Bayesian hierarchical models combining different study types and adjusting for covariate imbalances: a simulation...

Area of Science:

• Biostatistics
• Health Economics
• Clinical Epidemiology

Background:

• Bayesian hierarchical models are used to synthesize evidence from diverse study designs.
• Combining evidence from randomized and non-randomized studies presents challenges due to potential patient characteristic imbalances.
• Such imbalances can introduce bias into the synthesized results.

Purpose of the Study:

• To evaluate a novel Bayesian approach designed to mitigate bias from patient-level covariate imbalances.
• To assess the performance of this adjusted model when integrating data from both randomized controlled trials (RCTs) and non-randomized studies (NRSs).

Main Methods:

• Simulation studies were employed using generated data from RCTs and NRSs.
• Covariate imbalances were intentionally introduced into the NRS data.
• The proposed Bayesian hierarchical model, adjusted for imbalances, was compared against three alternative Bayesian synthesis methods.
• Simulations covered six scenarios varying imbalance impact and study proportions.

Main Results:

• The adjusted Bayesian hierarchical model consistently produced unbiased results across all six simulated scenarios.
• This model's estimates were closest to the true values compared to unadjusted or alternative Bayesian approaches.
• The model demonstrated robustness to variations in imbalance severity and the mix of study types.

Conclusions:

• The proposed hierarchical Bayesian method effectively adjusts for patient characteristic differences between study arms.
• This approach facilitates the optimal and unbiased synthesis of evidence from both RCTs and NRSs.
• It supports more informed healthcare decision-making by leveraging all available data robustly.