Meta-regression with partial information on summary trial or patient characteristics

Area of Science:

• Biostatistics
• Meta-analysis
• Statistical modeling

Background:

• Missing covariate data is a common challenge in meta-regression.
• Current methods like complete case or available case analysis can lead to biased results.
• Existing approaches often rely on the strong Missing Completely At Random (MCAR) assumption.

Purpose of the Study:

• To develop a Bayesian meta-regression model that effectively handles missing study-level covariate data.
• To propose a joint likelihood-based approach that utilizes all observed data.
• To relax the assumption from Missing Completely At Random (MCAR) to Missing At Random (MAR).

Main Methods:

• A Bayesian framework is employed for inference.
• The joint density is modeled as a factorization of the meta-regression model and the conditional density of covariates.
• Inter-relations between multiple covariates are explicitly modeled.

Main Results:

• The proposed model accommodates missing covariate data under the less restrictive Missing At Random (MAR) assumption.
• The approach is shown to be implementable in WinBUGS.
• Sensitivity and robustness analyses were conducted on two real datasets, evaluating the impact of the MCAR assumption.

Conclusions:

• The developed Bayesian meta-regression model provides a more robust method for handling missing covariate data compared to traditional approaches.
• The joint likelihood approach offers improved statistical efficiency and validity by utilizing all available information.
• The findings highlight the importance of considering the MAR assumption for more reliable meta-regression analyses.