Measuring and estimating interaction between exposures on dichotomous outcome of a population

TL;DR

This paper introduces a comprehensive method to measure and estimate various interaction effects between exposures on a dichotomous outcome using a single regression model, demonstrated through a study on Helicobacter pylori eradication.

Contribution

It proposes using conditional risk and five different measures of interaction, providing a unified approach with maximum-likelihood estimates and interval estimates from one regression model.

Findings

Multiple interaction measures can be estimated simultaneously.

Interval estimates are derived for each interaction measure.

Application to Helicobacter pylori eradication illustrates the method's utility.

Abstract

In observational studies for the interaction between exposures on dichotomous outcome of a population, one usually uses one parameter of a regression model to describe the interaction, leading to one measure of the interaction. In this article, we use the conditional risk of outcome given exposures and covariates to describe the interaction and obtain five different measures for the interaction in observational studies, i.e. difference between the marginal risk differences, ratio of the marginal risk ratios, ratio of the marginal odds ratios, ratio of the conditional risk ratios, and ratio of the conditional odds ratios. By using only one regression model for the conditional risk of outcome given exposures and covariates, we obtain the maximum-likelihood estimates of all these measures. By generating approximate distributions of the maximum-likelihood estimates of these measures, we obtain interval estimates of these measures. The method is presented by studying the interaction between a therapy and the environment on eradication of Helicobacter pylori among Vietnamese children.