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

Abstract

Introduction

Alcohol use disorder (AUD) is a major public health concern worldwide, with alcohol use during adolescence often leading to AUD in adulthood. Early identification of high-risk individuals is critical for reducing AUD risk. Absolute risk prediction models can help by providing individualised, time-specific risk assessments for the target population of adolescents and young adults.

Methods

We developed a deep learning model to provide personalised absolute risk estimates of developing AUD among adolescents or young adults who use alcohol, using data from the National Longitudinal Study of Adolescent to Adult Health. Predictor importance was assessed using Shapley Additive Explanations (SHAP) values. Model performance was evaluated using five-fold cross-validation (CV) with the area under the curve (AUC) and the ratio of expected to observed cases (E/O). The model was validated on an independent test dataset.

Results

Key predictors are biological sex, delinquency, and personality traits such as conscientiousness and extraversion. For predicting AUD risk within 6 years of first alcohol use, the model achieved AUCs of 0.72 in CV and 0.85 in independent validation, with E/O ratios of 1.03 and 1.28, respectively. In the test data, the weighted average AUC for 1- to 6-year prediction after first alcohol use was 0.86. These results indicate good discrimination and calibration performance.

Discussion and Conclusion

To our knowledge, the proposed model is the first deep learning model for absolute risk prediction of AUD. It can help identify high-risk adolescents and young adults, who may be then provided with timely and clinically appropriate interventions.