Validation of models that predict Cesarean section after induction of labor

OBJECTIVE: Models for the prediction of Cesarean delivery after induction of labor can be used to improve clinical decision-making. The objective of this study was to validate two existing models, published by Peregrine et al. and Rane et al., for the prediction of Cesarean section after induction of labor. METHODS: We studied consecutive women in whom labor was induced. In all women, we recorded maternal age, height, body mass index, parity, gestational age and the Bishop score prior to induction. Cervical length was measured by transvaginal ultrasound immediately prior to induction of labor. The primary end-point was delivery by Cesarean section. The calibration of the two prediction models was assessed by comparison of predicted and observed Cesarean delivery rates. The discriminative capacity of the models, i.e. the ability of the models to distinguish subjects who had Cesarean section from those who did not (discrimination), was assessed by receiver-operating characteristics (ROC) analysis. RESULTS: We included 240 women in the study, of whom 27 (11%) had Cesarean delivery. The capacity of cervical length in the prediction of Cesarean delivery was limited. In our study population, both prediction models overestimated the risk of Cesarean delivery. Calibration was better for the Peregrine et al. model than for the Rane et al. model, and the two models had areas under the ROC curve of 0.76 and 0.67, respectively. CONCLUSION: Current models that predict the occurrence of Cesarean section after induction of labor have only a moderate predictive capacity when applied within a Dutch practice. We do not recommend the use of these prediction models in clinical practice