Meta-analysis with sample-standardization in multi-site studies

Area of Science:

• Pharmacoepidemiology
• Biostatistics
• Causal Inference

Background:

• Multi-site pharmacoepidemiologic studies require robust methods for analyzing data across diverse networks.
• Standard meta-analytic techniques may yield inconsistent results when treatment effects vary by site.

Approach:

• This study conceptualizes a target population and estimand for multi-site studies.
• Sample-standardization is proposed as a meta-analytic method, estimating network-wide causal effects.
• The sample-standardization estimator is analytically compared with inverse-variance weighted meta-analyses.

Key Points:

• Sample-standardization provides a consistent estimator for network-wide causal treatment effects.
• Consistency is maintained even with heterogeneity of treatment effects across sites.
• Inverse-variance weighted meta-analyses can produce inconsistent estimators in the presence of such heterogeneity.

Conclusions:

• Sample-standardization is a valid approach for causal inference in multi-site studies, encompassing the entire network population.
• Clear specification of target population and estimand is crucial for selecting appropriate meta-analytic methods.