Attributing effects to interactions

Area of Science:

• Epidemiology
• Biostatistics
• Genetics

Background:

• Understanding the interplay between multiple exposures is crucial in epidemiological research.
• Existing methods may not adequately partition the total effect of an exposure into components related to interaction.

Purpose of the Study:

• To present a novel framework for estimating the proportion of an exposure's effect attributable to interaction with a second exposure.
• To provide methods for decomposing exposure effects, applicable on both difference and ratio scales.
• To guide identification of intervention targets for mitigating exposure effects.

Main Methods:

• Development of a statistical framework for effect decomposition.
• Utilizing standard regression models for component estimation.
• Discussion of alternative decompositions for dependent exposures.

Main Results:

• Demonstrated that the total effect of an exposure can be decomposed into a conditional effect and an interaction component when exposures are independent.
• The framework allows for the estimation of the proportion of the total effect attributed to the interaction.
• Illustrative example provided using genetic epidemiology data.

Conclusions:

• The proposed framework offers a robust method for dissecting exposure effects and quantifying interaction.
• This approach aids in identifying key variables for intervention to reduce the impact of primary exposures.
• Applicable in scenarios where direct intervention on the primary exposure is not feasible.