Time-series clustering of heart rate responses during CPET reveals distinct phenotypes in HFpEF

Background/Introduction Chronotropic incompetence (CI) is common in individuals with heart failure with preserved ejection fraction (HFpEF). However, current CI definitions rely on only one or two time-point measurements during cardiopulmonary exercise testing (CPET). As a result, they do not capture the full complexity of the heart rate (HR) response throughout exercise. Moreover, many definitions depend on peak HR metrics, which are highly sensitive to patient effort and may therefore misclassify chronotropic dysfunction. Purpose To identify distinct HR phenotypes in HFpEF using clustering of the HR and oxygen consumption (ml/kg/min) trajectories during CPET, and to determine how these phenotypes relate to conventional CI parameters, HFpEF likelihood scores, and major adverse cardiovascular events (MACE). Methods We retrospectively analysed data from 341 individuals with HFpEF (stage A–C) undergoing CPET in two Belgian centres. For both HR and VO2/kg increase, regression models (y = a + b·time + c·age) generated age-adjusted residual time series. These residuals were clustered using dynamic time warping with k-medoids, stratified by β-blocker use. Clusters were compared on conventional CI measures, clinical characteristics, HFpEF likelihood scores (H2FPEF and the morphological and functional HFA-PEFF domains) analysed as continuous variables, and prediction of MACE (heart failure hospitalisation, acute coronary event, or cardiovascular death) after a median follow-up of 3 years. Results Among patients not taking β-blockers (n = 162), three distinct clusters were identified. Clusters did not differ in age, overall fitness or HFA-PEFF score. Patients in cluster 3 showed the poorest chronotropic response, presented with a markedly flatter HR–VO2 slope, worse conventional CI measures, higher BMI, and higher H2FPEF scores, as confirmed by a positive regression coefficient after adjustment for age and obesity (p = 0.002). This low-chronotropy cluster was predictive of MACE (HR 2.91, p = 0.038). In patients using β-blockers (n = 179), three clusters were also observed. Cluster 3, characterized by the weakest chronotropic response showed a similarly distinct pattern. The association with higher H2FPEF scores remained significant after adjustment (p = 0.023), and this cluster predicted MACE (HR 3.95, p = 0.032). Conclusion Distinct HR phenotypes can be identified in patients with HFpEF using time-series clustering of CPET data. This approach offers an age-independent and more comprehensive evaluation of chronotropic function. In both β-blocker users and non-users, the phenotype with the poorest chronotropic response consistently showed worse conventional CI parameters, higher H2FPEF scores, and an increased risk of MACE, underscoring the value of the full chronotropic response in assessing the HFpEF likelihood and risk stratification.