A Deep Learning Model for Absolute Risk Prediction of Alcohol Use Disorder in Adolescents and Young Adults

Area of Science:

• * Public Health
• * Adolescent Medicine
• * Data Science

Background:

• * Alcohol use disorder (AUD) is a significant global health issue, with adolescent alcohol consumption often preceding adult AUD.
• * Early identification of at-risk individuals is crucial for mitigating AUD development.
• * Personalized, time-specific risk assessments using absolute risk prediction models are vital for adolescents and young adults.

Purpose of the Study:

• * To develop and validate a deep learning model for predicting the absolute risk of developing AUD in adolescents and young adults.
• * To identify key predictors of AUD risk within this demographic.
• * To assess the model's performance in terms of discrimination and calibration.

Main Methods:

• * A deep learning model was constructed using data from the National Longitudinal Study of Adolescent to Adult Health.
• * Shapley Additive Explanations (SHAP) were employed to determine predictor importance.
• * Model performance was evaluated using five-fold cross-validation (AUC, E/O ratio) and an independent test dataset.

Main Results:

• * Key predictors identified include biological sex, delinquency, and personality traits (conscientiousness, extraversion).
• * The model achieved an AUC of 0.72 in cross-validation and 0.85 in independent validation for predicting AUD risk within 6 years of first alcohol use.
• * Independent validation demonstrated strong performance with a weighted average AUC of 0.86 for 1- to 6-year predictions, indicating good discrimination and calibration.

Conclusions:

• * This study presents the first deep learning model for absolute AUD risk prediction.
• * The model effectively identifies adolescents and young adults at high risk for AUD.
• * Early identification facilitates timely and appropriate clinical interventions, potentially reducing AUD prevalence.