Researchers at the University of Florida have uncovered that machine learning algorithms designed to detect a common women’s health issue exhibited clinical bias towards tribal groups.
While artificial intelligence (AI) tools hold significant promise for improving healthcare delivery, experts and researchers are wary of their potential to perpetuate cultural disparities. This groundbreaking study, featured in the Nature publication Digital Medicine, marks the first examination of the fairness of these technologies concerning women’s health issues.
Ruogu Fang, an associate professor at the J. Crayton Pruitt Family Department of Biomedical Engineering and the lead researcher, emphasized the utility of machine learning in medical diagnostics but highlighted the concerning bias observed across different ethnic groups. Fang stated, “This bias is particularly troubling in the realm of women’s health, given the existing racial disparities.”
The study focused on evaluating the efficacy of machine learning in identifying bacterial vaginosis (BV), a prevalent condition affecting individuals of reproductive age with varying medical presentations across ethnicities.
Fang and co-author Ivana Parker, both faculty members at the Herbert Wertheim College of Engineering, collected data from 400 individuals, with 100 participants from each major racial group—white, Black, Asian, and Hispanic.
The research revealed disparities in accuracy based on ethnicity when assessing the performance of four machine learning models in predicting BV in asymptomatic individuals. Asian women experienced the highest rate of false negatives, while Hispanic individuals exhibited a greater incidence of false positives.
Parker, an assistant professor in biotechnology, noted, “The models exhibited optimal performance for white individuals but demonstrated subpar results for Asian women,” underscoring the unequal treatment of racial groups by machine learning algorithms.
The study not only enhances understanding of bacterial factors in women from diverse cultural backgrounds but also holds potential for improving tailored treatments based on ethnicity-specific considerations, despite the focus on AI’s predictive abilities for distinct racial demographics.
BV, one of the most common vaginal infections, can cause discomfort and pain due to an imbalance in healthy bacteria levels. The condition’s asymptomatic nature poses challenges in timely identification, despite the associated risks of adverse pregnancy outcomes, preterm births, and sexually transmitted infections.
The researchers advocate for the development of more equitable AI methodologies to mitigate healthcare disparities, as highlighted by their study’s findings.
