Paper Type
ERF
Abstract
This study explores whether AI can alleviate outgroup bias in medical diagnosis, specifically focusing on gender bias in hospital admissions for chest pain. Using the NHAMCS dataset from 2014-2018, the research examines the predictive power of off-the-shelf AI models and the implications of fair AI models. The results indicate that AI models can amplify existing gender bias in hospital admissions. While achieving reasonable accuracy, these models exhibit low recall scores and a decreased disparate impact ratio, indicating amplified bias. In the meantime, fair AI models can mitigate this bias, improving the disparate impact ratio, but potentially at the cost of male patients’ admission rates. This research provides practical insights for medical practitioners and policymakers interested in alleviating outgroup bias in medical diagnosis.
Paper Number
1278
Recommended Citation
Wongsiridech, Nipasri; Tangwaragorn, Pattharin; and Khern-am-nuai, Warut, "AI, Bias, and Fairness in Medical Diagnosis" (2025). AMCIS 2025 Proceedings. 4.
https://aisel.aisnet.org/amcis2025/health_it/sig_health/4
AI, Bias, and Fairness in Medical Diagnosis
This study explores whether AI can alleviate outgroup bias in medical diagnosis, specifically focusing on gender bias in hospital admissions for chest pain. Using the NHAMCS dataset from 2014-2018, the research examines the predictive power of off-the-shelf AI models and the implications of fair AI models. The results indicate that AI models can amplify existing gender bias in hospital admissions. While achieving reasonable accuracy, these models exhibit low recall scores and a decreased disparate impact ratio, indicating amplified bias. In the meantime, fair AI models can mitigate this bias, improving the disparate impact ratio, but potentially at the cost of male patients’ admission rates. This research provides practical insights for medical practitioners and policymakers interested in alleviating outgroup bias in medical diagnosis.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIGHEALTH