Establishment and evaluation of a predictive model for probability of death in patients with hepatitis B-related cirrhosis

Objective To establish and evaluate a mathematical model for predicating the probability of death for patients with hepatitis B-related cirrhosis and to provide references for selecting reasonable therapeutic regimen.Methods Data on clinical manifestation and follow-up survey of 1 386 patients with hepatitis B-related cirrhosis were collected and analyzed retrospectively.Logistic regression analysis was adopted to construct a multiple regression equation for predicting the mortality probability of the patients after hospitalization treatment.Receiver operation curve(ROC),Child-Turcotte-Pugh(CTP),and the score based on the model for end-stage liver disease(MELD)were used to evaluate the efficiency of the prediction model established.Results Seven clinical indicators of the patients(international normalized ratioINR,serum sodium contentNa,total bilirubinTBIL,creatinineCRE,spleen thickness,upper gastrointestinal bleeding,and hepatic encephalopathy)were included in the final multivariate regression equation.The established equation presents the sensitivity of 91.5%,specificity of 92.4%,correct index of 0.839,positive predictive value of 84.64%,negative predicative value of 96.0%,positive likelihood ratio of 12.04,and negative likelihood ratio of 0.09,respectively.For the utilization of CTP score,MELD score,and established regression model,the values of area under the ROC curve(95%confidence interval)were 0.786(0.762-0.789),0.825(0.794-0.864),and 0.912(0.875-0.931)for the prediction of death probability of the patients at the 3rd month of hospitalization,with significant difference among the three methods(Z=2.16,P=0.015).Conclusion A logistic equation was established with good sensitivity,specificity,and correct index for the prediction of death probability of patients with hepatitis B-related cirrhosis.