Interaction as departure from additivity in case-control studies: a cautionary note.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Interaction assessment traditionally relies on departures from additive models.
• Estimating fundamental interaction parameters is difficult in case-control studies.
• Surrogate measures like RERI, AP, and S are used but have limitations.

Purpose of the Study:

• To identify problems with surrogate interaction measures when controlling for covariates.
• To propose an alternative method for estimating interaction parameters.
• To evaluate the performance of the proposed method.

Main Methods:

• Critically analyzed surrogate interaction measures (RERI, AP, S) in the presence of covariates.
• Proposed and utilized a linear odds model for interaction assessment.
• Conducted a simulation study to compare models.

Main Results:

• RERI and AP vary across strata, unlike the fundamental interaction parameter.
• The synergy index (S) is invariant across strata but logistic regression models have misspecification issues.
• The linear odds model allows testing the fundamental interaction parameter and shows improved coverage in simulations, though bias remains a concern.

Conclusions:

• The linear odds model offers advantages over logistic regression for estimating interaction parameters in case-control studies.
• While the linear odds model improves coverage, bias in parameter estimation needs consideration.
• The synergy index (S) is a more stable measure across strata compared to RERI and AP.