Updating prognosis and therapeutic effect evaluation in cirrhosis with Cox's multiple regression model for time-dependent variables.

A multivariate Cox regression analysis with time-dependent variables has been performed on the data of 415 patients with cirrhosis included in a controlled clinical trial of 10-15 mg prednisone daily versus placebo. The analysis showed that a poor prognosis was associated with a low prothrombin index, marked ascites, GI bleeding, high age, high daily alcohol consumption, high bilirubin and alkaline phosphatase and low albumin values, little liver connective tissue inflammation, and poor nutritional status. Prothrombin index and ascites showed significant interaction with the treatment in such a manner that high prothrombin index and absence of ascites were associated with a beneficial effect of prednisone, whereas low prothrombin index and presence of ascites were associated with a harmful effect of prednisone treatment. The final model was validated in independent patients by comparing their actual survival with that predicted from the model, using a split-sample testing technique. The prognostic factors were combined with an index that can be used to update prognosis whenever changes occur in the clinical status of a patient during the course of the disease. The probability of surviving the next 3 or 6 months can be estimated from the prognostic index at any time during the course. The index may be of value for the correct timing of special therapeutic procedures such as liver transplantation.