The Role of Physical Activity in Predicting Nocturnal Hypoglycemia in Individuals with Type 2 Diabetes Mellitus

Background: Individuals treated with insulin face an elevated risk of nocturnal hypoglycemia, which is hard to detect due to reduced symptom awareness. While physical activity supports glycemic control, it may increase the risk of nocturnal hypoglycemia. This study aims to predict nocturnal hypoglycemia in insulin-treated individuals with Type 2 Diabetes Mellitus (T2DM) using a classification model based on daytime glucose patterns and physical activity. Methods: Data were collected from a randomized controlled trial involving insulin-treated individuals with T2DM (n=331). The intervention group (n=166) was analyzed using data from Continuous Glucose Monitoring (CGM), an activity tracker, and a connected insulin pen. Engineered features were extracted from data and used for training an XGBClassifier model with tuned hyperparameters. Model performance was evaluated on an independent test set and was assessed using sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Feature importance was analyzed using Shapley Additive Explanations (SHAP). Results: The model achieved a mean ROC-AUC of 0.81 ± 0.01 and PR-AUC of 0.19 ± 0.04. At a threshold of 0.45, specificity was 96.8%, sensitivity 24.5%, accuracy 95%, PPV 19.7%, and NPV 98%. SHAP analysis showed the top 10 features accounted for 73.8% of total importance, while physical activity features contributed only 7.5%. Conclusions: Prediction of nocturnal hypoglycemia appears promising, although physical activity features collected from wearable devices had a limited impact on performance. Additional studies are needed to confirm generalizability.