Objective To construct and validate a risk prediction model for cognitive decline in elderly patients with type 2 diabetes mellitus (T2DM).

Methods Elderly patients with T2DM who visited a tertiary hospital in Hefei from March to November 2025 were selected as the research subjects. Based on the literature review, a general information questionnaire for patients was compiled according to the actual clinical situation and research purpose. A total of 14 factors were included, and a cross-sectional survey was conducted. The physical frailty assessment, cognitive function assessment, depression status assessment, nutritional status assessment and dementia classification of patients were respectively evaluated by using the Frailty Phenotype Scale, Montreal Cognitive Assessment Scale, Short Form Geriatric Depression Scale, Mini-Nutritional Assessment Scale and Clinical Dementia Rating Scale. The included patients were divided into the modeling group and validation group according to the random number table method. The data of the modeling group were first analyzed by univariate analysis. The influencing factors with statistical significance were taken as independent variables and the occurrence of cognitive weakness was taken as dependent variables to construct a risk prediction model. Finally, the model was tested and evaluated.

Results A total of 430 patients were collected in the study, among whom 120 developed cognitive frailty, with an incidence rate of 27.91%. After analysis, four factors of the patient, namely age, glycated hemoglobin level, nutritional status and depression status, were the predictors of the model (P ＜ 0.01), and the logistic regression coefficient was the weight of each factor in the model. Finally, a nomogram prediction model for the risk of cognitive frailty was constructed. Evaluation of model discrimination: The AUC of the modeling group was 0.838, the sensitivity was 80.00%, the specificity was 94.17%, and the 95%CI was 0.782–0.894. The AUC of the validation group was 0.738, the sensitivity was 77.14%, the specificity was 91.95%, and the 95%CI was 0.634 to 0.843. The model calibration evaluation showed that the prediction accuracy was good (χ² = 7.67, P ＞ 0.05). The clinical decision curve showed that when the predicted probability was within the range of 0.01 to 0.80, the practical clinical application effect of the model was better.

Conclusions The predictive efficacy of the developed model for predicting cognitive decline risk in elderly T2DM patients is good, which can help medical staff assess the occurrence risk of cognitive frailty in elderly diabetic patients, and has high clinical application value.