Machine learning prediction of hypertension and diabetes in twin pregnancies using characteristics at prenatal care entry: a nationwide study

ABSTRACT Objectives To develop a prediction model for hypertensive disorders in pregnancy (HDP) and gestational diabetes (GDM) in twin pregnancies utilizing characteristics at the prenatal care entry level. Methods Cross‐sectional study using the US national live birth data between 2016 and 2021. The association of all prenatal candidate variables with HDP and GDM was tested with uni‐ and multi‐variable logistic regression analyses. Prediction models were built with generalized linear models using the logit link function and classification and regression tree approach (XGboost) machine learning (ML) algorithm. Performance was assessed with repeated 2‐fold cross‐validation and performance metrics we considered were area under the curve (AUC). P value <0.001 was considered statistically significant. Results A total of 707,198 twin pregnancies were included in the HDP analysis and 723,882 twin pregnancies for the GDM analysis. The incidence of HDP and GDM significantly increased from 12.2% in 2016 to 15.4% in 2021 and from 8.1% in 2016 to 10.7% in 2021, respectively. Factors that increase the risk of HDP in twin gestations are maternal age <20, age≥35, infertility, prepregnancy DM, non‐Hispanic Black population, obesity, and those with Medicaid insurance (p<0.001). Factors that more than doubled the risk are obesity class II and III (p<0.001). Factors that increase the risk of GDM in twin gestations are age <25, age≥30, history of infertility, prepregnancy hypertension, non‐Hispanic Asian population, non‐US nativity, and obesity (p<0.001). Factors that more than doubled the risk are maternal age ≥ 30 years, non‐Hispanic Asian, and class I, II, and III maternal obesity ( p<0.001). For both HDP and GDM, the performance of the ML and logistic regression model was mostly similar with negligible difference in terms of all tested performance domains. The AUC of the final ML model for HDP and GDM were 0.62±0.004, and 0.67±0.004, respectively. Conclusions The incidence of HDP and GDM in twin gestations is increasing. The predictive accuracy of the machine learning model for both HDP and GDM in twin gestations is similar to that of the logistic regression model. Both models had modest performance, well‐calibrated, and neither had a poor fit. This article is protected by copyright. All rights reserved.