Meta-analysis combining parallel and cross-over clinical trials. II: Binary outcomes

Area of Science:

• Biostatistics
• Clinical Trials
• Meta-Analysis

Background:

• Pooling binary outcomes from parallel and cross-over trials presents statistical challenges.
• Existing methods like Mantel-Haenszel and Peto use joint conditional probabilities, which can be problematic in cross-over designs.

Purpose of the Study:

• To compare different methods for pooling binary outcomes from parallel and cross-over trials.
• To identify the most suitable method for combined meta-analysis of parallel and cross-over trial data.

Main Methods:

• Comparison of odds ratio (OR) estimators derived from joint conditional probabilities versus marginal results in cross-over trials.
• Evaluation of the impact of between-period correlation on OR estimates.
• Assessment of computational similarity between methods for parallel and cross-over trials.

Main Results:

• Joint conditional odds ratio (OR) estimators differ from marginal ORs when between-period correlation exists in cross-over trials.
• Joint conditional ORs cannot be combined with OR estimates from parallel trials under correlation.
• The marginal OR estimate is independent of between-period correlation and adjusts the variance estimate for cross-over correlation.

Conclusions:

• The marginal odds ratio (OR) method is the preferred approach for pooling binary outcomes in combined meta-analyses involving both parallel and cross-over trials.
• This method's independence from between-period correlation and computational similarity across trial designs make it ideal for integrated analysis.