Can we improve the detection rate and interobserver agreement in capsule endoscopy?

Background: Data about strategies for improving the diagnostic ability of capsule endoscopy readers are lacking. Aim: (1) To evaluate the detection rate and the interobserver agreement among readers with different experience; (2) to verify the impact of a specific training (hands-on training plus expert tutorial) on these parameters. Methods: 17 readers reviewed 12 videos twice; between the two readings they underwent the training. The identified small bowel findings were described by a simplified version of Structured Terminology and classifies as clinically significant/non-significant. Findings identified by the readers were compared with those identified by three experts (Reference Standard). Results: The Reference Standard identified 26 clinically significant findings. The mean detection rate of overall readers for significant findings was low (about 50%) and did not change after the training (46.2% and 46.4%, respectively). There was no difference in the detection rate among readers with different experience. The interobserver agreement with the Reference Standard in describing significant findings was moderate (k= 0.44; CI95%: 0.39-0.50) and did not change after the training (k= 0.44; CI95%: 0.38-0.49) or stratifying readers according to their experience. Conclusions: Both the interobserver agreement and the detection rate of significant findings are low, regardless of the readers' experience. Our training did not significantly increase the performance of readers with different experience.