Abstract
BACKGROUND: Pulmonary arterial hypertension (PAH) remains a progressive and potentially fatal disease despite currently available treatments. Both sotatercept and selexipag have demonstrated clinical benefits in randomized controlled trials (RCTs); however, no direct head-to-head trial has compared these agents. We therefore conducted an indirect comparison using reconstructed individual patient data. METHODS: We performed a systematic search of PubMed, Scopus, and EMBASE to identify placebo-controlled RCTs evaluating sotatercept or selexipag in PAH. Kaplan-Meier curves from eligible trials were digitized and analysed using the artificial-intelligence (AI) algorithm IPDfromKM to reconstruct individual patient data. Only participants classified as WHO functional class II or III were included. Heterogeneity among pooled placebo arms was assessed, and treatment effects were estimated using Cox regression. Results were reported as hazard ratios (HRs) with 95% confidence intervals (CIs). RESULTS: Four relevant RCTs were identified (STELLAR, GRIPHON, HYPERION, and ZENITH). STELLAR, GRIPHON, and HYPERION satisfied the inclusion criteria, whereas ZENITH was excluded because it enrolled patients in WHO functional class III or IV. Using IPDfromKM, we reconstructed the six study arms from the three included trials. Based on reconstructed data, placebo arms showed no significant heterogeneity (likelihood ratio = 0.64; p = 0.70). Compared with pooled placebo, selexipag and sotatercept produced HRs of 0.26 (95% CI 0.18-0.38) and 0.57 (95% CI 0.47-0.70), respectively. The main indirect comparison demonstrated a statistically significant benefit for sotatercept over selexipag (HR = 0.45; 95% CI 0.29-0.70; p = 0.00036). CONCLUSIONS: AI-based reconstruction of individual patient data made it possible to compare the efficacy of therapies in the absence of direct head-to-head evidence. These findings suggest that sotatercept may reduce PAH-related events more effectively than selexipag, although the inference is derived from reconstructed and indirectly compared data.