Interobserver agreement in the physical diagnosis of alcoholic liver disease

Agreement among six physicians for 18 clinical signs in 50 alcoholic patients was prospectively studied. Twenty patients had alcoholic cirrhosis, 14 noncirrhotic alcoholic liver disease, and 16 alcoholics had no clinical or biochemical abnormalities. Agreement was assessed by kappa index for categorical variables and by intraclass correlation coefficient for the others. A good agreement was observed for ascites (r=0.75) and splenomegaly (r=0.75). It was fair for jaundice (r=0.65), Dupuytren's contracture (r=0.65),and vascular spiders (r=0.64). However, it was poor for white nails (r=0.27) and hepatic consistency (r=0.11). Agreement was better among senior physician's than junior physicians. In order to assess which signs contributed to differentiate the three groups of patients, a stepwise discriminant analysis was realized; it identified three variables: vascular spiders (P<0.001), splenomegaly (P<0.001), and abdominal wall collateral veins (P<0.01). These results suggest that studies based on physical findings must be cautiously considered.