Validation of prognostic models in primary biliary cirrhosis

Prognostic models in primary biliary cirrhosis have been validated for large population groups but the predictive value for individual patients has not been tested. We used data from ten deceased patients with primary biliary cirrhosis to test three prognostic models: the Shapiro model (bilirubin); the Christensen model (age, bilirubin, albumin, presence of cirrhosis or cholestasis, azathioprine treatment); and the Dickson model (age, bilirubin, albumin, prothrombin time, oedema). The predictive value of each model for individual patients was determined by assessing whether it would have accurately predicted appropriate timing of liver transplantation in patients prior to death. The Dickson model predicted that four of nine cases would have been considered for liver transplantation one year before death and one of seven cases two years before death. The Christensen model predicted that this procedure would have been considered in three of seven cases two years before death. The Shapiro model was demonstrated to be the least predictive of the three tested. Although none of the three models assessed was found to accurately predict survival, no model predicted a worse survival than actually occurred. Liver transplantation is indicated in those cases with a poor predicted survival.