Population-Level Predictive Variation in Machine Learning Diagnosis of Symptomatic Bacterial Vaginosis

Area of Science:

• Microbiology
• Computational Biology
• Health Disparities

Background:

• Bacterial vaginosis (BV) is a common vaginal condition impacting millions globally.
• The diverse vaginal microbiome complicates traditional diagnostic methods.
• Machine learning (ML) models show potential but may perpetuate health inequities.

Purpose of the Study:

• To assess the equitable performance of ML algorithms in predicting symptomatic BV across different ethnic groups.
• To identify ethnic variations in ML model accuracy for BV diagnosis.
• To explore ethnic differences in bacterial taxa associated with BV.

Main Methods:

• Utilized 16S rRNA sequencing data for microbiome analysis.
• Developed and evaluated ML algorithms for predicting symptomatic BV.
• Compared model performance across diverse ethnic cohorts.

Main Results:

• ML models demonstrated differential predictive performance across ethnic groups.
• Lower accuracy was observed for Black women when predicting BV.
• Significant bacterial taxa predictive of BV varied by ethnicity.

Conclusions:

• Existing ML models for BV diagnosis exhibit ethnic performance disparities.
• Addressing these disparities is crucial for equitable healthcare.
• Future studies require larger, diverse cohorts to refine diagnostic tools and mitigate health inequities.