Conditions for confounding of interactions.

Area of Science:

• Pharmaco-epidemiology
• Biostatistics
• Health Research Methods

Background:

• Pharmaco-epidemiology frequently examines drug-drug and drug-covariate interactions.
• The conditions leading to confounding bias in interaction estimates remain unclear.
• Understanding confounding is essential for reliable pharmaco-epidemiological research.

Purpose of the Study:

• To elucidate the conditions under which confounding bias affects interaction estimates in logistic regression.
• To explore how confounding influences the assessment of drug-drug and drug-covariate interactions.

Main Methods:

• Employed analytical derivations to investigate the conditions for confounding bias in interaction estimates.
• Utilized simulations to validate analytical findings and quantify the impact of parameters on bias.

Main Results:

• Failure to adjust for a risk factor (U) biases interaction estimates between exposures (E1, E2) if U's association with E1 differs across E2 strata.
• Confounding bias intensifies with higher confounder prevalence, stronger U-Y association, and greater heterogeneity in E1-U association across E2 strata.
• Variables not confounding main effects can significantly confound interaction effects.

Conclusions:

• Researchers should identify risk factors as potential confounders if their association with one exposure is modified by another.
• Proactive identification and adjustment for such confounders are necessary for accurate interaction studies in pharmaco-epidemiology.