| Background and objectives:Acute pancreatitis (AP) is a common disease with high mortality and morbidity worldwide, while the mortality of severe acute pancreatitis (SAP) is reported to be as great as 36-50%. Studies have shown that organ failure and infectious pancreatic necrosis (IPN) are determinants of mortality in acute pancreatitis. Early assessment of the severity of acute pancreatitis and prediction of IPN for guiding aggressive early treatment is important to improve the prognosis of patients and reduce the mortality. The revised Atlanta classification and definitions:mild acute pancreatitis (MAP), moderate severe acute pancreatitis (MSAP) or SAP, have not recommended multi-factorial scoring systems (such as APACHE II) or biological indicators as predict severity signs. The purpose of this study is to analyze and identify optimal multifactorial scoring systems or biological indicators for early prediction of MSAP, SAP and risk of death by establishing an AP database, then screenout indicators closely associated with IPN, aiming to build a multi-factor scoring system for the early prediction of IPN.Method:1. Establishment of acute pancreatitis database(1) The collection content of AP data and related standards are reference to the latest AP guidelines and consensus, and determined by the discussions of the gastroenterology experts in our hospital.(2) The AP database was established, including the data entry page of hospitalized patients and follow-up patients as well as automatically calculate function, then collected patient data continuously.2. Compare the value of the multi-factor scoring system with biological indicators in the early prediction of severity of acute pancreatitis(1) Selected the age of 18 to 85 years, onset 3 days hospitalized patients from the AP database as research subjects. Severity of AP were respectively classified by 1992 Atlanta Classification (referred to as the original Atlanta classification) and the revised Atlanta classification. The risk of death was defined as the patients died during hospitalization or patients in critical condition gave up treatment.(2) A test of significance in multi-factor scoring system and biological indicators were conducted among MAP, MSAP and SAP, the area under the receiver operating characteristic curve (AUC) was applied as a measure of the accuracy for predicting severity and risk of death of multiple scoring systems and biological indicators.3. Establishment of a multi-factor scoring system to predict infected pancreatic necrosis by multinomial Logistic regression and decision trees(1) Selected patients from the AP database were divided into three groups by the revised Atlanta classification:non-pancreatic necrosis (NPN) group, sterile pancreatic necrosis (SPN) group and infective pancreatic necrosis (IPN) group.(2) Clinical variables were selected in the clinical data that might be relevant to IPN, further to select significant difference variables among NPN, SPN and IPN as independent variables of the multinomial Logistic regression and decision trees analysis, and define the attributes and assignment.(3) Selected patients were randomly divided into modeling library (60%) and Verification Library (40%), selected NPN SPN and IPN as the dependent variable; Selected the ordinal Logistic regression analysis or multinomial Logistic regression analysis through the test of parallel lines, chose Stepwise Regression for screening independent variables and creating a multinomial Logistic regression prediction model; Selected decision trees and CHAID growing method for screening independent variables and establishing a CHAID model.(4) Establishment of a multi-factor scoring system based on the weight of the variable and the variable type of multinomial Logistic regression model, and created a multi-factor scoring system based on the variables of decision trees model. The accuracy of the two IPN multi-factor scoring system in predicting IPN were tested by calculating the area under the receiver operating characteristic(ROC)curve.Results:1. Establishment of acute pancreatitis database(1) AP database collected 269 indicators includes basic information, medical history, physical examination, laboratory tests, imaging studies, medical treatment, interventions, complications, condition assessment and prognosis and post-discharge follow-up data. The database has enrolled 1087 patients with AP in our hospital from January 2011 to December 2012.(2) We innovatively designed automatic function for AP database:self-test error data, automatic calculation of data, automatic scoring, automatic diagnosis of transient and sustained organ failure, automatic diagnosis of MAP, MSAP and SAP. In addition, it has a data query, data import and exchange, data statistical analysis and other functions.2. Compare the value of the multi-factor scoring system with biological indicators in the early prediction of severity of acute pancreatitis(1) Enrolled 708 patients can be divided into 278 patients with MAP (39.26%), 430 patients with SAP (60.74%) according to the original Atlanta classification; or divided into 215 patients with MAP (30.23%),324 patients with MSAP(45.90%) and 169 patients with SAP (23.87%) according to the revised Atlanta classification. A total of 41 patients (5.79%) died or in critical condition to give up treatment, the risk of death in patients with SAP was significantly higher than the MSAP patients (17.16% VS 3.7%, P<0.001)(2) ROC curve analysis showed that multi-factor scoring system (APACHE Ⅱ score, Ranson score, score BISAP), CTSI and biological indicators (HCT, BUN, Cr, CRP and PCT) had no value in predicting MSAP, their AUCs were 0.448 to 0.615.(3) ROC curve analysis showed that the multi-factor scoring systems and CTSI have predictive value in early prediction of SAP defined by original or revised Atlanta classification. On the first or second day after admission, AUCs for APACHE Ⅱ in the prediction of SAP were 0.752 and 0.772, respectively, AUCs for BISAP were 0.715 and 0.736. On the second day after admission, AUCs for Ranson,CTSI and CRP in the prediction of SAP were 0.718,0.654 and 0.707, respectively. The accuracy of other biological indicators was poor with AUCs lower than0.7.(4) On the first day of the admission, AUCs for APACHE Ⅱ, Ranson, BISAP, HCT, BUN and Cr in predicting the risk of death were 0.893, 0.818,0.559,0.875 and 0.844, respectively. Two days after admission, AUCs for APACHE Ⅱ, Ranson, BISAP,BUN, Cr, CRP and PCT in predicting the risk of death were 0.864,0.723,0.755,0.904,0.853,0.628 and 0.726, respectively. The accuracy of CTSI, which was 0.478, was poor in predicting the risk of death and the AUG.3. Establishment of a multi-factor scoring system to predict infected pancreatic necrosis by multinomial Logistic regression and decision trees(1) There were 521 cases (73.6%) with non-pancreatic necrosis,127 cases with SNP (39.26%) and 60 cases with MAP (39.26%) in enrolled 708 patients of acute pancreatitis. The risk of death in patients with SPN was 9.45%, while it was 20% in patients with IPN. R×C Chi-square difference test between SPN and IPN group is significant (P<0.001).(2) The respiratory rate, temperature, smoking history, CTSI, NEU, Cr, and PCT variables were screened out for predicting IPN by multinomial Logistic regression analysis with OR of 5.898,2.563,4.464,3.243,3.116,2.83 and 5.456, respectively. A multi-factor scoring system was established based on multinomial Logistic regression model, while a cutoff value of 9 points had the sensitivity of 95.09% and the specificity of 79.8% in predicting IPN (AUC=0.947)(3) The CTSI, NEU, and PCT variables were screened out for predicting IPN by decision trees analysis, while CTSI cutoff was 3 points (chi-square=145.78, P 0.001), PCT cutoff was 0.5ng/ml (chi-square=29.325, P<0.001), neutrophil cutoff was 85%(chi-square=12.229, P=0.001). A multi-factor scoring system was established based on decision trees model, a cutoff value of 3 points had a sensitivity of 85.71% and a specificity of 69.14%(AUC=0.869)(4) Based on Verification Library, ROC curve analysis revealed that the Logistic multi-factor scoring system had superior accuracy for in predicting IPN (AUC=0.945). Both decision trees multi-factor scoring system and PCT had considerable accuracy (AUC=0.869 and 0.864, respectively), but was lower than the Logistic multi-factor scoring system. AUCs for CTSI and APACHE Ⅱ in predicting IPN were 0.756 and 0.69, respectively.Conclusion:(1) We successfully established the AP database by using Epi Info7 software. The database has the following feature:self-test error, automatic scoring, automatic diagnosis, data exchange and statistical analysis etc. The database is simple, friendly, low cost and helpful in clinical and research.(2) Database-based study found that multi-factor scoring system and biological indicators had no predictive value for MSAP, but had predictive value for SAP and the risk of death. APACHE Ⅱ was the most accurate scoring system in the early prediction of SAP and the risk of death. BUN and Cr have high accuracy for predicting the risk of death.(3) Two multi-factor scoring systems for early predicting IPN were established based on multinomial Logistic regression and decision trees analysis. The Logistic multi-factor scoring system has the highest accuracy in early predicting IPN. It has clinical value for guiding the use of antibiotics and intervention in patients with IPN. |
PDF Full Text Request