A prognostic model based on integrated clinical indicators for recurrent acute pancreatitis in patients with hypertriglyceridemia: a retrospective cohort study

Figures

• 

  Fig. 1. Flowchart of patient selection for the primary logistic regression analysis and the competing-risk sensitivity analysis. A total of 287 patients entered the follow-up cohort and were included in the competing-risk sensitivity analysis. Among them, 9 patients who died during follow-up without documented recurrence were included only in the Fine–Gray competing-risk sensitivity analysis, whereas the remaining 278 patients constituted the primary logistic regression cohort

• 

  Fig. 2. Selection of predictors for AP recurrence using LASSO logistic regression. A LASSO coefficient profiles of candidate variables. As log(λ) increased, the regression coefficients gradually shrank toward zero. The numbers above the plot indicate the number of non-zero coefficients retained at each penalty level. B Cross-validation plot for identification of the optimal λ value. Red dots represent the mean binomial deviance, and gray error bars represent the standard error. The two vertical dashed lines correspond to lambda.min (left) and lambda.1se (right), respectively. The model corresponding to lambda.min (λ = 0.01299698), which yielded the minimum cross-validation error, was selected for subsequent analysis

• 

  Fig. 3. Nomogram for predicting AP recurrence. The nomogram integrates four independent predictors, including SMD, smoking history, ALB, and TyG. The score for each variable is determined according to the patient-specific value, and the sum of these scores yields the total score, which is then used to estimate the individualized probability of AP recurrence

• 

  Fig. 4. Example of individualized application of the nomogram. The figure shows the actual values of TyG, ALB, SMD, and smoking history for a representative patient and the corresponding points assigned to each variable. The total score was approximately 225, corresponding to a predicted probability of AP recurrence of approximately 94.5%, indicating a high risk of recurrence

• 

  Fig. 5. ROC curves of the combined model and individual predictors. The combined nomogram model achieved an AUC of 0.8964 (95% CI: 0.8597–0.9331), which was higher than those of SMD [0.8030 (95% CI: 0.7482–0.8578)], ALB [0.7731 (95% CI: 0.7198–0.8264)], TyG [0.7610 (95% CI: 0.7003–0.8216)], and smoking history [0.6008 (95% CI: 0.5375–0.6641)], indicating superior discriminatory performance of the combined model over any single predictor

• 

  Fig. 6. Calibration curve of the AP recurrence model. The figure shows the apparent curve, bias-corrected curve, ideal curve, and confidence limits (C.L.). Overall, the bias-corrected curve was close to the ideal curve, indicating good agreement between predicted and observed probabilities. Bootstrap resampling was repeated 1,000 times, with a mean absolute error of 0.021 and a sample size of 278. The Hosmer-Lemeshow test yielded χ² = 5.4053, df = 8, P = 0.7135, and the Brier score was 0.1187322

• 

  Fig. 7. Decision curve analysis. Across most clinically relevant threshold probabilities, the nomogram yielded a higher net benefit than the treat-all and treat-none strategies, indicating its potential utility in AP recurrence risk assessment and clinical decision-making