Predicting the risk of colonization of the birth canal in postpartum women by opportunistic pathogens and assessing the health threat to their newborns using machine learning methods

Colonization of the birth canal of postpartum women with resistant opportunistic microorganisms (OPs) is a key link in the development of infectious complications in the postpartum period and poses a threat to newborns. Classical methods of epidemiological analysis do not always allow taking into account complex interactions of risk factors. The aim of the study is to predict the risk of colonization of the birth canal of postpartum women with resistant strains of OPs and to assess the impact on outcomes in their newborns using machine learning.

We performed a retrospective analysis of 166 birth histories and exchange cards of pregnant women. Two analytical samples were formed: «Puerperants» (105 predictors) and «Newborns» (28 predictors). Classical statistics methods (Fisher’s exact test, relative risk) and five machine learning algorithms were used. The models were evaluated using Accuracy, Precision, Recall, F1-score, and ROC-AUC metrics; interpretation was performed using SHAP analysis.

The colonization rate was 60.2 % (100/166). Statistically significant risk factors were rural residence (RR = 1.68), low level of education (RR = 1.80), unemployment (RR = 1.74), unregistered marriage (RR = 1.76), history of abortions
(RR = 1.30), preterm birth (RR = 1.50), prolonged prenatal (RR = 1.65) and postpartum (RR = 3.43) hospitalization. The best predictive performance for the «Postpartum women» sample was shown by the Random Forest model (F1-score = 0.846; ROC-AUC = 0.842). Prematurity, diseases of the early neonatal period, and transfer to the intensive care unit were significantly more common in children born to mothers carrying resistant strains.

Colonization of the birth canal of postpartum women with resistant strains of OPs is associated with a complex of social, clinical and organizational factors. Machine learning methods allow constructing an effective predictive model and confirm the significant negative impact of maternal colonization on health of newborns.