Non-invasive diagnosis of liver cirrhosis: a comparison between four discriminant functions

We assessed the performance of 4 methods of discriminant analysis using as independent variables the age and 16 serum tests, for correctly identifying patients with liver cirrhosis among hospitalized patients affected by chronic liver disease without signs of liver failure; 290 patients entered this study: on the basis of laparoscopy with or without liver biopsy, 152 patients had a diagnosis of liver cirrhosis and 138 were classified as chronic hepatitic patients. Due to the non-multinormal distribution of the variables used and to the unequality of the variance-covariance matrices, we compared the following 4 methods: linear discriminant function, quadratic discriminant function, non-parametric discriminant function and logistic regression. The Receiver Operating Characteristic (ROC) analysis was used to compare diagnostic ability of the assessed methods: the quadratic discriminant function was the best performing method. The predictive ability of this function was compared to that reported for percutaneous liver biopsy, showing that this simple statistical method using age and biochemical tests can efficiently identify liver cirrhosis in the setting of chronic liver disease, reducing the need for invasive diagnostic procedures.