ObjectiveTo demonstrate the use of generalized additive logistic regression in the development of a risk-scoring system to predict cesarean delivery.MethodsWomen who delivered in the Prince of Wales Hospital, Hong Kong, from 1994 to 1995 were the subjects of our study. Cases included were term singleton pregnancies with cephalic presentation, excluding those requiring cesarean delivery before labor. The cases were divided randomly into two sets. The prediction models were developed from set A and tested on set B, and vice versa. Maternal demographic and obstetric variables were used as potential predictors. Two models were formed, one before and one after the onset of labor. The generalized additive logistic regression was used to achieve optimal dichotomization of continuous measurements, and the predictive models were then developed. The validating results were pooled, represented, and compared as areas under receiver operating characteristic (ROC) curves.ResultsThe first prediction model used maternal age, height, and weight at delivery as well as nulliparity, history of cesarean delivery, and the need for induction of labor. The second model had in addition the need for labor augmentation. The areas under the ROC curve for the models were 0.81 and 0.82, respectively, a statistically significant difference (z= 5.75,P< .001).ConclusionThe use of generalized additive logistic regression optimizes dichotomization of continuous measurements and facilitates the development of precise and reproducible prediction models. Generalized additive logistic regression appears to be a useful tool, and its use is commended.