Article Text
PDF
Abstract
Background Between 2000 and 2007, a demonstration pilot of biennial guaiac faecal occult blood test (GFOBT) screening was carried out in Scotland.
Methods Interval cancers were defined as cancers diagnosed within 2 years (ie, a complete screening round) of a negative GFOBT. The stage and outcome of the interval cancers were compared with those arising contemporaneously in the non-screened Scottish population. In addition, the gender and site distributions of the interval cancers were compared with those in the screen-detected group and the non-screened population.
Results Of the cancers diagnosed in the screened population, interval cancers comprised 31.2% in the first round, 47.7% in the second, and 58.9% in the third, although this was due to a decline in the numbers of screen-detected cancers rather than an increase in interval cancers. There were no consistent differences in the stage distribution of interval cancers and cancers from the non-screened population, and, in all three rounds, both overall and cancer-specific survival were significantly better for patients diagnosed with interval cancers (p<0.01). The percentage of cancers arising in women was significantly higher in the interval cancer group (50.2%) than in either the screen-detected group (35.3%, p<0.001) or the non-screened group (40.6%, p<0.001). In addition, the proportion of both right-sided and rectal cancers was significantly higher in the interval cancer group than in either the screen-detected (p<0.001) or non-screened (p<0.004) groups.
Conclusions Although GFOBT screening is associated with substantial interval cancer rates that increase with screening round, the absolute numbers do not. Interval cancers are associated with a better prognosis than cancers arising in a non-screened population, and GFOBT appears to preferentially detect cancers in men and the left side of the colon at the expense of cancers in women and in the right colon and rectum.
- Colorectal cancer
- screening
- cancer registries
- cancer epidemiology
- cancer prevention
- colorectal adenomas
- colorectal carcinoma
- Helicobacter pylori
- acid-related diseases
- non-ulcer dyspepsia
- genetic polymorphisms
- gastric neoplasia
- colorectal cancer genes
- microsatellite instability
- cell cycle
- gastrointestinal neoplasia
- molecular pathology
- screening
Statistics from Altmetric.com
- Colorectal cancer
- screening
- cancer registries
- cancer epidemiology
- cancer prevention
- colorectal adenomas
- colorectal carcinoma
- Helicobacter pylori
- acid-related diseases
- non-ulcer dyspepsia
- genetic polymorphisms
- gastric neoplasia
- colorectal cancer genes
- microsatellite instability
- cell cycle
- gastrointestinal neoplasia
- molecular pathology
- screening
Significance of this study
What is already known about the subject?
Population screening using guaiac faecal occult blood testing (GFOBT) reduces colorectal cancer-specific mortality.
Interval cancers arise in biennial GFOBT-based screening programmes.
What are the new findings?
Interval cancer rates are substantial, but, although they increase from round one to round three, this is due to a decrease in numbers of screen-detected cancers rather than an increase in interval cancers.
Interval cancers have a relatively good prognosis compared with cancers arising in a similar population that has not been offered screening.
When compared with screen-detected cancers, interval cancers are more likely to arise in women and in the right colon or rectum.
How might it impact on clinical practice in the foreseeable future?
Modifications to current screening algorithms will be necessary to reduce interval cancer rates, and, in FOBT screening, differential cut-off values may be necessary in men and women.
Introduction
In a cancer screening programme that involves repeat testing, the term ‘interval cancer’ is used to describe a cancer that is diagnosed after a negative screening test result and before the date of the next screening test.1 In the Scottish Bowel Cancer Screening Programme, which is based on biennial guaiac-based faecal occult blood testing (GFOBT) as the initial test, an interval cancer is defined as a cancer that is diagnosed within 2 years of a negative screening test result.2
While it is conceivable that an interval cancer may arise from normal mucosa in the interval between two screening tests, this is likely to be a rare event, and the majority of interval cancers are probably cancers that have been missed by the screening test either at an invasive or non-invasive (adenoma) stage. Thus the interval cancer rate is an important performance indicator of a population-based screening programme, reflecting the sensitivity of the test.
GFOBT has been shown to reduce disease-specific mortality in five independent population-based trials,3–7 but these studies provide little information on interval cancer rates. In Scotland, before the roll-out of the national screening programme, three rounds of a biennial GFOBT screening demonstration pilot were carried out between 2000 and 2007 for all residents of Grampian, Tayside and Fife aged between 50 and 69 years.2 In the present study, data from this pilot were used to examine the interval cancer rate and characteristics of interval cancers.
Methods
The demonstration pilot in Scotland was based on the Nottingham randomised trial,4 as this was considered by the UK National Screening Committee to be the most appropriate for the UK population. The methodology used for the pilot has been described in detail in earlier publications.2 8 9 Briefly, screening in Scotland was carried out in Tayside, Grampian and Fife (with a total population of about 1.3 million) and was based on biennial GFOBT using a test kit (Hema-screen; Immunostics Inc, Ocean, New Jersey, USA) biochemically identical with that used in the randomised trials carried out in Nottingham, England,4 Funen, Denmark5 and Göteborg, Sweden.6 The test was designed to collect two samples from each of three separate bowel motions on to guaiac gum-impregnated filter paper through windows in the test kit card. The card was developed and read within 14 days in a laboratory specifically set up for the pilot and accredited to ISO 15189 standards. If five to six windows were positive on the initial GFOBT (strong positive), colonoscopy was offered. If one to four windows were positive (weak positive), a further GFOBT was requested, and, if any of the subsequent windows were positive, a colonoscopy was offered. Minor changes to this algorithm took place during the pilot and these have been detailed elsewhere.2
Male and female residents in the three pilot NHS Boards aged between 50 and 69 years were identified by the Community Health Index and invited to participate by having a test kit and instructions sent in the mail. Community Health Index acts as a unique identification number for every person registered with a general practitioner or having some contact with the NHS in Scotland. It is made up of date of birth followed by a four-digit number as a unique identifier.
For the purposes of this study, the following definitions were used: a screen-detected cancer was defined as a cancer diagnosed at colonoscopy triggered by a final positive screening test result; an interval cancer was defined as a cancer diagnosed within 2 years of a negative test result; a missed cancer was defined as a cancer diagnosed within 2 years of a negative colonoscopy triggered by a positive test result; a non-screened cancer was defined as a cancer arising in the Scottish population not offered screening (ie, in a resident outside the pilot NHS Boards) during the same time period and in the same age range (50–69 years).
In order to categorise the cancers as above, all screened records from the demonstration pilot were linked with confirmed colorectal cancer records in the Scottish Cancer Registry (SCR) for the time periods relating to each round. This allowed the calculation of the time between the final screening test result and the date that the colorectal cancer was diagnosed. Cases where the individual defaulted before a complete GFOBT, refused investigation treatment, was declared unfit for colonoscopy, was undergoing symptomatic investigations, was admitted as an emergency, or died between screening investigations were categorised as ‘miscellaneous’, and were excluded from analysis.
Cancers arising in individuals who chose not to participate in the screening programme were not considered in this study. The prognosis associated with such cancers is already known to be particularly poor,4 presumably as they arise in people who are likely to have poor general health by virtue of their non-attendance.
The Dukes' staging system was used to describe the stage of the cancers studied as recorded by the SCR. Although polyp cancers were identifiable separately in the screen-detected group, it was not possible to do this reliably from the registry data. All polyp cancers were therefore included in the stage A category for the purpose of this analysis. Survival times were calculated for all individuals with colorectal cancers diagnosed within the time periods of the three rounds of the demonstration pilot (with life status as at 31 December 2009). For these groups, survival curves were estimated for all causes of death and for colorectal cancer deaths.
The SCR uses multiple sources, and routine indicators of data quality suggest that case ascertainment is very nearly complete. One study carried out in 1992 yielded an estimate of completeness of ascertainment of colorectal cancers of 98.5%,10 and a more recent study of breast cancer spanning the period 1978–2000 obtained similar results.11
Results
The demonstration pilot started on 29 March 2000, and the first round finished on 31 December 2002. The dates for the second and third rounds were 2 December 2002 to 30 April 2005 and 6 May 2005 to 20 July 2007, respectively. In the first round, 304 245 invitations were issued, and a final result was obtained in 167 415 (uptake of 55.0%); these numbers were 309 803 and 164 077 (uptake of 53.0%) in the second round and 317 864 and 175 853 (uptake of 55.3%) in the third. In the population that was screened and had a final test result, 618 cancers were diagnosed within 2 years of their test result date in those screened in the first round, 447 in the second round, and 389 in the third round. In the rest of Scotland, where screening was not being offered, 2416 cancers were diagnosed in the 50–69 age group in the time period occupied by the first round. These figures were 2196 and 2094 during the second and third rounds, respectively.
In the first round, 31.2% of all cancers diagnosed in the population that had a final screening test result were interval cancers. In the second and third rounds, these figures were 47.7% and 58.9%, respectively (table 1). These data are slightly different from those previously published2 because cancer registration is a dynamic process and information in the registry has been updated as new clinical information has become available.
Breakdown of all cancers diagnosed in the populations screened in the three rounds
When stage at diagnosis was examined, the distribution for the screen-detected cancers was significantly more favourable than in either the interval cancers or the non-screened cancers in all three rounds. When the interval cancers were compared with the non-screened cancers, the stage distribution was the same in the first round, but less favourable in the interval cancers in the second round owing to a significantly higher proportion of stage C cancers. However, this was reversed in the third round, with interval cancers displaying a more favourable distribution (table 2).
Dukes' stage at diagnosis of screen-detected, interval and non-screened cancers
The gender distributions in screen-detected, interval and non-screened cancers were similar for all three rounds, and in consequence the three rounds were pooled for further analysis. The proportion of cancers diagnosed in women in the screen-detected group was significantly lower than that seen in both the interval and non-screened groups, and the proportion of interval cancers diagnosed in women was significantly higher than that seen in the non-screened group (table 3).
Gender distribution of screen-detected, interval and non-screened cancers in all three rounds
Likewise, the anatomical site distributions of the screen-detected, interval and non-screened cancers were similar in the three rounds, so the rounds were analysed together. In the screen-detected group, the proportions of cancers arising on the right side of the colon and in the rectum were lower than in the interval group and non-screened group (ie, the non-screened general Scottish population), and the proportions of right-sided and rectal cancers in the interval group were higher than those seen in the non-screened group (table 4). When broken down by gender, right-sided cancers were significantly more common in women than in men, and rectal cancers were significantly more common in men than in women. This held for screen-detected, interval and non-screened cancers (table 5).
Anatomical site distribution of screen-detected, interval and non-screened cancers in all three rounds
Site and gender distribution of screen-detected, interval and non-screened cancers in all three rounds
Both overall and cancer-specific survival in the interval cancers was compared with that for the cancers arising in the non-screened population, and, in all three rounds, both were significantly better in the interval cancers (figures 1 and 2).
Comparison of overall survival of patients with interval cancers who had been screened in the first three rounds of the Scottish demonstration pilot with those in the same age range (50–69) who developed cancer in the non-screened population of Scotland in the same time periods. χ2 values and significance levels calculated by the Mantel–Cox log rank test.
Comparison of colorectal cancer-specific survival of patients with interval cancers who had been screened in the first three rounds of the Scottish demonstration pilot with those in the same age range (50–69) who developed cancer in the non-screened population of Scotland in the same time periods. χ2 values and significance levels calculated by the Mantel–Cox log rank test.
Discussion
In the Scottish demonstration pilot of colorectal cancer screening, the percentage of cancers diagnosed in the screened population that were true interval cancers rose from 31.2% in the first round to 58.9% in the third. It is important to appreciate, however, that the absolute numbers of interval cancers in each of the three rounds varied very little, suggesting that the main reason for the increase in the percentage of interval cancers over the three rounds was the previously documented fall in the positive predictive value of the GFOBT between the first round and the second and third rounds.2 Thus screening reduces the prevalence of cancers that are susceptible to detection by GFOBT screening, but has no effect on screening-resistant cancers.
It is also important to stress that this study was based on three separate rounds of screening and that each round was treated in isolation. Thus an interval cancer arising in the second round was defined on the basis of a negative test result in the second round irrespective of the result in the first round, or, indeed, of whether or not that individual had been screened in the first round. This is because the complexity of a dynamic screening programme12 is such that any other approach would be impractical.
The rate of interval cancers in this study is very much in keeping with the little that is already known. In previous population-based controlled studies that were carried out over multiple rounds, interval cancers were defined as cancers diagnosed after a negative GFOBT result within the interval between two screenings for rescreened subjects and at any time for those who were not rescreened, and included those where further investigation (usually colonoscopy) was negative or had been refused. By these criteria, the interval cancer rate was 51.3% in the Nottingham Study,4 55.2% in the Funen Study,5 and 59.3% in the Burgundy Study.7 Putting all three rounds of the Scottish pilot together and applying the same criteria, the interval cancer rate was comparable: 52%. Interestingly, in Finland where population screening is being rolled out on a randomised basis, the interval cancer rate after the first round was 27.3%,13 very similar to the first round figures from the Scottish pilot.
It could be argued that cancers diagnosed as a result of a positive screening test result after a negative result in the previous round may behave in a manner similar to interval cancers, as it is likely they were present, in some form, when the previous test was completed. Indeed, we have shown that cancers diagnosed at first and second incidence screening do have a less favourable stage distribution than those diagnosed at prevalence screening,12 supporting this contention. However, GFOBT screening has only been shown to be effective when used in an iterative manner, and this phenomenon appears to be an unavoidable consequence of a programme that is based on repeated invitations. The most important question is whether or not true interval cancers, which arise fairly quickly after a negative test result, are particularly advanced or aggressive, as this may negate the value of the screening programme.
Although, as expected, the stage at diagnosis was more favourable in the screen-detected cancers than in either the interval or non-screened cancers, the differences between the stage distribution of interval cancers and non-screened cancers were small and inconstant between rounds. Furthermore, despite these small differences, both overall and colorectal cancer-specific survival were significantly better for interval cancers than for cancers arising in the non-screened population. This is very likely to be due to selection bias since, by definition, interval cancers arise in people who have engaged in the screening process and are therefore likely to represent a population that is generally healthier than average. Again, this finding is not novel; similar findings were reported by the Nottingham,4 Funen5 and Burgundy7 studies. However, we do believe that this is worth emphasising, as it reinforces the concept that individuals who develop interval cancers are not significantly disadvantaged compared with a population that has not been offered screening. Theoretically, this could have been the case if false negative test results produced a substantial ‘certificate of health’ effect leading to delayed diagnosis caused by false reassurance. It is also conceivable that interval cancers may have different biology from screen-detected cancers, which makes them less likely to bleed, and it is reassuring that this does not appear to translate into a particularly aggressive phenotype.
It is already known that both the positivity rate and the cancer detection rate of GFOBT screening are higher in men than in women.14 15 It is also well established that both the incidence and prevalence of colorectal cancer is higher among men than among women,16 17 and this could at least partially explain these findings. However, the data from the Scottish pilot presented here demonstrate that the proportion of screen-detected cancers in women was less than in both interval cancers and cancers arising in the non-screened population. In addition, the proportion of interval cancers in women was greater than that seen in the non-screened population. Thus, in comparison with the non-screened population, there is a paucity of women in the population with screen-detected cancers, and the population with interval cancers shows an excess of women, strongly suggesting that GFOBT is less sensitive for cancer in women than in men.
In a recent study of men and women undergoing both quantitative faecal immunochemical testing (FIT) and colonoscopy, as part of the German Colorectal Cancer Screening Programme, it was found that, at any cut-off for haemoglobin concentration, sensitivity and positive predictive value were substantially higher, and the specificity and negative predictive value were substantially lower, among men than among women.18 Furthermore, the Finnish screening programme has reported lower test sensitivity in women than in men using incidence cancer data.19 Thus there can be little doubt that the sensitivity of ‘blood in the stool tests’ for cancer is substantially lower in women than in men, and the advent of quantitative faecal haemoglobin analyses poses the question of whether the cut-off level used to trigger colonoscopy should be lower in women than in men.
On examination of the anatomical site of screen-detected cancers, the Scottish pilot data indicated that the proportion of screen-detected cancers in the right colon and the rectum was less than in both the interval cancers and the cancers arising in the non-screened population. Furthermore, the proportion of interval cancers in the right colon and rectum was greater than that in the population that had not been offered screening. Thus it would appear that GFOBT is less sensitive for cancers on both the right side of the colon and the rectum than in the left side of the colon.
The data presented here must be interpreted with caution since the non-screened population in this study was not randomly selected and may be biased by geographical differences throughout Scotland. However, there are good theoretical reasons why they should represent an accurate finding; haemoglobin from cancers in the right colon will have had more time to be degraded than that from left-sided cancers, and haemoglobin from rectal cancers is more likely to remain within intact erythrocytes and therefore not detectable by un-rehydrated GFOBT.
Analysis of the breakdown of cancer site by gender showed that, in the population not offered screening, right-sided cancers were less common in men than in women, and this was also observed in the screen-detected and interval populations. This raises the possibility that the effect of gender on the sensitivity of GFOBT may be mediated by tumour site, and, of course, vice versa. However, the association between gender and rectal cancer operates in the opposite direction, making it more likely that the dominant effect is related to gender.
It is not clear why the sensitivity of GFOBT should be less in women than in men, but there are various possibilities. First, the average haemoglobin concentration is lower, on average, in women than in men. However, gender differences in the haemoglobin concentration of blood are quite small and disappear 10 years after the menopause.20 Other possibilities include the documented lower colonic transit in women than in men,21 and it is also known that women in Scotland have a higher residue diet than men, which may dilute the haemoglobin concentration in faeces.22
In summary, the interval cancer rate increased steadily from the first round to the third, but the absolute number of interval cancers varied very little and was in keeping with previous reports. Also, relative to the non-screened population, interval cancers had a better prognosis despite little difference in stage distribution. Finally, GFOBT tended to underdiagnose cancers in women relative to men and may also tend to miss right-sided and rectal cancers.
These findings have extremely important implications for population-based screening programmes. Because GFOBT is not specific for human haemoglobin, it is likely that FIT, which is specific, will be more accurate. Indeed, a randomised trial from the Netherlands has demonstrated greater sensitivity for cancer with FIT than with GFOBT, albeit that the FIT positivity rate was over twice that of the GFOBT.23 However, it is reasonable to hypothesise that, even using the same percentage positivity as GFOBT, FIT should produce fewer false positive results and thereby reduce the number of interval cancers. Quantitative FIT also affords the possibility of setting differential cut-off levels for men and women.
However, the high interval cancer rate that has become accepted as an inevitable consequence of GFOBT screening may require a more radical approach. Currently, in the UK, the positivity rate of GFOBT screening is around 2% for prevalent screening, dropping during incidence screens.2 If interval cancer rates are to be substantially improved, it will be necessary to use a faecal haemoglobin cut-off concentration that will give a higher percentage positivity (and therefore higher colonoscopy rates) or to introduce endoscopic screening in some form. In population screening, the most practical approach for the latter is currently one-off flexible sigmoidoscopy, which has been shown to reduce colorectal cancer mortality and the incidence of left-sided cancer for up to 12 years.24 Flexible sigmoidoscopy as an adjunct to testing for blood in faeces may provide an appropriate solution, and this strategy deserves evaluation in a population screening context.
References
Footnotes
The STROBE checklist has been used in the preparation of this document.
Funding The pilot was funded by the Scottish Government Health Department and the analysis was supported by a grant from the Chief Scientist Office (grant No CZH/6/4), Scottish Government Health Department to establish a Bowel Screening Research Unit. The University of Dundee acts as the sponsor, and administers the grant that supports the unit. All authors are independent of the funders in terms of freedom to publish.
Competing interests None.
Ethics approval Ethics approval was not sought for the demonstration pilot. This was a decision made by the UK National Screening Committee and endorsed by the UK Departments of Health on the grounds that screening for colorectal cancer using faecal occult blood is of proved efficacy, and the study constituted evaluation of the feasibility of introducing a screening programme into the NHS. Permission to access and analyse the anonymised data presented in this paper was granted by the Community Health Index Advisory Board and the Privacy Advisory Committee, National Services Scotland.
Provenance and peer review Not commissioned; externally peer reviewed.