2E-03 - Patient-Reported Outcome Endpoint Emulation in Heart Failure External Control Study

Background: External control arm (ECA) study designs have gained prominence in recent years and have been leveraged in medical product development programs where randomized controlled trials (RCTs) are not feasible or ethical to conduct. However, ECAs are not without limitations. ECA studies often rely on real-world data, yet, in real-world settings, patient-reported outcome (PRO) data is often not collected in a standardized manner and may be incomplete or prone to bias (e.g., selective reporting). As such, studies using ECAs to evaluate PROs may have to consider alternative sources of external data.

Objectives: This study aims to use a target trial emulation approach to evaluate the utility and methodological challenges of ECA studies for analyzing PRO data.

Methods: We performed an emulation analysis of a PRO endpoint, as assessed in the EMPEROR-Reduced trial, using contemporaneous data from the EMPEROR-Preserved trial placebo arm as the external control. Adults with heart failure (HF) aged ≥18 years with ejection fraction 31% to 54% and N-terminal-pro-brain natriuretic peptide ≥ 600 pg/ml on empagliflozin or placebo were eligible for inclusion. We adjusted for imbalance in the measured baseline covariates between treatment groups using propensity score methods. The outcome of change from baseline in the clinical summary score (CSS) of the Kansas City Cardiomyopathy Questionnaire (KCCQ) and 95% confidence interval (CI) at week 52 was estimated using inverse probability of treatment weighting (IPTW)-adjusted mixed model for repeated measures.

Results: A total of 526 subjects on empagliflozin and 628 external control subjects were included in this emulation study. After IPTW-adjustment, the change from baseline in the KCCQ-CSS at Week 52 comparing empagliflozin to the external control placebo group was modest and did not meet statistical significance (0.56; 95% CI: -3.98, 5.10). The IPTW-adjusted effect estimate was generally consistent with the EMPEROR-Reduced trial result for the target subset population (0.54; 95% CI: -2.52, 3.59; Standardized Difference = -0.01), leading to the same conclusion. Sensitivity analyses supported the robustness of the findings.

Conclusion: We successfully replicated the evaluation of a PRO endpoint in an ECA study. Challenges and lessons learned are used to formulate requirements for relevant ECA data sources and propose best practices. Ensuring robust strategies for the collection, quality, validation, and analysis of PRO data in ECA studies is crucial for the interpretation of findings.