We extensively addressed the question of whether FOBT, the most common non-invasive tool for CRC screening, shows differential sensitivity for detecting left- vs
right-sided advanced colorectal neoplasia within a large screening study conducted in average-risk subjects. Our analysis of data on an ELISA-based immunochemical FOBT supports the hypothesis that FOBT is more sensitive for detecting left-sided advanced neoplasia than right-sided advanced neoplasia. The magnitude of the difference in sensitivities (about 13 percentage points) at a specificity of 95% is similar to the findings reported by Morikawa et al (2005)
regarding an agglutination-based immunochemical FOBT (11) and is also supported by another studies reporting on site-specific test performance of a guaiac-based FOBT (Ahlquist et al, 2008
However, the primary focus of the aforementioned studies was on overall test performance; the authors did not provide further information on the characteristics of the subgroups with left- and right-sided advanced neoplasia. We focused, for the first time, on this stratification according to anatomical site and explored potential factors that could explain a higher sensitivity of FOBT for left-sided neoplasia.
Specifically, we described the subgroups with left- and right-sided advanced neoplasia with respect to parameters that were reported to correlate with a higher sensitivity of FOBT, such as male gender, current NSAID use and adenoma characteristics (Ciatto et al, 2007
; Morikawa et al, 2007
; Levi et al, 2009
; Rozen et al, 2009
). The proportion of men, the proportion of current NSAID users, the proportion of subjects with large adenomas (
cm in diameter) as well as the proportion of subjects with more than one adenoma was higher in the subgroup with right-sided advanced neoplasia than in the subgroup with left-sided advanced neoplasia. Thus, the distribution of these parameters could not explain the observation of a higher sensitivity for left-sided advanced neoplasia. In contrast, the proportion of subjects with pedunculated adenomas was strikingly higher in the subgroup with left-sided advanced neoplasia (0.47 vs
0.14). Although information on adenoma shape was often missing, it is unlikely that this difference is caused by information bias. The latter would have occurred if the likelihood of reporting on pedunculated shape had been differential with respect to the location of adenomas, which doesn't seem plausible. The proportion of participants for whom information on adenoma shape was missing altogether was the same in the subgroup with left- and right-sided advanced neoplasia. The higher proportion of pedunculated adenomas in the left colon is consistent with autopsy studies that reported on the shape of adenomas according to anatomical site (Blatt, 1961
; Eide and Stalsberg, 1978
; Rickert et al, 1979
; Williams et al, 1982
). The comparison of median faecal haemoglobin levels in subjects with pedunculated adenomas vs
subjects with otherwise shaped adenomas as well as logistic regression analyses adjusted for site supported the importance of pedunculated shape regarding the site difference in test sensitivity. The proportion of CRC cases was also higher in the subgroup with left-sided neoplasia, but the respective numbers were small and their exclusion did not change the observed differences in site-specific test performance.
In the primary analysis of ROC curves, we observed a higher sensitivity for individuals with left-sided advanced neoplasia at all cutoff levels, including cutoff levels yielding specificities that are typically required in the screening setting (i.e., well above 90%). In sensitivity analysis, we explored whether the difference in sensitivities according to anatomical site was due to the location of the advanced neoplasm itself. For that purpose, we restricted the analyses of ROC curves to individuals with one advanced colorectal neoplasm and no other adenomas. Even though this restriction created an artificial setting that does not reflect the distribution of neoplasms in the natural setting, it avoided interfering effects that could result from more than one adenoma in the same individual, which are potentially distributed over different anatomical sites. Interestingly, the difference in diagnostic performance according to anatomical site in this restricted setting occurred only when sensitivity was increased by shifting the cutoff level to lower levels (i.e., lower specificity), but was not observed at higher, clinically relevant cutoff levels (i.e., at levels yielding specificities well above 90%). This finding could be interpreted as follows: when the cutoff level is high, which means that only individuals with a relatively strong source of bleeding test positive, it doesn't matter whether the advanced neoplasm is in the right or in the left colon. When the cutoff level is lowered, which allows individuals with a weaker source of bleeding to test positive as well, then sensitivity is higher for left-sided than for right-sided advanced neoplasia. In other words, our results suggest that weak sources of bleeding are more likely to occur (possibly due to pedunculated shape of adenomas) or to be detected (possibly due to degradation of haemoglobin) in the left colon than in the right colon, while the location doesn't matter when there is a strong source of bleeding.
An explanatory model that synthesises the different findings could be as follows: a certain proportion of advanced neoplasms present by themselves a relatively strong source of bleeding and their likelihood of being detected through FOBT does not depend on the anatomical site. Another proportion of advanced neoplasms is weak sources of bleeding and do not lead by themselves to a positive FOBT at clinically relevant cutoff levels. However, in conjunction with other adenomas, which may also represent weak sources of bleeding, haemoglobin levels may mount up leading to a positive FOBT at clinically relevant cutoff levels (i.e., at levels yielding specificities well above 90%). This additive mechanism may cause the higher sensitivity for left-sided neoplasia at clinically relevant cutoff levels since our findings suggested that the detection (and/or occurrence) of weak sources of bleeding is more likely in the left colon than in the right colon.
Irrespective of the reasons explaining the higher sensitivity of FOBT for left-sided advanced neoplasia, the finding itself would be of clinical relevance. While colonoscopy has already been questioned regarding protection from right-sided colorectal neoplasia (Brenner et al, 2007
; Lakoff et al, 2008
; Baxter et al, 2009
), our study supports the hypothesis that FOBT, the most common non-invasive tool for CRC screening, also shows lower performance for right-sided neoplasia. Our findings may stimulate further diagnostic research in the field. They may also provide valuable information for modelling analyses to estimate the potential effects of site-specific test performance on the programmatic sensitivity and the effectiveness of FOBT-based screening programmes in which FOBT is repeated at frequent intervals (e.g., annually or biennially). In particular, current strategies that combine flexible sigmoidoscopy and FOBT could be affected by a lower sensitivity of FOBT for right-sided lesions (Pignone et al, 2002
; Zauber et al, 2008
). Such strategies have been suggested to be as effective as colonoscopy in some scenarios, but those findings might change if site-specific test performance of FOBT is taken into account. Generally, site-specific performance should be a focus in the optimisation of current screening tests as well as in the development and evaluation of novel screening tests, such as computed tomographic colonography (Heresbach et al, 2011
) and colon capsule endoscopy (Sacher-Huvelin et al, 2010
Our study offered good opportunity for evaluating site-specific test performance due to the fact that all study participants underwent colonoscopy irrespective of the FOBT result and due to a setting that reflects the average-risk target population of CRC screening. Using a symptomatic study population or using a clinical follow-up instead of colonoscopy to estimate diagnostic performance could bias site-specific estimates of test sensitivity since the presence of symptoms or the likelihood of clinical manifestation may depend on anatomical site. The prospective design, the careful application of exclusion criteria as well as the high level of experience and qualification among gastroenterologist who participate in the German screening colonoscopy programme were favourable in terms of minimising potential sources of bias.
Our study has also important limitations. First, the group with ‘advanced neoplasia' was comprised for the most part individuals with advanced adenomas, while the number of individuals with CRC was low. Given that advanced adenomas are considered to be precursors to CRC, our results thus support site-specific effects with respect to prevention from CRC through FOBT. To investigate whether there are also site-specific effects with respect to early detection of invasive CRC, very large sample sizes would be needed since the prevalence of undetected CRC is inherently low in an average-risk study population. From a theoretical point of view, site-specific effects might be less important for invasive CRCs, which, if they bleed, have been reported to go along with comparatively high faecal haemoglobin levels (Levi et al, 2007
). Second, although we had a reasonable sample size of participants with advanced neoplasia, further stratification according to parameters that have been suggested to impact faecal haemoglobin levels and calculation of site-specific test performance for each of the subgroups would not have been possible due to sample size limitations. We, therefore, only assessed whether these parameters were differentially distributed across the subgroups with left- and right-sided advanced colorectal neoplasia. There were also sample size limitations regarding the more detailed stratification of sensitivities according to anatomical site. Although the point estimates of these analyses tentatively suggest a progressive increase in sensitivity from the right to the left side of the colon (excluding the rectum), the CIs of these estimates were large. The comparatively low sensitivity for rectal lesions could be due to the exclusion criterion ‘visual bleeding', which may apply more often to subjects with rectal neoplasms that bleed than to subjects with neoplasms at other parts of the colon (where it may be less likely that the blood is detected visually). Third, it would be interesting to investigate whether site-specific test performance is affected by the number of stool samples that are tested per person or by different sampling techniques, which, however, we could not assess in our study.
We did not consider site-specific performance of FOBT with respect to any adenoma, but focused on advanced adenomas because they are considered to be the most clinically relevant precursor lesions. This focus also minimised potential misclassification bias due to miss rates at colonoscopy, which may be site specific, but which have been shown to be generally low for advanced neoplasia (van Rijn et al, 2006
; Heresbach et al, 2008
In conclusion, the immunochemical FOBT in our study was more sensitive for detecting subjects with left- vs right-sided advanced colorectal neoplasia. Our findings may stimulate further research in the field as well as modelling analyses to estimate the potential effect of site-specific test performance on the programmatic sensitivity and the effectiveness of annual or biennial FOBT-based screening programmes, in particular with respect to protection from right-sided CRC.