Abstract: Objective To analyze the risk factors for hypoglycemia in the elderly patients with type 2 diabetes mellitus combined with cognitive impairment, and to establish a prediction model for hypoglycemia based on decision tree. Methods A total of 261 elderly patients with type 2 diabetes mellitus combined with cognitive impairment in the Affiliated Geriatric Hospital of Nanjing Medical University from February 2022 to June 2023 were enrolled in this study by convenient sampling method. Decision tree model was established based on CART algorithm. Receiver operating characteristic (ROC) curve was drawn to evaluate the model. Results In the decision tree, the group with a history of hypoglycemia and HbA1c≥8.3% had the highest risk of 80% of hypoglycemia in the next 6 months, followed by the group with a history of hypoglycemia, HbA1c < 8.3%, insulin use and disease duration ≥20 years, with a risk of 74%. The area under ROC curve (AUC) of the training set was 0.805, with the sensitivity of 0.779, the specificity of 0.725, and the accuracy of 0.840. The AUC of the validation set was 0.722, with the sensitivity of 0.668, the specificity of 0.766, and the accuracy of 0.739. Conclusions The decision tree prediction model established in this study can better predict the risk of hypoglycemia in the elderly type 2 diabetic patients with cognitive impairment in the next 6 months, and can provide a scientific and effective tool to assess the risk of hypoglycemia in clinic.

Key words: aged, type 2 diabetes mellitus, cognitive impairment, hypoglycemia, decision tree, prediction model