Additive risk versus additive relative risk models

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Assessing causal interaction is vital in epidemiology for understanding the interplay of multiple risk factors.
• Nonadditivity is often used as a criterion for detecting causal interaction.
• The choice of statistical model (additive risk vs. additive relative risk) can impact the assessment of interaction.

Purpose of the Study:

• To clarify the distinction between additive risk models and additive relative risk models in the context of causal interaction.
• To demonstrate that additive relative risk models do not reliably indicate the absence of causal interaction in stratified studies.
• To establish the necessity of using additive risk models for a complete assessment of causal interaction.

Main Methods:

• Theoretical analysis of additive risk and additive relative risk models.
• Examination of causal models, including those proposed by Rothman.
• Consideration of implications for stratified studies and matched case-control studies.

Main Results:

• Additive relative risk models do not consistently correspond to the absence of causal interaction in stratified analyses.
• The presence of nonadditivity in relative risk models does not necessarily equate to causal interaction.
• Additive risk models are required for a comprehensive evaluation of causal interaction.

Conclusions:

• Researchers should utilize additive risk models for accurate assessment of causal interaction.
• Findings highlight potential misinterpretations of causal interaction when relying solely on additive relative risk models.
• Matched case-control studies may necessitate external data for fitting appropriate additive risk models.