Abstract
BACKGROUND: This study developed a Causal Graphical Model (CGM) to analyze Bacterial Vaginosis (BV), a condition caused by an imbalance in the vaginal microbiota, whose bacterial composition varies among women. While previous studies used variable selection, clustering, and association rules to identify BV-associated bacteria, these approaches lack visual tools to explore causal relationships and determine which are the most relevant. In contrast, the CGM generated in this study allows visualization of associated bacteria and their causal links, thereby identifying those most influential. METHODS: Path Analysis (PA), a statistical structural equation modeling method, was used to construct the CGM, with emphasis on observable variables and to assess direct and indirect effects through correlations and covariances. PA was applied to an already-collected third-party dataset related to BV diagnosis, consisting of data from 132 pregnant women between 4 and 24 weeks of gestation. RESULTS: The CGM, built using a theoretical model based on the Spearman correlation matrix, was validated through statistical metrics and by a clinical-biological expert. The resultant model highlights bacteria influencing BV diagnosis, specifically Mycoplasma hominis (Mh), Atopobium vaginae (Av), Gardnerella vaginalis (Gv), Megasphaera Type 1 (MT1), and Bacteria Associated with Bacterial Vaginosis Type 2 (BVAB2). Among them, MT1 and BVAB2 showed the strongest association with BV. CONCLUSIONS: The CGM effectively identifies causal associations among bacteria related to BV.