We retrospectively collected data from patients with T2DM who underwent SIIT at the Second Affiliated Hospital of Zhejiang Chinese Medical University (Xinhua Hospital of Zhejiang Province) between March 2017 and March 2024. This study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement and the principles of the Declaration of Helsinki, and was approved by the Ethics Review Committee of the Second Affiliated Hospital of Zhejiang Chinese Medical University (approval number: Lunshen 2023yan No. 089-01). All enrolled patients were exempted from providing informed consent due to the retrospective nature of the study.
Primary diagnosis of T2DM (E11.900) or T2DM with poor glycemic control (E11.600x051).
Age between 18 and 90 years, gender unlimited.
Patients must have a hospitalization duration of 7 days or more, during which they receive SIIT.
The number of effective fingertip blood glucose monitoring per day is greater than or equal to 6 times, and the earliest and latest measurement time interval must be more than 12 hours. The number of hospitalization days with effective fingertip blood glucose test must account for more than 80% of the total hospitalization days.
The first measurement date of the GV index must be no later than 2 days after admission, the last measurement of the GV index must be no earlier than 3 days before discharge, and the interval between any two consecutive GV index measurements must not exceed 2 days.
Patients must have results for HbA1c test conducted within 2 months prior to the admission date, and other laboratory tests are accepted only if conducted within 1 week before admission or within 2 days after admission.
For patients who are re-hospitalized during the study period, only the first diagnosed case will be included.
Other types of diabetes: type 1 diabetes, gestational diabetes, late-onset autoimmune diabetes in adults, etc.
Patients who did not undergo routine blood and lipid tests in the laboratory examination 2 weeks before admission and within 2 days after admission;
Patients diagnosed with hypoglycemic coma, hyperglycemic hyperosmolar syndrome coma, diabetic ketoacidosis, and other diabetes mellitus with severe acute complications;
Grade III (severe) or above hypertension: systolic blood pressure ≥ 160 mmHg or diastolic blood pressure ≥ 100 mmHg;
Patients with severe renal impairment: chronic renal failure azotemia stage, chronic renal insufficiency uremia stage, end-stage renal disease, blood urea nitrogen (BUN) ≥ 100 mg/dl, blood creatinine (SCR) ≥ 5mg/dl, acute renal failure or receiving dialysis treatment;
Patients with severe liver function damage: acute viral hepatitis, EB viral hepatitis, drug-induced liver injury, severe alcoholic liver disease, liver failure, hepatic encephalopathy, ALT or AST increase more than 5 times the upper limit of normal, etc;
There are acute major cardiovascular events such as acute myocardial infarction, acute stroke, acute heart failure in the diagnosis;
Diagnosis of other diseases that affect blood glucose control, such as: acute and chronic pancreatitis, pancreatic cancer, pancreatectomy and other pancreatic diseases; Cushing’s syndrome (Cushing’s Syndrome), acromegaly (Acromegaly) and other endocrine diseases; or fungal infection of the urinary tract, Candida vaginitis, urinary tract infection, herpes zoster, syphilis and other infections and stress; autoimmune diseases such as Graves’ disease;
Diagnosed with malignant tumors such as prostate malignant tumor, lung malignant tumor, Kaposi’s sarcoma, etc., or pituitary adenoma, parathyroid tumor, adrenal adenoma, pheochromocytoma, etc., which cause direct glucose or insulin;
Excluding the uncertainty of Chinese medicine treatment such as inability to determine the dose of Chinese medicine and the method of use.
The electronic medical record data, including diagnoses, medical orders, laboratory tests, and fingertip blood glucose records, were directly exported from the hospital’s scientific research management system. A standardized spreadsheet was designed to organize the retrospective data. Each patient was identified using a unique visit ID, and diagnostic information was coded accordingly. If a patient lacked certain laboratory test results, these data were extracted independently by two researchers from clinical records and cross-checked to minimize bias. Finally, the original datasets were merged into a master table, where each row represented a unique patient (patient ID) and each column contained continuous and categorical variables describing patient characteristics.
In this study, indicators related to blood glucose fluctuations during hospitalization were defined as research endpoints. Based on the actual measurement time points of patients, and combined with the hospital’s work and rest schedule, we summarized the time range of multiple daily measurements of patients as follows: post-breakfast (05:00-06:59), the second pre-breakfast segment (07:00-08:59), post-breakfast (09:00-10:59), before lunch (11:00-12:59), post-lunch (13:00-14: 59), post-lunch segment 2 (15:00-16:59), pre-dinner (17:00-18:59), post-dinner (19:00-20:59), pre-bedtime segment 1 (21:00-22:59), pre-bedtime segment 2 (23:00-00:59), early morning segment 1 (01:00-02:59), early morning segment 2 (03:00-04:59).
In the real-world, patients may not be able to measure their fingertip blood glucose every day during their stay in the hospital according to a strict measurement schedule, due to a combination of subjective factors such as patient compliance and a variety of objective factors such as consultative examinations. To minimize data bias, strict inclusion criteria for GV index were formulated (refer to #4 and #5 in the inclusion criteria). TIR (blood glucose range: 3.9-10.0mmol/L) and the time in tight range (TITR) (blood glucose range: 3.9-7.8mmol/L) were calculated based on fingertip glucose values at least 6 times per day as an indicator for assessing daily glucose fluctuations.
Patients with poorly controlled T2DM generally require 7–10 days of SIIT during hospitalization; therefore, achieving TIR ≥ 70% within 10 days was considered the primary outcome. Because TIR only represents the proportion of time with glucose levels between 3.9- 10.0 mmol/L, we also analyzed TITR as a subgroup variable to further assess glycemic control during hospitalization.
After manual inquiry and supplementation, height, weight, HbA1c, fasting C-peptide (FCP) and fasting insulin (FINS) were still missing. Because the number of missing values was small, we chose to impute these missing data. To minimize bias, multiple imputation using predictive mean matching was applied for height (36 missing values), weight (32), HbA1c (13), FCP (83), and FINS (67). Continuous variables, including age, sex, length of stay, and FBG, as well as dichotomous variables such as the presence of chronic diabetic complications, obesity, fatty liver, metabolic syndrome, and hypertension, were included in the regression-based imputation model. This procedure generated five separate datasets (five imputations) (
Comparison of multiple imputation results. Comparison of multiple imputation results. Time0 represents the data before imputing missing values, and Time1–Time5 represent the results of the five imputations.
Patients were categorized into four groups based on their TIR and time in tighter range (TITR) within the first 10 days of hospitalization. Initially, patients were divided into two main cohorts according to whether they achieved TIR ≥70% during the 10-day treatment period. Subsequently, to further explore the factors associated with stricter glycemic control (3.9–7.8 mmol/L), a subgroup analysis was conducted within the TIR-passing cohort (Cohort 2). Specifically, patients in Cohort 2 were further classified according to whether they achieved both TIR ≥70% and TITR ≥50%. The grouping framework is summarized in
Patient grouping overview.
| Grouping level | Cohort/Subgroup name | Grouping criteria | Description |
|---|---|---|---|
| Primary grouping (by TIR) | Cohort 1 – TIR failing group | TIR < 70% within 10 days of treatment | Patients who did not achieve TIR ≥70% during the 10-day hospitalization period. |
| Cohort 2 – TIR passing group | TIR ≥ 70% within 10 days of treatment | Patients who achieved TIR ≥70% during the 10-day hospitalization period. | |
| Subgroup analysis (within Cohort 2) | Subgroup 1 – TIR ≥70% & TITR ≥50% | TIR ≥70% and TITR ≥50% within 10 days | Patients with both good overall glucose control and stable glucose levels within the tighter range (3.9–7.8 mmol/L). |
| Subgroup 2 – TIR ≥70% but TITR <50% | TIR ≥70% but TITR <50% within 10 days | Patients achieving general TIR targets but failing to meet the tighter glucose control standard. |
All clinically collected variables were analyzed for differences using SPSS version 27, and candidate variables showing significant differences between the two groups (P < 0.05) were selected. Non-normally distributed data were analyzed using the Mann–Whitney U test, and binary categorical variables were compared using the chi-square test. Normally distributed data were expressed as mean ± standard deviation, skewed data as median (interquartile range), and categorical data as frequency (percentage). A P-value of < 0.05 was considered statistically significant.
After identifying the key variables, we constructed both logistic regression and random forest models using the scikit-learn library in Python version 3.11.4. For the logistic regression model, L2 regularization (regularization strength C = 1.0) was applied, and the “lbfgs” optimizer was used for training, with the maximum number of iterations set to 1000. The dataset was randomly split into training and testing sets in a 7:3 ratio, with a random seed of 42 to ensure reproducibility.
For the random forest model, an ensemble learning approach was adopted by constructing multiple decision trees for prediction. The random seed was also set to 42 to maintain consistency and reproducibility. The dataset for this model was randomly divided into training and testing sets in an 8:2 ratio, and the SMOTE technique was applied to address the issue of class imbalance.
The performance of both models was evaluated using multiple metrics, including the area under the receiver operating characteristic (ROC) curve (AUC), calibration curves, the Hosmer–Lemeshow goodness-of-fit test, and decision curve analysis (DCA). After assessing the performance of both models, the logistic regression model was selected as the final model for statistical analysis. To further verify the model’s reproducibility, the logistic regression model was revalidated using Zstats 2025 and R version 4.5.1. In addition, the model’s performance was comprehensively evaluated based on the F1 score, recall, specificity, accuracy, and AUC.