Development of a machine learning algorithm for early detection of opioid use disorder

Area of Science:

• Medical informatics
• Computational psychiatry
• Public health

Background:

• Opioid use disorder (OUD) impacts millions globally, with frequent diagnostic delays.
• Early identification of OUD is crucial for effective intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning model for the early prediction of OUD.
• To assess the utility of machine learning algorithms in identifying individuals at risk for OUD.

Main Methods:

• Analysis of 10 million medical insurance claims (2006-2018) from 550,000 patients.
• Utilized 436 predictor variables across demographics, conditions, diagnoses, medications, costs, and episode counts.
• Employed Word2Vec and Gradient Boosting algorithms for predictive modeling.

Main Results:

• The prediction model achieved a high c-statistic of 0.959, with 0.85 sensitivity and 0.882 specificity.
• Key predictors included opioid use days, prescription overlaps, and co-prescribing of benzodiazepines and muscle relaxants.
• Significant differences were observed in diagnoses like intervertebral disc disorders and pain disorders.

Conclusions:

• The developed algorithm can reduce the time to OUD diagnosis by an average of 14.4 months.
• Early OUD diagnosis holds potential for reducing morbidity, healthcare costs, addiction severity, and mortality.