Background: Within the framework of a chemoprevention trial on stomach cancer, two substudies based on repeat measurement were undertaken to evaluate reliability of histological diagnoses of gastric precancerous lesions.
Methods: A subgroup of 45 subjects received two endoscopies separated by a period of one month. The two biopsies were reviewed by a single pathologist. A second subsample of 50 subjects had a single endoscopy and the biopsy results were reviewed by two pathologists. Agreement between the two diagnoses was assessed by Cohen's Kappa and by repeat frequency.
Results: When the same samples were reviewed by the pathologists involved in the trial, agreement was very high for advanced lesions (repeat frequency = 0.96 for intestinal metaplasia and 1.00 for dysplasia) but lower for less advanced lesions (repeat frequency = 0.73 for superficial gastritis and chronic gastritis, 0.65 for atrophic gastritis). When the same pathologist reviewed two sets of biopsies taken less than 2 months apart, the combination of random observer error and biopsy sampling error gave rise to quite low agreement, especially for early lesions, mainly attributable to biopsy sampling error. Comparison of diagnoses made at routine reading and at review by the same pathologist and by different pathologists showed substantial overall agreement with the exception of one pathologist for whom agreement was moderate.
Conclusions: These results confirm that misclassification of histological diagnosis may be a relevant problem in chemoprevention trials of stomach cancer, more so when baseline diagnosis is taken into account in the analysis to estimate progression and regression rates of precancerous lesions. Further, the results suggest that misclassification is limited to early lesions, while diagnostic reliability of severe lesions is quite high.