Abstract
A systematic review following PRISMA principles searched PubMed, Embase and grey literature from January 2020 to April 2025.Studies were included if they examined AI/ML tools in triage or risk stratification with disaggregated ethnic data.One reviewer conducted screening, extraction and quality assessment.Eight studies met inclusion criteria, comprising cohort studies (n=62,482 to n=249,829), systematic reviews and cross-sectional analyses examining stroke prediction, heart failure, atrial fibrillation, cardiovascular disease risk and emergency department triage across diverse populations.AI/ML models demonstrated moderate to high discrimination (AUC 0.63-0.86)but showed consistently poorer performance for ethnic minority populations, with AUC differences of 0.05-0.12reported across multiple studies.Only 38% of studies reported performance differences by demographic group, and merely 25% conducted comprehensive subgroup analyses.Three-quarters relied on internal validation, 88% lacked geographical diversity and none provided evidence of improved clinical outcomes following AI implementation.Models trained on predominantly white populations showed reduced accuracy in other ethnic groups.Emergency department data revealed white patients received more acute triage scores despite often requiring less intensive care.AI triage and prioritisation systems demonstrate measurably inferior performance for ethnic minority populations, with critical methodological shortcomings undermining claims of equitable healthcare improvement.The evidence reveals a concerning pattern of algorithmic bias that may exacerbate existing health disparities.Key recommendations include mandatory demographic subgroup performance reporting, external validation in ethnically diverse settings before clinical deployment, integration of social determinants alongside clinical variables and post-implementation monitoring with ethnicity-disaggregated outcomes.Regulatory oversight should precede widespread AI triage system adoption to prevent institutionalisation of healthcare inequities.Future research must prioritise real-world implementation studies examining long-term effects on health equity across diverse populations.