Prediction of Type 2 Diabetes Mellitus According to Glucose Metabolism Patterns in Pregnancy Using a Novel Machine Learning Algorithm

Purpose We aimed to predict future development of type-2 diabetes mellitus (T2DM) in women according to their glucose screening test results during pregnancy and several maternal and fetal characteristics, using a machine learning algorithm. Methods Machine learning was used to predict future development of T2DM following pregnancy in women who delivered at our university-affiliated tertiary medical center between 2007 to 2014. Data was retrieved from women’s medical records, and included: age, parity, gravidity, oral glucose tolerance test results—both the 1-h, 50-g, glucose challenge test (GCT) and the 3-h, 100-g, oral glucose tolerance test (OGTT), gestational age at delivery and birthweight. We used an XGBoost algorithm that fits the training data using decision trees. Results Six thousand and ninety-two women who had both GCT and OGTT data available were reviewed. We demonstrated an accuracy rate of 91% in predicting future T2DM, with a specificity and sensitivity of 74% each. The prediction model achieved an area under the curve of 0.85. The most predictive parameters were patient’s age at the index pregnancy and neonatal birthweight. Conclusions A state-of-the-art machine learning algorithm presents promising ability to predict T2DM following pregnancy using simple parameters such as GCT and OGTT values. The algorithm provides an opportunity to identify at-risk patients who may benefit from early assessment and intervention.