Introduction

Computed tomography (CT)-colonography has been studied for screening for (precursors of) colorectal cancer (CRC) and the Multisociety Task Force on Colorectal Cancer has indicated CT-colonography as an acceptable technique for CRC screening [1, 2]. However, recently the National Institute of Health has published a statement regarding CRC screening concluding that there is still lack of information regarding the use of CT-colonography as screening technique in an average risk population [3]. Also other guidelines state that there is insufficient evidence yet [4, 5].

Several meta-analyses have been published on the diagnostic value of CT-colonography including both average risk and high risk individuals, but no meta-analysis has been published including average risk individuals only [610]. Individuals are considered to be at average risk if they have no symptoms, no personal history of CRC, adenomatous polyps or inflammatory bowel disease and no family history of advanced neoplasia [11]. By including studies containing high risk populations, the diagnostic value of CT-colonography in an average risk population might be overestimated. It is known that the estimated diagnostic value of a technique depends on factors such as disease prevalence and spectrum.

Therefore, the aim of this meta-analysis was to estimate the diagnostic value of CT-colonography to detect (advanced) adenomas and CRC in an average risk population aged 50–75 years.

Materials and methods

Literature search

Articles were obtained from the electronic databases PubMed, Embase and Cochrane, without restrictions with respect to the publication date and language. Lists of synonyms for CT-colonography were produced (Fig. 1) and combined using the Boolean operator “OR”. The same was done for colonoscopy. Both search results were combined, using the Boolean operator “AND”. By reading title and abstract of all retrieved articles, two observers identified possible relevant papers, based on the inclusion and exclusion criteria described below. The remaining articles were retrieved as full-text articles and independently checked by two reviewers. Disagreement regarding inclusion was resolved by consensus. Reference lists of the final selection of articles were checked manually to identify other relevant papers. If additional information of an article considered for inclusion was needed due to incomplete data or description of the methods, the corresponding authors were contacted.

Fig. 1
figure 1

Flowchart of search strategy, search date: 21st of July, 2010

Inclusion and exclusion criteria

Inclusion criteria were prospective, randomized trials or cohort studies, in humans ≥50 years, in which at least 50 predominantly asymptomatic average risk subjects (≥95%) underwent CT-colonography and completed colonoscopy for verification within 3 months. In addition, eligible studies needed to report the detection of colorectal polyps (adenomatous and non-adenomatous), advanced neoplasia and CRC and should include true-positive (TP), false-positive (FP), true-negative (TN) and false-negative (FN) values. Studies that included predominantly high risk subjects (symptomatic, history of hereditary CRC, personal history of polyps, CRC or IBD) were excluded, as well as studies that performed CT-colonography as a consequence of incomplete colonoscopy or studies that only performed colonoscopy after positive findings on CT-colonography.

Quality assessment

Systematic assessment of quality and documentation of relevant data of the selected articles was performed independently by two reviewers, using a standardized form. To grade the study quality, the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool was used with special focus on the characteristics of the included study population, index test and reference test [12]. We assessed whether the inclusion and exclusion criteria were described clearly and would result in a representative screening cohort. In addition,the presence of a disease progression bias and a verification bias was determined: did all participants receive their reference test <3 months? Furthermore, we assessed whether the index test did not form part of the reference standard, whether all subjects received the same reference test, if the test results of both test were interpreted without knowledge of the other test results and whether withdrawals or uninterpretable test results were reported. Results are presented in Appendix 1.

Study population

The following patient characteristics were documented: number of asymptomatic and symptomatic subjects, sex ratio, mean or median age with age range and CT-colonography indication.

Imaging features

The following characteristics were documented regarding the imaging features of CT-colonography: bowel preparation, dietary restrictions, tagging and bowel distention, use of spasmolytical drugs and type of CT-system and CT-parameters, the positioning of the patient and the use of intravenous contrast medium during CT-colonography. For colonoscopy, the type of bowel preparation and dietary restrictions were documented.

Imaging and diagnostic criteria

The following characteristics were documented regarding image analysis of both CT-colonography and colonoscopy: number of diagnostic examinations, number and experience of CT-colonography readers and endoscopists, reading strategy on CT-colonography, use of segmental unblinding or second look colonoscopy, determination of size on CT-colonography and during colonoscopy and histopathological confirmation.

Data extraction

For the analysis per patient, 2 × 2 contingency tables were constructed to be able to calculate the sensitivity and specificity values for the following type of lesions: all polyps, adenomatous polyps, advanced adenomas (defined as an adenoma with >25% villous features, size ≥10 mm and/or high grade dysplasia [13]), advanced neoplasia and CRC. For each type of lesion, except for CRC, data were collected using the following thresholds: 6–9 mm, ≥6 mm and ≥10 mm, based on the associated potential CRC risk [1416].

For the analysis per polyp, we extracted TP and FN findings to calculate the sensitivity of all polyps, (advanced) adenomas and advanced neoplasia for the same thresholds.

If needed, a request for additional data was send to the corresponding author. If possible, the following matching algorithm was used: the lesion should be at least <50% margin of error in size and should be found in the same or adjacent segment.

Statistical analysis

Heterogeneity of sensitivity or specificity was assessed using I2 statistics [17]. If I2 values were >25%, we considered these data significantly heterogeneous, and random-effects analyses were performed. In case of I2-values <25%, fixed-effects approaches were used.

For the per-patient analyses, we used bivariate models [18] to obtain summary estimates of sensitivity and specificity with 95% confidence intervals. For the per-polyp analyses, we used univariate models to obtain summary estimates of sensitivity with 95% confidence intervals. All analyses were performed using SAS software (SAS 9.2 procNlmixed, SAS Institute, Cary, NC, USA).

Publication bias was examined by constructing funnel plots.

Per patient

The x-axis consisted of the natural logarithm of the diagnostic odds ratio \( \left( { = \left( {{\hbox{TP}} \times {\hbox{TN}}} \right)/\left( {{\hbox{FN}} \times {\hbox{FP}}} \right)} \right) \). On the y-axis, we plotted the number of patients.

Per-lesion

The x-axis consisted of the sensitivity and on the y-axis, we plotted the number of patients.

Egger’s regression tests were used to examine the asymmetry of the funnel plots. A significant regression coefficient (P < 0.05) indicates an association between sample size and the diagnostic values.

Results

The initial search yielded 1,841 articles (Fig. 1). By excluding doubles, 1,021 articles remained. After screening on title and abstract, 1,008 articles were excluded. The most frequent reasons for exclusion were study design, study population (i.e. high risk) or non-related to CT-colonography or screening for polyps and CRC (i.e. IBD, MR-colonography). After assessment of 13 full text publications, seven articles were excluded, because they included only (n = 2)[19, 20] or predominantly (n = 4) high risk participants (16.7%, 37.0%, 76.6% and 80.4%, respectively)[2124] or because colonoscopy was only offered to a small selection of the participants (n = 1)[25]. Finally, six articles were included in this systematic review describing the results found in five prospective cohort studies [2631]. Screening of title and abstract of references and related articles did not result in additional relevant articles.

Patient characteristics

Patient characteristics are outlined in Table 1. We included five studies with in total 4,086 patients (54% male). Four studies [2628, 30, 31] did have a study population of over 200 average risk subjects, the largest population comprised 2,249 average risk subjects [27]. The smallest study had a population of 68 participants at average risk [29]. All studies provided a clear description of patient characteristics and the inclusion and exclusion criteria. Three studies included high risk subjects: 2.6% [30, 31], 5.2% [28] and 11.3% [27], respectively. The corresponding authors of these papers were contacted to obtain data concerning average risk patients only. This succeeded in two out of three studies, resulting in a total of four datasets containing data of average risk subjects only [2629] and one study including 2.6% high risk participants [30, 31]. Resulting in a total of 4,086 participants, of which 37 were at high risk (0.9%). The mean age varied between 55 and 60.5 years, the minimal age was 50 in four studies [2629].

Table 1 Patient characteristics included studies

Bowel preparation and CT-colonography procedure

Bowel preparation and CT-colonography procedure are outlined in Appendix 2.

Three studies used an extensive bowel preparation predominantly based on 4 liters polyethylene glycol [2628] combined with a clear liquid diet [26], a low-residue diet [28] or dietary restrictions depending on the institutional standard of the clinical centres where the examinations were done [27]. The remaining two studies both used a more limited preparation based on sodium phosphate [2931]. One study combined this with a clear liquid diet [30, 31], the dietary restrictions of the other study were not specified [29].

Three studies used oral tagging [26, 27, 30, 31], one study did use intravenous contrast medium [28]. Of one study it was not specified whether the participants received tagging [29]. Bowel preparation was the same for colonoscopy, as both colonoscopy and CT-colonography were performed on the same day in all studies.

Bowel distension methods varied between the studies. Two studies used (primarily) automated CO2 insufflation, combined with butylscopolamine bromide (Buscopan, Boehringer, Ingelheim, Germany) [26] or glucagonhydrochloride (GlucaGen, Novo Nordisk A’S, Bagsvaerd, Denmark) as spasmolytical drug [27]. Three studies used manual room air [2831]. In one study no spasmolytical drug was administered [28], it was not specified whether spasmolytical drugs were used in the remaining two studies [2931]. Two studies used at least 4 slice CT equipment [2931], two studies used at least 16 slice CT [27, 28] and one study used 64 slice CT [26].

Study characteristics

Study characteristics are outlined in Appendix 3. All participants received CT-colonography and colonoscopy on the same day. Different reference standards were used. One study used the colonoscopy results without knowledge of the CT-colonography findings [29], two studies used the colonoscopy result after segmental unblinding as reference [26, 30, 31], one study used colonoscopy (followed by a second look colonoscopy if lesions ≥10 mm reported on CTC were missed on the initial colonoscopy) combined with histopathology as reference [27] and another study used the histopathology results of the polyps that were removed during colonoscopy after segmental unblinding [28]. It is unclear whether there were any withdrawals in the selected studies. Uninterpretable results of CT-colonography or colonoscopy (outlined in Table 1) were reported and excluded from the analyses in two studies [26, 30, 31].

Image analysis

The characteristics of the readers and the reading strategy are outlined in Appendix 3. The minimal experience of the CT-colonography readers was specified in four out of five studies, and varied between 25 and 100 examinations [2628, 30, 31]. In one study the only reader had 5 years of reading experience [29]. Two studies used 2D read as primary reading strategy [28, 29], two studies used 3D read [26, 30, 31] and one study used both reading strategies at random [27]. None of the included studies specified whether CAD was used. The experience of the endoscopists and use of different scopes of the included studies was not specified in most studies [27, 2931]. One study had been done by gastroenterologists with a minimum experience of 1,000 colonoscopies [26], while the gastroenterologists in another study had a prior experience of 3,000 colonoscopies [28].

Size measurement of the polyp was done by the use of an open biopsy forceps [26, 28, 29], by a calibrated linear probe [30, 31] or determined by the pathologist [27]. In all studies histopathology confirmation was available.

Data extraction

Four studies used a matching algorithm almost the same as the one described in the methods [26, 2831]. These studies considered a CT-colonography finding to correspond with a colonoscopy lesion, if it was found in the same or adjacent segment. In addition it should be at least <50% margin of error in size [28], in the same or adjacent size category [26, 30, 31] or should have a size difference of <4 mm [29] to be considered as a true positive. The fifth study [27] used a different matching algorithm: one or more lesions should be in the same size category, irrespective of location. Of this study new data were requested and received, using the matching algorithm as specified in the methods section.

Per patient data for each of the different size categories regarding all polyps and adenomas respectively, could be obtained in three respectively four of the five studies (Table 2). Per polyp data for each of the different size categories regarding all polyps could be obtained in all studies while per polyp data for adenomas could be obtained in four studies and per polyp data of advanced adenomas and CRC in three of the five studies (Table 3).

Table 2 Results regarding all polyps, (advanced) adenomatous polyps, colorectal cancer, advanced neoplasia: per patient
Table 3 Results regarding all polyps, (advanced) adenomatous polyps, colorectal cancer and advanced neoplasia: per polyp

Corresponding I2 values for heterogeneity are reported in Tables 2 and 3. The results of individual studies are shown in forest plots (Figs. 2 and 3).

Fig. 2
figure 2

Forest plot of per patient sensitivity and specificity, including sensitivity and specificity estimates, for all polyps. Summary statistics: estimated sensitivities for polyps ≥6 mm and ≥ 10 mm were 75.9% (95% CI 62.3–85.8) and 83.3% (95% CI 76.8–89.0), corresponding estimated specificities 94.6% (90.4–97.0) and 98.7% (97.6–99.3)

Fig. 3
figure 3

Forest plot of per patient sensitivity and specificity, including sensitivity and specificity estimates, for adenomas Summary statistics: estimated sensitivities for adenomas ≥6 mm and ≥10 mm were 82.9 (73.6–89.4) and 87.9 (82.1–92.0), corresponding estimated specificities 91.4 (84.1–95.5) and 97.6 (95.0–98.9)

Data analysis per patient

All polyps

Estimated sensitivities for polyps ≥ 6 mm and ≥ 10 mm (regardless of histology) were 75.9% (95%CI 62.3–85.8) and 83.3% (95%CI 76.8–89.0), while corresponding specificities were 94.6% (95%CI 90.4–97.0) and 98.7% (95%CI 97.6–99.3).

Adenomas

Estimated sensitivities for adenomas ≥ 6 mm and ≥ 10 mm were 82.9% (95%CI 73.6–89.4) and 87.9% (95%CI 82.1–92.0), while corresponding specificities were 91.4% (95%CI 84.1–95.5) and 97.6% (95%CI 95.0–98.9).Estimated sensitivities of all polyps and adenomatous polyps of 6–9 mm are available in Table 2.

Advanced adenomas, CRC and advanced neoplasia

Estimated results for the detection of advanced adenomas, advanced neoplasia and CRC were not calculated, as a consequence of the small number of participants with these findings (Table 2).

Data analysis per polyp

All polyps

Estimated sensitivities for polyps ≥ 6 mm and ≥10 mm (regardless of histology), were 74.3% (95%CI 61.6–83.3) and 83.7% (95%CI 76.6–89.0).

Adenomas

Estimated sensitivities for adenomas ≥ 6 mm and ≥ 10 mm were 80.0% (95%CI 66.9–88.7) and 85.9% (95%CI 80.4–90.0).

Advanced adenomas

Estimated sensitivities for advanced adenomas ≥ 6 mm and ≥10 mm were 83.9% (95%CI 77.6–88.7) and 83.3% (95%CI 77.1–88.8). Estimated sensitivities for polyps and (advanced) adenomas of 6–9 mm, are presented in Table 3.

Advanced neoplasia and CRC

Estimated sensitivities for advanced neoplasia and CRC by CT-colonography were not calculated, as a consequence of the small number of CRCs (n = 6) that were detected in the included studies. In all studies, no CRCs were missed (Table 3).

Publication bias

The data points in the funnel plots are symmetrically distributed in a funnel shape suggesting the absence of publication bias (Appendix 4a5b). In addition, the Egger’s regression tests showed no associations between sample size and diagnostic values (data not shown).

Discussion

This systematic review demonstrates an estimated per patient sensitivity and specificity of CT-colonography for the detection of adenomas ≥ 6 mm of 82.9% (95%CI 74–89%) and 91.4% (95%CI 84–96%) in asymptomatic screening participants. The estimated per patient sensitivity and specificity for adenomas ≥ 10 mm, were 87.9% (95%CI 82–92%) and 97.6% (95%CI 95–99%). The estimated per patient sensitivities for all colorectal polyps were slightly lower. All six CRCs were detected by CT-colonography.

As we obtained additional data of the studies in which high risk participants were excluded [27, 28], the study results might not be identical to previously published data. In addition, the results of Johnson et al. [27] are different then published before, as we used a different matching algorithm then the one that was used in their study, resulting in lower sensitivities and higher specificities.

There are a few explanations available for the substantial variability between studies in sensitivity and specificity. The largest study [27] (n = 2,249 participants), did not report lesions <5 mm found on CT-colonography (while a colonoscopy lesion of 6 mm could match a CTC lesion of 3 mm) and performed no second look colonoscopy for colonoscopy negative CTC lesions <10 mm . Obviously, both factors will probably result in a lower sensitivity for medium sized adenomas and a less prominent difference in the detection of adenomas ≥ 10 mm compared to the studies of Graser [26] and Pickhardt [30, 31]. The second explanation could be the use of primary 2D or primary 3D read: those studies with the highest sensitivities for the detection of adenomas used primary 3D read [26, 30, 31]; the other studies used primary 2D read [28, 29] or both methods randomly [27]. However, there is conflicting evidence regarding the possible difference of sensitivity when using primary 2D or 3D read [32, 33].

To our knowledge this is the first meta-analysis in which the diagnostic value of CT-colonography is compared to colonoscopy for the detection of (adenomatous) polyps and CRC in an average risk population. Previously, at least five systematic reviews [610] were published describing the diagnostic value of CT-colonography in general (not specified for (advanced) adenomas), including both average risk and high risk populations. By comparing our results to the estimated sensitivities per patient for polyps 6–9 mm and ≥10 mm published previously, we found lower sensitivities, especially when looking at polyps of 6–9 mm. Estimated sensitivities per patient for polyps 6–9 mm published before were 59%, 70%, 84% and 86%, respectively [68, 10], while we calculated an estimated sensitivity of 68.1%. Estimated sensitivities per patient for polyps ≥ 10 mm were 76%, 85%, 88% and 93%, respectively [68, 10], while we calculated an estimated sensitivity of 83.3%. The fifth meta-analysis reported results using different thresholds [9].

Our study has several strengths. We aimed to use data on average risk participants only and collected data regarding all polyps, (advanced) adenomas and CRC. This provided the possibility to estimate the diagnostic value of CT-colonography for adenomas and CRC in a screening setting. In order to perform an unbiased study selection, two reviewers independently selected possible relevant articles.

Our study also has several limitations. Although we tried to include only individuals at average risk, we could not obtain these data from one study [30, 31]. Therefore, 37 individuals (0.9%) at high risk were included. However, it is assumable that this will be daily practice in screening and it is unlikely that this small number will have a substantial impact on the results.

Secondly, participants of two studies comprised the majority of included participants, which might give the impression that this meta-analysis is actually a two study meta-analysis. However, the results of the two largest studies were heterogeneous and, moreover, were not at one end of the spectrum of the sensitivity or specificity range. Therefore it is unlikely that the larger studies skewed the results in one direction (of higher or lower values). Furthermore, sensitivity and specificity estimates were calculated using statistical analyses in which the individual studies are weighted by number of included participants [18].

Thirdly, we did not calculate the negative predictive value (NPV) because the prior probability of a negative outcome was high [34].

Fourthly, it is known that colonoscopy is not 100% sensitive for colorectal lesions and therefore no perfect reference standard [35]. Using the colonoscopy results after segmental unblinding and compared with histology, would be the best reference standard.

Fifthly, because of limited data we were not able to calculate estimated sensitivities per patient for the detection of advanced adenomas, advanced neoplasia and CRC.

In summary, this meta-analysis of prospective studies studying the diagnostic value of CT-colonography compared to colonoscopy in an average risk population, shows that CT-colonography has a good sensitivity for (advanced) adenomas ≥ 10 mm. For (advanced) adenomas ≥ 6 mm sensitivity is somewhat lower.