Fairness of ML Models for Opioid Use Disorder Treatment Assessed

VERA Brief

AI-generated. Grounded in the article and its cited sources.

A new study published on arXiv investigated machine learning models used to predict treatment outcomes for opioid use disorder. The research found that these models can show performance differences between patient subgroups, even when overall accuracy is good.

Key facts

• Machine learning models are being developed to predict treatment retention and identify patients at risk of premature discontinuation for medications for opioid use disorder.
• A study evaluated the fairness of these models across patient subgroups defined by race, ethnicity, age, and sex.
• The research found that models can exhibit performance disparities between subgroups, even when overall predictive performance is satisfactory.
• Bias mitigation techniques reduced performance gaps but did not eliminate them completely.
• Algorithmic fairness is a critical consideration for MOUD treatment decision support systems.