Our results indicate that size estimation of polyps found at CT colonography varies from other measures of polyp size. Furthermore, analyses of polyps stratified by categories of proposed clinical importance demonstrates that classification of polyps into incorrect polyp reference categories (even when estimates fall within 95% limits of agreement) may occur.
Prior work regarding the reliability of colonoscopic, as well as radiologic, measurements of polyp size merits consideration and is consistent with our findings. One study, examining polyp measurement in a training set of 16 polyps identified with multi-detector CT, employing open biopsy forceps measurement at colonoscopy as the gold standard, demonstrated wide 95% limits of agreement for CT colonography size measurement and open biopsy forceps 34
. The authors reported variation of −0.9 to 0.5 cm, even for experienced radiologists.
Yeshwant and others18
retrospectively studied optimal modes of size measurement in a subset of 79 polyps from one of the three large multicenter trials of CT colonography10
. A research assistant, trained by a radiologist, determined linear size of 79 polyps and volume of 251 polyps in conjunction with the same radiologist. Automated linear and volume measurements of 251 polyps were also taken using reading software. Size measurement by linear probe at colonoscopy was the reference standard. The main analyses included determination of mean size measurement error, as well as a confidence analysis, in which the confidence with which a given polyp would be characterized as being greater than 1cm in size was determined. Overall, linear measurement of polyps employing a 3D display was reported to be the most accurate, with mean measurement error of −0.14cm +/− 0.68 cm, and the confidence with which a polyp would be called greater than 1cm for polyps actually 1 to 1.2 cm in size ranging from 50 to 75% and for polyps actually 0.6 to 0.9 cm in size ranging from 0 to 15%. Fractional differences for linear 3D measurements averaged 21% +/− 22.
More recently, Burling and others reported an analysis of measurement methods, including automated polyp measurement applied by a radiologist and a radiology technician, to a study set of 50 polyps from 33 patients. Measurements obtained by different viewing modes of CT colonography were compared to measurements obtained by either comparison to open biopsy forceps or linear probe at colonoscopy. For all 6 measurement methods employed, Bland-Altman analysis suggested variability of − 0.3cm to 0.8 cm 95% of the time, with a reported tendency towards overestimation of polyp diameter except when using an abdominal window display19
One group has compared polyp size measurement by CT colonography to gross pathology measurement20, 35
. Using a resected, sewn off colon specimen from a single patient with familial adenomatous polyposis syndrome, polyp size was measured by CT colonography, followed by caliper measurement of the polyps within the gross specimen. Two observers (a radiologist and a radiology technician) recorded polyp size measurement using 2D and 3D views, as well as automated polyp measurement software. Optimal limits of agreement were achieved by the radiologist, with the resected colon fully-insufflated: −0.71 to 0.31cm on 2D measurement, and −0.19 to 0.32 cm on 3D measurement. The proportion of polyps correctly categorized into groups of clinical significance ranged 75 to 100% for 8 polyps 0.1 to 0.5 cm in size, 18 to 64% for 11 polyps 0.6 to 0.9 cm in size, and was 100% for the single polyp greater than 1 cm in size. Inter- and intra-observer 95% limits of agreement suggested variability in size estimation of up to 2.5 mm.
Under experimental conditions employing a glass colon model containing 24 artificial polyps, Young and others demonstrated the optimal potential for polyp size measurements by CT colonography36
. Six reviewers measured these polyps employing two of three different types of CT colonography reading workstations. The three optimal methods of measurement (two dimensional axial or multiplanar analysis in lung windows, as well as three-dimensional tissue cube analysis) had small mean absolute errors ranging from 0.48 to 0.76 mm and standard deviations ranging from 0.51 to 0.79 mm. Underestimation of polyp size was consistently noted for all but two of 14 methods of measurement employed. While this analysis highlights optimal performance of CT colonography size measurement in an experimental model, results from ex-
colon analyses18, 34, 35
, as well as our analysis, are consistent with our conclusion that polyp size measurements at CT colonography show insufficient agreement with polyp size measurement by colonoscopy or pathology to be considered the primary factor for risk stratification of polyps detected by this modality.
The reliability of polyp size measurement by colonoscopy also appears to be problematic. In an analysis of 100 polyps employing pre-fixation pathology as the “gold standard”, open biopsy forceps measurement was associated with a mean difference of 0.1cm, with a 95% confidence interval for mean difference versus actual size of 0.06 to 1.4 cm28
. Linear probe measurement was reported to be the closest measurement to gold standard pre-fixation pathology, though the endoscopy nurse measuring pre-fixation size was not blind to investigator measurement by probe and forceps, raising the possibility of measurement bias. In a study comparing open biopsy forceps measurement to pre-fixation pathology among 31 polyps, the mean difference was 0.16 cm, and gastroenterologists consistently overestimated polyp size37
. In a study of 33 snared polyps, employing visual estimation alone, without requiring open biopsy forceps comparison, Shoen reported that endoscopists were within a measurement error of 0.3 cm of pre-fixation measurement only 80% of the time 32
. Data from a large trial that included visually estimated size measurement of 1795 patients with a single polyp found at sigmoidosocopy confirmed by colonoscopy demonstrated high variability in polyp size estimation38
. Utilizing in vitro models, two groups found consistent underestimation of polyp size by endoscopist measurement employing open biopsy forceps as a guide 39, 40
. These studies raise a number of important points regarding polyp size variation, limitations include small sample sizes in many cases, and the absence of formal agreement assessment of agreement in others. One interpretation of these data (in conjunction with ours) is that the inherent variability of size measurement at colonoscopy supports polypectomy for all polyps encountered at colonoscopy, as is currently recommended standard practice41
Polyp size appears to be proportional to risk for advanced neoplasia, and has been intensely studied. While the prevalence and subsequent incidence of cancer in individuals with polyps 1cm or more in size is consistently reported to be high 42–46
, and while the majority of reports have not identified cancer in polyps less than 0.5 cm in size42, 47–51
, uncertainty regarding the rate of cancer and worrisome histology in polyps sized 0.5 to 0.9 cm exists. Zero to 10% of adenomas 5 to 9mm in size may have high grade dysplasia47, 48, 52–54
, with most estimates for high grade dysplasia ranging from 3 to 7%, and for invasive cancer from 0 to 2.5%47, 52, 54
. However, the exact prevalence of advanced histology in polyps less than 1cm in size is difficult to discern based on these studies, as many only included adenomas (rather than all polyps) in their calculations, and most did not include only a screening population. Nonetheless, it may be generally accepted that size is an important, but incomplete predictor of underlying histology.
Our study has potential limitations. As has been typical of the other larger multicenter trials of CT colonography for colorectal neoplasia detection, a polyp matching algorithm was employed to match polyps identified on CT colonography to those identified at colonoscopy. Because in this study, the algorithm defined a true match as a polyp identified by both CT colonography and colonoscopy in the same or adjacent segment of the colon if size estimates agreed within 50%, but did not acknowledge a match if size differed by more than 50% or location by more than 1 adjacent segment, bias, most likely towards more favorable agreement between CT colonography and pre-fixation measurement, may exist.
Compared to the study that reported the greatest sensitivity in a large number of patients 10
, the study from which we used data for the current analysis had lower sensitivity and specificity for detection of polyps 12
. It could be argued that because readers did not detect lesions which great accuracy that they also did not accurately measure polyp size. We have considered the possibility that purported poor reader ability reflects a systematic lack of expertise that could bias the current investigation. In order to address this possibility, we considered an analysis by individual study site within the original trial. However this was not possible due to limitations in sample size at many study sites. In a post-hoc analysis of the single study site that contributed the most polyps to our main analysis, and had the highest sensitivity 42 out of the 82 polyps from the study site with the highest sensitivity for polyps greater than 6mm in size (>80%), Bland-Altman limits of agreement for polyp size estimation by CT colonography were not substantially different from our overall results (95% limits of agreement 47% lower to 50% higher than pre-fixation size). Furthermore, we are unaware of data linking lack of ability to detect lesions with any type of lack of ability to measure polyps with electronic tools available for current workstations.
It is important to emphasize that although we found measures of polyp size by CT colonography to vary substantially compared to pre-fixation polyp size, it is possible that in some settings, CT colonography may more accurately size lesions. Use of multislice CT scanners may reduce polyp distortion, and barium and iodine stool and fluid tagging may improve polyp detection and characterization55
. Additionally, improved imaging reconstruction software may be associated with improved sensitivity for polyps and improved polyp characterization. In one study of a subset of 79 polyps from an earlier published cohort, the authors noted that 3D reconstruction appeared to be more accurate than 2D reconstruction for size estimation, and that measurements made in two dimensions are inherently inaccurate because they require the reader to choose an optimal reading frame for measurement18
. Choosing an optimal reading frame requires that the reader recognize the plane of largest dimension for the polyp, and, if not optimally applied, may add to measurement error. Size analysis with 3D reconstruction was not performed in our study, perhaps resulting in less favorable estimates than may be expected with improvements in CT colonography software. Overall, both technology and reader related factors may affect estimates of agreement of polyp size measurement by CT colonography with pre-fixation measures. Nonetheless, it is notable that our estimates of the agreement of polyp size measurement by CT colonography with pre-fixation and colonoscopic measurement are similar to those from smaller studies employing more experienced readers and improved technology18, 19, 34
We attempted to explore potential bias with our analysis of agreement of CT colonography and pre-fixation polyp size for participants with only one polyp identified in the same segment. Of 14 polyps with perfect segmental match, 6 (43%) would have been classified into a size category of proposed clinical relevance different than that indicated by pre-fixation pathology (2 underestimated; 4 overestimated). Furthermore, in a post-hoc analysis, we also performed a “one segment” analysis in which agreement was determined for those polyps which had no other polyps present in the same segment at colonoscopy. Of the 82 polyps identified by colonoscopy and CT, 54 polyps did not have a polyp in the same segment. The 95% limits of agreement suggested that estimates of size by CT were 63% lower to 71% higher than pre-fixation estimates. Thus, both “one polyp” and “one segment” analyses are consistent with our overall observation that polyp estimation by CT colonography varies compared to other measures. The individual measuring pre-fixation size was not by protocol blinded to the measurement as determined by comparison to open biopsy forceps; this may have biased our estimation of size by colonoscopy favorably, and may have had a positive or negative impact on limits of agreement determination for CT colonography with pre-fixation size. Polyps may shrink after electrocautery and removal28, 32, 37
; despite this, a trend towards underestimation of polyp size by CT colonography compared to prefixation size with increasing polyp size was observed. Lastly, our estimates of the agreement of CT colonography measurements compared to open biopsy forceps may be biased in favor of colonoscopy measurement, given that in many cases, the same biopsy forceps used for measurement may have been used for polypectomy. Despite this, significant variation between colonoscopic and pre-fixation measurement was noted.
The finding that the measurement of polyp size by CT colonography has substantial variation in agreement with pre-fixation pathology measurement has several important clinical implications. First, current recommendations for management of polyps detected by CT colonography rely heavily on estimates of polyp size. A consensus statement recommended that polyps less than 0.6 cm not be reported or resected, that patients with 1 to 2 polyps 0.6 to 0.9 cm be followed by a subsequent surveillance CT colonography in up to three years or be referred for colonoscopy, and that patients with 3 or more polyps 0.6 to 0.9 cm or any polyp ≥1 cm be referred for colonoscopy21
. However, our analyses suggest that misclassification of a hypothetical polyp 0.7 cm in size as greater than 0.9 cm, or less than 0.6 cm using CT colonography may occur with moderate frequency. The degree to which such misclassification may occur leads us to speculate that polyp size estimation by CT colonography may be insufficiently reliable to be used as a primary factor to guide clinical management. This, in context with the fact that histology offers complimentary (if not more prognostic) risk information and, in consideration of recent modeling data which suggests excess cancer risk in patients with small polyps triaged to surveillance by CT colonography56
, raises the possibility that individuals with polyps suspected on CT colonography examination should be referred for diagnostic colonoscopy, with polypectomy if polyps are confirmed. In context of clinical investigation, our findings call into question the reliability of polyp matching algorithms that do not allow matching of polyps detected at CT and colonoscopy if size estimation differs by more than 50%. Removing size as an important criterion, or improving size estimation, might result in more favorable estimates of sensitivity for polyps by CT colonography. Future studies of CT colonography should continue to pursue novel methods of size estimation that may be more be accurate and easier to follow over time, such as volume measurement57
and use of optimal viewing modes36
, and place these advances in context of other clinical assessments of risk for colorectal cancer to more precisely delineate which individuals with polyps detected by CT colonography may forgo colonoscopy.