitalization using the HeartInsight score alone, we additionally performed ROC curve analyses for HF hospitalization prediction using the presence or absence of HeartInsight score alerts, mean HeartInsight score, MAX–mean, and HeartInsight score SD.
ROC curve analysis demonstrated that HeartInsight score SD showed the highest discriminative ability for HF hospitalization, with an AUC of 0.673 (95% confidence interval [CI]: 0.501–0.844), compared with HeartInsight alerts (AUC 0.635, 95% CI: 0.487–0.738), mean HeartInsight score (AUC 0.641, 95% CI: 0.469–0.813), and MAX–mean (AUC 0.598, 95% CI: 0.380–0.816). The optimal cutoff value for HeartInsight score SD was 15.145, with a sensitivity of 70.0% and specificity of 66.7%.
Furthermore, DeLong's test was performed to compare the ROC curves. Although HeartInsight score SD numerically showed a higher AUC than MAX–mean, HeartInsight alerts and mean HeartInsight score, no statistically significant differences were observed in any comparison (SD vs. MAX–mean: p = 0.169; SD vs. HeartInsight alerts: p = 0.529; SD vs. mean score: p = 0.701).
3.3. Seattle Heart Failure Model
No significant differences were observed in SHFM score or 1‐year survival rate between the HF hospitalization and non‐hospitalization groups (SHFM score: 0.193 ± 0.6 vs. 0.086 ± 0.4, p = 0.612; 1‐year survival rate: 93.3% ± 4.7% vs. 93.1% ± 2.4%, p = 0.866).
4. Discussion
The two groups displayed significant differences in the maximum HeartInsight score and the difference between the maximum and mean values [8, 9]. However, in both groups, the daily HeartInsight scores remained below the default threshold, suggesting a high number of false negatives and low sensitivity. Notably, only two patients progressed to hospitalization for HF with threshold elevation, whereas eight patients were hospitalized for HF despite remaining below the threshold. These findings indicate the limitations of predicting hospitalization for HF based solely on the HeartInsight score threshold [10].
However, the results suggest that observing daily score variability may enable the prediction of hospitalization for HF [5, 13, 14].
No significant correlation was observed between the observation period and HeartInsight score SD (R 2 = 0.054, p = 0.073). These findings suggest that HeartInsight score variability was not simply influenced by the duration of observation. Rather, increased variability may reflect fluctuations in heart failure status and could therefore have clinical relevance for predicting HF hospitalization.
In the analysis of variability, patients with variability greater than the cutoff value displayed a higher proportion of HF hospitalizations, and variability was strongly influenced by mean ventricular rate variability, activity, and thoracic impedance. However, there were cases in which hospitalization for HF occurred despite no threshold elevation and variability below the cutoff value, as well as cases in which threshold elevation occurred but hospitalization did not occur when the SD was below 14.7, indicating that the prediction accuracy was limited when using a single indicator.
In the group classification analysis using the HeartInsight score threshold and SD, Group II exhibited the highest risk of HF deterioration, with several cases progressing to hospitalization without threshold elevation, suggesting the need for optimization of threshold settings. By contrast, in Group III, alerts owing to threshold elevation were observed; however, hospitalization did not occur, indicating that a reduction in false positives remains a challenge. Although stable or unstable trends were observed in Groups I and IV, the association with hospitalization was not uniform, suggesting that long‐term observation of variability is important. Although the HeartInsight algorithm is not publicly disclosed and its clinical interpretation remains difficult, combining physiological parameters such as heart rate variability and thoracic impedance may improve predictive accuracy [6, 7, 15, 16].
This study has several important limitations. First, the sample size was relatively small, with only 61 patients included in the analysis and 10 HF hospitalization events observed during follow‐up. Therefore, the statistical power of the present study was limited, and the results should be interpreted with caution.
In addition, because this was a single‐center retrospective study with a limited sample size, the generalizability of the findings to broader HF populations remains uncertain. Accordingly, the clinical utility of HeartInsight score variability cannot be conclusively determined based on the present findings alone, and larger multicenter studies are warranted to further validate its usefulness for predicting HF hospitalization.
4.1. Study Limitations
This study had a relatively small sample size of 61 patients and only 10 cases of hospitalization for HF, resulting in limited statistical power. Additionally, the calculation method for the HeartInsight score and the contribution of each parameter have not been publicly disclosed, making it difficult to clarify how changes in individual physiological parameters affect the score in each case. Furthermore, patients with atrial fibrillation were excluded, resulting in the exclusion of a population at high risk for HF deterioration, and caution is required when generalizing the results for the prediction of HF hospitalization. Future studies should include prospective multicenter collaborations and long‐term follow‐up to validate the optimal utilization of HeartInsight, reevaluate threshold settings, and assess comprehensive evaluations incorporating multiple parameters.
5. Conclusion
An increase in the HeartInsight score above the predefined threshold alone is insufficient to reliably predict hospitalization for HF. However, continuous monitoring of day‐to‐day fluctuations in scores may enable the prediction of HF deterioration. Furthermore, although alerts are not triggered when the threshold is not exceeded, careful assessment of score variability is crucial, even in the absence of threshold elevation.
Author Contributions
All authors contributed to the conception and design of this study. Material preparation, data collection, and analysis were performed by T.S., H.U., N.M., M.T., Y.M., K.H., S.H., M.Y., R.N., T.T., M.M., M.M., K.H., N.T., K.I., and H.T. The first draft of the manuscript was written by T.S., and all authors commented on previous versions of the manuscript. All the authors have read and approved the final version of the manuscript.
Funding
This study was partially supported by a research grant from the University of Fukui, Japan.
Ethics Statement
This retrospective study of human participants was performed in compliance with the ethical principles established by the Institutional and National Research Committees, as well as the 1964 Declaration of Helsinki and its subsequent revisions or equivalent ethical guidelines. The study protocol was reviewed and approved by the Institutional Review Board (Human Investigation Committee) of the University of Fukui Hospital.
Consent
Owing to the retrospective nature of the study, the need for informed consent was waived for administrative data.
Conflicts of Interest
Dr. Hiroshi Tada has received lecture fees and speaker honoraria from Daiichi Sankyo Company Ltd., Novartis Pharma K.K., Medtronic Japan Co. Ltd., BIOTRONIK Japan Inc., Bristol Myers Squibb, and Boston Scientific Japan K.K. He has also been awarded research funding for investigator‐initiated studies unrelated to the present manuscript from Abbott Medical Japan LLC, Daiichi Sankyo Company Ltd., Nippon Boehringer Ingelheim Co. Ltd., Otsuka Pharmaceutical Co. Ltd., Eli Lilly Japan K.K., and Marubun Tsusho K.K. The other authors declare no conflicts of interest.
Acknowledgments
The authors would like to express their sincere gratitude to the cardiologists for their valuable guidance and advice in clinical practice and to the medical clerks for their cooperation in remote monitoring management. The authors also thank the clinical engineers for their support in balancing the clinical duties and research activities. The authors thank Ms. Hiromi Nishimura, Ms. Motoko Oku, Ms. Mari Kurata, Ms. Yoshiko Kurose, and Ms. Chiharu Kawahara (Administrative Assistants, Department of Medical Engineering Device Management) for their technical and administrative assistance. We would also like to thank Editage (www.editage.jp) for English language editing.
Data Availability Statement
The data used in this study cannot be shared publicly to protect the privacy of individuals who participated in the study. The data will be shared upon reasonable request to the corresponding authors.
References
- 1. Ferrick A. M., Varma N., Mittal S., et al., “2023 HRS/EHRA/APHRS/LAHRS Expert Consensus Statement on Remote Monitoring of Cardiovascular Implantable Electronic Devices,” Heart Rhythm 20 (2023): e263–e321. [Google Scholar]
- 2. Japanese Heart Rhythm Society (JHRS) , Statement on Remote Monitoring of Cardiac Implantable Electronic Devices (2022 Revision) (JHRS, 2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Nishii N., “Remote Monitoring of Cardiac Implantable Electronic Devices,” Journal of Arrhythmia 30 (2014): 395–412. [Google Scholar]
- 4. Varma N., “Remote Monitoring of ICDs and CRTs,” Journal of Arrhythmia 29 (2013): 144–152. [Google Scholar]
- 5. Liu P. S. and Tse H. F., “Implantable Sensors for Heart Failure Monitoring,” Journal of Arrhythmia 29 (2013): 314–319. [Google Scholar]
- 6. Scholte N. T. B., Gürgöze M. T., Aydin D., et al., “Telemonitoring for Heart Failure: A Meta‐Analysis,” European Journal of Heart Failure 44 (2023): 2911–2926. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Zito A., Restivo A., Ciliberti G., et al., “Heart Failure Management Guided by Remote Implant‐Based Monitoring Strategies: A Meta‐Analysis,” International Journal of Cardiology 388 (2023): 131163. [DOI] [PubMed] [Google Scholar]
- 8. Santini L., Adamo F., Danisi N., et al., “Optimizing Remote Heart Failure Management: First Experiences With the HeartInsight Score for Implanted Defibrillators,” Journal of Arrhythmia 40 (2024): 643–646. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Bulava A., De Sousa J., Guédon‐Moreau L., et al., “Evaluating an Alert‐Based Multiparametric Algorithm for Predicting Worsening Heart Failure Hospitalisations (HeartInsight),” Open Heart 12 (2025): e003474. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Botto G. L., Sinagra G., Bulava A., et al., “Predicting Worsening Heart Failure Hospitalizations in Patients With Implantable Cardioverter‐Defibrillators: Is it all About Alerts? A Pooled Analysis of Nine Trials,” Europace 26 (2024): euae032. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Zanotto G. and Capucci A., “HeartInsight: From SELENE HF to Implementation in Clinical Practice,” European Heart Journal Supplements: Journal of the European Society of Cardiology 25, no. Suppl C (2023): C337–C343. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Shiraishi Y., Kohsaka S., Nagai T., et al., “Validation and Recalibration of the Seattle Heart Failure Model in Japanese Acute Heart Failure Patients,” Journal of Cardiac Failure 25 (2019): 561–567. [DOI] [PubMed] [Google Scholar]
- 13. Nishii N., Sakata Y., Murohara T., et al., “Prediction of Heart Failure Events Based on Physiologic Sensor Data in Defibrillator Patients,” ESC Heart Failure 11 (2024): 3322–3331. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Zwaenepoel B. A. C., Kluft A., Feijen M., et al., “Impact of Multisensor CIED‐Based Heart Failure Monitoring on Outcomes,” Current Heart Failure Reports 22 (2025): 21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. D'Onofrio A., Solimene F., Bulava A., et al., “A Multiparametric Heart Failure Score at Baseline is Associated With Long‐Term Outcome in Patients With Remotely Monitored Implantable Cardioverter‐Defibrillators: A Pooled Analysis of 9 Clinical Trials,” Heart Rhythm 22 (2025): 1055–1063. [DOI] [PubMed] [Google Scholar]
- 16. De Lathauwer I. L. J., Nieuwenhuys W. W., Hafkamp F., et al., “Remote Patient Monitoring in Heart Failure: A Comprehensive Meta‐Analysis of Effective Programme Components for Hospitalization and Mortality,” European Journal of Heart Failure 27 (2024): 1670–1685. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data used in this study cannot be shared publicly to protect the privacy of individuals who participated in the study. The data will be shared upon reasonable request to the corresponding authors.