Regression-based model for predicting preterm birth using vaginal lactobacilli and routine clinical data.

backgroundPreterm birth (PTB) is a leading cause of neonatal morbidity and mortality. Although increasing evidence links vaginal microbiota to preterm birth, clinically practical models integrating microbial and clinical indicators are lacking. We aimed to develop a model that combines vaginal microbiota with routine clinical data for prenatal PTB risk assessment.

methodsWe conducted a retrospective, single-centre cohort study including 4,558 pregnant women who delivered between 2019 and 2023 at the Women's and Children's Hospital of Chongqing Medical University. Clinical variables included blood counts, biochemical tests, and vaginal microbiota. After excluding incomplete and non-hospital cases, we excluded clinical variables with missing rates greater than 50% to improve data quality and the accuracy of model predictions. For variables with less than 50% missingness, we applied the MissForest algorithm for non-parametric imputation, 1,228 participants were included in the final analysis for model development. Logistic, LASSO, and Poisson regression models were internally validated, with no external validation performed.

resultsReduced vaginal Lactobacillus, vaginal fluid leakage, abnormal blood glucose, and elevated Cystatin C were significantly associated with an increased risk of preterm birth (Lactobacillus reduction OR = 1.723; PROM OR = 3.297; abnormal glucose OR = 1.401; Cystatin C OR = 4.298; all P < 0.05). When the logistic prediction model combining indicators reflecting vaginal microecological imbalance (such as reduced Lactobacillus, vaginal cleanliness, microbial density, fungi, hydrogen peroxide, and N-acetyl-β-D-glucosaminidase) with clinical variables including cervical cerclage (performed for CL ≤ 25 mm), vaginal fluid leakage, blood glucose, uric acid, and Cystatin C, the model demonstrated moderate predictive performance for preterm birth (AUC = 0.753), suggesting that integrating vaginal microbiota with routine clinical parameters may improve the accuracy of prenatal risk assessment for preterm birth.

conclusionsThis study constructed a prediction model for preterm birth integrating routine prenatal examination indicators and vaginal microbiota information. The model provides a preliminary, feasible, and low-cost framework approach for prenatal risk stratification in pregnancy. However, at this stage, it should be regarded as an exploratory research prototype that requires further multicenter validation across diverse clinical settings and populations before potential clinical application.