A Novel Dual-Outcome Risk Calculator for Trial of Labor After Cesarean

Objective: To develop and validate a multivariable prediction model and clinically actionable risk score for vaginal birth after cesarean (VBAC) success using machine learning, and to integrate neonatal morbidity outcomes into a decision-analytic framework for trial of labor after cesarean (TOLAC) counseling. Methods: We performed a retrospective cohort study of 1,418 consecutive TOLAC cases at a single tertiary care center in California from 2019 through 2025. Multivariable logistic regression and four machine learning algorithms (logistic regression, random forest, gradient boosting, extreme gradient boosting) were trained using 5-fold stratified cross-validation. A cumulative risk score (negative 1 to 7 points) was constructed from independently significant predictors. Neonatal intensive care unit (NICU) admission rates and uterine rupture rates were evaluated across risk strata. Results: The overall VBAC rate was 76.7% (1,087/1,418). Penalized logistic regression achieved the highest cross-validated AUC (0.71, 95% CI 0.67 to 0.75). A parsimonious multivariable logistic model used for score derivation had an AUC of 0.70 (95% CI 0.67 to 0.73). Independent predictors of failed TOLAC included induction of labor (adjusted odds ratio [aOR] 1.93, 95% CI 1.48 to 2.52), hypertensive disorders (aOR 1.60, 95% CI 1.19 to 2.15), diabetes mellitus (aOR 1.71, 95% CI 1.19 to 2.47), obesity (body mass index [BMI] 30 or greater; aOR 1.46, 95% CI 1.11 to 1.90), maternal age of 40 years or older (aOR 1.49, 95% CI 0.89 to 2.50), and gestational age of 41 weeks or greater (aOR 2.22, 95% CI 1.40 to 3.52). Prior vaginal delivery was independently protective (aOR 0.37, 95% CI 0.28 to 0.48). The cumulative risk score stratified VBAC success from 89.1% (score negative 1) to 37.8% (score 4 or higher). NICU admission rates increased concordantly from 31.7 to 200.0 per 1,000 across risk strata negative 1 through 4 or higher (Spearman rho 0.94, P for trend = .005). Uterine rupture occurred in 28 cases (1.97%) and was associated with severe maternal morbidity (10.7% vs 0.7%; odds ratio 16.56, P < .001) but was not predicted by any antepartum risk factor. Exclusion of patients with risk scores of 3 or higher (11.3% of the cohort) improved overall VBAC success to 80.0% (P = .04) and reduced NICU rates to 66.0 per 1,000. Conclusion: A machine learning to derived cumulative risk score incorporating prior vaginal delivery as a protective factor identifies TOLAC candidates with poor VBAC prognosis and elevated neonatal morbidity, providing an evidence-based tool for individualized delivery counseling. Uterine rupture remains unpredictable by antepartum characteristics.