Estimating additive interaction in 2-stage individual participant data meta-analysis.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Individual participant data (IPD) meta-analysis enhances power for studying interaction and effect modification.
• Additive interaction is more relevant for public health than multiplicative interaction but lacks established methods in IPD meta-analysis for binary or time-to-event outcomes.
• Existing methodological literature does not adequately address additive interaction within 2-stage IPD meta-analysis.

Purpose of the Study:

• To describe a valid method for estimating additive interaction, specifically the Relative Excess Risk due to Interaction (RERI), in 2-stage IPD meta-analysis.
• To address the limitations of directly pooling study-level RERI estimates.
• To provide a practical procedure for estimating additive interaction in complex epidemiological studies.

Main Methods:

• A 3-step procedure is proposed: 1) estimate exposure and product term effects within each study, 2) pool study-specific estimates using multivariate meta-analysis, and 3) calculate an overall RERI with a 95% confidence interval.
• The method is illustrated using data from the PSYchosocial factors and Cancer (PSY-CA) consortium, examining the interaction between depression and smoking on smoking-related cancer risk.
• The procedure is designed for binary or time-to-event outcomes within IPD meta-analysis.

Main Results:

• The proposed 3-step procedure provides a valid approach for estimating additive interaction (RERI) in 2-stage IPD meta-analysis.
• Direct pooling of study-level RERI estimates can yield invalid results, highlighting the need for the proposed method.
• The application to depression, smoking, and cancer risk demonstrates the feasibility and utility of the method.

Conclusions:

• The developed 3-step procedure offers a robust framework for estimating additive interaction in IPD meta-analysis, particularly for binary and time-to-event data.
• This methodology enhances the ability to investigate effect modification relevant to public health.
• The findings have implications for future meta-analyses, including those based on published data.