Six studies (n=1,277 patients, range 57 to 700) were included in the review; two studies were randomised trials, one was a prospective non-randomised trial and three were prospective cohorts. All studies had an overall QUADAS score of 11 or 12. No study provided clear reporting of uninterpretable test results or withdrawals. Half of the included studies did not report sufficient detail of the reference standard to permit replication. One study did not provide a clear description of selection criteria. The remaining QUADAS criteria were met by all studies.
The pooled incremental yield for the detection of dysplasia for chromoendoscopy over white light endoscopy was 7% (95% CI 3.2 to 11.3; six studies; n=1,277 patients).
The difference in the proportion of all dysplastic lesions detected by targeted biopsies was 44% (95% CI 28.6 to 59.1; six studies; n=1,277 patients) in favour of chromoendoscopy. The difference in proportion of flat lesions detected was 27% (95% CI 11.2 to 41.9; four studies; n=1,118 patients) in favour of chromoendoscopy.
Chromoendoscopy reduced the lesions detected by random biopsy by -40% (95% CI -52.8 to -27; four studies; n=1,075 patients) compared with white light endoscopy.
The pooled weighted difference in procedure time between chromoendoscopy and white light endoscopy was 11 minutes (95% CI 10 minutes 15 seconds to 11 minutes 43 seconds; four studies; n=1,120 patients),.
For detection of all dysplastic lesions, meta-regression indicated that the use of magnification endoscopes, study design (patients acting as their own controls versus two separate groups undergoing chromoendoscopy and white light endoscopy), and pancolonic or targeted dye spraying in chromoendoscopy were significant predictors of yield. For detection of any dysplasia using tagetted biopsies, meta-regression indicated that the dye used (methylene blue or indigo carmine), the number of endoscopists (single or multiple), study design (patients acting as their own controls versus two separate groups undergoing chromoendoscopy and white light endoscopy), and QUADAS score (≥12 or <12) were significant predictors of yield.
Sensitivity analyses did not significantly alter results.
There was no evidence of publication bias.