Predicting adolescent alcohol and other drug problems using electronic health records data
Recommended Citation
Chi FW, Alexeeff S, Ahmedani B, Boscarino JA, Waitzfelder B, Dugan R, Frankland T, Hu Y, Loree A, and Sterling S. Predicting adolescent alcohol and other drug problems using electronic health records data. J Subst Abuse Treat 2022; 132:108487.
Document Type
Article
Publication Date
1-1-2022
Publication Title
Journal of substance abuse treatment
Abstract
IMPORTANCE: Alcohol and other drug (AOD) use problems may cause significant burden on affected adolescents and their families, yet treatment providers often do not identify these problems early enough.
OBJECTIVE: To develop, and internally and externally validate a multivariable prediction model of adolescent AOD problems using child- and maternal-level predictors before age 12, and child-level predictors between ages 12 to 18, as recorded in the electronic health records (EHR).
DESIGN: A retrospective cohort study conducted time-to-event analyses using Cox proportional hazards models.
SETTING AND PARTICIPANTS: 41,172 children born between 1997 and 2000 at four health plans (Kaiser Permanente Hawaii, KPHI; Kaiser Permanente Northern California, KPNC; Geisinger Clinic, GC; and Henry Ford Health System, HFHS) who had continuous membership since birth and linkable maternal records in the health plan.
OUTCOMES: AOD use problems between ages 12 to 18, defined as either: 1) having a contact with the AOD treatment program or 2) receiving a non-tobacco AOD diagnosis in an inpatient or outpatient encounter.
EXPOSURES: Candidate predictor variables include demographics, socioeconomic status, and clinical diagnoses of the children and the mothers.
RESULTS: Overall, 1400 (3.4%) adolescents had an AOD disorder between ages 12 to 18; the median follow-up time post-age 12 was 5.3 years. The research team developed two final prediction models: a "baseline" model of 10 child-level and 7 maternal-level predictors before age 12, and a more comprehensive "time-varying" model, which incorporated child risk factors after age 12 as time-varying covariates in addition to the baseline model predictors. Model performance evaluation showed good discrimination performance of the models, with the concordance index improved for the time-varying model, especially for prediction of AOD events in late adolescence.
CONCLUSIONS AND RELEVANCE: This study identified a number of child and maternal characteristics and diagnoses routinely available in EHR data as predictive of risk for developing AOD problems in adolescence. Further, we found that risk of developing problems varies significantly by the timing and persistence of the risk factors. Findings may have potential clinical implications for prevention and identification of adolescent AOD problems, but more research is needed, especially across additional health systems.
PubMed ID
34098206
Volume
132
First Page
108487
Last Page
108487
