Re: Predicting successful vaginal birth after Cesarean section using a model based on Cesarean scar features examined using transvaginal sonography

Abstract

Several models that estimate the probability of successful vaginal birth after Cesarean section (VBAC) have been published and many achieve reasonable predictive performance in terms of discrimination and calibration1. Naji et al.2 presented an innovative and interesting prediction model by introducing novel predictors derived from sonographic measurement of the Cesarean scar. This model was developed within a cohort of just 131 women with one previous Cesarean section. The final model consists of four variables: maternal age, prior VBAC, residual myometrial thickness (RMT) and change in RMT from the first to the second trimester. It is notable that the presented model has extraordinary results regarding discriminative performance, with an area under the receiver–operating characteristics curve (AUC) of 0.94, close to the theoretical maximum of 1. These positive results can be attributed to the introduction of the Cesarean scar variables, since their addition leads to a remarkable improvement in the AUC from 0.62 to 0.94. However, the introduction of Cesarean scar variables deserves further attention. The association of these variables with the probability of successful trial of VBAC was reported by Naji et al. in an earlier study using the same data3. There is no plausible mechanism that explains why these variables have such an impact on the probability of success. Furthermore, rather than choosing predictors based on observed significant relations to outcome variables in the same dataset, contemporary methodological guidelines for prediction research state that predictors should be chosen based on preselection4, a method that results in higher external validity and less overfitting. Therefore, we would like to emphasize the importance of external validation of these predictors in other data. Additionally, the need for external validation is shown when looking into studies on inter- and intraobserver validity of the RMT measurements. These studies use cut-off values of 2.4–3.5 mm for evaluating reproducibility and state that overall interobserver differences are ≤ 1 mm for 77.5–88% of observers5, 6. However, in the model RMT is entered in millimeters while predicted probability increases per millimeter with a coefficient (beta) of 1.44. Therefore, a variability of 1 mm between measurements compromises the model's performance. To be more explicit, the 32-year-old patient without a previous VBAC, with an RMT of 2.7 mm and an RMT decrease of 1.5 mm, has a predicted probability of successful trial of VBAC of 36%; however, if 3.7 mm instead of 2.7 mm had been measured for the RMT, the predicted probability would have increased to 71%. In conclusion, we think that the findings of Naji et al. hold promise. However, these data should be confirmed before introduction into practice. We hope that this brief commentary makes readers aware of the preliminary nature of such optimistic results, and the drawbacks of prediction research in general. We would also like to emphasize the significance of (prospective) external validation and impact studies. To date, no impact studies have been published on the prediction of successful trial of VBAC, and there have been only a few external validation studies. As application of the model for counseling might lead to different birth preferences, resulting in a different selection of women undergoing trial of VBAC, prospective evaluation is essential. E. N. C. Schoorel*†, S. M. J. van Kuijk‡, J. G. Nijhuis†, L. J. M. Smits‡, H. C. J. Scheepers† on behalf of the cesarean Section IMPLEmentation (SIMPLE) II study group †GROW-School for Oncology and Developmental Biology, Department of Obstetrics and Gynecology of Maastricht University Medical Centre+, P.O. Box 5800, 6202 AZ Maastricht, The Netherlands; ‡Department of Epidemiology, Caphri School for Public Health and Primary Care, Maastricht University Medical Centre+, Maastricht, The Netherlands *Correspondence. (e-mail: [email protected])