Pre- and post-training session evaluation for interobserver agreement and diagnostic accuracy of probe-based confocal...

Current diagnostic modalities for indeterminate biliary strictures offer low accuracy. Probe-based confocal laser endomicroscopy (pCLE) permits microscopic assessment of mucosal structures by obtaining real-time high-resolution images of the mucosal layers of the gastrointestinal tract. Previously, an interobserver study demonstrated poor to fair agreement even among experienced confocal endomicroscopy operators. Our objective was to assess interobserver agreement and diagnostic accuracy upon completion of a pCLE training session.

Forty de-identified pCLE video clips of indeterminate biliary strictures were sent to five endoscopists at four tertiary care centers for scoring. Observers subsequently attended a teaching session by an expert pCLE user that included 20 training clips and rescored the same pCLE video clips, which were randomized and renumbered.

Pre-training interobserver agreement for all observers was 'fair' (Κ: 0.31, P-value: <0.0001) and diagnostic accuracy was 72% (55-80%). Post-training interobserver agreement for all observers was 'substantial' (Κ: 0.74, P-value: <0.0001) and diagnostic accuracy was 89% (80-95%). Using a paired t-test, we observed an increase of 17% (95% CI 7.6-26.4) in post-training diagnostic accuracy (t = 5.01, df = 4, P-value 0.007).

Interobserver agreement and diagnostic accuracy improved after observers underwent training by an expert pCLE user with a specific sequence set. Users should participate in such training programs to maximize diagnostic accuracy of pCLE evaluation.