Abstract: Objective To develop a severity prediction model and an online calculator for early prediction of the severity of acute pancreatitis based on simple and readily available clinical indicators for patients with acute pancreatitis within 24 h of admission to the hospital. Methods  The clinical data of 378 patients with acute pancreatitis admitted to Beijing Shijitan Hospital, Capital Medical University, from January 2012 to May 2022 were retrospectively collected and analyzed. According to the Atlanta classification criteria in 2012, 226 patients with mild AP were treated as the mild group (MAP group), and 152 patients with moderate and severe AP were treated as the non-mild group (NMAP group). High risk factors related to the severity of AP were screened through univariate analysis and binary Logistic regression analysis. On this basis, the Logistic regression prediction model is constructed. The receiver operating characteristic  (ROC） curve is used to screen the critical value of the model prediction, calculate the authenticity of the sensitivity and specificity evaluation model prediction, and calculate the consistency between the prediction results of the evaluation model in Kappa and the actual results. Results  A total of 378 patients with AP were included in this study, 252 (66.7%) were males and 126 (33.3%) were females, including 136 cases of moderately severe acute pancreatitis (MSAP) and 16 cases of severe acute pancreatitis (SAP). Univariate analysis revealed that age, diabetes mellitus, hypertension,  heart rate (HR）， white blood cell (WBC）， red blood cell distribution width (RDW）， neutrophils/lymphocyte ratio (NLR）， D-dimer，fibrinogen (FIB），aspartate aminotransferase (AST），  lactate dehydrogenase (LDH）, amylase (AMY）， blood glucose (Glu）， blood urea nitrogen (BUN） and  albumin (ALB） were correlated with the severity of acute pancreatitis (P <0.05), and the above indicators were included in the multivariate analysis, establishing a prediction model equation of Y=－9.487 + 0.363 × RDW (%) + 0.525 × FIB (g/L) + 0.086×Glu (mmol/L)+0.417×LDH (U/L)+0.248×AMY (U/L). The AUC of the prediction model was 0.825, which was greater than the AUC of the correlation index and BISAP score.Calibration curves showed good consistency between predicted risk of NMAP and actual risk of occurrence. The clinical decision curves suggested that when the threshold probability is 0.4-1.0, using this model to predict and identify the development of AP into NMAP and take corresponding treatment measures can make patients benefit in clinical practice. The critical value of NMAP is predicted to be 0.321 according to the maximum point screening model of Yoden index. When the critical value ≥ 0.321 was used to predict NMAP, the sensitivity of model prediction was 69.5%, and the specificity was 86.2%; Kappa value=0.53, indicating that the prediction results of the model have certain authenticity and moderate consistency with the actual results, and have certain clinical application value, but the rate of missed diagnosis is relatively high. Conclusion  The prediction model established based on simple and easily accessible clinical indicators RDW, FIB, Glu, LDH and AMY within 24 h of admission can be used to predict the severity of acute pancreatitis at an early stage, but the rate of missed diagnosis is relatively high, which needs to be further improved.

Key words: acute pancreatitis, severity, early prediction model, online calculator