Mimicking Clinical Trials Using Real-World Data: A Novel Method and Applications

Area of Science:

• Biostatistics
• Health Informatics
• Clinical Trials

Background:

• Simulating individual-level trial data from summary statistics is crucial for meta-analysis, external control arms, and real-world data calibration.
• Real-world data (RWD) correlations may not accurately reflect trial-specific correlation structures, potentially biasing analyses with nonlinear models or effect modifiers.

Approach:

• Developed an iterative algorithm using copula and resampling, estimating propensity scores for trial enrollment likelihood.
• Validated the algorithm via Monte Carlo simulations comparing trial and RWD covariate distributions.
• Applied the method to an actual trial and SEER-Medicare data for generalizability and external control arm feasibility.

Key Points:

• The algorithm's approximated correlations closely matched true trial correlations, outperforming RWD correlations.
• Successfully reproduced the joint distribution of characteristics for an actual trial.
• Simulated data yielded similar standardized mean differences (SMD) and overlapping confidence intervals for survival estimates compared to individual-level trial data.

Conclusions:

• The iterative algorithm offers a feasible approach for simulating individual-level data from summary trial data.
• Further validation with larger sample sizes is recommended to confirm the approach's robustness.