A variety of tests have been proposed for colorectal cancer (CRC), giving rise to uncertainty regarding the optimal approach. The efficacy and effectiveness of different tests are related to both screenee participation and the detection rate.
To perform a meta-analysis on adherence and detection rates of CRC screening tests.
Relevant publications were identified by MEDLINE/EMBASE and other databases for the period 1999–2012. A previous systematic review was used for the period before 1966–1999. RCTs and controlled studies including a direct comparison of the uptake rates among different options for CRC screening were included. Adherence and detection rates for advanced neoplasia and cancer were extracted. Risk for bias was ascertained according to CONSORT guidelines. Forrest plots were produced based on random-effect models.
Fourteen studies provided data on 197 910 subjects. Endoscopic strategies were associated with a lower participation (RR: 0.67, 95% CI: 0.56, 0.80) rate, but a higher detection rate of advanced neoplasia (RR: 3.21, 95% CI: 2.38, 4.32) compared with faecal tests. FIT was superior to g-FOBT with regard to both adherence (RR: 1.16, 95% CI 1.03, 1.30) and detection of advanced neoplasia (RR: 2.28, 95% CI 1.68, 3.10) and cancer (RR: 1.96, 95% CI: 1.2, 3.2).
The superior accuracy of endoscopy compared with faecal tests minimised any impact of the participation rate in determining the detection rate of advanced neoplasia in a screening setting.
Colorectal cancer (CRC) represents a major cause of morbidity and mortality in Western countries, and results in a substantial economic burden due to costs for surgery, chemotherapy and terminal care.[1, 2]
Several screening tests are effective in reducing CRC incidence and/or mortality, and population screening is widely recommended in Europe and in the United States.[3, 4] Among the available options, screening with guaiac-based faecal occult blood test (g-FOBT) was associated with a 13–18% CRC-mortality reduction in large, randomised studies (5–8). This mortality reduction resulted primarily from detection of early stage CRC.[5-8] The immunochemical faecal occult blood test (FIT) is now recognised to have better sensitivity than and similar specificity to g-FOBT for detecting advanced neoplasia, as FIT specifically detects human haemoglobin. Faecal tests have the advantage of being relatively simple and safe screening tests, suitable for a mass screening programme. However, because of poor sensitivity for premalignant lesions, faecal tests need to be repeated every 1–2 years. Therefore, high compliance to repeated testing is required to achieve long-term effectiveness with faecal tests.
Unlike faecal tests, endoscopic screening (flexible sigmoidoscopy (FS) and optical colonoscopy (OC)) is claimed to be highly effective in preventing both CRC incidence and mortality. Incidence reduction with endoscopic screening is attributed to high sensitivity for premalignant polyps, including advanced adenomas.[10, 11] Two high-quality randomised trials demonstrated the efficacy of flexible sigmoidoscopy in reducing CRC incidence and mortality by 31–33% and 38–43% in screen-attendant average-risk subjects.[12, 13] Although not yet validated in randomised trials, screening OC has also been found effective in reducing CRC incidence and mortality in both cohort and case–control studies.[14, 15] Because of its potential for preventing incident CRC, endoscopic screening might be performed only once in a lifetime or repeated infrequently at 5–10 years intervals.[3, 4, 11] However, endoscopic screening is perceived as more invasive and embarrassing than faecal tests, and has a small, but definite, risk of complications.
The effectiveness of screening depends not only on the sensitivity for colorectal neoplasia but also on population attendance. Low participation rates dilute the intrinsic efficacy of CRC screening tests, reducing the overall yield for advanced neoplasia in a population setting. The impact of adherence on the eventual effectiveness of any screening strategy has been confirmed by simulation modelling, in which apparently large differences in efficacy were shown to be reversed by small gradients in adherence rates.[17, 18] Therefore, a societal decision maker confronted with a choice of alternative tests for preventing CRC incidence and/or mortality should be expected to choose the strategy with the most efficient compromise between adherence and efficacy, i.e. the highest effectiveness. Uncertainty in this choice will appear to be mainly related to the technical and procedural differences among the available tests, primarily between faecal tests, on one hand, and endoscopic strategies, on the other. The aim of this meta-analysis was to compare the participation rates among different CRC screening options, and to assess the effect of such differences on the detection rates for advanced neoplasia and cancer.
The methods of the analysis and inclusion criteria were based on PRISMA recommendations. The present review represents a nested analysis of a more general systematic review on the methods to increase participation in screening programmes, including colorectal/breast/cervical cancer screening. Such a general review was conceived as an update of a previously published systematic review by Jepson et al. in 2000, including articles published between 1966 and 1999. We decided to limit the present review to CRC screening, because the availability of different types of screening tests results in greater uncertainty regarding the optimal approach.
We considered all randomised control trials (RCTs) and controlled studies for the period from January 1999 (i.e. not included in Jepson et al. analysis) to February 2012 in which adherence rates of different options for CRC screening were directly compared. As our meta-analysis was conceived as un update of the previous analysis, the systematic review by Jepson et al. was adopted for retrieving the studies published before 1999. Studies reporting efficacy data without information on adherence were excluded, as were those without direct comparison between two or more type of tests (i.e. when the only control group was no screening), those in which the participant was offered a choice among multiple different strategies or with procedures not currently recommended, and those performed in a nonscreening setting. If there was any suspicion of cohort overlap between studies, only the most recent study was considered for inclusion.
A literature search was performed in February 2012. Relevant publications were identified by the following database: MEDLINE, EMBASE, PsycINFO, NHS CRD DARE, Cochrane Database of Systematic Reviews, Cochrane Controlled Trials Register. To identify eventual grey literature, all the official website of the European Health Ministers, IARC and European Union were searched. Additional references were located through searching the bibliographies of related papers. There were no language restrictions, and both published and unpublished studies were included if they met the criteria of this review, which are described below. Medical terms early detection, screening, depistage, neoplasm, cancer, oncologic, neoplasia, colorectal, colon, compliance, adherence, adhesion, attendance and participation were used in the search.
Titles and abstracts of papers were screened by one reviewer, with a random sample (5%) of included and excluded papers checked by a second reviewer. Any disagreements were resolved through discussion. Studies were independently prescreened for relevance by two reviewers (using the full report of the study). Any disagreements were resolved through discussion. Data were extracted from relevant studies by one reviewer and checked by a second reviewer. Data from the included studies were extracted into extraction tables. Any disagreements were resolved through discussion with a third reviewer. Information was also recorded relating to the methodological quality of each study.
Data collection process and list of items
From each report, reviewers independently abstracted (i) year of publication; (ii) country where the study was performed; (iii) whether it was a single- or multi-centre study; (iv) study design (RCT/controlled); (v) population source; (vi) number of patients included; (vii) number of patients randomised/eligible; (viii) mean age and (ix) sex distribution; (x) types of interventions compared; (xi) type of initial invitation; (xii) means to increase adherence; (xiii) uptake for each intervention at intention-to-treat analysis (ITT); and, when available, (xiv) detection rate for advanced neoplasia (cancer or adenoma > 10 mm or with high-grade dysplasia or villous component); and/or (xv) cancer at both per-protocol (PP; i.e. test-related detection rate unadjusted by adherence) and ITT (i.e. detection rate adjusted by adherence) analyses for each intervention.
Risk of bias in individual studies
Information was recorded relating to the methodological quality of each study. Quality assessment was performed using CONSORT checklist for RCTs and for clustered RCTs, as appropriate.
This study aimed to address the following questions:
When equally offered to a screening population, do the screening tests differ with regard to adherence to CRC screening?
Do differences in adherence between tests affect the eventual detection rate of advanced neoplasia/cancer in a screening population?
Planned methods of analysis
Data for the difference in uptake and detection rates among different interventions were extracted where possible. Only studies with two or more relevant interventions (i.e. direct comparisons only) were included. Relative risks (RR) and 95% CI were calculated only for RCTs; data for all other studies were reported descriptively. Data on uptake/detection rate were entered into the RevMan 5.0 software (Cochrane library, RevMan, Version 5.1, The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2011) and a random-effects model was used. Forrest plots were separately built up for g-FOBT and FIT, to provide a head-to-head comparison among individual tests. A P-value of less than 0.05 was considered statistically significant. Heterogeneity was assessed using the I2 statistic. 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. Values of I2 below 25%, between 26% and 74%, and above 75% were considered to represent low, moderate and high heterogeneity respectively.
To explore the effect of adherence on the detection rate of advanced neoplasia/cancer, we created for each comparison between two tests, the following Forrest plots: (i) comparison between the adherence rates of the two tests; (ii) comparison between the detection rates for advanced neoplasia/cancer of the two tests by assuming a hypothetical 100% adherence (PP analysis; i.e. detection rate in those actually screened); (iii) comparison between the detection rates for advanced neoplasia/cancer adjusted for the adherence rate (ITT analysis; i.e. impact on the whole invited population in terms of detected lesions). It was postulated that adherence would meaningfully affect the detection rate when the apparent superiority of one test at PP analysis was reversed at ITT analysis, because of the effect of an adherence gradient.
A flow diagram of this meta-analysis, with the number of papers retrieved, included and excluded, as well as the reasons for exclusion, is shown in Figure 1. In summary, 6846 studies were identified by the included databases, 2656 were dealing with colorectal cancer. After nonpertinent papers were removed, 63 were regarded as relevant. Of these, 12 fulfilled the inclusion criteria, while 51 were excluded (Appendix S1), either because they did not compare different type of tests, used obsolete or not officially recommended tests or protocols, or focused on specific subgroup populations. Four other relevant studies were found in the Jepson et al. systematic review, of which two were excluded (1 study mostly included CRC relatives, 1 study because of duplication). Finally, 14 published papers were selected for inclusion in the meta-analysis.[23-36]
Characteristics of the included studies
The main characteristics of the included studies are provided in Tables 1-3. As shown in Table 1, all the included studies were RCTs, and all but one were multicentre. All but three studies were performed in European countries. The study population was invited from the general population or population registers in all but one case, in which subjects were extracted from a Veteran Affairs database. The invitation consisted of a letter in all the studies. In some studies employing faecal tests, the test (i.e. Haemoccult) was also sent by mail, while in other studies, it was not.
Table 1. Main characteristics of the included studies
Year of publication
Population source (No. of GP register)
Type of invitation
g-FOBT, guaiac-faecal occult blood test; GP, General Practitioner; UK, United Kingdom; US, United States of America; w/out, without.
A total of 197 910 subjects were randomised in the selected studies (Table 2). Of these, 182 459 were considered eligible (i.e. in some studies, exclusion occurred after randomisation). The number of subjects enrolled for study ranged from 404 to 53 302 with a median of 7855. All studies enrolled subjects older than 50 years with age ranges similar among the selected series. Both genders were also similarly distributed among the series, with the exception of one study in which most of the subjects were male. As shown in Table 3, all but one study also reported information on the detection rate for advanced neoplasia and/or cancer. The definition of advanced neoplasia was limited to ≥ 10 mm adenomas in 4 series, while the remaining studies included smaller adenomas with unfavourable histology.
Five studies compared g-FOBT with FIT, including 59 729 randomised subjects. Faecal tests were compared with endoscopic strategies (with or without the addition of faecal tests) in 10 series (153 192 subjects). In detail, faecal tests were compared with FS in 6 studies (66 871 subjects), with a combination of faecal tests and FS in 4 studies (21 093 subjects), and with OC in 4 studies (87 419 subjects). A combination of FS and faecal test was also compared with FS and OC in two studies (5423 subjects), while FS was compared with OC in one study (18 477 subjects).
The overall rate of attendance among the selected studies was 33.3% (60 767/182 459 eligible subjects). Overall rates of advanced neoplasia and cancer at PP and ITT analyses were 3.3% (2014/61 882), 0.2% (205/97 339), 1.1% (2014/182 055), and 0.1% (205/174 051) respectively (Table 3).
Risk of bias in individual studies (Appendix S2)
Randomisation procedures were detailed for all but two studies.[23, 34] These procedures may be regarded as valid as randomisation was performed before the invitation on the whole target population without any knowledge about screening attitudes and CRC risk of the subjects. The allocation procedures were poorly described across all the studies. This may be partially explained by the fact that randomisation and allocation were performed at the same time on the list of target population in most of the studies. This should have prevented any risk of allocation bias. None of the studies masked the intervention to either the target population or the physicians, while two studies masked the final assessment of histological results.[27, 33] Only one study asked for consent before randomisation. Four studies reported using a cluster randomisation,[26, 28, 33, 35] but only two accounted for it in the analysis.[26, 28] Although 6 studies adopted algorithms in the randomisation that forced people from the same household or at the same address to be allocated to the same arm,[23, 24, 27, 30, 33, 36] none of them accounted for it in the analysis (the effect on variance of this clustering is usually irrelevant, as usually only one subject per household and rarely two are eligible). Although in one study a cross-over between the two groups (FIT and OC) was allowed for noncompliant subjects, it was limited to a marginal fraction of the study population.
Synthesis of results (see also Appendix S3)
g-FOBT vs. FIT
As shown in Figure 2 and Appendix S5, FIT resulted in a higher uptake compared with g-FOBT (RR: 1.16; 95% CI: 1.03, 1.3). Inter-study heterogeneity (I2) was 96%. Such heterogeneity appeared to be related only to one series, its exclusion resulting in a I2 equal to 0%. The detection rate for advanced neoplasia and cancer with FIT was also superior to g-FOBT at both PP (RR: 1.94, 95% CI: 1.37, 2.76, I2: 56%; RR: 1.67, 95% CI: 1.01, 2.8, I2: 0%) and ITT analyses (RR: 2.28, 95% CI: 1.68, 3.10, I2: 43%; RR: 1.96, 95% CI: 1.2, 3.2, I2: 0%).
Endoscopy vs. g-FOBT/FIT
As shown in Figure 3 and Appendix S5, endoscopic tests were associated with a lower uptake compared with faecal tests (RR: 0.67, 95% CI: 0.56, 0.80; I2: 99%). However, the detection rate for advanced neoplasia with endoscopic strategies was superior at both PP (RR: 4.65, 95% CI: 3.67, 5.90, I2: 64%) and ITT analyses (RR: 3.21, 95% CI: 2.38, 4.32, I2: 77%). Similarly, the detection rate for cancer with endoscopic strategies was superior at PP (RR: 2.08, 95% CI: 1.40, 3.10, I2: 23%), with a borderline significance at ITT analysis (RR: 1.58, 95% CI: 0.97, 2.56, I2: 44%; P = 0.07). Nevertheless, there is a decreasing trend in the RR for detection rate at increasing severity of the lesion. When subgrouping according to the type of faecal tests, endoscopy was superior to both g-FOBT and FIT in terms of detection rate for advanced neoplasia at both PP and ITT analyses. Although one series was reported twice (i.e. FS vs. g-FOBT and FS vs. FIT), the duplication of the sample size did not affect the results. Indeed, when excluding either of the two comparisons, the study results were practically unchanged (see Appendix S4).
Sigmoidoscopy (FS) vs. g-FOBT/FIT
Sigmoidoscopy produced a nonsignificant lower participation rate compared with faecal tests (RR: 0.78, 95% CI 0.59, 1.04; I2: 99%). The detection rate for advanced neoplasia was higher with sigmoidoscopy compared with g-FOBT/FIT at both PP (RR: 3.96, 95% CI 2.86, 5.49, I2: 62%) and ITT analyses (RR: 3.2, 95% CI 1.97, 5.19, I2: 82%). On the other hand, the detection rate for cancer was not different between the two groups at both PP and ITT analyses.
Sigmoidoscopy with g-FOBT/FIT (FS + g-FOBT/FIT) vs. g-FOBT/FIT
g-FOBT/FIT resulted in higher attendance compared with FS + g-FOBT/FIT (RR: 0.62, 95% CI: 0.42, 0.91, I2: 99%). The detection rate for advanced neoplasia was superior with FS + g-FOBT/FIT at both PP (RR: 5.10, 95% CI 2.51, 10.35, I2: 45%) and ITT analyses (RR: 3.12, 95% CI 1.62, 6.01, I2: 38%), while the detection rate for cancer was superior at PP analysis (RR: 3.45, 95% CI 1.40, 8.50), but not at ITT analysis.
Colonoscopy (OC) vs. g-FOBT/FIT
g-FOBT/FIT resulted in higher attendance compared with OC (RR: 0.57, 95% CI 0.42, 0.78, I2: 99%). On the other hand, detection rate for advanced neoplasia was superior for OC at both PP (RR: 5.91, 95% CI 3.35, 10.42, I2: 76%) and ITT analyses (RR: 3.56, 95% CI 1.79, 7.09, I2: 83%), whilst only a trend was found for cancer at both PP and ITT analyses.
Two comparisons – namely, OC vs. FS and FS + g-FOBT vs. FS – were performed only in one study each.[24, 30] Corresponding Forrest plots are reported in Appendix S3.
According to our analysis of nearly 200 000 randomised subjects, the higher uptake of faecal tests is unable to reverse the superiority of endoscopic strategies in detecting advanced neoplasia in a screening setting. Thus, although faecal tests produced a 32% increase in the participation rate relative to endoscopy, the absolute difference in participation was less than 10%. These results indicate that more user-friendly tests are needed before adherence can provide a significant advantage in terms of efficacy. In detail, the detection rate of advanced neoplasia with endoscopy was 3 fold higher than that of faecal tests. This large difference suggests that offering faecal tests instead of endoscopy procedures would displace a substantial proportion of subjects from the potential identification of advanced neoplasia. Indeed, two recent high-quality trials have shown the efficacy of endoscopy – namely, sigmoidoscopy – in preventing CRC. It can be argued that faecal tests mainly aim at down-staging prevalent cancers rather than at detecting advanced adenomas. However, our review showed that endoscopy tended to be superior to faecal tests also at detecting prevalent cancers. In other words, the advantage of endoscopy compared with faecal tests is not limited to prevention of incident CRC, but would include the possibility of higher mortality reduction related to down-staging of prevalent cancer. On the other hand, the included studies compared endoscopic procedures with only the first round of faecal tests, and faecal tests might detect additional cases of advanced neoplasia or cancer at subsequent rounds. Also, the detection rate of faecal tests may have been partly affected by suboptimal compliance with posttest colonoscopy.
The superiority of endoscopy over faecal tests shown by the head-to-head comparison between the two types of strategies was further confirmed by the internal comparison between faecal tests and the addition of sigmoidoscopy to such tests. Despite lower adherence, the addition of sigmoidoscopy to faecal tests increased the detection rate for advanced neoplasia by 3–5 fold. However, considerations other than the efficacy – such as budget, human resources and ethical equity issues – may affect the final choice between endoscopy and faecal tests.
Our analysis found that FIT was superior to g-FOBT in terms of both population uptake and the detection rates for both advanced neoplasia and cancer. Of note, when excluding the only study which required three stool samples rather than one, there was no heterogeneity in our estimate of attendance, indicating that FIT superiority is quite independent from the population setting. The 2-fold higher detection rate for both advanced neoplasia and cancer at ITT analysis would support a preference for FIT vs. g-FOBT, as recommended by the American College of Gastroenterology. This preference is also reinforced by the marginal difference in cost between the two tests.
Our analysis was unable to draw definitive conclusions regarding the comparison between sigmoidoscopy and colonoscopy, because of the availability of only one direct comparison. However, when considering sigmoidoscopy and colonoscopy compared with faecal tests, colonoscopy appears associated with a greater increase in detection rate for advanced neoplasia compared with sigmoidoscopy at both PP and ITT analyses. However, this was only partially confirmed by the only available direct comparison of the two endoscopic tests, and therefore further comparative studies between the two tests are needed.
There are limitations to our analysis. The quality assessment identified multiple pitfalls in the design of the included RCTs, such as lack of masking or informed consent. However, such biases appear to be mainly due to the design of pharmacological trials for which the CONSORT check list was designed, rather than with methodological deficiencies of the included screening series. Secondly, there was heterogeneity in several of the study estimates, especially when dealing with the uptake rate. The large size of the trials and the strong effect of cultural, social and organisational context on participation probably explain most of the inter-study heterogeneity. However, the main differences observed in our analysis were large and hence the observed heterogeneity is unlikely to have public health relevance. We acknowledge that our results might be reversed when comparing a multiple round strategy using faecal tests with once-in-a-life endoscopy. Indeed, FIT has been shown to have a relatively high positive predictive value for advanced neoplasia at subsequent rounds, although reduced compared with the initial round.[38, 39] Finally, the end-point of this study was the detection of advanced neoplasia. Although the latter has been universally adopted as a critical endpoint when validating a CRC screening technique, its role in CRC carcinogenesis has not been fully elucidated. Certainly, some of the undetected advanced neoplasia would not progress to malignancy before the repetition of the screening test.
Our results could have significance for the design of national colorectal cancer screening programmes and the approach to screening alternatives taken in screening guidelines. For example, countries that can afford to offer either endoscopic or faecal tests (but not both) would choose endoscopic tests if the goal is to maximise detection of colorectal neoplasia. Similarly, countries that can afford to provide both endoscopic screening and faecal testing might create a programme that encourages endoscopic screening, with faecal testing reserved for screening candidates who decline endoscopy. In the US, there is no national colorectal screening programme, and the US screening guidelines differ in their recommendations for how clinicians should approach the availability of multiple screening options. Thus, some guidelines present multiple options for screening with no preference for one of the screening options. For example, the US Preventive Services Task Force recommends that colonoscopy, sigmoidoscopy, barium enema and FOBT are all appropriate options for screening. Our data suggest that presentation of multiple options could reduce the impact of screening if FOBT is presented to patients as an equivalent alternative to colonoscopy or sigmoidoscopy, as undergoing FOBT could displace patients from endoscopic screening who might be willing to undergo endoscopy. With regard to efficacy, our results favour an approach of offering endoscopic screening first, and reserving alternative screening tests for candidates who decline screening. These conclusions are tempered by the reality that some countries may only be able to afford FOBT for screening. In addition, more effective screening tests such as colonoscopy have greater risks, and the risks of procedures may be of great importance to patients.
According to our meta-analysis, the substantially higher accuracy of endoscopy minimised any impact of higher participation rates for faecal tests in the detection rate of advanced neoplasia in a screening setting.
Declaration of personal interests: None. Declaration of funding interests: This study was partially funded by the Italy Ministry of Health, through a project coordinated by the Agenzia Nazionale per i Servizi Sanitari and conducted by Laziosanità: ‘Strumenti e metodi per il governo dei processi di innovazione tecnologica, clinica ed organizzativa nel SSN – Un sistema integrato di ricerca’, sub-project ‘Analysis of the impact of professional involvement in evidence generation for the HTA process’ grant no. I85J07000080001.