Mitigating Algorithmic Bias in AI-Driven Tele-Triage Systems: Ensuring Equitable Diagnostic Accuracy for Underrepresented Demographic Datasets

• Adaan Ahsun

  Covenant University

  Author

The rapid integration of artificial intelligence into tele-triage systems promises to enhance emergency department efficiency and diagnostic accuracy, yet mounting evidence reveals that these systems may perpetuate or amplify existing healthcare disparities across demographic groups. Algorithmic bias in AI-driven triage represents a critical threat to equitable healthcare delivery, particularly for underrepresented populations historically marginalized in clinical datasets. This study investigates the sources, manifestations, and mitigation strategies for algorithmic bias in tele-triage systems through a mixed-methods approach combining retrospective analysis of MIMIC-IV-ED data (N=18,714 patient encounters) with prospective simulation of debiasing interventions. Our findings demonstrate that state-of-the-art large language models exhibit significant demographic bias, with gender flip rates ranging from 9.9% to 43.8% across evaluated models and systematic undertriage of female patients presenting with identical clinical conditions. Implementation of a comprehensive fairness governance framework incorporating demographic blinding, continuous monitoring with ΔF1 ≤ 0.05 thresholds, and retrieval-augmented generation corpus rebalancing reduced bias metrics by 78.4% while maintaining overall diagnostic accuracy at 89.4%. These results establish that algorithmic bias in tele-triage is both measurable and mitigable through systematic intervention. The study contributes a validated framework for equitable AI deployment in emergency telemedicine, with implications for clinical practice, regulatory policy, and responsible AI development.

• PDF

Copyright (c) 2026 Adaan Ahsun (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.