Reconstructed IPD sotatercept meta-analysis

Systematic Review and Individual Patient Data Meta-Analysis of Sotatercept for Pulmonary Arterial Hypertension Using Stratified Covariate-Reconstructed Survival (SCRS) Authors: Tyler Pitre, MD Division of Respirology, Department of Medicine, University Health Network, Toronto, ON Dena Zeraatkar, PhD Health Research Impact and Evidence, McMaster University, Hamilton, ON 1. Introduction 1.1 Rationale Pulmonary arterial hypertension (PAH) is a progressive, life-threatening disease characterized by elevated pulmonary vascular resistance (PVR), right ventricular failure, and premature death. Despite advances in endothelin receptor antagonists (ERAs), phosphodiesterase-5 inhibitors (PDE5i), soluble guanylate cyclase stimulators (sGCs), and prostacyclin pathway agents, long-term outcomes remain poor. Contemporary registries report 5-year survival rates of approximately 50–65% even with combination therapy. Sotatercept (Winrevair; Merck/MSD) is a first-in-class activin signaling inhibitor that targets the underlying proliferative vasculopathy of PAH by rebalancing growth-promoting and growth-inhibiting signaling through the activin receptor type IIA-Fc (ActRIIA-Fc) pathway. Sotatercept was approved by the FDA in March 2024 based on the STELLAR phase 3 trial, and has since demonstrated consistent efficacy across the PULSAR phase 2 trial, the HYPERION phase 3 trial (newly diagnosed PAH on initial combination therapy), and the ZENITH phase 3 event-driven trial (higher-risk PAH). The SPECTRA single-arm open-label trial and the Study020 Japanese bridging study provide additional supportive evidence. Standard aggregate data meta-analysis cannot address cross-trial population heterogeneity, particularly when trials enroll systematically different patient populations. Individual patient data (IPD) meta-analysis is the gold standard for addressing these limitations, but obtaining true IPD from sponsors requires protracted negotiation and may not be feasible. We have developed Stratified Covariate-Reconstructed Survival (SCRS), a validated method for reconstructing IPD with preserved covariate-outcome associations from published Kaplan-Meier curves and baseline tables (Pitre & Zeraatkar, 2026). SCRS extends the standard Guyot algorithm by augmenting reconstructed time-to-event data with synthetic patient-level covariates through stratified assignment and within-stratum rank-based coupling. SCRS has been validated using the ASPIRE PAH registry (N=773, 83% C-index recovery) and 1,300 Monte Carlo synthetic trials (96% median C-index recovery, 87–100% win rate over stratified random assignment). 1.2 Objectives Primary Objective: To estimate the pooled treatment effect of sotatercept added to background PAH therapy versus placebo on time to clinical worsening (or first morbidity-mortality event) across randomized controlled trials in PAH, using reconstructed IPD. Secondary Objectives: (1) To assess heterogeneity of treatment effects across trials using both one-stage and two-stage IPD meta-analysis with Rubin’s rules pooling across multiple SCRS reconstructions. (2) To conduct subgroup analyses using two-stage pooling of published subgroup-specific hazard ratios, supplemented by SCRS consistency validation. (3) To build a pooled prognostic model identifying predictors of clinical worsening and to estimate treatment benefit by baseline risk stratum. (4) To characterize cross-trial population heterogeneity and perform matching-adjusted indirect comparisons (MAIC) where population differences are substantial. (5) To evaluate the safety profile of sotatercept through Mantel-Haenszel pooling of adverse events of special interest. (6) To test the proportional hazards assumption and, if violated, estimate restricted mean survival time (RMST) at clinically relevant landmarks. 2. Methods 2.1 Eligibility Criteria Study Designs: Randomized controlled trials (RCTs) of any phase (2, 2b, or 3) comparing sotatercept to placebo (with or without background PAH therapy). Single-arm open-label studies will be included for descriptive and safety pooling but excluded from comparative efficacy meta-analysis. Participants: Adults (≥18 years) with confirmed WHO Group 1 pulmonary arterial hypertension, diagnosed by right heart catheterization (mPAP >20 mmHg, PVR >2 Wood units, PAWP ≤15 mmHg). Eligible etiologies include idiopathic, heritable, drug/toxin-induced, and PAH associated with connective tissue disease, congenital heart disease (repaired), HIV, or portal hypertension. Intervention: Sotatercept administered subcutaneously at any dose (0.3 or 0.7 mg/kg) every 21 days, added to background PAH therapy. Comparator: Placebo administered subcutaneously every 21 days, added to background PAH therapy. 2.2 Trials Expected to Be Included Based on the sotatercept clinical development program, the following trials will be identified and processed: Trial Phase N Design Population Primary EP IPD Plan PULSAR + STELLAR (pooled) 2 + 3 ~429 RCTs, pooled per Hoeper ERJ 2025 Stable PAH, double/triple Rx 6MWD / CW composite SCRS Tier 1: rich subgroup data available HYPERION 3 320 RCT 1:1 Newly diagnosed, initial combo Rx Clinical worsening SCRS Tier 1: rich subgroup data available ZENITH 3 ~172 RCT 1:1, event-driven Higher-risk PAH (WHO FC III/IV) Morbidity- mortality SCRS Tier 1: KM + NaR available SPECTRA 3b 163 Single-arm open-label Stable PAH, monotherapy PVR change No comparator; safety pooling only Study020 3 ~70 RCT 1:1 Japanese PAH PVR change Extraction only; not in IPD MA (bridging study) Note: PULSAR and STELLAR will be pooled as a single IPD unit based on the Hoeper et al. (ERJ 2025) pooled analysis, which provides combined subgroup forest plots, event counts, and baseline characteristics. Study020 will not contribute to the IPD meta-analysis due to its bridging-study design and limited subgroup data. 2.3 Outcomes Primary outcome: Time to first clinical worsening event, using each trial’s own composite endpoint definition. Endpoint definitions will NOT be harmonized across trials; instead, each trial will contribute its protocol-specified composite (STELLAR: 5-component clinical worsening; HYPERION: 4-component clinical worsening including lung transplant; ZENITH: 3-component morbidity-mortality). This approach preserves the integrity of each trial’s primary analysis and avoids post-hoc redefinition of endpoints. Safety outcomes: Any AE, serious AE, AE leading to discontinuation, AE leading to death, and adverse events of special interest (AESI): erythrocytosis/increased hemoglobin, thrombocytopenia, telangiectasia, and bleeding events. SPECTRA (single-arm) will be included in safety pooling as a separate descriptive stratum. 2.4 Data Extraction Data extraction will be performed using the Total Extraction Checklist v1.0, a 15-phase structured protocol covering approximately 100 items per trial across: trial identity, study design, eligibility criteria, endpoints, statistical methods, patient flow, baseline characteristics, efficacy results, hemodynamics, safety, subgroup analyses, sensitivity analyses, patient-reported outcomes, additional data, and quality checks. Each extraction will produce a comprehensive Excel workbook. 2.5 IPD Reconstruction via SCRS For trials reporting Kaplan-Meier curves with number-at-risk tables for time-to-event endpoints, individual patient data will be reconstructed using the SCRS pipeline: Phase 0 (Guyot Reconstruction): KM curves will be digitized from published figures. The Guyot algorithm will invert the KM estimator to recover individual event and censoring times. Phase 1 (Virtual Cohort Generation): Virtual cohorts will be generated with covariate values sampled from distributions consistent with published baseline tables. Inter-covariate dependencies will be modeled using a Gaussian copula (NORTA method) with correlation structure informed by PAH registry data. Phase 2 (Rank-Based Coupling): Patients will be classified by WHO functional class stratum (the sole stratification variable; see Section 2.6). Survival times from Phase 0 will be assigned with rank-based coupling (default cs = 0.30), aligning event timing with a composite prognostic score derived from baseline 6MWD, PVR, NT-proBNP, and age. Event allocation ratios per stratum will be derived from published subgroup forest plots. Phase 3 (Multiple Imputation): B = 50 independent reconstructions will be generated with different random seeds and pooled using Rubin’s rules to provide valid standard errors and account for reconstruction uncertainty. Variance will be decomposed into within-imputation and between-imputation components, with Fraction of Missing Information (FMI) reported. 2.6 Stratification Variable WHO functional class will be used as the sole stratification variable for SCRS coupling across all trials. Specifically: WHO FC II vs III for PULSAR+STELLAR and HYPERION; WHO FC III vs IV for ZENITH. This variable was selected because it is (a) the strongest available prognostic factor with published subgroup-specific hazard ratios across all trial populations, and (b) consistently reported in all trial forest plots. Additional stratification variables (PVR, background therapy) were considered but will not be used due to minimum cell size constraints in multi-variable stratification grids. 2.7 IPD Quality Metrics Each IPD reconstruction will be evaluated using: (a) RMSE of reconstructed vs published survival function (target 10% from published values will be flagged and re-digitized. 2.8 Risk of Bias Risk of bias for each included RCT will be assessed using the Cochrane Risk of Bias tool version 2 (RoB 2). Domains assessed will includ