OC.01.3: A TRAINING STRATEGY TO IMPROVE DETECTION RATE AND INTEROBSERVER AGREEMENT IN CAPSULE ENDOSCOPY (CE) FOR READERS WITH DIFFERENT EXPERIENCE: RESULTS OF A MULTICENTRE STUDY

Capsule Endoscopy (CE) is the cornerstone for the evaluation of the small bowel. The efficacy of this technique depends on the ability of the readers in recognizing and reporting mucosal abnormalities. At present data about possible strategies for improving the diagnostic ability of readers in reviewing CE videos are lacking. to verify whether a training, similar to that proposed by the ASGE for credentialing and granting privileges for CE, is effective in improving both detection rate and interobserver agreement among CE readers with different experience in reading CE videos. 17 readers have been involved in the study. Of them, 6 had previously evaluated less than 50 CE videos, six 50-99 videos and 5 more than 100 videos. For the study all readers reviewed 30 CE videos, divided into two sets of 15 each, reporting small bowel findings (SBF) by means of the CE Structured Terminology. These videos were also reviewed by 3 experts, who represented the reference standard (RS) and categorized SBF as “clinically significant” (class A) or “not clinically significant” (class B). After reviewing all 30 videos, during an 8 hour meeting, the SBF identified by the readers in the first set of 15 videos were collectively discussed and compared with those identified by the RS (training session-TS). After the TS the readers reviewed again the second set of videos. The number of findings identified before and after the training session were compared by means of Wilcoxon test for paired and unpaired data; the evaluation of the agreement between each reader and the RS was performed by means of Cohen's kappa statistic. In the second set of 15 videos the RS identified 89 SBF. Before the TS each reader identified a mean±SD of 74±45 SBF; a mean+SD of 35+11 of these SBF matched those identified by the RS. There was no difference among readers with different experience (p=0.9). After the TS both the overall number of SBF per reader (mean±SD: 85±47) and the mean number of SBF matching those identified by the RS (mean±SD: 38±12) slightly increased, although not significantly (p=0.1 and p=0.2 respectively). This figure remains the same even after stratifying readers according the experience or taking into account only class A findings. The overall agreement with the RS in describing SBF was low and remains closely similar before and after the TS; (k=0.14; CI 95%: 0.12-0.16 and k=0.15; CI 95%: 0.12-0.17 respectively). Moreover no difference was found even after stratifying readers according the experience or taking into account only class A findings (before: k=0.33; after k=0.34). a training strategy combining hands-on practice and experts' tutorial, does not improve the detection rate or the interobserver agreement in describing SBF identified during CE, regardless of the previous experience in reading CE videos.