Results of a recent study found significantly lower detection rates of clinically significant findings in different capsule endoscopy reading modes and speeds than previously reported.
Researchers observed 17 physicians with varying levels of interpretation experience (from 23 to >1,000 capsule procedures) to determine detection rates in capsule endoscopy readings. Each capsule endoscopist read 24 clips, 18 of which included abnormal lesions, on two different reading modes and using two speeds. The modes and speeds were SingleView at 15 frames per second (fps), SingleView 25 fps, QuadView 20 fps, and QuadView 30 fps. Pathology detection rates correlated with reading mode, lesion type, reader experience, and timing order were examined.
Researchers said three of the reading modes had similar, but not high, detection rates: SingleView 15 (45%), QuadView 20 (47%), and QuadView 30 (43%). SingleView 25 resulted in the lowest detection rate (26%). Of the four lesion types present, angioectasias had the highest detection rate (69%), followed by masses/polyps (46%), ulcers/erosions (38%), and blood (17%). Angioectasias had a significantly higher detection rate than ulcers/erosions (P=.037) and blood (P=.004).
“The overall detection rates for the different reading modes were alarmingly low,” researchers said, “ … and not influenced by increasing experience.”
The highest detection score of 78% was attained by two participants, each of whom had been in practice less than five years and had read less than 300 capsules. The lowest score (17%) was recorded by an endoscopist who had read more than 300 capsules.
The researchers said interpretation practices for images generated by capsule endoscopies, which typically include 40,000 to 60,000 frames for one study, are not standardized. Capsule endoscopy technology has limitations, in part because there is “significant interobserver variability for interpretation.”
“Standardizing reading methods for capsule endoscopy is likely to result in improved overall detection rates and improved reader agreement,” researchers concluded.