A simple model to predict acute on chronic pancreatitis: A multicenter study

To the Editor: Chronic pancreatitis (CP) is a progressive, long-standing inflammatory disease of the pancreas that leads to irreversible structural damage, including fibrosis, calcification, and ductal abnormalities such as pancreatic duct dilation, strictures, and the formation of varying numbers of intraductal stones.[1] One of the serious complications of CP is acute-on-chronic pancreatitis (ACP), which refers to acute exacerbations in patients with established CP. However, due to the lack of a consistent definition of ACP across studies, its reported incidence remains ambiguous. Notably, hospital admissions related to ACP have been increasing annually.[2] Frequent acute exacerbations significantly impair patients’ quality of life and impose a substantial economic burden. Therefore, early recognition and timely intervention are essential to reduce the frequency of acute episodes and extend the interval between them. This study aimed to develop and validate a simple, clinically applicable model for predicting ACP using multicenter clinical data from across China. We hypothesized that a straightforward and effective predictive tool, incorporating clinical and biochemical parameters, could aid clinicians in identifying patients at high risk of ACP. We conducted a multicenter, retrospective study using secondary data collected from January 2010 to January 2024. A total of 542 cases from hospitals in Beijing were included, of which 377 (70%) were used for model training and 165 (30%) for internal validation. An additional 106 cases from a regional hospital were used for external validation (Supplementary File 1, https://links.lww.com/CM9/C688). The study was approved by the Ethics Committee of the First Medical Center of the General Hospital of the PLA (No. S2024-515-01). Given the retrospective nature of the study, informed consent was waived. At the time of initial diagnosis, patients with CP were divided into two groups: those who experienced ACP episodes (ACP group) and those who did not (No-ACP group). Among the 377 patients included in the analysis, 115 (30.5%) had experienced at least one ACP episode. Patients in the ACP group had a younger average age compared with those in the No-ACP group (46.6 ± 15.2 vs. 51.9 ± 14.2 years). Gender distribution was relatively balanced between the two groups, with a slightly higher proportion of females in the ACP group (23.5% vs. 19.5%). Differences in body mass index (BMI) were also observed. A significantly higher proportion of underweight patients was found in the ACP group (15.7%) compared with the No-ACP group (9.2%), whereas obesity was uncommon in both groups. ACP was more frequently observed in patients from North China, with 66.1% of ACP cases originating from this region, compared with 47.7% in the No-ACP group. A history of pancreatopathy was also more common in the ACP group (91 patients [79.1%] vs. 139 patients [53.1%]). Furthermore, hereditary or congenital etiologies were significantly more frequent in the ACP group (9.6%) than in the No-ACP group (1.9%), suggesting that genetic factors may play a key role in the development of ACP [Supplementary Table 1, https://links.lww.com/CM9/C688]. Univariable and multivariable logistic regression analyses identified pancreatopathy as a significant risk factor for ACP (odds ratio [OR] 4.83, 95% confidence interval [CI]: 2.78–8.37, P <0.001). Among imaging findings, pancreatic duct strictures (OR 0.06, 95% CI: 0.02–0.21, P <0.001), pancreatic pseudocysts (OR 0.13, 95% CI: 0.05–0.36, P <0.001), and sinistral (left-sided) portal hypertension (OR 4.21, 95% CI: 1.70–10.44, P = 0.002) were significantly associated with ACP [Supplementary Table 2, https://links.lww.com/CM9/C688]. To further improve the model’s predictive performance, eight different models were developed and compared using receiver operating characteristic (ROC) curve analysis. The key variables included pancreatopathy, carbohydrate antigen 19-9 (CA19-9) >35 U/mL, pancreatic duct stenosis, sinistral portal hypertension, and pancreatic pseudocyst. Model 1, which included all candidate variables, demonstrated the highest predictive accuracy, with an area under the ROC curve of 0.850 (95% CI: 0.810–0.885). Models 2–5 each incorporated four variables, while Models 6–8 included three variables [Supplementary Table 3, https://links.lww.com/CM9/C688]. Among all models, Models 1, 2, 5, and 7 exhibited the highest AUROC (area under the receiver operating characteristic curve) values [Supplementary Table 4, https://links.lww.com/CM9/C688]. The DeLong test revealed no statistically significant difference in AUROC between Models 1 and 5 in the training cohort (P = 0.175). Considering its balance of accuracy and simplicity, Model 5 comprising pancreatic disease history, CA19-9 >35 U/mL, pancreatic duct stenosis, and pancreatic pseudocyst was selected as the final predictive model. This model integrates clinical, biochemical, and imaging parameters, and a corresponding nomogram was developed to facilitate its clinical application [Supplementary Figure 1, https://links.lww.com/CM9/C688]. Using calibration curves, the simplified model demonstrated good agreement between predicted and observed probabilities of ACP in the internal validation cohort [Figure 1A]. External validation showed a similar, though slightly lower, predictive accuracy compared with the Beijing cohort [Figure 1B], suggesting that regional differences in clinical practice or patient populations may affect the model’s generalizability. The model exhibited high predictive accuracy in both internal and external validations, with an area under the ROC curve of 0.828 (95% CI: 0.768–0.888) for internal validation [Figure 1C] and 0.761 (95% CI: 0.673–0.849) for external validation [Figure 1D].Figure 1: Accuracy of simple model between (A and B) internal set and (C and D) external set to predict ACP. ACP: Acute-on-chronic pancreatitis; AUC: Area under the curve.Decision curve analysis indicated a positive net benefit (NB) in the internal validation cohort. Assuming a diagnostic threshold probability of 25% for initiating intervention in patients predicted to have ACP, the model would benefit 13 out of every 100 patients assessed. Although the external validation also showed a positive NB, it was lower than that observed in the internal validation cohort [Supplementary Figure 2, https://links.lww.com/CM9/C688]. This study presents a simple and clinically applicable model for predicting the occurrence of ACP, based on key clinical variables including a history of pancreatic disease, elevated CA19-9 levels, pancreatic duct stenosis, and pancreatic pseudocysts. The model demonstrated high predictive accuracy in internal validation, although its performance was moderately reduced in external validation. In this section, we discuss the significance of these findings, the limitations of our study, and future directions for refining and implementing this predictive model. The four predictors identified in our model—pancreatopathy, elevated CA19-9, pancreatic pseudocyst, and pancreatic duct strictures—are all supported by significant biological and clinical plausibility as risk factors for ACP. Patients with CP often have preexisting structural and functional pancreatic impairments, which increase their susceptibility to acute exacerbations. These findings underscore the importance of regular imaging surveillance for high-risk patients, particularly those with known pancreatic disease who are prone to ACP development. Elevated CA19-9 levels may reflect ongoing pancreatic inflammation or biliary obstruction, both of which substantially contribute to acute exacerbations in CP.[3] Pancreatic pseudocysts, a common complication of CP, arise from pancreatic duct rupture and leakage of pancreatic juice. Some pseudocysts communicate with the pancreatic duct, which can relieve intraductal pressure and potentially reduce the risk of acute pancreatitis.[4] Another key predictor identified was pancreatic duct strictures. Chronic inflammation can lead to stricture formation, which obstructs the outflow of pancreatic enzymes, causing increased intraductal pressure. This elevated pressure damages acinar tissue and contributes to recurrent acute episodes.[5] This finding suggests that patients with imaging evidence of ductal strictures should be considered at high risk for ACP. Such patients may benefit from early endoscopic or surgical interventions aimed at relieving strictures to prevent further complications. Internal validation demonstrated that our model had good predictive performance, with an area under the ROC curve of 0.828, indicating strong discrimination in identifying patients at risk for ACP compared with those without ACP. The AUROC was somewhat lower in external validation (0.761), suggesting that the model’s generalizability may be somewhat limited when applied to different populations and clinical settings. These discrepancies may stem from variations in patient characteristics or healthcare practices, as well as regional differences in diet and lifestyle factors that influence the prevalence and management of CP and ACP. For instance, we observed that ACP patients originating from Northern China were more likely to develop ACP, potentially related to regional dietary habits such as high-fat intake or alcohol consumption, both established as risk factors for pancreatitis. The observed differences in model accuracy between internal and external validation cohorts highlight the need for further refinement of the original nomogram before it can be developed into a broadly applicable predictive tool. Our model is a simple, user-friendly tool designed to predict ACP, enabling clinicians to identify high-risk patients with CP who may benefit from more frequent monitoring or early interventions. For instance, patients with elevated CA19-9 levels or imaging findings suggestive of ductal strictures can be targeted for more aggressive management strategies, such as closer biochemical surveillance (e.g., every 2–3 months), timely endoscopic procedures, or surgical evaluation when indicated, due to their increased risk of developing ACP. Importantly, the model relies on clinical predictors commonly available in most healthcare settings, making it feasible for adoption even in resource-limited environments. This accessibility allows the tool to be useful not only in tertiary care centers but also in smaller hospitals and clinics where advanced diagnostic technologies may not be readily available. In summary, this study presents a reliable and easy-to-use predictive model for ACP that integrates key clinical and biochemical factors. Prospective studies are warranted to validate and refine the model across diverse populations and healthcare settings. Although further research is needed to enhance its accuracy and broaden its applicability, this model represents a significant advancement in the clinical management of ACP. Facilitating early identification of high-risk patients holds promise for improving clinical outcomes and reducing the overall burden of ACP on patients and healthcare systems alike. Acknowledgements The authors thank First Medical Center of General Hospital of PLA, Third Medical Center of General Hospital of PLA, Fourth Medical Center of General Hospital of PLA, Fifth Medical Center of General Hospital of PLA, Sixth Medical Center of General Hospital of PLA, Seventh Medical Center of General Hospital of PLA, Eighth Medical Center of General Hospital of PLA, Tianjin Medical University General Hospital, Affiliated Hospital of Nankai University, Shanxi Provincial People’s Hospital, The First Hospital of Shanxi Medical University, The First Affiliated Hospital of Anhui Medical University, Dazhou City Central Hospital, The First Hospital of Lanzhou University, and The Second Hospital of Hebei Medical University for collecting the clinical data. Conflicts of interest All authors disclosed no financial relationships. Funding This study was supported by General Program of the Health Bureau of the Logistics Support Department under the Central Military Commission (No. 24BJ218).