Correspondence
Accuracy of CT Colonography for Colorectal Cancer Screening
N Engl J Med 2008; 359:2842-2844December 25, 2008
- Article
To the Editor:
Johnson et al. (Sept. 18 issue)1 discuss the use of computed tomographic (CT) colonography as a noninvasive option in screening for colorectal cancer, with optical colonoscopy and histologic review serving as the reference standard. Two important points merit discussion. First, sensitivities that are reported in the “per-polyp” analysis for lesions measuring 5 to 9 mm in diameter are deceiving since values are reported as “greater than or equal to” a certain size. With this method, a higher sensitivity for larger lesions falsely elevates the sensitivity for smaller ones. Using data abstracted from Table 3 and Table 4 in the article, we calculated the number of adenomas or cancers of each specified size (Table 1Table 1
Per-Polyp Analysis of Sensitivity of CT Colonography for the Detection of Adenomas and Cancers, According to a Specified Diameter.). For example, for the specified size of 5 mm, 32 of 104 known lesions were detected by CT colonography, yielding a sensitivity of 31%. Our analysis revealed a sensitivity of 46% (113 of 246) for adenomas measuring 5 to 9 mm and of 57% (81 of 142) for those measuring 6 to 9 mm in size in aggregate. This factor is important because the field is maturing to the point at which specific treatment is predicated on the size of individual lesions.2 The second critical point has to do with the generalizability of the study's results, since only 15 of 20 highly experienced radiologists were allowed to participate. If CT colonography were widely implemented, could the average radiologist participate?
2 ReferencesDon C. Rockey, M.D.
Samir Gupta, M.D.
University of Texas Southwestern Medical Center, Dallas, TX 75390
don.rockey@utsouthwestern. edu 1
Johnson CD ,Chen M-H ,Toledano AY , et al. Accuracy of CT colonography for detection of large adenomas and cancers. N Engl J Med 2008;359:1207-1217
Full Text | Web of Science | Medline2
Zalis ME ,Barish MA ,Choi JR , et al. CT colonography reporting and data system: a consensus proposal. Radiology 2005;236:3-9
CrossRef | Web of Science | Medline
To the Editor:
Johnson et al. may have overestimated the accuracy of CT colonography for detection of large adenomas and cancers for two reasons. First, they made no mention of the proportion of patients in whom colonoscopy to the cecum was documented. Second, they did not state the frequency of flat or depressed adenomas detected at colonoscopy. As noted by Fletcher in the accompanying editorial,1 such lesions may comprise up to about 10% of precancerous adenomas, which are recognized with increasing frequency by skilled endoscopists in the United States2,3 and are quite often missed by CT colonography.
3 ReferencesClaude Matuchansky, M.D.
Paris VII University, 75010 Paris, France
claude.matuchansky@wanadoo. fr 1
Fletcher RH . Colorectal cancer screening on stronger footing. N Engl J Med 2008;359:1285-1287
Full Text | Web of Science | Medline2
Soetikno RM ,Kaltenbach T ,Rouse RV , et al. Prevalence of nonpolypoid (flat and depressed) colorectal neoplasms in asymptomatic and symptomatic adults. JAMA 2008;299:1027-1035
CrossRef | Web of Science | Medline3
Lieberman D . Nonpolypoid colorectal neoplasia in the United States: the parachute is open. JAMA 2008;299:1068-1069
CrossRef | Web of Science | Medline
To the Editor:
The high estimates of sensitivity and specificity of CT colonography that are reported by Johnson et al. concern us for several reasons. Primarily, when assessing sensitivity and specificity, there are underlying assumptions of independence of test results and of measurement errors conditional on the true outcome. Although the tests were conducted separately by radiologists and endoscopists, since both tests are visual inspections of the colon, it is very likely that the measurement errors of both tests (when determining a continuous trait such as lesion diameter) are strongly correlated. This, in turn, can substantially inflate sensitivity and specificity.1 Moreover, the clustering of data for patients according to radiologist (for the CT colonography) or endoscopist (for the colonoscopy) has not been adequately addressed, thus resulting in possible underestimation of the variance.2 Because of these methodological issues, the estimates of sensitivity and specificity and their confidence intervals must be interpreted with caution.
2 ReferencesRinku Sutradhar, Ph.D.
Lawrence Paszat, M.D.
Linda Rabeneck, M.D.
Institute for Clinical Evaluative Sciences, Toronto, ON M4N 3M5, Canada
rinku.sutradhar@ices. on. ca 1
Brenner H . How independent are multiple `independent' diagnostic classifications? Stat Med 1996;15:1377-1386
CrossRef | Web of Science | Medline2
Beam CA . Analysis of clustered data in receiver operating characteristic studies. Stat Methods Med Res 1998;7:324-336
CrossRef | Medline
Author/Editor Response
With respect to the comments made by Rockey and Gupta regarding categories of polyp size for reporting sensitivity, we used the same size categories that had been used in another, similar, report published in the Journal.1 Participating radiologists came from 15 national sites, which were a mixture of small and large academic and private practices. Although 4 of these radiologists had previous experience with CT colonography, 11 of them participated in a training program, and all readers participated in a testing program conducted before the study began. Training and experience have been shown to improve the accuracy of interpreting CT colonography.2 The American College of Radiology has developed a national quality registry for CT colonography in which practitioners can enter their data and have them compared with those of other sites nationally.3 Participation in this database with standards of minimum performance should be a requirement for payment for this type of screening in the future. Such a requirement would be a powerful incentive for practices to engage radiologists who have this skill set.
In response to Matuchansky: 40 of 2600 participants who were enrolled in our study had incomplete or no colonoscopy data. Only 10 of 2531 participants did not have colonoscopy data documented to the cecum because of previous resection. A total of 2.4% of patients were classified as having a flat lesion (i.e., in which the ratio of height to width was ≤50%), representing 2.9% of adenomas.
In response to Sutradhar et al.: optical colonoscopy was the reference standard, and CT colonography was the only diagnostic test evaluated. The article by Brenner4 cited by the correspondents addresses multiple diagnostic tests used together, whereas our study had one diagnostic test and one reference standard. We used the data from CT colonography and the reference standard to compute sensitivity and specificity in the ordinary way, as have most other investigators measuring the performance of CT colonography. With respect to the clustering of data according to radiologist, the variances from both participants and radiologists were taken into account when we estimated the variance of the estimated sensitivity across radiologists.
4 ReferencesC. Dan Johnson, M.D.
Mayo Clinic Arizona, Scottsdale, AZ 85259Mei-Hsiu Chen, Ph.D.
Brown University, Providence, RI 02912Alicia Y. Toledano, Sc.D.
Biostatistics Consulting, Providence, RI 029061
Pickhardt PJ ,Choi JR ,Hwang I , et al. Computed tomographic virtual colonoscopy to screen for colorectal neoplasia in asymptomatic adults. N Engl J Med 2003;349:2191-2200
Full Text | Web of Science | Medline2
Fletcher JG, Johnson CD, Toledano A, et al. ACRIN 6664: lessons for CT colonography (CTC) training and certification (abstract). In: Proceedings of the Radiological Society of North America 91st Scientific Assembly and Annual Meeting Program, Chicago, November 27–December 2, 2005.
3
American College of Radiology. CT colonography registry. (Accessed December 4, 2008, at https://nrdr.acr.org/portal/CTC/Main/page.aspx.)
4
Brenner H . How independent are multiple `independent' diagnostic classifications? Stat Med 1996;15:1377-1386
CrossRef | Web of Science | Medline
- Citing Articles (1)
Citing Articles
1
D. Boone, S. Halligan, R. Frost, C. Kay, A. Laghi, P. Lefere, E. Neri, J. Stoker, S.A. Taylor. (2011) CT colonography: Who attends training? A survey of participants at educational workshops. Clinical Radiology 66:6, 510-516
CrossRef
