Prediction Model for Vaginal Birth After Induction of Labor in Women With Hypertensive Disorders of Pregnancy.

OBJECTIVE: To identify characteristics associated with vaginal delivery compared with cesarean delivery after labor induction among women with hypertensive disorders of pregnancy and to develop and validate a prediction model to assist in clinical care.

METHODS: We studied a retrospective cohort of women with singleton pregnancies who had hypertensive disorders of pregnancy and who underwent induction of labor at 34 weeks of gestation or more from January 1, 2002, to March 31, 2013. Exclusion criteria included spontaneous labor, prelabor cesarean delivery, and known fetal anomalies. The study cohort was randomly divided into two groups; 70% of pregnancies were used to identify characteristics associated with vaginal delivery and develop a prediction model, and 30% were used to internally validate the model. Candidate predictors were limited to those associated with cesarean delivery and were available to a practitioner at time of induction. Stepwise backward logistic regression was used to build the most parsimonious model predicting cesarean delivery. Hosmer-Lemeshow test was used to assess goodness-of-fit. Model discrimination was evaluated using the concordance index and displayed through the area under the receiver operating characteristic curve (AUC).

RESULTS: Of the 1,357 women meeting study criteria, 974 (71.8%) had a vaginal delivery. The final model consisted of eight variables: maternal age, body mass index, gestational age, intrapartum magnesium sulfate for seizure prophylaxis, need for cervical ripening, prior cesarean delivery and cervical dilation, and effacement. Model calibration and discrimination were satisfactory with Hosmer-Lemeshow test P=.35 and with a 95% CI, an AUC of 0.76 (0.73-0.79). Among those with predicted probability of cesarean delivery of 20% or less, 89.5% had a vaginal delivery. Internal validation demonstrated similar discriminatory ability.

CONCLUSION: Using information available before labor induction, and contingent on future external validation, our model can help women better understand their likelihood of vaginal delivery success when undergoing induction of labor for hypertensive disorders of pregnancy.