A RECENT study has revealed that machine learning (ML) tools used to diagnose bacterial vaginosis (BV) may perform inconsistently across different ethnic groups, raising concerns about equity in women’s health diagnostics.
Using 16S rRNA sequencing data to predict symptomatic BV, researchers found that ML models were less accurate in diagnosing the condition in Black women compared to other ethnic groups. This discrepancy appears to stem from differences in the composition of the vaginal microbiome, as the key bacterial taxa used to predict BV varied significantly by ethnicity.
The study highlights the importance of population-level diversity in training data for AI-driven diagnostic tools. The authors call for further research using larger, more representative datasets to improve model fairness and reduce diagnostic disparities. As machine learning becomes increasingly integrated into clinical practice, ensuring equitable performance across diverse populations will be critical for improving health outcomes and promoting justice in healthcare delivery.
Reference
Ojo DP et al. Population-level predictive variation in machine learning diagnosis of symptomatic bacterial vaginosis. npj Womens Health. 2025;DOI: 10.1038/s44294-025-00092-w.
