Interaction and intervention modeling: predicting and extrapolating the impact of multiple interventions

Methods called interaction and intervention modeling are presented. Interaction modeling examines the interactions between variables as the basis for predicting the impact of multiple variables on a target population and on populations with difference distributions of risk factors. Intervention modeling incorporates these interactions and aims to extrapolate the impact of multiple interventions to new populations. The aim is to develop methods that will be useful for modeling and comparing intervention strategies using existing data and standard statistical methods.

Traditional hypothesis testing methods used for randomized clinical trials and cohort studies and extrapolating the results to new populations are compared with interaction and intervention modeling methods. Interaction and intervention modeling utilizes the same data as the traditional approach but examines the impact of multiple simultaneous interactions and allows extrapolation of the results to populations with different prevalences and distributions of risk factors. An example using real data demonstrates the potential of interaction and intervention modeling to predict the impact of multiple interacting variables and to compare the impact of alternative interventions.

The methods outlined take into account the impact of the magnitude of the relative risks, prevalence of risk factors, and interaction of risk variables when predicting the impact on a new population or extrapolating the results of one or more interventions on a new population. Traditional methods that do not take into account interactions are shown to produce different conclusions from the intervention modeling approach that incorporates interactions. The impact of the intervention modeling approach compared with the traditional approach will be quite variable depending on the prevalence of the risk factors and their extent of interaction.

Studies designed to test a hypothesis treat most variables as potential confounding variables adjusting for their impact and their interactions as part of the analysis using traditional regression methods. Interaction and intervention modeling focuses on the interactions themselves and allows comparison of the effectiveness of alternative interventions.