A Preliminary Study On The Data Model Construction For Predicting Pancreatic Necrosis In Acute Pancreatitis

Objective:To explore the factors of pancreatic necrosis in acute pancreatitis,then construct the data prediction model of pancreatic necrosis in acute pancreatitis,followed by evaluating and verifying.Methods:A retrospective analysis was performed for the data of patients with acute pancreatitis from January,2009,to August,2018,in the First Affiliated Hospital of Guang Xi Medical University.According to abdominal CT scan,patients were divided into pancreatic necrosis group and non-pancreatic necrosis group.The basic medical history information,etiology,laboratory examination results within 24～48 hours of admission,and volume status were compared between the two groups.（The standard was as below:（1）reach-goal group:average Arterial pressure>65mm Hg,BUN<7.14mmol/L and0.35≤HCT≤0.44）;（2）non-reach-goal group which failed to meet the standard:mean arterial pressure≤65mm Hg or BUN≥7.14mmol/L or HCT>0.44 or HCT<0.35).Single factor analysis was used to screen out correlative predictors,then use lasso regression to screen variables into a multivariate binary logistics regression model,and finally screen out independent related factors that affect pancreatic necrosis in patients with acute pancreatitis.The last step was further incorporating it into the clinical prediction model for nomogram.And the model was evaluated by the degree of differentiation,calibration,and clinical effectiveness.Results:Among 980 patients,there were 668（68.2%）males,312（31.8%）females;618（63.1%）Mild acute pancreatitis,and 267 Moderate severe acute pancreatitis（27.2%）,Severe acute pancreatitis 95（9.7%）.It was found that serum triglycerides（TG）,serum calcium(Ca²⁺),C-reactive protein（CRP）,and hypovolemia were independent risk factors for pancreatic necrosis in Acute pancreatitis.Cholelithiasis was a protective factor of pancreatic necrosis relative to other etiologies（such as hyperlipidemia,alcohol,etc.）,and pancreatic necrosis prediction model can be established based on the above independent predictors.The sensitivity of the model was 0.724.The specificity was 0.799,and the C index was 0.798（95%CI:0.765-0.831）,indicating that the model had predictive ability.The best cut-off point of ROC curve was 0.195,the sensitivity and specificity of the model were 0.724 and 0.799,respectively,and it was further proved that the model was reliable and effective.The Hosmer-Lemeshow goodness-of-fit test showed that the calibration curve wasx²=10.843,P=0.21,the calibration curve showed that there was good consistency between the calibration curve and the ideal curve,which further indicated that the model had good fitting validity and high predictive value.The prediction model has a net benefit rate greater than 0 in the range of 0-0.76,indicating that it has good clinical effectiveness.The internally verified C index is 0.793,which proves that this model is stable and reliable.Conclusions:Independent predictors for pancreatic necrosis in acute pancreatitis are biliary cause,TG,Ca²⁺,CRP,and volume status on admission.The clinical prediction model constructed based on the above independent predictors can predict the risk of pancreatic necrosis.This model is stable,reliable,and has good discriminative,calibration and clinical validity,which maybe use to predict the occurrence of pancreatic necrosis in the early stage of acute pancreatitis.