A Comparative Effectiveness Study on Opioid Use Disorder Prediction Using Artificial Intelligence and Existing Risk...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Opioid use disorder (OUD) is a significant cause of mortality in the U.S., imposing substantial burdens on individuals, families, and healthcare systems.
• Artificial intelligence (AI) offers potential for developing automated prediction tools for OUD by leveraging available healthcare data.

Purpose of the Study:

• To develop and evaluate AI-based models for predicting OUD.
• To compare the predictive performance of AI models against existing clinical tools, including the unweighted Opioid Risk Tool (ORT).

Main Methods:

• A retrospective study utilizing a dataset of 474,208 patients over 10 years.
• Development of various AI models, including a transformer-based model, for OUD prediction.
• Comparison of AI model performance (AUC) against logistic regression, random forest, xgboost, LSTM, transformer, and the unweighted ORT model on 100 test sets (47,396 patients).

Main Results:

• The proposed transformer-based AI model achieved a higher predictive accuracy (AUC = 0.742 ± 0.021) compared to other AI models and the unweighted ORT (AUC = 0.559 ± 0.025).
• AI models, particularly the transformer-based approach, demonstrated superior performance in predicting OUD compared to established clinical methods.

Conclusions:

• AI algorithms show significant promise in enhancing the prediction of OUD.
• Integrating AI into clinical practice can assist healthcare providers in risk stratification and managing patients undergoing opioid therapy.