Adverse Childhood Experiences and Alcohol-Related Outcomes: A Machine Learning Study

Adverse childhood experiences (ACEs) are a highly prevalent risk factor for a multitude of negative health outcomes (e.g., alcohol dependence; Pilowsky et al., 2009; Rhee et al., 2019; Khosravani et al., 2019) and are commonly represented by two overarching categories: abuse (e.g., emotional) and household dysfunction (e.g., parent substance use). ACE-related literature often applies a cumulative sum-score of ACE categories and a tacitly agreed upon cut-score of four or more ACEs which is consistently related to severe negative outcomes (Anda et al., 2020; Campbell, 2020; Hughes et al., 2017; Lacey & Minnis, 2020), including alcohol-related outcomes (e.g., Baiden et al., 2022). The present study explored the commonly applied standards that assume ACEs carry equal predictive weight on health outcomes alongside the potentially inherent interrelations among ACEs. In a college sample (n = 373), we assessed the individual and interactive influences of a series of ACEs (as measured by the ACE-IQ; World Health Organization [WHO], 2018) on three alcohol-relevant outcomes (i.e., consequences of alcohol misuse, frequency of binge drinking behavior, and age of regular alcohol consumption onset), across four ACE-IQ scoring strategies (i.e., binary, ordinal, standard ACE categories, and factors identified in Gette et al., 2021). Given their robust performance, ability to navigate non-linear interactions, and improved model evaluation strategies (Lantz, 2019; Orrù et al., 2020), we used common machine learning algorithms (i.e., random forest and elastic net) to test for ACE interactions, compare classification performance of these more complex techniques with traditional logistic regression approaches, and calculate empirically derived cut-point scores. Our work did not identify any particular strengths between analytic strategy (AUC: 0.57-0.60), scoring approach (AUC: 0.54-0.61), nor outcome of interest (AUC: 0.58-0.63), thereby supporting current practices used in the greater ACE literature.