Flexible sigmoidoscopy in colorectal cancer screening: implications of different colonoscopy referral strategies

Colorectal cancer (CRC) is a common, yet largely preventable disease. Although colonoscopy is the gold standard in CRC and advanced adenoma detection, disadvantages such as higher costs, discomfort and complication rates as well as lower adherence limit its use as primary screening method. Flexible sigmoidoscopy (FS) reliably detects colonic neoplasms in the distal colon and rectum, which is where the majority of all colorectal neoplasms occur [1]. Proximal neoplasms can be detected by FS only indirectly if detection of distal lesions is followed by colonoscopy, as recommended in CRC screening guidelines [2‐4]. Several randomized clinical trials (RCTs) demonstrated that FS screening reduces CRC incidence and mortality [5‐8]. FS screening is offered in Italy, UK and in the USA [9].

The RCTs on effectiveness of FS screening differed in their colonoscopy referral criteria [5‐8]. While the UK Flexible Sigmoidoscopy Trial [10] imposed the highest thresholds for colonoscopy referral, less restrictive criteria were imposed in the SCORE trial (Italy) [5], the PLCO trial (USA) [6] and the NORCCAP trial (Norway) [11]. Accordingly, colonoscopy referral rates and detection rates of proximal colonic neoplasms differed between these trials [12].

More restrictive referral strategies imply lower detection rates than less restrictive criteria, but also lower numbers of colonoscopies needed (NCN) to detect one relevant proximal finding [12]. Comparative evaluations of various potential colonoscopy referral strategies within the same study population are sparse [13‐15]. Differential performance of the same strategies in men and women can be expected due to the typically higher prevalence of colorectal neoplasia among men. In Germany, FS is not routinely conducted for primary CRC screening, despite proven effectiveness and lower costs and effort compared to colonoscopy screening. We used colonoscopy findings in a large German CRC screening population to compare the expected impact on diagnostic performance (sensitivities and NCN) of 12 referral strategies following screening FS in men and women.

We estimated diagnostic performance of a once-only FS for the detection of advanced colorectal neoplasms by modeling different colonoscopy referral criteria following FS in a German CRC screening population. Without colonoscopy referral, assuming that the descending colon is not reached, FS would detect 62% of AN in men and 59% of AN in women. At the upper end, with FS reaching the entire descending colon and colonoscopy referral following the US (PLCO) criteria, 57% of all men and would undergo colonoscopy, yielding sensitivities of 92% for AN. In women, colonoscopy following any neoplasm or hyperplastic polyp would yield the highest number of colonoscopies (25% of all women) and NCN (25.3) and achieve 76% sensitivity for AN. Compared to referral due to distal AN only (NCN 15.6), the number of colonoscopies to detect one proximal AN would strongly increase, and a large share of this increase in sensitivity could be achieved with much lower colonoscopy referral rates using less extensive referral strategies.

We found major sex differences in sensitivity (higher in men) and the NCN to detect one proximal AN (higher in women) that might suggest considering sex-specific referral strategies. Although sensitivity could be increased among women by using less restrictive referral strategies compared to men, this would further aggravate the gender discrepancies in the NCN. For example, assuming FS to reach and visualize the descending colon, referral of women with any distal neoplasm or HPP would approximately yield the same overall sensitivity for detecting any AN (76%) as referral of men with any distal advanced neoplasm only (77%) (see Table 3). However, such an approach would require 25.3 colonoscopies to detect one additional proximal AN among women, three times the corresponding number among men (8.4, see Table 4). To achieve a better use of colonoscopy resources, quite a contrary approach may make sense: Referring men to colonoscopy after detecting any distal neoplasm, but women only after detecting a distal AN would not only result in quite comparable NCN for both sexes (12.4 and 15.6, respectively, see Table 4), but also lead to a higher number of detected proximal AN (156 + 27 = 183 vs. 102 + 74 = 176), despite an overall much lower number of colonoscopies (1941 + 422 = 2363 vs. 854 + 1874 = 2728) compared to the aforementioned “equal sensitivity” scenario.

Several earlier studies had reported an increased risk of proximal AN in the presence of specific distal findings [20‐22], supporting the use of colonoscopy referral strategies based on such findings, but only few studies (one each from Japan, Spain and China) have explicitly compared expected performance of various referral strategies in terms of detection of proximal AN [13‐15]. Like our study, they reported increased sensitivity for proximal AN with more extensive referral strategies, and two of them [14, 15] also reported major sex differences. Castells et al. [14] obtained virtually identical sensitivities for proximal AN using the UK, SCORE and NORCCAP criteria (22, 31 and 37%, respectively). Similar to our study, sensitivities were consistently higher in men than in women. The US strategy, where the largest sex-specific differences could be expected, was not examined, though. In the study of Wong et al. [15], estimated achievable sensitivities for proximal AN using the SCORE and NORCCAP strategies were very similar to those obtained in our study, with 31 and 38%, respectively. Compared to our study, they found lower sensitivities when using the UK criteria (14 vs. 25%) or the US criteria (42 vs. 55%). Our study expands the evidence from these previous studies in several important respects, in that we included a much larger number of potential referral strategies (12 compared to 2, 3 or 4), along with a larger number of participants with proximal AN (581 compared to 319, 212, and 132, respectively). Furthermore, we provided, for the first time, detailed sex specific analyses for different assumptions regarding the reach of FS. The mean age of the so far largest study population from Japan (48 years) had been below the typical age at which FS is recommended and conducted, and this study had assessed only two referral strategies (any distal AN versus any distal neoplasm) for both sexes combined [13]. Our study is the first to explicitly quantify the “incremental NCN”, i.e., the number of additional colonoscopies needed to detect one additional proximal AN when comparing more extensive to more restrictive referral strategies. Such “incremental NCN”, which we reported for comparisons of referral strategies to the most restrictive strategy under investigation could be derived analogously for comparisons between any other pairs of more restrictive and more comprehensive referral strategy from the data presented in this paper and may be a parameter of particular relevance for delineating the study population that should be referred.

In contrast to a previous study indicating that histology-defined AAs discovered during FS are stronger predictors for proximal AN than large distal adenomas [21], we did not find pronounced differences in sensitivity when using different definitions of AA for referral strategies. However, referral based on size-defined AN tended to perform worse in terms of NCN and ΔNCN than referral based on histology-defined AN. Compared to other primary colonoscopy screening populations [23‐27], the CRC detection rate was somewhat higher in our study (1.4%). Similarly, estimated sensitivities of FS for AN detection were mostly higher than previously estimated from meta-analysis results [28]. Possible reasons for the relatively high CRC detection rate include the substantially older age of the study population compared to other studies, conduction of the study in a high incidence country and a high incidence region within Germany [29], and exclusion of subjects with a previous colonoscopy in the past 5 years who have very low CRC detection rates [17]. Despite stringent in- and exclusion criteria applied in our study, it cannot be ruled out entirely that some subjects who underwent screening colonoscopy had other symptoms that motivated them to visit a gastroenterology practice. However, our study thereby accurately reflects the setting in which subjects are recruited in the German colonoscopy screening program. This self-selection might have further contributed to the somewhat higher CRC prevalence in our study. One consequence are consistently lower NCNs than those found in a previous study [27].

Via adenoma removal during FS, a share of CRCs can actually be prevented. Compared to screening colonoscopy, some proximal neoplasms are inevitably missed, thus limiting the preventive potential of FS-only screening. Results from the UK Flexible Sigmoidoscopy Screening Trial with its comparably restrictive referral criteria indicate that even a single FS can achieve a significant and long-lasting reduction of distal CRC incidence and mortality [8]. Reductions of proximal colon cancer incidence were not seen, suggesting that much higher colonoscopy referral rates would be needed to have a significant effect on proximal colon cancer incidence. Whether expanding the colonoscopy referral rates would be the best way to enhance sensitivity and effectiveness of FS based screening appears questionable in the light of our results. Other approaches, such as conduction of a single FIT first, using a positive FIT result as referral criterion to colonoscopy, followed by a once-only FS in FIT negatives [28, 30] are potentially more promising. In the NORCCAP study, CRC incidence and mortality of FS screening alone were also compared to a combination with FIT [7]. The study found no significant differences in detection rates of adenomas or CRC and statistically non-significant reductions in CRC mortality in both groups. Incidence was higher and mortality lower in the group with combined FIT and FS screening compared to sole FS screening. In a previous approach of modeling performance of FS, alone and combined with FIT [28], no individual patient data was available and thus, performance of FS could not be examined for different referral strategies. Estimated overall sensitivities of FS for AN detection were 60, 68 and 72%, assuming colonoscopy referral rates of 0, 20 and 30%, respectively. In the present study, estimated sensitivities for AN were somewhat higher, with 66, 74 and 79% at comparable referral rates (0, 22 and 36% using no referral, any AN or any neoplasm as referral criteria, respectively). Another recent study [30] investigated performance of a single FS alone and combined with FIT for the widely recommended referral criterion due to any distal neoplasm. With sensitivities of 86% and 72% of FS alone for CRC and AA, respectively, estimated accuracy of FS was similar to that of the present study. Additional conduction of FIT was estimated to increase sensitivities for CRC to 100% and sensitivities for AA to at least 72% and up to 82%, depending on the FIT cutoff.

Our study has several strengths. To our knowledge, it is the most comprehensive analysis of colonoscopy referral criteria, examining 12 different criteria in a very large population of participants of screening colonoscopy. It is the first to explicitly quantify the incremental benefit and burden of extending a relatively restrictive colonoscopy referral strategy following FS. Sensitivities for proximal and for any AN were investigated. All included participants underwent full colonoscopy, ensuring that sensitivities of FS are not overestimated due to missed proximal neoplasms at an incomplete colonoscopy. Participants underwent colonoscopy for primary screening, not for clarification of symptoms. Thus, potential overestimation of sensitivity of FS if symptomatic subjects with a presumably higher prevalence of FS-reachable findings had been included should have been avoided. To our knowledge, our study is the by far largest of its kind, with more than 15,000 participants, including over 1500 AA and 200 CRC cases. For all participants, detailed colonoscopy data were available, including location, size and histopathologic features of every finding. These comprehensive data allowed us to provide precise gender-specific estimates of sensitivity, NCN and the NPV for a range of FS-based screening strategies. Assuming a more limited reach of FS did not alter any of the results materially.

Our study also has limitations. Sigmoidoscopy results were derived from colonoscopy. Although a common approach in studies investigating sensitivity of FS [13‐15, 21, 22], performance of FS might differ from that of colonoscopy, e.g. due to different bowel preparation procedures (enema vs. oral bowel cleansing) or the absence of sedation. Evidence in this matter is inconclusive, however [31]. On the other hand, the more convenient preparation procedure of FS might yield higher participation rates which might offset a somewhat smaller sensitivity in the distal colon and rectum compared to colonoscopy. Furthermore, also colonoscopy may miss neoplasms, mainly in the proximal colon. Participants with incomplete colonoscopy (0.9%, N = 164/19,261) were excluded from our analyses. Not reaching the coecum will slightly diminish sensitivity of colonoscopy in screening practice. Another factor potentially influencing sensitivity of FS is conduction of the procedure by non-gastroenterologists. Although FS, in contrast to colonoscopy, is frequently conducted by general practitioners and may even by conducted by nurses [32], detection rates are expected to be similar to those of gastroenterologists [33, 34]. Nevertheless, a certain fraction of FS exams will not be completed in practice, e.g. due to pain or insufficient bowel preparation. Those factors might reduce insertion depth and thereby sensitivity of FS somewhat, although our sensitivity analyses indicated that results were very similar even when only the rectum and sigmoid colon were assumed to be reached and visualized by FS. Full compliance to follow-up colonoscopy due to the examined referral criteria was assumed. Partial non-adherence would reduce the gain in achievable sensitivities compared to FS without colonoscopy referral. For example, assuming only 75% adherence rate to colonoscopy follow-up after recommended referral would reduce the 24 percentage points increase in sensitivity for AN achievable by applying the US criteria to men to approximately 18 percentage points. In some of the investigated strategies, we assumed that distal AN discovered during FS would be removed immediately without colonoscopy follow-up. Although advanced adenomas can in principle be removed during FS, the American Society of Gastrointestinal Endoscopists (ASGE) recommends polypectomy during FS only after adequate bowel preparation [35]. Finally, our analyses did not consider sessile serrated polyps which were not systematically detected and recorded by many endoscopists during the earlier years of study recruitment.

In conclusion, FS applied with a moderately restrictive colonoscopy referral strategy, such as referral only due to advanced distal findings, would likely achieve high sensitivities for detection of ANs in women, and even higher sensitivities in men, in a large German CRC screening population and require considerably fewer colonoscopies than primary screening colonoscopy. Even for moderately restrictive referral strategies, the share of false-negatives, i.e. of missing a proximal AA or even proximal colon cancer if FS is not followed by colonoscopy, was very low in our study. However, for any referral strategy, much higher numbers of colonoscopies to detect one proximal AN would be needed for women than for men, and these numbers would substantially increase with increasingly extensive referral strategies. Restricting colonoscopy referral to those with any advanced distal neoplasm rather than the commonly recommended referral of those with any distal neoplasm, at least among women, should be considered. A drawback of such a strategy would be a potentially higher number of interval cancers. On the other hand, even a single FIT detects the majority of proximal colon cancers and a significant proportion of proximal advanced adenomas [28, 36]. To avoid an increase in the number of interval cancer cases, additional conduction of FIT prior to FS, with colonoscopy referral of FIT-positives and conduction of FS among FIT-negatives might be an alternative.