Predictive modeling for bacterial vaginosis in a Tanzanian cohort of women living with HIV

IntroductionThe vaginal microbiome is an important factor affecting both HIV risk and reproductive health in women living with HIV, and women of African descent are disproportionately affected. Here we use 16S rRNA sequencing data to predict BV in HIV-positive women living in Africa and delineate microbial features that are important for accurate prediction in this cohort compared to two HIV negative cohorts. MethodsOur study population was comprised of a cohort of HIV-positive women living in Tanzania, and two cohorts of HIV-negative women living in the United States. Using OTU variables from data in the Tanzanian cohort, we used random forest, logistic regression, support vector machine, and multi-layer perceptron algorithms to predict BV outcome. To evaluate our models, we compared predicted BV outcome to BV outcome determined using Nugent scoring. ResultsIn evaluating model performance, all four models predicted BV outcome better for the HIV-negative cohorts than the HIV-positive cohorts. Overall, models were better at predicting negative BV outcomes for low Nugent scores than positive BV outcomes for high Nugent scores. Evaluation of significant predictor variables of BV in each cohort revealed that shared features existed between the Tanzanian and Symptomatic cohorts, but not among all three cohorts. Upon comparing performance of models in predicting BV outcome for Black women in HIV-positive and HIV-negative cohorts, we observed that all four models perform better at predicting BV in the HIV-negative cohorts. ConclusionsThe lower predictive performance observed for this Tanzanian HIV positive cohort, coupled with the difference in microbial communities important for accurate prediction suggest that distinct microbial communities in women who are living with HIV may create challenges that affect the accuracy of BV diagnosis and treatment. This study highlights the need for diagnostic tools that consider unique biological and epidemiological factors of populations to address health disparities.