Prediction Models for the Decompensation in Patients with Cirrhosis

J Health Info Stat > Volume 39(2); 2014 > Article

J Health Info Stat 2014;39(2):57-64.

간경변증환자의 비대상성 간경변증으로의 진행 예측모형

곽민정

Prediction Models for the Decompensation in Patients with Cirrhosis

Min Jung Kwak

ABSTRACT

Objectives:
This study is intended to find the significant prognostic factors for the prediction of hepatic decompensation with cirrhosis. Also, this study provides the proper cut off value for significant factors.
Methods:
232 patients with cirrhosis were investigated retrospectively from 1996 to 2010. Logistic regression and odds ratio estimates are used to find the most significant factors and the effects of those factors for the decompensation with cirrhosis. The decision tree model is adopted to find the proper cut off point for the most significant factor. The Cox’s proportional hazard regression model is used to consider the time to decompensation with prognostic factors.
Results:
The result of logistic regression shows that H/L ratio and prothrombin time are significant measures for decompensation. Moreover, H/L ratio is the most significant factor with AUROC 0.84 with odd ratio 29.94 in simple logistic regression. Also, Cox’s proportional hazard regression model using progression time to decompensation supports this result. ALT, Prothrombin time, H/L ratio and AST/ALT ratio show significant results and H/L ratio is the most significant prognostic factor in survival analysis. From decision tree model, 71.0% of the patients with H/L ratio above 0.5 are progressed to decompensation. At cut off value 0.5, sensitivity and specificity are 75.9% and 81.4%, respectively.
Conclusions:
Both the results of logistic regression and survival analysis show that H/L ratio is an important parameter to predict the progression to a decompenation state.

Key words: Decompensation, Cirrhosis, H/L ratio, Logistic regression, Survival analysis