Researchers from the University of Florida have discovered diagnostic biases in machine learning algorithms used for diagnosing bacterial vaginosis (BV). Nature’s Digital Medicine journal published this article.
The study, led by Ruogu Fang and Ivana Parker, analyzed data from 400 women across four ethnic groups: white, Black, Asian, and Hispanic. The study investigated the performance of four machine learning algorithms in diagnosing asymptomatic bacterial vaginosis using 16S rRNA sequencing data from women of Asian, Black, Hispanic, and white ethnicities.
They found that machine learning models’ accuracy in predicting bacterial vaginosis varied among these ethnicities. Hispanic women experienced the highest rate of false-positive diagnoses, while Asian women had the most false negatives. The models performed best for white women and worst for Asian women.
This study highlights the challenges in machine learning applications in medical diagnostics, particularly regarding ethnic disparities. The research also aids in understanding how bacteria-related factors affect women from various ethnic backgrounds, potentially leading to better treatments. This study emphasizes the need for more equitable and effective methods in developing AI tools to reduce healthcare biases.
Ref: Celeste C, Ming D, Broce J, et al. Ethnic disparity in diagnosing asymptomatic bacterial vaginosis using machine learning. Npj Digit Med. 2023;6(1):211. doi:10.1038/s41746-023-00953-1