Aim: To determine the accuracy of computed tomography colography (virtual colonoscopy) in detecting colorectal polyps and colorectal cancer.
Design: Blinded comparison of virtual colonoscopy (initially supine-only scans and later supine plus prone scans) with the criterion standard of conventional colonoscopy.
Subjects and setting: 100 patients aged 55 years or over referred to a public teaching hospital for colonoscopy, July 1997 to January 2000, because of colonic symptoms or a family history of bowel cancer.
Main outcome measures: Presence and size of polyps and other lesions; certainty of polyp identification on virtual colonoscopy (on 100-point visual analogue scale); sensitivity and predictive values of virtual colonoscopy.
Results: Conventional colonoscopy identifed 121 polyps in 47 patients; 28 of these polyps, in 19 patients, were identified by virtual colonoscopy. Sensitivity of virtual colonoscopy for detecting polyps (using supine plus prone scans) was 73% for polyps with diameter > or = 10 mm (95% CI, 39%-94%) and 19% for smaller polyps (95% CI, 10%-31%) (P < 0.001); corresponding figures for supine-only scans were 57% (95% CI, 18%-90%) and 11% (95% CI, 4%-24%), respectively. Ten polyps identified at virtual colonoscopy were considered false-positive findings (8%). The value of finding a polyp on virtual colonoscopy (with thresholds of 5 mm for diameter and 30 points for certainty score) was assessed as a predictor of finding a polyp (diameter > 5 mm) on conventional colonoscopy. Positive and negative predictive values were 88% and 89%, respectively, for supine plus prone scans.
Conclusion: Although virtual colonoscopy shows potential as a diagnostic tool for colorectal neoplasia, it is currently not sufficiently sensitive for widespread use.