Nayan M, Salari K, Bozzo A, et al. - Variation in performance by race was shown by a machine learning (ML) algorithm trained to predict survival post-radical prostatectomy; this was found to be true irrespective of whether the algorithm was trained in a naturally race-imbalanced, race-specific, or synthetically race-balanced sample.

• A total of 68,630 patients receiving radical prostatectomy were identified from the National Cancer Database; 57,635 (84%) Caucasians, 8,173 (12%) African-Americans, and 2,822 (4%) non-Caucasian, non-African-American (NCNAA).

• An Extreme Gradient Boosting (XGBoost) classifier was trained to predict 5-year survival in different training samples.

• F1 scores for the classifier trained in the naturally race-imbalanced sample were estimated to be 0.514, 0.511, 0.545, and 0.378 in the race-imbalanced, Caucasian, African-American, and NCNAA test samples, respectively.

• For all race subgroups, similar performance of F1 scores of classifiers trained in the race-specific or synthetically race-balanced samples was evident relative to training in the naturally race-imbalanced sample.

• Thorough assessment of ML algorithms in race subgroups is important before clinical deployment to avert potential differences in care.