BACKGROUND Although the accurate estimation of tumor size is essential for proper patient selection for endoscopic resection in early gastric cancer (EGC), no study has been conducted to date on tumor size estimation. We aimed to evaluate the accuracy of endoscopic visual estimation of tumor size of EGC.

METHODS In 508 EGC patients that underwent endoscopic resection, endoscopic visual estimations were performed retrospectively by independent two endoscopists using still images. Data were compared with pathologic measurements as gold standard. Inter-observer agreement was determined using the Bland-Altman method and intra-class correlation coefficients (ICC). Measurement discrepancies were presented as differences between measurements.

RESULTS The ICC between the two endoscopists was 0.915 (95 % CI 0.900-0.928). Mean endoscopic estimates for both endoscopists were significantly lower than mean pathologic measurements (1.50 and 1.67 vs. 1.80 cm, P < 0.001). Absolute differences between average endoscopic estimates and pathologic measurements were found to be acceptable in most cases: an absolute difference of <0.4 cm was found for 80 % (404/508) of cases. Bland-Altman plot showed that 94 % of cases lay within the 95 % limits of agreement. Measurement discrepancy was proportional to tumor size and increased for an undifferentiated histology.

CONCLUSIONS Endoscopic visual estimations were found to show reliable agreement with pathologic measurement in EGC patients undergoing endoscopic resection, together with good inter-observer agreement. Further prospective study is needed to confirm the validity of this method.