Allowing for uncertainty due to missing data in meta-analysis--part 2: hierarchical models

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Modeling

Background:

• Missing data in randomized trials poses significant analytical challenges.
• Standard methods may not adequately account for the uncertainty introduced by missing outcomes.
• The missing-at-random assumption is often difficult to justify.

Purpose of the Study:

• To propose a novel hierarchical model for analyzing data from multiple randomized trials with missing outcomes.
• To introduce and model the informative missing odds ratio (IMOR) to quantify departures from the missing-at-random assumption.
• To assess prior correlations between IMORs across different trial arms and studies.

Main Methods:

• Development of a hierarchical Bayesian model.
• Incorporation of informative missing odds ratios (IMORs) with realistic prior distributions.
• Fitting the model using Monte Carlo Markov Chain (MCMC) techniques.
• Application and validation on three diverse datasets.

Main Results:

• The proposed model effectively captures the additional uncertainty arising from missing data.
• Demonstrated the practical application of the hierarchical model across different datasets.
• Identified conditions under which information about the IMOR can be learned from the data.

Conclusions:

• The hierarchical model provides a robust framework for handling missing data in meta-analyses of randomized trials.
• The IMOR offers a quantifiable measure of missing data bias.
• The approach enhances the reliability of statistical inferences in the presence of missing outcomes.