Fifteen single-centre trials (719 patients; 2,290 specimens) were included in the review. The average QUADAS score was 11.2; all trials avoided differential verification bias and incorporation bias, and all of them blinded assessors to the reference standard and index test results.
The 15 trials showed that confocal endomicroscopy had high sensitivity (94%, 95% CI 88 to 97) and specificity (95%, 95% CI 89 to 97) for discriminating between colorectal neoplasms and non-neoplasms. The area under the curve was 0.97. There was no evidence of a threshold effect using the Spearman coefficient.
Subgroup analyses showed that real-time endomicroscopy sensitivity (96%) and specificity (97%) were significantly higher than video assessment (sensitivity 85%, specificity 82%, p<0.001). This was confirmed by the summary receiver operating characteristic curve.
The results from other subgroup and sensitivity analyses were reported. The authors stated that heterogeneity was relatively high across the trials, but some subgroup analyses had lower levels.
Four randomised controlled trials compared confocal endomicroscopy with chromoendoscopy: sensitivity and specificity were similar for comparisons between magnifying or virtual chromoendoscopy and real-time or video confocal endomicroscopy.
Adverse events were briefly discussed. There was no evidence of publication bias in the funnel plot.