Preterm birth etiological pathways: a Bayesian networks and mediation analysis approach

Abstract

Background The aim of this study was to determine the mediating effect of spontaneous preterm birth (PTB) main predictors that would allow to suggest etiological pathways. Methods We carried out a case–control study, including sociodemographic characteristics, habits, health care, and obstetric data of multiparous women who gave birth at a maternity hospital from Tucumán, Argentina, between 2005 and 2010: 998 women without previous PTB who delivered at term and 562 who delivered preterm. We selected factors with the greatest predictive power using a penalized logistic regression model. A data-driven Bayesian network including the selected factors was created where we identified pathways and performed mediation analyses. Results We identified three PTB pathways whose natural indirect effect was greater than zero with a 95% confidence interval: maternal age less than 20 years mediated by few prenatal visits, vaginal bleeding in the first trimester mediated by vaginal bleeding in the second trimester, and urinary tract infection mediated by vaginal bleeding in the second trimester. The effect mediated in these pathways showed greater sensitivity to confounders affecting the variables mediator–outcome and exposure–mediator in the same direction. Conclusion The identified pathways suggest PTB etiological lines related to social disparities and exposure to genitourinary tract infections.