Predicting Adolescent Substance Use in a Child Welfare Sample: A Multi-Indicator Algorithm

Abstract

Given the risk of substance use (SU) among adolescents in the child welfare system, identification of risk for prospective impairing SU behaviors is a significant public health priority. We sought to quantify the incremental validity of routine multi-informant assessments of adolescent psychological distress (i.e., the Child Behavior Checklist and Youth Self-Report) and a commonly used SU screening protocol (i.e., the CRAFFT) to predict SU at 18 and 36 months after baseline in a nationally representative child welfare sample (N = 1,054; Mage = 13.72). We used receiver operator characteristics and reclassification analyses to develop our algorithms. We found that a battery consisting of baseline CRAFFT scores, self-reported delinquent behavior, and parent-reported rule-breaking behavior provided an incrementally valid prediction model for SU behavior among females, while baseline CRAFFT scores and self-reported delinquent behavior incrementally predicted SU for males. Results suggest that leveraging existing assessments within the child welfare system can improve forecasting of SU risk for this population.