Competing‐risks model for prediction of small‐for‐gestational‐age neonate from maternal characteristics and medical history

Background The established method of identifying a group of women at high risk of delivering a small‐for‐gestational‐age (SGA) neonate, requiring increased surveillance, is use of risk scoring systems based on maternal demographic characteristics and medical history. Although this approach is relatively simple to perform, it does not provide patient‐specific risks and has an uncertain performance in predicting SGA. Another approach to predict delivery of a SGA neonate is to use logistic regression models that combine maternal factors with first‐trimester biomarkers. These models provide patient‐specific risks for different prespecified cut‐offs of birth‐weight percentile and gestational age (GA) at delivery. Objectives First, to develop a competing‐risks model for prediction of SGA based on maternal demographic characteristics and medical history, in which GA at the time of delivery and birth‐weight Z ‐score are treated as continuous variables. Second, to compare the predictive performance of the new model for SGA neonates to that of previous methods. Methods This was a prospective observational study in 124 443 women with singleton pregnancy undergoing routine ultrasound examination at 11 + 0 to 13 + 6 weeks' gestation. The dataset was divided randomly into a training and a test dataset. The training dataset was used to develop a model for the joint distribution of GA at delivery and birth‐weight Z ‐score from variables of maternal characteristics and medical history. This patient‐specific joint Gaussian distribution of GA at delivery and birth‐weight Z ‐score allows risk calculation for SGA defined in terms of different birth‐weight percentiles and GA. The new model was then validated in the test dataset to assess performance of screening and we compared its predictive performance to that of logistic regression models for different SGA definitions. Results In the new model, the joint Gaussian distribution of GA at delivery and birth‐weight Z ‐score is shifted to lower GA at delivery and birth‐weight Z ‐score values, resulting in an increased risk for SGA, by lower maternal weight and height, black, East Asian, South Asian and mixed racial origin, medical history of chronic hypertension, diabetes mellitus and systemic lupus erythematosus and/or antiphospholipid syndrome, conception by in‐vitro fertilization and smoking. In parous women, variables from the last pregnancy that increased the risk for SGA were history of pre‐eclampsia or stillbirth, decreasing birth‐weight Z ‐score and decreasing GA at delivery of the last pregnancy and interpregnancy interval < 0.5 years. In the test dataset, at a false‐positive rate of 10%, the new model predicted 30.1%, 32.1%, 32.2% and 37.8% of cases of a SGA neonate with birth weight < 10 th percentile delivered at < 42, < 37, < 34 and < 30 weeks' gestation, respectively, which were similar or higher than the respective values achieved by a series of logistic regression models. The calibration study demonstrated good agreement between the predicted risks and the observed incidence of SGA in both the training and test datasets. Conclusions A new competing‐risks model, based on maternal characteristics and medical history, provides estimation of patient‐specific risks for SGA in which GA at delivery and birth‐weight Z ‐score are treated as continuous variables. Such estimation of the a‐priori risk for SGA is an essential first step in the use of Bayes' theorem to combine maternal factors with biomarkers for the continuing development of more effective methods of screening for SGA. Copyright © 2020 ISUOG. Published by John Wiley & Sons Ltd.