Computer Modeling to Predict Cesarean Delivery in Term Nulliparas [28N]. (May 2016)

• Record Type: Journal Article
• Authors: Howbert, J. Jeffry; Souter, Vivienne; Kauffman, Ellen; Sitcov, Kristin
• Is Part Of: Obstetrics and gynecology. Volume 127(2016)Supplement 1
• Publication Date: 2016-05
• Subjects: Obstetrics -- Periodicals; Gynecology -- Periodicals; 618
• DOI: 10.1097/01.AOG.0000483497.81885.89
• Languages: English
• ISSNs: 0029-7844

Abstract

Abstract : INTRODUCTION: Statistical and machine learning approaches to prediction of clinical outcomes are being studied in multiple areas of medicine. Our aim was to develop a computer model to predict cesarean delivery (CD) in term nulliparas. METHODS: This study used chart-abstracted data on consecutive deliveries, collected over a 3-year period (2012–2014) by the Obstetrics Clinical Outcomes Assessment Program (OB COAP), a multicenter, clinician-led, quality initiative. Analysis was restricted to singleton, term nulliparas who labored. Variables known at presentation to labor and delivery were included. Logistic regression modeling was used to predict the outcome of interest, CD. Prediction accuracy was evaluated using 10-fold cross validation, a standard machine learning technique that divides data into subsets, for model training and testing. RESULTS: 11, 395 of 54, 904 total deliveries met criteria for inclusion. Of 24 pre-admission variables initially explored using logistic regression, a subset of 8 contributed to prediction of CD: race, weight, height, age, gestational week, cervical dilation, gender of baby, and pregnancy related hypertension. Comparison of known outcomes with those predicted by this model under 10-fold cross validation gave a ROC (receiver operator curve) plot with AUC (area under the curve) of 0.754 (0.744–0.764, 95% CI). CONCLUSION: This demonstrates that readily ascertained pregnancy variables can be used in a computer model to predict the likelihood of CD with moderate accuracy. This approach has the potential to provide individualized risk assessment and help inform clinical decisions, such as the most appropriate birth setting.