Various screening tests in current use reduce the incidence and rate of death from colorectal cancer (7, 8). Despite the proven efficacy of these tests, however, patient adherence to screening guidelines is low: Only 30% to 45% of persons eligible for screening undergo such tests. Low adherence rates are believed to be due to poor public awareness and poor public acceptance of current screening techniques (9-13).

An increasingly popular screening test for colorectal cancer is computed tomographic (CT) colonography, also known as CT colography or virtual colonoscopy. Computed tomographic colonography was first described in 1994 as a radiographic technique in which thin-section images of pneumocolon could be reconstructed by sophisticated software into high-resolution 2- and 3-dimensional images (14). Over time, improvements in hardware and software have allowed faster scanning, reduced exposure to radiation, and better imaging. Newer modes of imaging (called fly-through) can produce results that resemble endoscopic images and permit sophisticated characterization of detected lesions (15-17). Early studies primarily used the spiral CT scanner, which has limitations in spatial resolution that can make small polyps more difficult to detect (17). The multidetector CT scanner has permitted rapid acquisition of finer images, obtained during a single breath-hold, that can greatly improve image quality and spatial resolution (17, 18). Many aspects of this technology are under study, including software that assists in detection of lesions, refinements in image reconstruction, and stool tagging (19-21). The latter development relies on ingestion of contrast material over several days or hours, after which software digitally subtracts residual solid and fluid fecal material from the acquired images, creating a "virtually clean" mucosal surface (22, 23). This technique may improve sensitivity and may someday obviate the need for bowel cleansing before examination.

Although it is touted as a less invasive screening method than flexible sigmoidoscopy or colonoscopy, CT colonography typically requires full bowel cleansing and insufflation of air through the rectum (24). Studies have suggested that CT colonography may be similar, and in some cases preferable, to colonoscopy in terms of comfort and acceptability, but no convincing difference between these 2 approaches has been demonstrated (25-31). If virtual colonoscopy is found to have equivalent test characteristics, improve patient adherence, and be safer or less expensive than colonoscopy, it may be more cost-effective and become the screening method of choice (32, 33).

Studies of the test characteristics of CT colonography have had mixed results. Pickhardt and colleagues used CT colonography in 1233 patients and found a sensitivity of 93.9% for adenomatous polyps larger than 8 mm (25). Other studies have had less favorable results, with sensitivities as low as 55% for polyps larger than 10 mm, raising concerns about the overall test performance of CT colonography when used in a broader range of settings (34). Various reasons for these discrepant results have been offered, but the source of this heterogeneity has not been fully explored (16, 35, 36). Such assessment is needed because patients and providers look to this technology in the hope of improving screening rates (29).

We systematically reviewed the literature to assess the test performance of CT colonography compared with colonoscopy or surgery, to define characteristics of these studies, and to attempt to explain the sources of conflicting results.

Methods

Top Methods Discussion Author & Article Info References

Study Identification and Selection

We searched the PubMed, EMBASE, and MEDLINE databases and the Cochrane Controlled Trials Register for all relevant articles published in the English language between 1975 and February 2005 by using the Medical Subject Headings or text words virtual colonoscopy, CT colonography, CT colography, or CT pneumocolon. The title and abstract of potentially relevant studies and review articles were screened for appropriateness before retrieval of the full articles. Two reviewers independently searched the literature. Inclusion criteria were a prospective, blinded design (in which results of CT colonography were interpreted independently of findings on colonoscopy or during surgery); enrollment of adult patients who were to undergo CT colonography after a full bowel preparation, followed by complete colonoscopy or surgery; and use of at least a single-detector CT scanner, with colon insufflation by air or carbon dioxide, scan intervals no greater than 5 mm, and use of both 2-dimensional and 3-dimensional views during scan interpretation.

Study Quality

Two observers independently extracted data on test characteristics; study setting; patients; and components of methodologic quality that may be associated with bias in test accuracy studies, including disease severity, disease prevalence, prospective design, relevant clinical sample (as opposed to a diagnostic case–control study), enrollment of a series of consecutive patients, assurance that all patients underwent reference testing, performance and interpretation of the index test without knowledge of the results of the reference test, and performance and interpretation of the reference test without knowledge of the results of the index test (33). A piloted standardized data extraction sheet was used, and disagreements were resolved by consensus.

Data Abstraction

We abstracted characteristics of the study (design, country, year, reference standard, and type of contrast used), patients (demographic and risk for colorectal cancer), scanners (manufacturer, type of viewer, type of contrast, software, and hardware), and study quality. Sensitivity and specificity were calculated per patient, per polyp, and for polyps of 3 size categories: smaller than 6 mm, 6 to 9 mm, and larger than 9 mm. When data on test performance were reported for 2 or more separate CT colonography readers, we calculated an average value. When possible, we excluded data on double readings. If a study reported data related specifically to adenomas instead of polyps, in general, we abstracted only the data for adenomas. For studies that performed retrospective analysis (for example, fly-through imaging in the study by Cotton and associates [34]), we abstracted only data on CT colonography findings before colonoscopy. If data could not be extracted or calculated from the manuscript with confidence, none were entered. Two reviewers independently abstracted data, and disagreements were resolved by consensus.

Statistical Analysis

Pooled sensitivities and specificities on a per-patient basis were combined and weighted according to sample size. Confidence intervals for each study were calculated by using exact binomial methods in a random-effects model. We focused our analysis on per-patient data because this is the most important perspective for a screening test, whereas per-polyp data emphasize the ability of CT colonography to find colonic lesions. That is, the latter analysis assesses the performance of the technology rather than its utility as a screening tool. Heterogeneity was assessed by using the I² statistic (37). The I² statistic provides an estimate of the amount of variance due to heterogeneity rather than chance and is based on the traditional measure of variance, the Cochrane Q statistic. Potential threshold effects were assessed by using the Spearman statistic and by creating receiver-operating characteristic curves according to the method of Moses and coworkers (38). Heterogeneity was assessed by performing stratified analyses when the potential confounding variable was dichotomous or categorical, by plotting the weighted effect size against the potential confounding variable when that variable was continuous, and by applying meta-regression methods in either case (39). Subgroup analyses were done by year of publication, imaging technique (2-dimensional imaging with 3-dimensional confirmation only when a lesion was noted, 3-dimensional imaging with 2-dimensional confirmation, 2-dimensional imaging with concomitant 3-dimensional imaging, or fly-through technology), collimation width and reconstruction interval (in millimeters), type of scanner (single-detector, multidetector, or mixed), and use of a contrast agent (yes or no). When collimation or reconstruction thickness was given in half-millimeter increments, we rounded the values up to the next whole number. The meta-regression analysis used the restricted maximum likelihood method and was performed by using indicator variables to assess differences among the strata. All analyses were performed with Stata software, version 8.2 (Stata Corp., College Station, Texas).

Data Synthesis

Our final pool of eligible studies (Appendix Figure) included 33 prospective studies involving 6393 patients that compared CT colonography to the reference standard of colonoscopy or surgery (22, 25, 34, 40-69). Studies originated from 7 different countries, but most were done in the United States (64%). The average number of participants in a study was 248 (range, 20 to 1233). The mean age of participants was 61.9 years; 63.6% of participants were male, and 74% were at high risk for colorectal cancer. Sixteen studies used single-detector scanners, 13 used multidetector scanners, and 4 used both single-detector and multidetector scanners. Fifteen studies used 2-dimensional imaging, with 3-dimensional imaging on selected slices at the discretion of the radiologist; 14 studies used dedicated 2-dimensional and 3-dimensional imaging; and 2 studies used fly-through imaging with 2-dimensional reconstruction. The average collimation was 4 mm (range, 1 to 5 mm), and the average reconstruction interval was 1.86 mm (range, 1 to 5 mm). Tables 1 and 2 show detailed information from individual studies.

View larger version (31K): [in this window] [in a new window]   Appendix Figure. Map of the literature search and selection process. The QUORUM (Quality of Reporting of Meta-Analyses) guidelines for reporting of meta-analyses were used. CRC = colorectal cancer; CT = computed tomographic.

Computed tomographic colonography was compared to various reference standards, including standard colonoscopy, segmental unblinded colonoscopy (after each colon segment is examined, the results of CT colonography are revealed to the endoscopist and discrepant segments are reexamined), optimized colonoscopy (in which videotapes of the endoscopy are reviewed in comparison with discrepant CT colonography findings), and surgical findings or results of double-contrast barium enema (when subsequent colonoscopy was not or could not be performed). Several studies used a combination of these reference standards.

Table 2 shows biases that may be present in the studies, as defined by Whiting and colleagues (70). One important source of bias was differences in disease severity or prevalence among studies. Because the baseline risk of the study participants may have been apparent to the investigators, clinical review bias was probably present in many of the studies. In addition, because the reference standards varied not only among studies but among segments from a single patient, bias could result from differential verification of findings or might be considered incorporation bias in some cases. Although most studies did not define observer variability, those that did had a range of values. The studies that used consensus readings may not be representative of typical CT image-reading practice and may have increased bias. In several studies, many persons interpreted the images (from CT colonography or colonoscopy), but most studies involved few readers or a single reader. The nature of these studies precluded quantitative comparisons of quality.

Sensitivity of CT Colongraphy

Per-patient sensitivity for CT colonography varied from 21% to 96% (Figure 1). The overall pooled sensitivity for CT colonography was 70% (95% CI, 53% to 87%). Sensitivity increased progressively as polyp size increased: It was 48% (CI, 25% to 70%) (range, 14% to 86%) for detection of polyps smaller than 6 mm, 70% (CI, 55% to 84%) (range, 30% to 95%) for polyps 6 to 9 mm, and 85% (CI, 79% to 91%) (range, 48% to 100%) for polyps larger than 9 mm. Each of these analyses was statistically heterogeneous (P < 0.001 for each), and most of the variance was attributable to between-study heterogeneity. The I² statistic was 96.7% for polyps smaller than 6 mm, 93.1% for polyps 6 to 9 mm, and 85.2% for polyps larger than 9 mm. Appendix Tables 1 and 2 show data from individual studies.

View larger version (22K): [in this window] [in a new window]   Figure 1. Reported per-patient sensitivities in the included studies, by polyp size. Summary statistics: For polyps <6 mm, 0.48 (95% CI, 0.25 to 0.70); for polyps 6–9 mm, 0.70 (CI, 0.55 to 0.84); for polyps >9 mm, 0.85 (CI, 0.79 to 0.91).

View this table: [in this window] [in a new window]   Appendix Table 1. Test Performance of Computed Tomographic Colonography, by Per-Patient Analysis

View this table: [in this window] [in a new window]   Appendix Table 2. Test Performance of Computed Tomographic Colonography, by Per-Polyp Analysis

We found several potential sources for this heterogeneity. First, studies that used thinner slices for collimation appeared to have better sensitivity, and meta-regression of data from 19 studies suggested that every 1-mm increase in collimation width decreases sensitivity by 4.9% (CI, 0.8% to 7.1%). Second, the 7 studies that used multidetector scanners and that reported overall sensitivity had homogenously high sensitivity (95% [CI, 92% to 99%]; I² = 40%; P > 0.2) (Figure 2). This sensitivity was higher than that in the 9 studies reporting overall sensitivity in which a scanner with a single-detector was used (82% [CI, 76% to 92%]), although the latter results were heterogeneous (I² = 87.1%; P < 0.001). The 10 studies that used 2-dimensional imaging, with confirmation by 3-dimensional imaging only when considered necessary, yielded a sensitivity of 81.9% (CI, 71% to 91%) (I² = 87.5%; P = 0.02), whereas the 6 studies that used standard 2-dimensional imaging and concomitant 3-dimensional imaging had a pooled sensitivity of 91% (CI, 83% to 99%) (I² = 53.1%; P = 0.06) and the 2 studies that used fly-through technology had a pooled sensitivity of 99% (CI, 95% to 100%) (I² = 47.6%; P = 0.17) (Figure 3).

View larger version (21K): [in this window] [in a new window]   Figure 2. Reported per-patient sensitivities in the included studies, by type of scanner.

View larger version (25K): [in this window] [in a new window]   Figure 3. Reported per-patient sensitivities in the included studies by mode of imaging.

Analysis of year of publication, type of scanner hardware or software, thickness of the reconstruction interval, use of contrast (bowel, intravenous, or none), and patient characteristics (age, sex, and high or average risk) yielded no other source of heterogeneity. We found no evidence of a threshold effect between sensitivity and specificity when the Spearman statistic was calculated or receiver-operating characteristic curves were constructed.

Specificity of CT Colonography

In contrast to the broad range of sensitivities reported, per-patient specificity was more consistent across polyp sizes (Figure 4). Overall, CT colonography was 86% specific (CI, 84% to 88%) (I² = 92.6%; P = 0.001) on the basis of data from 14 studies. Specificity improved as polyp size increased, and the results were homogenous within each strata. Only 4 studies reported specificity for detection of polyps smaller than 6 mm, and the pooled specificity from these studies was 91% (CI, 89% to 95%) (I² = 47.1%; P = 0.15). For polyps 6 to 9 mm in size (6 studies), specificity was 93% (CI, 91% to 95%) (I² = 50%; P = 0.07) and increased to 97% (CI, 96% to 97%) (I² = 41.8%; P > 0.2) for polyps larger than 9 mm (15 studies). Appendix Tables 1 and 2 show data from individual studies.

View larger version (27K): [in this window] [in a new window]   Figure 4. Reported per-patient specificities in the included studies, by polyp size.

Discussion

Top Methods Discussion Author & Article Info References

We found that CT colonography is highly specific, particularly for polyps greater than 9 mm in size. However, the reported sensitivities for CT colonography vary widely, even for larger polyps. Before any screening method can be recommended for general use, it must be demonstrated to be highly and consistently sensitive in a variety of settings. The inability of our meta-analysis to clearly explain why the reported sensitivities vary so widely suggests that CT colonography needs further refinement before it can be recommended for general use in screening for colorectal cancer.

Our analysis revealed some factors that account for the wide range of sensitivities. First, scanners that used thinner collimation had higher sensitivity. Every 1-mm increase in collimation width decreased the subsequent sensitivities by almost 5%. That is, if scanners with 1-mm slices had 98% sensitivity, increasing the collimation width to 2 mm would decrease sensitivity to 93%. Second, scanners that used multiple detectors rather than single detectors were more sensitive. Finally, the mode of imaging also appeared to be important: The more recently developed fly-through technology had a sensitivity of 99%. However, this latter finding must be interpreted with caution because it is based on data from only 2 studies and considerable heterogeneity was found for the other types of imaging used. These results suggest that CT colonography is promising as a screening test for colorectal cancer. Before it is put into general use, however, it must be shown to be reliably sensitive and questions about the optimal technological characteristics of the technique must be settled. Our results are not definitive, but rather suggestive of avenues of pursuit in refinement of this method.

Our conclusions differ from those of another recent systematic review that favored use of CT colonography (71). That review included data from 14 trials and 1324 patients, compared with the 33 trials and 6393 patients in our analysis. The investigators reported a summary sensitivity of 87% for detection of polyps 6 to 9 mm and 88% for detection of polyps larger than 10 mm, and a specificity of 95% for detection of polyps larger than 10 mm. Their conclusion that "the specificity and sensitivity of CT colonography are high for polyps larger than 10 mm" does not take into account the sources of heterogeneity in the sensitivities reported among the studies that they included. Until new technology, particularly for screening tests, can be demonstrated to be consistently reliable, we believe it cannot be recommended for general use.

Other sources for the wide range in reported sensitivities may exist. Previous reports have implied that the differences in test performance among studies of CT colonography is related to the CT colonography technology used, the type of contrast medium, the mode of imaging, and the expertise of the radiologists reading the images. The available data are sufficient only to suggest that multidetector scanners, mode of imaging, and low collimation width affect test performance. Many other possible sources of false-negative results exist, including limitations in technology and technique, insufficient resolution, poor bowel distention, poor preparation, breath-hold artifacts, misinterpretation of stool or folds, sessile or flat polyps, paired lesions, software limitations, and errors in reading (perceptive errors) (45, 46, 55, 58, 63, 72-75). Whether a study used CT colonography to detect all polyps (including hyperplastic polyps) or adenomas only may also affect test performance, because CT colonography may have a higher sensitivity for detection of adenomas (76). Delineation of these possible sources of heterogeneity requires a more sensitive technique than meta-analysis. We abstracted information on these study characteristics, but our ability to discriminate whether any of these is the source of heterogeneity is limited. In addition, we could not evaluate such factors as the expertise of the radiologists reading the CT colonography scans. Our findings cannot be taken to mean that none of these other variables are the source of heterogeneity, only that the current data do not allow us to clearly demonstrate that one of these characteristics is the explanation.

Our analysis has limitations. First, 18 of the studies used colonoscopy as the gold standard, yet colonoscopy may miss more than 10% of small polyps, up to 10% of large polyps, and up to 5% of colorectal cancers (64-68). Eleven studies used segmental unblinded colonoscopy or optimized colonoscopy so that CT colonography and augmented colonoscopy could be used in tandem, to maximize overall detection of lesions. However, even these methods do not ensure that each segment of the colon is examined multiple times, so that no lesion is missed by either method. Studies that used segmental unblinded colonoscopy (in which results from CT colonography are revealed after the endoscopist has examined each colonic segment) or optimized colonoscopy (in which video images from colonoscopy are reviewed for discrepant results) demonstrated that CT colonography found polyps (and several tumors or masses) that were missed on blinded colonoscopy (25, 34, 40, 46). In addition, only 3 studies were designed to evaluate a true screening population: persons who are at average risk for colorectal cancer (Table 2) (25, 41, 64). In 1 of these studies, the rate of false-negative findings of polyps larger than 9 mm was similar to the rate of false-negative results with colonoscopy reported in studies of tandem colonoscopy, although other studies have not shown such favorable results (77).

Second, the power to elucidate sources of heterogeneity is limited by the information reported in each article and by the relatively small number of included articles. Finally, meta-regression using summary covariates for each article has limited accuracy. Finer evaluation of the source of heterogeneity would require procurement of patient-level data.

Computed tomographic colonography is very specific, particularly for detection of polyps larger than 9 mm. In studies that used a multidetector scanner, low collimation, and an optimal mode of imaging, the sensitivity of CT colonography to detect polyps larger than 9 mm was the highest and most consistent. However, results were inconsistent when other technical approaches were used and smaller polyps were present. Acceptable techniques for colorectal cancer screening should have consistently high sensitivity over specificity so that preneoplastic polyps are effectively ruled out in patients with a negative result. Although some studies have reported high sensitivities for CT colonography, the range among all studies is broad (as low as 21% overall, and as low as 48% for polyps greater than 9 mm).

Until the source of this heterogeneity is more clearly explained and CT colonography is demonstrated to be consistently and reliably sensitive, it cannot be recommended for general use. However, the technology shows much promise in this regard. Refinement of CT scanners, improved patient preparations, and evolving software for CT colonography will probably improve diagnostic accuracy. For the time being, CT colonography should be used in research protocols or when other accepted screening methods are not appropriate.