Baseline clinical characteristics of the study cohort

Between 2000 and 2018, 1189 patients were implanted successfully with CRT-P (46%) and 1414 with CRT-D (54%) at our center (Table 1). The CRT-D group comprised significantly younger (67 (60–73) vs. 69 (62–76) years, p < 0.001) and less symptomatic patients (NYHA functional class III/IV: 51 vs. 61%, p < 0.001) and a higher proportion of men (81 vs. 67%, p < 0.001). More patients had ischemic etiology (58 vs. 38%, p < 0.001) and experienced ventricular arrhythmias previously (37 vs. 11%, p < 0.001) in the CRT-D group than in the CRT-P group. CRT-D patients had wider QRS complexes (160 (140–170) vs. 160 (150–180) ms, p = 0.003), lower LVEF (28 (23–32) vs. 30 (25–35)%, p < 0.001), and larger LV diameters than CRT-P patients. Beta-blockers (90 vs. 87%, p = 0.018), MRAs (73 vs. 62%, p < 0.001), and amiodarone (32 vs. 21%, p < 0.001) were administered more frequently in the CRT-D group.

Table 1 Baseline clinical characteristics of the entire study cohort

A total of 1273 (49%) patients with ischemic etiology and 1330 (51%) patients with non-ischemic etiology were included in our retrospective database. From the former, 456 (36%) patients were implanted with CRT-P and 817 (64%) with CRT-D, whereas from the latter, 733 (55%) patients were implanted with CRT-P and 597 (45%) with CRT-D. The baseline clinical characteristics of the ischemic and non-ischemic subgroups are presented in Supplemental Tables 2 and 3, respectively.

All-cause mortality of CRT-D and CRT-P patients

Over the median follow-up of 8.0 (5.3–12.1) years, a total of 1572 patients died in our cohort. Based on Kaplan–Meier estimates, 5- and 10-year mortality were 45 (43–47)% and 71 (68–73)% in the entire cohort, 43 (40–46)% and 71 (67–74)% in patients with CRT-D, and 48 (45–50)% and 71 (68–74)% in those with CRT-P, respectively (Supplemental Table 4). CRT-D was not associated with a greater survival benefit than CRT-P in univariable Cox regression (unadjusted HR 0.94, 95% CI 0.85–1.03, p = 0.220). However, after adjustment for relevant clinical covariates (i.e., age, sex, etiology of heart failure, and history of ventricular arrhythmia), CRT-D was superior to CRT-P (adjusted HR 0.83, 95% CI 0.74–0.92, p < 0.001) in terms of all-cause mortality (Fig. 1A). When sequential multivariable models were built by adding these covariates to the univariable model in a stepwise manner (Supplemental Table 5), we found both etiology and history of ventricular arrhythmia to be negative confounding variables (i.e., not adjusting for them would result in the underestimation of the true strength of the association between the type of the implanted device and all-cause mortality). As shown by absolute risk reductions calculated at each year of the follow-up, CRT-D was associated with better survival than CRT-P only in the 2-to-6-year interval after implantation (Fig. 2A, Supplemental Table 4). In the sensitivity analysis, device type was found to be a significant predictor of the composite endpoint in the multivariable (adjusted HR 0.85, 95% CI 0.76–0.95, p = 0.003) but not the univariable Cox model (unadjusted HR 0.97, 95% CI 0.88–1.07, p = 0.498) (Supplemental Fig. 1A).

Fig. 1

Kaplan–Meier estimates of the time to death from any cause in the entire study cohort, ischemic patients, and non-ischemic patients. Univariable and multivariable Cox proportional hazards models were used to compute hazard ratios with 95% confidence intervals. Each multivariable model included the following features: device type, age, sex, history of atrial fibrillation, and history of ventricular arrhythmia. CI confidence interval, CRT-D cardiac resynchronization therapy defibrillator, CRT-P cardiac resynchronization therapy pacemaker, HR hazard ratio

Fig. 2

Absolute risk reduction associated with CRT-D vs. CRT-P at each year of the follow-up. Absolute risk reductions and their confidence intervals were calculated based on Kaplan–Meier estimates. Absolute risk reduction was considered significant if the value 0 fell outside of its confidence interval. ARR absolute risk reduction, CRT cardiac resynchronization therapy, CRT-D cardiac resynchronization therapy defibrillator, CRT-P pacemaker, pp percentage point

Among those with ischemic etiology, 882 patients reached the primary endpoint over the median follow-up duration of 7.9 (5.1–12.2) years, and 5- and 10-year mortality were 54 (51–57)% and 80 (77–82)%, respectively (Supplemental Table 4). In this subset of patients, CRT-D was associated with a lower risk of death compared to CRT-P according to both univariable (unadjusted HR 0.84, 95% CI 0.74–0.97, p = 0.014) and multivariable analysis (adjusted HR 0.86, 95% CI 0.75–0.99, p = 0.043) (Fig. 1B). Based on absolute risk reduction, CRT-D was associated with better survival than CRT-P only in the 2-to-7-year interval following implantation (Fig. 2B, Supplemental Table 4). Sensitivity analysis showed that the type of device is a significant predictor of the composite endpoint only in the univariable (unadjusted HR 0.87, 95% CI 0.76–0.99, p = 0.046) but not the multivariable Cox model (adjusted HR 0.88, 95% CI 0.76–1.01, p = 0.071) (Supplemental Fig. 1B).

In patients with non-ischemic etiology, 690 individuals died over 8.4 (5.5–12.0) years. Five-year mortality was 36 (34–39)%, whereas 10-year mortality was 61 (58–65)% (Supplemental Table 4). The survival benefit from CRT-D vs. CRT-P was confirmed by both univariable and multivariable analysis (unadjusted HR 0.82, 95% CI 0.70–0.96, p = 0.013 and adjusted HR 0.78, 95% CI 0.66–0.93, p = 0.005) (Fig. 1C). We also observed that patients with CRT-D exhibited significantly lower mortality than those with CRT-P for up to 9 years after the implantation (Fig. 2C, Supplemental Table 4). In terms of the composite endpoint, CRT-D was associated with better outcomes than CRT-P based on the univariable (unadjusted HR 0.86, 95% CI 0.74–0.99, p = 0.046) and the multivariable analysis as well (adjusted HR 0.81, 95% CI 0.69–0.96, p = 0.013) (Supplemental Fig. 1C).

In addition, we also assessed the association of the device type with all-cause mortality in patients with and without a history of atrial fibrillation (Supplemental Fig. 2). Multivariable Cox regression analysis revealed that CRT-D was associated with better survival than CRT-P only in the latter (adjusted HR 0.77, 95% CI 0.67–0.89, p < 0.001) but not in the former subgroup of patients (adjusted HR 0.91, 95% CI 0.77–1.07, p = 0.260).

Clinical characteristics and outcomes of the TDA-derived phenogroups

Using TDA, we created a circular network containing 2359 patients in which we could delineate five phenogroups showing distinct patterns of clinical characteristics and outcomes (Figs. 3 and 4, Table 2, Supplemental Tables 6–10). Etiology appeared to be an important factor, as three phenogroups (1, 2, and 3) included almost exclusively patients with non-ischemic etiology, whereas the other two phenogroups (4 and 5) comprised predominantly ischemic patients (Table 2). As expected, CRT-D was chosen more frequently in the latter two phenogroups than in the non-ischemic ones (Table 2).

Fig. 3

The topological network with the five identified phenogroups of CRT patients. The topological network was created using twenty-five features (metric: normalized correlation, lenses: 2 × multidimensional scaling [resolution: 29, gain: 1.6, equalized]). Each node represents a collection of similar patients, and two nodes are connected if they have at least one patient in common. Nodes are color-coded based on all-cause mortality. The topological network was divided into five regions (i.e., phenogroups) based on all-cause mortality. CRT cardiac resynchronization therapy, CRT-D cardiac resynchronization therapy defibrillator, CRT-P cardiac resynchronization therapy pacemaker

Fig. 4

Clinical characteristics of the five phenogroups identified using topological data analysis. AF atrial fibrillation, CRT-D cardiac resynchronization therapy defibrillator, DM diabetes mellitus, HTN hypertension, LBBB left bundle branch block, LVEF left ventricular ejection fraction, MRA mineralocorticoid receptor antagonist, NYHA New York Heart Association

Table 2 Baseline clinical characteristics of the phenogroups identified using topological data analysis

When we compared the non-ischemic phenogroups (Table 2), we found that patients in phenogroup 1 were more likely to be women (40 vs. 30%, p = 0.006), had smaller LV end-diastolic and end-systolic diameters (61 (56–68) vs. 64 (59–71) mm, p = 0.001 and 52 (45–58) vs. 54 (47–61) mm, p = 0.025, respectively), and a higher percentage of them took MRAs than in phenogroup 2 (76 vs. 65%, p = 0.003). In phenogroup 3, a higher percentage of patients presented with an NYHA functional class III or IV than in phenogroup 2 (66 vs. 57%, p = 0.041). Among the three non-ischemic groups, phenogroup 3 showed the lowest LVEF, the longest QRS duration, and the lowest rate of LBBB (Table 2).

When the two dominantly ischemic phenogroups were compared (Table 2), we observed that patients in phenogroup 5 were older (70 (63–76) vs. 67 (60–75) years, p = 0.009), had lower hemoglobin concentration (14 (12–15) vs. 14 (14–15) g/dL, p = 0.020), presented more frequently with LBBB (69 vs. 57%, p = 0.001), and were more likely to have ischemic etiology than individuals in phenogroup 4 (99 vs. 84%, p < 0.001).

Differences could also be observed in all-cause mortality between the phenogroups (Fig. 5, Supplemental Table 11). Although the survival of the three non-ischemic phenogroup was not significantly different from each other and the two dominantly ischemic phenogroups exhibited similar survival, all three non-ischemic phenogroups showed significantly better survival than those including dominantly ischemic patients (Fig. 5, Supplemental Table 11). In the sensitivity analysis, we also found similar results regarding the composite endpoint (Supplemental Fig. 3, Supplemental Table 12).

Fig. 5

Kaplan–Meier estimates of the time to death from any cause in the five phenogroups identified using topological data analysis. CRT cardiac resynchronization therapy, TDA topological data analysis

ICD-related survival benefit in the different phenogroups

The survival of CRT-D and CRT-P patients were plotted and compared in each phenogroup (Fig. 6). From the three non-ischemic phenogroups, CRT-D was superior to CRT-P only in phenogroup 2 (unadjusted HR 0.65, 95% CI 0.51–0.83, p < 0.001, adjusted HR 0.61, 95% CI 0.47–0.80, p < 0.001) (Fig. 6B). From the two ischemic phenogroups, CRT-D was associated with a significantly lower risk of all-cause mortality compared to CRT-P only in phenogroup 5 (unadjusted HR 0.80, 95% CI 0.69–0.93, p = 0.005, adjusted HR 0.84, 95% CI 0.71–0.99, p = 0.033) (Fig. 6E). In the sensitivity analysis, similar trends could be observed regarding the associations between the device type and the composite of all-cause death, heart transplantation, and left ventricular assist device implantation (Supplemental Fig. 4).

Fig. 6

Kaplan–Meier estimates of the time to death from any cause in CRT-D and CRT-P patients in the five phenogroups identified using topological data analysis. Univariable and multivariable Cox proportional hazards models were used to compute hazard ratios with 95% confidence intervals. Each multivariable model included the following features: device type, age, sex, history of atrial fibrillation, and history of ventricular arrhythmia. CI confidence interval, CRT-D cardiac resynchronization therapy defibrillator, CRT-P cardiac resynchronization therapy pacemaker, HR hazard ratio