339: Future Type-2 Diabetes Prediction Following Pregnancy - Using a Novel Machine Learning Algorithm

We aimed to predict diabetes mellitus type 2 following pregnancy in women with and without gestational diabetes (GDM) using machine learning abilities. Machine learning was used to predict diabetes mellitus type 2 in patients after pregnancy, among women who delivered in our medical center (2007-2014). All women were followed for a median period of 64 months (± 32). Data was retrieved from the medical records of women, with the following baseline characteristics: maternal age, parity, gravity, oral glucose test results - both glucose challenge test (GCT) and oral glucose challenge test (OGTT), fetal birth weight and presence of GDM. Positive outcome was defined as pathological 75gram OGTT postpartum or diagnosis of diabetes mellitus later in life by reviewing patient's medical records. We used XGBoost algorithm, that fits the training data using decision trees. We were able to rate the factors according to their influence on the prediction. We assigned a weight to each class (positive / negative), to overcome potential biases. For the machine training phase 66% of the cohort was randomly selected. We have then used the rest of the samples to evaluate our model's accuracy. The baseline parameters were fed into the trained model and the predicted outcome was compared to the actual outcome. 73,959 were collected of whom 6,091 women had both GCT and OGTT available. The incidence outcome – Type 2 diabetes following pregnancy - was 5.8%. On the randomly picked samples used to evaluate our model, we demonstrated an accuracy rate of 91% in predicting future diabetes mellitus. The specificity and sensitivity were both 74%. The most predictive parameter was OGTT value at 60 minutes A state-of-the-art machine learning algorithm presented promising ability to predict type 2 diabetes mellitus following pregnancy using simple parameters such as OGTT values. The algorithm provides an opportunity to identify at-risk patients who may benefit from early assessment and intervention