Racial and Ethnic Disparities in Predictive Accuracy of Machine Learning Algorithms Developed Using a National Database for 30-Day Complications Following Total Joint Arthroplasty

Comprehensive Summary

This literature review dives into racial and ethnic disparities in predictive accuracy of machine learning (ML) algorithms which were designed to forecast 30-day complications following total joint arthroplasty (TJA). Researchers utilized a national database comprising 267,194 TJA patients from 2013 to 2020, organized by race and Hispanic or non-Hispanic ethnicity. The analysis involved histogram-based gradient boosting (HGB) and random forest (RF) models, which showed high accuracy in the non-Hispanic White group (AUC=0.86). However, predictive accuracy declined in minority groups, with AUC values ranging from 0.67 in the American-Indian population to 0.79 in Asian non-Hispanics. The research highlighted significant calibration and accuracy issues, especially within smaller-sized minority cohorts.

Outcomes and Implications

These findings showcase critical concerns about equity in healthcare AI, highlighting that ML algorithms trained predominantly on non-Hispanic White populations perform poorly for racial and ethnic minority groups. This racial disparity may showcase existing healthcare inequalities,