Machine learning-based clustering in idiopathic PAH: mortality is driven by hemodynamic severity in vascular phenotype vs. frailty in cardiometabolic co-morbidity group
C Kaymaz, B Kultursay, H C Tokgoz, S Tanyeri, C Bulus, M Kibar, C Erdem, A Vezir, A Kucuk, I Akkus, M Bulut, C E Cagliyan, A Hakgor, I H Tanboga, N OzdemirAbstract
Objective
Idiopathic pulmonary arterial hypertension (IPAH) is characterized by significant clinical heterogeneity. This study aimed to identify distinct patient phenotypes using machine learning-based cluster analysis and to evaluate their differences in therapeutic response and long-term survival.
Methods
A total of 297 patients with IPAH were analyzed. Hierarchical agglomerative clustering was performed based on baseline demographics, invasive hemodynamics, and comorbidities. Longitudinal changes in 6-minute walk distance (6MWD), NT-proBNP, multiparametric risk scores and echocardiographic parameters (PASP, TAPSE/PASP) were evaluated over 12 months.
Results
Two distinct clusters were identified. Cluster 1 (n=198; "Vascular Phenotype") comprised younger patients (median age 40 years) with low comorbidity burden but severe hemodynamic impairment. Cluster 2 (n=99; "Cardiometabolic Phenotype") consisted of older patients (median age 65 years) with a high prevalence of obesity (78%), diabetes (73%), and hypertension (57%). Female predominance was significantly higher in Cluster 2 (p=0.012). While WHO Functional Class, ESC/ERS, REVEAL 2.0, REVEAL 2.0 lite, and FPHN risk status were comparable, Cluster 1 exhibited significantly lower TAPSE, S’, and TAPSE/PASP ratio, alongside a higher rate of pericardial effusion compared to Cluster 2 (p<0.001 for all). Furthermore, Cluster 1 was associated with lower LVEDP (p=0.032) and cardiac index (p<0.001), but higher mean pulmonary arterial pressure, pulmonary vascular resistance, and pulmonary-to-systemic vascular resistance ratio (p<0.001 for all). Triple combination therapies were more frequently utilized in Cluster 1 (41.6% vs. 25.5%, p=0.048). Median follow-up was 763 (227-1993) days. Cluster 1 showed a greater increase in 6MWD (median 75m vs. 42m, p<0.001) and significantly lower NT-proBNP levels at final assessment (median 55 vs. 145 pg/mL, p=0.017). Despite these phenotypic differences, five-year survival was similar between groups (Log-rank p=0.452). Neither unadjusted (HR: 0.868; 95% CI: 0.60–1.26, p=0.452) nor adjusted (HR: 0.711; 95% CI: 0.47–1.07, p=0.099) Cox regression analyses showed a significant mortality difference.
Conclusion
IPAH patients can be phenotypically stratified into a "classic" vascular group and a "cardiometabolic" group. Although the first group demonstrates a more robust functional and neurohumoral response to PAH therapy, the similar survival rates suggest that mortality is driven by hemodynamic severity in the younger vascular phenotype group versus metabolic frailty in the older comorbid group.FigureFor image description, please refer to the figure legend and surrounding text.Figure2For image description, please refer to the figure legend and surrounding text.