br to to to and to were
.20, .21 to .40, .41 to .60, .61 to .80, and .81 to 1.00 were defined as slight agreement, fair agreement, moderate agreement, substantial agreement, and almost perfect agreement, respectively. All analyses were conducted using the statistical package R version 3.3.3 (R Project for Statistical Computing, Vienna, Austria) and GraphPad Prism 6 (GraphPad Software, Inc, San Diego, Calif).13
Participant demographics are presented in Table 2. Two hundred six participants (56%) had been registered as board-certified fellows of the Japan Gastroenterological Endoscopy Society.
Accuracy and AUC/ROC comparison between tests 1 and 3
Figure 3 shows the mean accuracy of tests 1 and 3 in 365 participants. The mean accuracy of test 3 was significantly higher than that DAPT of test 1 (.68 and .61, respectively; P <
.01). Table 3 shows the AUC/ROC result of tests 1 and 3 in the study subjects. The cNRI analysis of all lesions
showed significant improvement of reclassification when participants underwent e-learning. The IDI analysis demonstrated that the e-learning system improved the diagnostic ability. According to the analysis, depending on the lesion’s characteristics, high AUC/ROC was demonstrated in depressed and small lesions (<10 mm). The cNRI analysis revealed remarkable e-learning improvement in both depressed and small lesions. In contrast, no significant e-learning improvement was observed in elevated and flat lesions.
The e-learning group underwent e-learning between tests 1 and 2. Meanwhile, the non–e-learning group under-went e-learning between tests 2 and 3. There was a time difference between the 2 groups taking e-learning until test 3 was implemented. The average number of days from the e-learning study to test 3 was 82.6 (standard de-viation, 5.9) in the e-learning group and 27.1 (standard de-viation, 20.2) in the non–e-learning group, which was significantly longer in the e-learning group. Although there was a difference in the time when e-learning was taken, as described above, there was no statistically significant differ-ence in accuracy in test 3 of both groups (e-learning group, 67.4%; non–e-learning group, 68.4%).
Table 4 shows the results of comparison of the AUC/ROC of DL, MV, and MS evaluated by M-NBI between tests 1 and
Table 5 shows interobserver agreement of 365 participants in tests 1 and 3. Fleiss’ kappa value analyses of all lesions were slightly improved after e-learning (from .27 to .34).
According to the analysis, depending on the characteris-tics of the lesions, a low kappa value was demonstrated in flat and inflammatory lesions in test 1. The interpretations of these lesions were improved into fair agreement after e-learning.
In endoscopic diagnosis of gastric epithelial lesions, dif-ferentiation between cancerous and noncancerous lesions is important. Regarding the qualitative diagnosis of gastric epithelial tumor, the pathologic evaluation by biopsy sam-pling is the criterion standard. In recent years the number of patients taking antithrombotic medication has been increasing. Moreover, increasing medical cost is a problem
4 GASTROINTESTINAL ENDOSCOPY Volume -, No. - : 2019 www.giejournal.org
Ikehara et al Learning effect after e-learning training in endoscopic diagnosis
Figure 2. Examples of the representative magnifying endoscopy with narrow-band images used in the tests. A, Example of a cancerous lesion. Based on the vessel-plus-surface classification system, Buoyant density lesion had an irregular microvascular pattern and an irregular microsurface pattern with a demarcation line. B, Example of a noncancerous lesion. Based on the vessel-plus-surface classification system, this lesion had a regular microvascular pattern and a regular microsurface pattern with a demarcation line.
TABLE 2. Characteristics of the 365 participants
Accredited by JGES
No. of endoscopies conducted
Use of M-NBI in screening endoscopy
No. of cases for which M-NBI was used for differential
No. of cases for which M-NBI was used for diagnosis of
Values are n (%) unless otherwise defined.
SD, Standard deviation; JGES, Japan Gastroenterological Endoscopy Society; M-NBI, magnifying endoscopy with narrow-band imaging.
that cannot be ignored. Therefore, the number of biopsy specimens should be kept to a minimum. It has been shown that diagnostic performance for gastric lesions is improved by M-NBI diagnosis7 and could lessen the biopsy samples needed to diagnose each cancer.14
Furthermore, establishing an efficient learning system to widely educate the large number of endoscopists on the detailed process of M-NBI is important. We have previously reported that the e-learning system on the web is effective for improving the diagnostic performance of endoscopists.8 In our previous study, e-learning was shown to improve the accuracy of cancer and noncancer discrimination by M-NBI.8 However, whether the learning effect by e-learning is different depending on ordinary endoscopic findings such as macroscopic type and lesion size has not been examined. Hence, the feature of the lesion with poor learning effect by e-learning was not clarified.