Risk factor screening and predictive modeling of Time-in-Range in patients with T2DM undergoing SIIT therapy

Yujia Liu¹Wanting Liu¹Zhifeng Jin¹Nengjuan Li²Lingyong Cao¹Jiang Hu²Shuyuan Lin^1*

• ¹Zhejiang Chinese Medical University, Hangzhou, China
• ²The Second Affiliated Hospital of Zhejiang Chinese Medical University, Hangzhou, China

Objective: To identify the risk factors that influence the time in range (TIR) of blood glucose during hospitalization in patients with type 2 diabetes mellitus (T2DM) undergoing short-term intensive insulin therapy (SIIT), and to establish a predictive model for in-hospital blood glucose fluctuations based on real-world data. Methods: Retrospective data of T2DM patients who were admitted to the Second Affiliated Hospital of Zhejiang Chinese Medicine University for SIIT between 2017 and March 2024 were collected. Random allocation was used to divide the dataset into a training set and a validation set at a ratio of 7:3. Prediction models were constructed separately using logistic regression and random forest algorithms. Additionally, a nomogram was developed for facilitating clinical application. Results: A total of 796 T2DM patients who received SIIT were included, with 651 achieving TIR ≥70% within 10 days of hospitalization. Increasing age, fasting blood glucose (FBG), and use of glinides had a negative effect on achieving TIR ≥70%. In contrast, female sex and higher lymphocyte count were associated with increased likelihood of achieving TIR ≥70%. In the subgroup analysis, FBG, the presence of diabetic nephropathy (DN), and the occurrence of major adverse cardiovascular events (MACE) were found to potentially reduce the risk of achieving both TIR ≥70% and TITR ≥50% within 10 days of hospitalization. For model performance evaluation, the logistic regression model demonstrated slightly superior predictive accuracy (F1 score = 0.89, AUC = 0.80) compared with the random forest model (F1 score = 0.84, AUC = 0.72) on the full sample. After applying undersampling, the model's ability to correctly identify negative cases improved, with specificity increasing to 0.53. Conclusion:This study, based on real-world data, developed a machine learning model (including logistic regression and random forest) to predict the achievement of TIR during hospitalization. The model not only identifies key clinical factors influencing blood glucose fluctuations, but also provides quantifiable decision support for personalized glucose management. This model has the potential to offer new insights and methods for early identification of high-risk patients and optimization of SIIT treatment strategies in clinical practice.