Examining the Reliability of Model-Based Meta-Analysis (MBMA) Outcomes: A Simulation Study

Model-based meta-analysis (MBMA) can be utilized to synthesize literature data and predict drug efficacy, particularly suitable for constructing external comparator arms for non-randomized controlled trials (NRCTs). This study evaluated the reliability of MBMA by comparing covariate models generated through MBMA to individual patient data. A pharmacodynamic covariate model, commonly employed in MBMA, was used to set true parameter values and simulate data across various scenarios. The reliability of MBMA models was assessed by comparing estimated to true parameter values and identifying optimal conditions for MBMA use. Linear and nonlinear covariate models were evaluated in 24 scenarios, focusing on the relative deviations of parameter estimates from their true values. Evaluation metrics included minimization successful rate, covariate introduction rate, and the accuracy of parameters such as E_max, ET₅₀, and covariate influences. Both model types showed similar reliability in most scenarios. Notably, model performance significantly improved when the number of included trials was 10 or more, the distribution of covariates exceeded 66.6% of its median, and the covariate impact coefficient was greater than 0.15. The study identified critical factors and thresholds that influence the accuracy of MBMA modeling. Enhanced accuracy in synthetic control analysis using MBMA was achieved under specified conditions, highlighting the effectiveness of MBMA in NRCT applications.