Application of a biochemical and clinical model to predict individual survival in patients with end-stage liver disease

Table 3 Contributions of different variables to survival prediction at 104 wk, results from multivariate Cox regression models¹

Variable	Variable χ2	Regression coefficient	Hazard ratio	95% CI for Hazard ratio		P value	c-statistic	BIC
Ascites	53.90	2.310	10.2	3.78	28.1	< 0.0001	0.76	2014.15
+ Ln (creatinine)	63.43	1.370	3.99	1.57	10.9	0.006	0.83	1988.15
+ BEV	65.71	1.195	3.25	1.01	9.77	0.048	0.85	1970.65
+ HE	68.91	0.909	2.50	0.915	6.88	0.070	0.89	1961.89
+ Ln (bilirubin)2	70.11	0.349	1.46	0.66	3.33	0.427	0.90	1951.77

¹Estimated from Cox proportional hazards models.

²Biochemical (bilirubin and creatinine) and Clinical (ascites, encephalopathy and bleeding esophageal varices) Model; BEV: Bleeding esophageal varices; HE: Hepatic encephalopathy; BIC: Bayesian Information Criterion. Ln was used to normalize distributions and improve the fit for individual predictors. Hazard ratio for quantitative variables are expressed for 1 relevant unit increase of log. + indicates the addition of each variable separately to the model with ascites only. χ² is the likelihood ratio statistic for each group of variables when added to the model. The risk prediction was based on data from the model derivation cohort (n = 170) at 104 wk follow-up.