Causal Inference for Meta-Analysis and Multi-Level Data Structures, with Application to Randomized Studies of Vioxx

Area of Science:

• Biostatistics
• Epidemiology
• Pharmacovigilance

Background:

• Meta-analysis is crucial for synthesizing evidence but can be complicated by heterogeneity.
• Understanding sources of variation in treatment effects across studies is essential for accurate conclusions.

Purpose of the Study:

• To develop a novel framework for meta-analysis and multi-level data structures.
• To codify and examine sources of heterogeneity in treatment effects across studies or settings.

Main Methods:

• Proposed a framework considering potential outcomes for each subject under each treatment.
• Identified four sources of heterogeneity: response inconsistency, nonequivalent treatment grouping, nonignorable treatment assignment, and subject composition variability.
• Examined the impact of these sources on conditional and unconditional treatment effects.

Main Results:

• The framework provides a structured approach to identify and analyze heterogeneity in treatment effects.
• Re-analysis of Vioxx cardiovascular risk data from 29 randomized trials illustrated the framework's utility.
• Demonstrated how different sources of heterogeneity influence the homogeneity of treatment effects.

Conclusions:

• The developed framework enhances the rigor of meta-analyses by systematically addressing sources of heterogeneity.
• This approach is valuable for re-analyzing complex datasets, such as drug safety studies.
• Improves the interpretation of treatment effects in multi-study settings.