Estimation of a common effect parameter from follow-up data when there is no mechanistic interaction

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Stratified analysis is crucial for estimating effect parameters, but homogeneity assumptions can be measure-dependent.
• Different effect measures may lead to conflicting conclusions regarding homogeneity, impacting research validity.
• Existing methods may not adequately address measure-specific homogeneity in stratified analyses.

Purpose of the Study:

• To develop a novel stratified analysis method based on the sufficient component cause model.
• To introduce and define a specific effect measure, the 'peril ratio', for assessing homogeneity.
• To provide a method for pooling strata when peril ratio homogeneity is established.

Main Methods:

• Utilized the sufficient component cause model to conceptualize mechanistic interactions.
• Developed a stratified analysis framework centered on the peril ratio as the effect measure.
• Derived formulas for estimating a common peril ratio under conditions of homogeneity.

Main Results:

• Demonstrated that peril ratio remains constant across strata when exposure and stratifying variable lack mechanistic interaction.
• Presented practical formulas for estimating this common peril ratio.
• Validated the method through re-analysis of three real-world datasets.

Conclusions:

• The proposed peril ratio method offers a robust approach to stratified analysis, ensuring measure consistency.
• Researchers can confidently pool strata using the provided formulas when peril ratio homogeneity is confirmed.
• This method enhances the reliability and interpretability of findings from stratified epidemiological studies.