Missing and spurious interaction in additive, multiplicative and odds ratio models

TL;DR

This paper demonstrates that commonly used additive, multiplicative, and odds ratio models for interactions in epidemiology are biased, resulting in false interactions and missing true ones, highlighting the need for improved modeling approaches.

Contribution

The study reveals biases in traditional interaction models, showing they can produce spurious results and overlook genuine interactions, which challenges their widespread use.

Findings

Traditional models are biased and produce spurious interactions.

These models can miss true interactions in epidemiological data.

The paper calls for reconsideration of current interaction modeling approaches.

Abstract

Additive, multiplicative, and odd ratio neutral models for interactions are for long advocated and controversial in epidemiology. We show here that these commonly advocated models are biased, leading to spurious interactions, and missing true interactions.