Predicting preterm birth using machine learning techniques in oral microbiomeOriginal paper
What was studied?
This study asked whether the prenatal oral microbiome can predict preterm birth. Researchers compared oral microbiomes of women who delivered preterm versus full-term and built a machine-learning classifier. Samples came from a single mouthwash rinse taken within 24 hours before delivery. They used 16S rRNA sequencing, DESeq2 to find differentially abundant taxa, and a random forest model with cross-validation to predict outcome.
Who was studied?
Participants were singleton pregnant women admitted for delivery at Jeonbuk National University Hospital in South Korea between 2019 and 2021. After exclusions, 59 women were analyzed, 30 with preterm and 29 with full-term birth. Maternal clinical characteristics did not differ between groups, except prelabor rupture of membranes was more common in preterm cases. No participant smoked or had periodontal disease.
What were the most important findings?
Of 465 genera and species examined, 25 taxa differed between groups after removing membrane-rupture confounders, with 22 enriched in full-term birth and only 3 in preterm birth. This pattern points to loss of protective species in preterm cases. Gestational age correlated negatively with the shift toward preterm-enriched taxa. A random forest model using the 9 most important taxa reached a balanced accuracy of 0.765 plus or minus 0.071.
What are the greatest implications of this study?
The findings suggest the oral microbiome carries potential biomarkers for preterm birth, supporting a possible bloodstream route in addition to the vaginal ascending route. A simple mouthwash sample could one day aid risk screening. Because most predictive taxa were enriched in full-term births, overall oral community balance may matter more than any single pathogen. The study was small and single-center, so larger multi-center validation is needed before clinical use.