Value Of Self-built Model And Classical Models In Predicting Prognosis Of Patients With Liver Cirrhosis And Rebleeding Risk Of Patients With Esophageal Varices

Background and objective: There is a high incidence of liver disease in China, and there is a high morbidity in patients with liver cirrhosis. Currently, evaluations of prognosis on patients with liver cirrhosis are mostly based on Child-Pugh classification in our country, but some inherent problems of this system limit its clinical application. The predictive value of MELD model in patients with liver cirrhosis has been accepted in recent years by scholar, but MELD model still has some shortcomings. Logistic regression analysis is a method which studies relationship between binary classification observations and some influencing factors. Its main purposes are: First, to look for risk factors; Second, to predict the occurrence of a disease or how much the probability of a situation according to logistic regression model; Third, to estimate probability of someone belongs to a particular disease or situation. In this study, the factors that affect prognosis of patients with decompensated liver cirrhosis were selected by univariate and multivariate logistic regression analysis, and the logistic model (self-built model) was created. And then value of self-built model and classical models (MELD model,MELD-Na model and CTP) in predicting prognosis of patients with decompensated liver cirrhosis and rebleeding risk of patients with esophageal variceal bleeding at 3-month and 1-year were compared.Methods: From January 2003 to December 2008, a total of 1086 hospitalized patients with decompensated liver cirrhosis in our hospital, including 365 patients with esophageal variceal bleeding, were retrospectively evaluated and followed up at least for 1-year in this study. MELD-Na score,MELD score,CTP score were calculated on entry. According to the patient's survival results in 3 months, the factors that affect prognosis of patients with decompensated liver cirrhosis in 3 months were selected by univariate and multivariate logistic regression analysis and self-built model was created. Receiver operating characteristics curves (ROC) and the area under ROC curves (AUC) were used to determine the ability of four models for predicting prognosis and rebleeding risk in 3-month and 1-year, and Z-test was used to compare their predictive ability. Results:1. Compared with normal serum sodium group, hyponatremia group has a higher CTP score,MELD score,MELD-Na score,mortality in 3 months and 1 year and incidence of hepatic encephalopathy(P<0.05);2. Compared with survival group,death group has a higher CTP score,MELD score,MELD-Na score, and has a lower serum sodium(P<0.05);3. The significant positive variables for death outcome were ALT,TBIL,BUN,Cr,INR,serum sodium and hepatic encephalopathy by multivariate Logistic regression analysis in patients with decompensated cirrhosis;4. The AUC of four models for predicting prognosis in patients with decompensated cirrhosis in 3 months and 1 year are over 0.8, indicating that four models have good predictive value for patients with decompensated liver cirrhosis in 3 months and 1 year, and self-built model has the highest AUC (0.906); the self-built model is better than MELD model and CTP in predicting prognosis for patients with decompengsated liver cirrhosis in 3 months (P<0.05). There are no statistics difference between self-built model and MELD-Na model (P>0.05), but the AUC of self-built model is large than that of MELD-Na model. The self-built model is better than CTP in predicting prognosis for patients with decompengsated liver cirrhosis in 1 year(P<0.05), and there are no statistics difference among self-built model,MELD-Na model and MELD model(P>0.05),but the AUC of self-built model is larger than that of MELD-Na model and MELD model;5. The AUC of self-built model,MELD-Na model and MELD model for predicting prognosis in patients with decompensated liver cirrhosis and esophageal variceal bleeding in 3 months and 1 year are over 0.7, predictive value of self-built model is better than that of CTP(P<0.05), but there are no statistics difference among self-built model,MELD-Na model and MELD model (P>0.05);6. The AUC of self-built model,MELD-Na model and MELD model for predicting rebleeding risk in patients with decompensated liver cirrhosis and esophageal variceal bleeding in 3 months and 1 year are over 0.75, predictive value of self-built model is better than that of CTP(P<0.05), but there are no statistics difference among self-built model,MELD-Na model and MELD model(P>0.05). Conclusion:1. Four models can exactly predict the prognosis of patients with decompensated cirrhosis in 3 months and 1 year. Self-built model and MELD-Na model are better than CTP, and self-built model has the largest AUC.2. Self-built model,MELD-Na model and MELD model can exactly predict the prognosis of patients with decompensated cirrhosis and esophageal variceal bleeding in 3 months and 1 year, and self-built model and MELD-Na model are better than CTP. There are no statistics difference among self-built model,MELD-Na model and MELD model; the AUC of CTP is under 0.7, indicating that its predictive value is poor for patients with decompensated cirrhosis and esophageal variceal bleeding.3. Self-built model,MELD-Na model and MELD model can exactly predict rebleeding risk of patients with decompensated cirrhosis and esophageal variceal bleeding in 3 months and 1 year. Self-built model and MELD-Na model are better than CTP, but there are no statistics difference among self-built model,MELD-Na model and MELD model; the predictive value of CTP is poor for these patients.