Aims According to the classification established by the Japanese Society for Oesophageal Disease, early oesophageal cancer can be subdivided into six successive layers of the mucosa or submucosa, which influences the treatment strategy and prognosis of the individual patient. However, the reproducibility of this classification in terms of inter- and intraobserver variability is unclear.

Methods Histological slides from 105 surgical resection specimens of patients who had undergone oesophagectomy for early oesophageal adenocarcinoma were reviewed independently by three gastrointestinal pathologists, and were classified according to the Japanese criteria (m1/m2/m3/sm1/sm2/sm3 tumours). Inter- and intraobserver variation was determined by κ-statistics.

Results The interobserver reproducibility was good between pathologist 1 and 2 (κ=0.61, 95% CI 0.55 to 0.67), and moderate between pathologist 1 and 3 (κ=0.51, 95% CI 0.45 to 0.57) and between pathologist 2 and 3 (κ=0.50, 95% CI 0.38 to 0.61). The intraobserver agreement as assessed by the expert pathologist was good (κ=0.76), with a 95% CI that was interpreted as good to very good (0.67 to 0.85). Most agreement was achieved at the lower (m1) and upper site (sm2, sm3) of the spectrum, whereas the m2 tumours reflected the most discrepant stage. The majority of the observed discrepancy included the variation in one substage only.

Conclusions The reproducibility of the Japanese classification is good in terms of inter- and intraobserver variability when grading early oesophageal adenocarcinoma on surgical resection specimens. The present data confirm that dedicated gastrointestinal pathologists with broad experience are preferred when grading the resection specimens of patients with early oesophageal adenocarcinoma.