Family, Neighborhood and Individual Factors Linked to Early Sipping as Identified by Machine Learning Analysis

• In Event: Poster Session 02
  In Poster Session: PS 02 Topic - Developmental Psychopathology

Abstract

Background
Early sipping of alcohol is an important early marker of problematic alcohol use that predicts heavy drinking, alcohol-related problems and onset of drug use in adolescence, as well as likelihood of substance abuse in adulthood. Previous studies suggest that environmental factors such as access to substances, parental approval, family environment, religiosity and cultural factors influence age of first alcohol sipping (Donovan & Molina, 2014; Jackson, 2015; Rolando, 2012). However, there are no studies parsing these complex multifactorial mechanisms that drive early initiation of alcohol use. Here, we present personalized modelling of adult supervised incipient alcohol use across three domains, namely, 1) environmental, 2) neurocognitive & affect, and 3) health & psychopathology factors. We hypothesized that environmental factors would be most predictive of early sipping.

Methods
Using the Adolescent Brain and Cognitive Development study sample (ABCD; N = 10,708; N = 1490 with early sipping, 42% female), we predicted alcohol sipping by 10 years of age with a gradient tree boosting classifier (XGBoost), excluding those that drank accidentally, furtively, or for religious reasons. To both optimize hyperparameter settings and evaluate the area under the receiver operating characteristic curve (AUROC) without bias, we used 5-fold nested cross-validation. We then employed SHapley Additive exPlanations (Lundberg, 2018) to rank the unique contribution of the top predictors in each model (environmental: 370 predictors; neurocognitive/affect: 119 predictors; health/psychopathology: 182 predictors) using algorithms derived from cooperative game theory. Planned future analyses include adding features for brain function (task/resting fMRI).

Results
As hypothesized, the environmental model (AUROC = 0.736) outperformed the health/psychopathology (AUROC = 0.660) and neurocognitive/affective (AUROC = 0.628) models. The number one ranked feature from our environmental model was the child’s ease of access to alcohol, with easy access in the household increasing the likelihood of early sipping. Religiosity of the family was a protective factor, while raising kids to be independent was a risk factor. A number of neighborhood characteristics, such as residential density, high average mortgage rate, high unemployment, high air pollution and high risk of lead exposure, were also associated with higher likelihood of early alcohol sipping. From the health/psychopathology model, features related to pubertal development (e.g., tall height and increased body hair) were highly predictive, suggesting that more physically developed children were more likely to sip alcohol. A higher score on either the prodromal psychosis or ADHD symptom scale was a mental health risk factor for early sipping. Lastly, our neurocognitive/affect model indicated that low inhibitory control (Stop Signal Task) and measures of high sensation & reward seeking were predictive of childhood alcohol use.

Conclusion
Family and neighborhood factors were most predictive of early onset of alcohol use during childhood. However, physical development, mental health and neurocognitive factors that have been widely implicated in addiction theories, such as inhibitory control and reward seeking, also predicted early first alcohol use. We anticipate that as the ABCD study children grow older, these individual factors will play an increasingly important role for predicting emergence of regular alcohol use patterns and onset of alcohol misuse.

Authors

• Gregory Niklason, University of Minnesota - Twin Cities
  Presenting Author

• Andrea M Maxwell, University of Minnesota - Twin Cities
  Non-Presenting Author

• Sisi Ma, University of Minnesota - Twin Cities
  Non-Presenting Author

• Eric Rawls, University of Minnesota - Twin Cities
  Non-Presenting Author

• Erich Kummerfeld, University of Minnesota - Twin Cities
  Non-Presenting Author

• Anna Zilverstand, University of Minnesota - Twin Cities
  Non-Presenting Author