Comprehensive Summary
This Literature Review investigates the performance of machine learning (ML) algorithms performance in racial and ethnic minority patients. To achieve this, the study focuses on predicting 30-day complication rates following total joint arthroplasty (TJA), using data from a national outcomes database containing 267,194 patients who underwent TJA between 2013 and 2020. Two ML algorithms, histogram-based gradient boosting (HGB) and random forest (RF), were evaluated for discrimination, calibration, and accuracy in predicting complications. While both models performed well in non-Hispanic White patients (AUC = 0.86), their predictive accuracy decreased significantly in minority populations, with the poorest performance observed in the American Indian cohort (AUC = 0.67 to 0.68).
Outcomes and Implications
The literature review highlights the limitations of modern ML algorithms in healthcare when applied to underrepresented racial and ethnic minority groups. While the models proved accurate in performance in non-Hispanic White patients, their reduced predictive accuracy in minority groups shows the impact of imbalanced data representation in healthcare datasets. The study emphasizes the need for equity in approaches to ML model development and ensuring that datasets adequately represent all demographic groups to improve predictive accuracy and support equitable healthcare outcomes for all.