Composite Categorical Regression for Correlated Categorical Exposures: Application to Adverse Childhood Experiences and Their Health Effects.

Adverse childhood experiences (ACEs), which often co-occur and are associated with physical and mental health problems later in life, are of critical concern in child protection efforts worldwide. This article is motivated by the need to comprehensively understand the impact of ACEs on adulthood depression, which presents several challenges: ACEs are highly correlated categorical exposures and their effects on health are interrelated and complex. To our knowledge, no statistical methodology effectively characterizes their joint effect while identifying the key contributing ACEs. In this article, we propose the Composite Categorical Regression (CCR) method as a novel approach to study correlated categorical exposures. The joint effects are captured by a constrained weighted sum of their dummy variables, and the relative contribution of each exposure is measured by aggregating the weights of its dummy variables. Key exposures are identified when the aggregated weight exceeds a data-adaptive threshold. We propose a CCR algorithm to estimate the CCR parameters by solving a convex optimization problem with constraints on the weights. Simulation studies show that the CCR method performs well. Application of the CCR method to Behavioral Risk Factor Surveillance System data aligns with known harmful ACEs and reveals a new insight: the under-recognized role of caregiver verbal violence in adulthood depression.