A simulation-based evaluation of synthetic control arm methodology in cardiovascular trials using propensity score matching

Abstract

Background: Conventional randomized controlled trial (RCTs) in cardiovascular medicine are expensive, ethically constrained, and logistically demanding. The synthetic control arm (SCA) methodology leverages real-world inspired simulated patient data and propensity score matching to construct an external comparator, enabling all enrolled participants to receive active treatment. Despite increasing regulatory acceptance, cardiovascular applications of this methodology remain methodologically under-characterized.

Methods: A simulation study was conducted involving 75 patients: 40 in the treatment arm receiving a combined antihypertensive and lipid-lowering agent, and 35 in the SCA control arm derived by propensity score matching on ten baseline clinical covariates. Primary endpoint was MACE at 12 months. Secondary endpoints were change in LDL cholesterol and systolic blood pressure. Statistical analyses included independent samples t-tests, Chi-squared tests, Cohen's d, odds ratios, and Kaplan-Meier survival analysis.

Results: Propensity score matching produced acceptable covariate balance across nine of ten variables (all SMD <0.3), with one residual imbalance in baseline systolic BP (SMD=0.54; p=0.024). Treatment significantly reduced LDL cholesterol (−29.4±10.3 versus −5.0±5.1 mg/dl; p<0.001; d=−2.94) and systolic blood pressure (−10.6±3.6 versus −3.3±3.5 mmHg; p<0.001; d=−2.06). MACE occurred in 32.5% of treated versus 40.0% of SCA control patients (OR=0.72; p=0.664), non-significant owing to insufficient power at this sample size.

Conclusion: The SCA design successfully demonstrates the feasibility of SCA methodology under controlled simulation assumptions, with strong performance for surrogate endpoints but limited power for clinical endpoints in a cardiovascular simulation context, demonstrating robust detection of surrogate endpoint effects. The MACE non-significance is attributable to sample size limitations, not methodological failure. Adequate power for a 7.5 percentage-point MACE difference requires approximately 645 patients per arm. These findings provide a structured methodological framework for future SCA applications in cardiovascular pharmacoepidemiology.