Abstract： Objective:To construct a risk prediction model for infection with Klebsiella pneumonia (KP) for patients with severe acute pancreatitis (SAP).Methods:Retrospective analysis was done on the clinical data of 109 SAP patients who were admitted to Shanghai General Hospital, between March 2016 and December 2021. Patients were classified into infection group ( n=25) and non-infection group ( n=84) based on the presence or absence of KP infection, and the clinical characteristics of the two groups were compared. The least absolute shrinkage and selection operator (LASSO) algorithm was used to reduce the dimension of the variables with statistical significance in univariate analysis. A nomogram prediction model was created by incorporating the optimized features from the LASSO regression model into the multivariate logistic regression analysis. Receiver operating characteristic curve (ROC) was drawn and the area under curve (AUC) was calculated; and consistency index (C-index) were used to assess the prediction model's diagnostic ability. Results:A total of 25 strains of KP were isolated from 109 patients with SAP, of which 21(84.0%) had multi-drug resistance. 20 risk factors (SOFA score, APACHEⅡ score, Ranson score, MCTSI score, mechanical ventilation time, fasting time, duration of indwelling of the peritoneal drainage tube, duration of deep vein indwelling, number of invasive procedures, without or with surgical intervention, without or with endoscopic retrograde cholangiopancreatography (ERCP), types of high-level antibiotics used, digestion disorders, abnormalities in blood coagulation, metabolic acidosis, pancreatic necrosis, intra-abdominal hemorrhage, intra-abdominal hypertension, length of ICU stay and total length of hospital stay) were found to be associated with KP infection in SAP patients by univariate analysis. The four variables (APACHEⅡ score, duration of indwelling of the peritoneal drainage tube, types of high-level antibiotics used, and total length of hospital stay) were extracted after reduced by LASSO regression. These four variables were found to be risk factors for KP infection in SAP patients by multiple logistic regression analysis (all P value <0.05). Nomogram prediction model for KP infection in SAP was established based on the four variables above. The verification results of the model showed that the C-index of the model was 0.939, and the AUC was 0.939 (95% CI 0.888-0.991), indicating that the nomogram model had relatively accurate prediction ability. Conclusions:This prediction model establishes integrated the basic clinical data of patients, which could facilitate the risk prediction for KP infection in patients with SAP and thus help to formulate better therapeutic plans for patients.