Applying machine-learning models to successful vaginal birth after two cesareans

Vaginal birth after two cesarean deliveries (VBAC2) is increasingly recognized as a reasonable and, in selected cases, preferable option. Current evidence suggests that successful and safe VBAC2 can be achieved, particularly among carefully selected candidates, including grand multiparous women. However, the factors that reliably predict success and reduce maternal and neonatal risk in this population remain insufficiently defined.

This retrospective observational study included 541 women with a history of two previous cesarean deliveries who attempted vaginal birth between 2005 and 2022 at a single tertiary teaching hospital. Machine learning models were applied to predict the safety and success of VBAC2 using maternal demographic and clinical characteristics. The primary outcome was successful VBAC2 without uterine rupture. Secondary outcomes included low Apgar score, admission to the Neonatal Intensive Care Unit (NICU), and the need for postpartum blood transfusion.

A total of 541 women attempted a trial of labor, and 78.9% achieved a successful VBAC. Uterine rupture occurred in 2.0%. The XGBoost Classifier performed best with a sensitivity of 83%. The most important factors associated with successful VBAC were the time interval since the last cesarean delivery, followed by the number of previous vaginal births and cervical dilatation at admission.

Grand multiparous women had significantly higher VBAC2 success rates, while uterine rupture rates remained low. Machine learning models can help predict the likelihood of safe vaginal delivery after two cesarean deliveries, thereby supporting informed decision-making regarding the preferred mode of delivery in this unique population. Predictive modeling in this study showed that grand multiparous women had high VBAC2 success rates (78.9%) with a low risk of uterine rupture (2%); key predictors included prior vaginal birth and cervical dilatation.