Predicting mental and behavioral health service utilization among child welfare-involved caregivers: A machine learning approach

Abstract

Caregiver substance misuse (SM) and mental illness (MI) are risk factors for child abuse and neglect and are associated with more intensive CPS involvement including increased risk of foster care placement and multiple re-reports. This study examines the prevalence of SM and MI among 929 CPS-involved caregivers during the early phases of CPS involvement and explores the extent to which family and CPS-case characteristics predict referral and service receipt. We used a machine learning approach to identify the strongest predictors of SM and MI service receipt by comparing the predictive strength of random forest and logistic regression models. Results indicate a high prevalence of self-reported need for SM (13%) and MI (34%) services among caregivers. Nearly one-quarter (23.5%) of caregivers with SM needs and 34% of caregivers with MI needs did not receive needed services. Frequent contact with CPS workers, adverse experiences in adulthood, and court involvement were strong predictors of both SM and MI service uptake. Findings suggest the need for consistent screening for SM and MI among primary caregivers at the early stages of CPS-involvement, as well as enhanced referral practices. Machine learning applications for applied social science researchers are also discussed.