Development and efficacy validation of a T3cDM risk model for pancreatitis patients based on multivariate logistic...

Type 3c diabetes mellitus (T3cDM) is increasingly recognized with the rising incidence of pancreatitis, yet its risk profile and early diagnostic tools remain insufficient. Identifying independent risk factors and establishing reliable prediction models are essential for improving early detection and clinical management.

A retrospective study was conducted on 864 patients with pancreatitis admitted to Zhengzhou Central Hospital from January 2020 to December 2022. They were randomly divided into a training set (n = 604) and a validation set (n = 260) using a 7:3 ratio. Collect baseline data, laboratory indicators, and outcomes of T3cDM through case follow-up. Single factor and multiple factor logistic regression were used to screen potential risk factors, and after excluding collinearity and clinically unrelated variables, a multiple factor logistic regression model was constructed; Evaluate model performance using receiver operating characteristic (ROC) curve, calibration curve, and clinical decision curve (DCA); Construct a column chart and dynamic prediction model based on independent risk factors, and use tenfold cross validation to verify the stability of the model. In addition, 172 patients from external hospitals were included for external validation. Furthermore, Kaplan-Meier (K-M) curve analysis was performed to evaluate the survival characteristics of T3cDM in patients with pancreatitis.

In 864 patients with pancreatitis, the cumulative incidence rate of T3cDM was 16.67% (144/864). The baseline data of training set and verification set were balanced (P > 0.05), and there was no significant selection bias. Multivariate logistic regression showed that a history of pancreatitis (OR = 4.301, 95%CI: 2.370-7.804, P < 0.001), alcohol consumption factors (OR = 4.542, 95% CI: 1.669-12.360, P = 0.003), body mass index (BMI) (OR = 1.209, 95% CI: 1.038-1.409, P = 0.015), blood potassium (K⁺) (OR = 2.119, 95% CI:1.440-3.938, P = 0.018), maximum blood glucose(Max Glu) (OR = 1.079, 95% CI:1.015-1.146, P = 0.014), glycated hemoglobin (HbA1c%) (OR = 1.401, 95% CI: 1.145-1.716) Triglycerides (TG) (OR = 1.022, 95% CI: 1.006-1.039, P = 0.008), triglyceride glucose index (TyG) (OR = 1.802, 95% CI: 1.248-2.603, P = 0.002) are independent risk factors for T3cDM in patients with pancreatitis. The column chart model constructed based on the above factors has a training set ROC curve area under the curve (AUC) of 0.857 (95% CI: 0.815-0.899), a sensitivity of 80.3%, and a specificity of 80.4%; The AUC of the validation set is 0.773 (95% CI: 0.691-0.855), with a sensitivity of 72.0% and a specificity of 69.0%. The calibration curve shows that the average absolute error of the training set is 0.023, and the validation set is 0.017. The predicted risk is highly consistent with the actual risk; The DCA curve suggests that the net benefit of the model is significantly higher than that of the "full intervention" or "no intervention" strategies within the risk threshold of 0.05 ~ 0.80. External validation further confirmed the robust predictive performance (AUC = 0.904 [95%CI: 0.838-0.971]). Stratified analysis demonstrated that the model exhibited good predictive efficacy across different age and gender groups. Age-stratified analysis demonstrated that the model exhibited favorable performance across three patient cohorts: young adults (18-44 years), middle-aged individuals (45-59 years), and elderly patients (60-75 years), with AUC values of 0.801, 0.956, and 0.903, respectively. Furthermore, the K-M curve showed no significant difference in T3cDM survival curves between the training and validation sets (Log rank P = 0.126, HR = 0.781, 95% CI: 0.545-1.120), and the overall population had a T3cDM free survival rate of approximately 32% at 60 months of follow-up.

The incidence rate of T3cDM in patients with pancreatitis is high. Previous pancreatitis history, drinking inducement, BMI, K⁺, Max Glu, HbA1c%, TyG, etc. are independent risk factors for T3cDM; The prediction model and column chart constructed based on the above factors have good discrimination, calibration, and clinical practicality, and can be used as an effective tool for T3cDM risk assessment in patients with pancreatitis. Furthermore, external validation further supports the model's applicability.