Variable selection when estimating effects in external target populations

Area of Science:

• Epidemiology
• Biostatistics

Background:

• External validity is crucial for generalizing research findings to target populations.
• Estimating effects in new populations requires careful consideration of effect measure modifiers (EMMs).
• The impact of including non-EMMs in adjustment sets on these estimates is not well understood.

Purpose of the Study:

• To evaluate how including non-EMMs affects the estimation of transported risk differences (RDs).
• To assess the influence of covariates differing between populations, associated with outcomes, or modifying RDs on estimate precision and bias.

Main Methods:

• Simulations were used to model the inclusion of non-EMMs with varying characteristics.
• Covariates were analyzed based on differences between trial and target populations, outcome association, and RD modification.
• Estimation methods included outcome modeling and inverse odds weighting.

Main Results:

• Including non-EMMs that differ in distribution between populations reduced estimate precision, irrespective of outcome association.
• Non-EMMs associated with selection did not exacerbate bias caused by omitting necessary EMMs.
• Adjusting for all outcome-associated variables can lead to imprecise treatment effect estimates in external populations.

Conclusions:

• Careful selection of adjustment variables is necessary for valid external validity estimation.
• Over-adjustment with non-EMMs can compromise the precision of transported effect estimates.
• Understanding covariate roles is key to balancing validity and precision in epidemiologic research.