Observer Variation in Assessment of Liver Biopsies of Alcoholic Patients

Publisher Website
Google Scholar
Abstract
The aim was to construct a questionnaire analyzing pathological features possibly present in alcoholic liver disease, to assess its interobserver variation and to determine the influence of technical data on this variation. A total of 764 inpatients drinking 90 g (median) of pure alcohol per day for 25 years was observed; 402 patients were excluded because of associated nonalcoholic disease, refusal or contraindication to biopsy, leaving 362 patients included. Two pathologists independently analyzed each biopsy and completed a questionnaire including 41 items. Coefficient of concordance between observers was evaluated with the kappa statistic (κ). The prevalence of 14 lesions was low, equal to or under 10%, leading to a nonsignificant concordance. For the 27 remaining features, two had an almost perfect degree of concordance (κ > 0.81): presence of hepatocellular carcinoma and cirrhosis. Three had a substantial coefficient of concordance (κ > 0.61): fibrous septa, size of cirrhotic nodules, and liver cell regeneration. Nine had a moderate (κ > 0.41), 11 a fair (κ > 0.21), and two a slight (κ < 0.21) coefficient of concordance. In terms of final diagnosis of alcoholic liver disease the concordance was substantial for cirrhosis with acute alcoholic hepatitis (κ= 0.77), cirrhosis without alcoholic hepatitis (κ= 0.75), acute alcoholic hepatitis without cirrhosis (κ= 0.65) and normal liver (κ= 0.64). Concordance was moderate for steatosis (κ= 0.47) and slight for fibrosis alone (κ= 0.16). For technical data, there was a trend for a higher concordance for the plastic‐embedded biopsies than for the paraffin‐embedded (p < 0.08) and for the samples containing more than six portal tracts than for the smaller biopsy samples (p < 0.10). In conclusion interobserver variations suggest that assessment of liver biopsies in alcoholic liver disease has to be carefully performed, in duplicate using clear definitions and a simplified questionnaire, for clinical research.