Using machine learning to predict opioid misuse among U.S. adolescents.

Area of Science:

• Public Health
• Data Science
• Adolescent Health

Background:

• Opioid misuse is a significant public health concern among U.S. adolescents.
• Accurate prediction of opioid misuse is crucial for early intervention and prevention strategies.
• Machine learning (ML) offers potential for improving prediction models in public health.

Purpose of the Study:

• To evaluate and compare the prediction performance of three ML techniques against penalized logistic regression for adolescent opioid misuse.
• To identify the most effective ML algorithm for predicting rare outcomes in large population datasets.

Main Methods:

• Utilized data from the 2015-2017 National Survey on Drug Use and Health (N=41,579 adolescents).
• Developed prediction models using artificial neural networks, distributed random forest, and gradient boosting machine.
• Compared ML models with penalized logistic regression using area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC), with AUPRC as the primary metric.

Main Results:

• Opioid misuse prevalence was 3.7% in the adolescent sample.
• All four models demonstrated similar prediction performance based on AUROC (0.809-0.815).
• Distributed random forest achieved the highest AUPRC (0.172), outperforming penalized logistic regression (0.162), gradient boosting machine (0.160), and artificial neural networks (0.157).

Conclusions:

• Machine learning techniques are effective for predicting adolescent opioid misuse, particularly for rare outcomes.
• Distributed random forest demonstrated superior predictive performance in this context.
• ML models offer a promising approach for identifying at-risk adolescents for targeted interventions.