An ExplainableFair Framework for Prediction of Substance Use Disorder Treatment Completion

TL;DR

The paper tackles fairness and explainability in healthcare AI by introducing ExplainableFair, a framework that first optimizes predictive performance and then applies in-processing bias mitigation to balance outcomes across race and sex groups. It integrates SHAP-based explanations to reveal how feature importances shift during fairness optimization, enabling clinical interpretability of the bias-reduction process. On a substance use disorder treatment completion task, the approach achieves AUROCs around $0.86$ while substantially reducing Equalized Odds Disparity (EOD) for both race ($ ext{EOD} o0.0298$) and sex ($ ext{EOD} o0.0282$), with only modest losses in sensitivity and specificity. This combination of performance, fairness, and explanation supports more trustworthy and actionable AI assistance for clinical decision-making and resource allocation.

Abstract

Fairness of machine learning models in healthcare has drawn increasing attention from clinicians, researchers, and even at the highest level of government. On the other hand, the importance of developing and deploying interpretable or explainable models has been demonstrated, and is essential to increasing the trustworthiness and likelihood of adoption of these models. The objective of this study was to develop and implement a framework for addressing both these issues - fairness and explainability. We propose an explainable fairness framework, first developing a model with optimized performance, and then using an in-processing approach to mitigate model biases relative to the sensitive attributes of race and sex. We then explore and visualize explanations of the model changes that lead to the fairness enhancement process through exploring the changes in importance of features. Our resulting-fairness enhanced models retain high sensitivity with improved fairness and explanations of the fairness-enhancement that may provide helpful insights for healthcare providers to guide clinical decision-making and resource allocation.

Figures (6)

• Figure 1: The ExplainableFair framework.
• Figure 2: Most important features before fairness optimization.
• Figure 3: Most important features after race-fair optimization.
• Figure 4: Most important features after sex-fair optimization.
• Figure 5: Most changed features (by ranking) during race fairness optimization