629: MFMU Calculator Versus Machine Learning Techniques for Predicting VBAC Success – Which Model Works Best?

Better prediction of vaginal birth after cesarean (VBAC) may improve patient selection for trial of labor after cesarean (TOLAC). Machine learning (ML) techniques, overcoming nonlinear interactions, have recently been used to improve the prediction of different health outcomes. We sought to develop a ML model for prediction of VBAC success and compare its performance with the MFMU VBAC calculator. All consecutive singleton TOLAC deliveries from a tertiary academic medical center between 2017-2018 were included. We analyzed and compared the following ML algorithms: decision trees, linear discriminant analysis, partial least squares, generalized linear model with LASSO regularization, boosted logistic regression, linear and radial SVM, random forest, stochastic and extreme gradient boosting and neural networks. For developing the ML models, we classified prior arrest of dilatation prior to the second stage of labor and prior arrest of descent as different features. We used a nested cross-validation approach to determine a robust accuracy for each algorithm. Mean and standard deviation of the AUC of the new models developed were calculated and Wilcoxon ran sum test p-values were calculated in comparison to the MFMU calculator. 1006 TOLAC deliveries were included in the analysis with an observed VBAC rate of 86% (861/1006). The ML algorithm that performed best was partial least squares with an AUC of 0.735±0.047, not significantly different from the MFMU calculator (0.733±0.048, p-value 0.55). Prior arrest of descent significantly decreased the probability of VBAC success (OR 3.11 95% CI 1.93-5.03), while prior arrest of dilatation was not significantly different between the groups (table). In addition, maternal height was significantly associated with VBAC success while maternal weight at admission did not contribute to VBAC prediction (figure 1). The partial least squares ML model performed as well as the MFMU VBAC calculator and identified a prior arrest of descent as increasing the odds of TOLAC failure.View Large Image Figure ViewerDownload Hi-res image Download (PPT)