Reporting of race and ethnicity in studies of artificial intelligence in pediatric urology

Abstract

Introduction: While an increasing number of artificial intelligence (AI) models are being developed in pediatric urology, the extent of race/ethnicity reporting among these studies is unclear. Our objective was to evaluate the inclusion and quality of race/ethnicity reporting in AI models in pediatric urology. Methods: We conducted a secondary analysis of studies included in the AI in PEDiatric UROlogy (AI-PEDURO) collaborative living scoping review and repository. We examined racial/ethnic groups reported, their proportional representation, use of race/ethnicity as a predictor, conducting of stratified analyses by race/ethnicity, data collection methods, bias evaluation, and discussions of implications on equity. Results: Of 81 studies in the AI-PEDURO repository, six (7.4%) reported race/ethnicity. Five studies included White and Black patients, representing 4824/7968 (60.5%) and 1377/7968 (17.3%) of the pooled cohort, respectively. Asian patients were included in three studies and represented 178/6861 (2.6%). Two studies reported Native Hawaiian or other Pacific Islander and Hispanic or Latino patients, representing 20/6704 (0.3%) and 1236/6704 (18.4%), respectively. One study included American Indian or Alaska Native patients, representing 69/6604 (1.0%). Mixed patients were included in three studies and represented 103/7711 (1.3%). Race/ethnicity was a predictor variable in 4/6 studies. None of these six studies conducted stratified analyses of model performance across race/ethnicity subgroups, reported race/ethnicity data collection methodologies, examined algorithmic biases, discussed implications on equity, or examined socioeconomic status or geographic residence. Conclusions: Race/ethnicity reporting is poor in most AI studies in pediatric urology. Standardized reporting may help ensure fairness and generalizability of models across diverse pediatric urology populations.

Ähnliche Arbeiten