Clinical effectiveness

Thirty studies met the inclusion criteria for the systematic review of test accuracy. These assessed NBI (24 studies^20,54–78), i-scan (five studies^77,79–82) and FICE (three studies^78,83,84). Two of these studies assessed two of the technologies of interest in this diagnostic assessment (NBI and i-scan;⁷⁷ and NBI and FICE⁷⁹). Using the QUADAS criteria, we assessed that the results of the studies are likely to be at a low risk of bias. The evidence we identified meets the decision problem for this diagnostic assessment, but there is comparatively little evidence for two of the three technologies being considered (i-scan and FICE). Most of the available evidence evaluated the diagnostic accuracy of NBI for assessing diminutive colorectal polyps. The FICE evidence base was particularly limited. We did not identify any FICE studies that assessed the diagnostic accuracy of endoscopists’ real-time high-confidence evaluations of diminutive polyps, whereas we found evidence in relation to high-confidence assessments made with NBI and i-scan. Some of the included studies explicitly referred to a DISCARD strategy, whereas others did not.

Most of the included studies reported high sensitivity and specificity (with some exceptions), showing that endoscopists had a high probability of correctly identifying adenomas and hyperplastic polyps when using NBI, i-scan or FICE (sensitivity and specificity results are discussed in more detail below; see Table 65). NPV (that is, the probability that patients who are diagnosed by VCE as having a hyperplastic polyp truly do not have an adenoma) was more variable across the NBI studies than the FICE or i-scan studies. There was especially little variation in this outcome across the i-scan studies, in which NPV ranged from 93% to 96.30% for all characterisations and 94.74% to 100% for high-confidence characterisations. Of the three technologies, i-scan had the most consistently favourable results on this outcome. The greater heterogeneity found among the NBI studies may in part be explained by the larger pool of evidence available for NBI than for i-scan and FICE. In additional, two of the FICE studies were conducted by the same research group, which may have reduced heterogeneity. The heterogeneity in the NBI results may have also been as a result of variability in the prevalence of adenomas in the populations included in the studies. When prevalence is increased, the result is a decrease in the NPV. The more favourable NPV results found for i-scan and variability among the NBI studies may also be explained by the endoscopists’ experience in these studies. We note that a range of endoscopists was involved in the NBI studies; some were less experienced in conducting colonoscopy generally and had little or no experience using NBI, while others were very experienced endoscopists who also had extensive experience of using NBI. By contrast, three^77,79,80 of the five^77,79–82 i-scan studies included endoscopists with prior experience of i-scan and all the studies were conducted in single centres, often described as academic or specialist centres. The NPV results found in the i-scan studies may therefore not reflect the accuracy that might be achieved by endoscopists working in more generalist or community settings. On the other hand, the large evidence base for NBI may have captured the variability in this outcome that may be observed in practice, where it is likely endoscopists with a range of experience will carry out colonoscopy (although we note that the ESGE guidance recommends that only experienced and adequately trained endoscopists should undertake VCE for the real-time assessment of polyps³¹).

Table 65 summarises the key sensitivity and specificity results from the review and the meta-analyses, which we now discuss in more detail. Meta-analysis was conducted where possible, but the technologies were not assessed head to head in the meta-analyses (as this was not within the decision problem for the assessment, derived from the National Institute for Health and Care Excellence scope), so we cannot comment on how the technologies directly compare with each other statistically.

TABLE 65

Summary of key results

For all characterisations of polyps (regardless of confidence level) in the whole colon, the i-scan (one study⁷⁷) and FICE (three studies^78,83,84) results were in the same range of values obtained from the NBI studies (17^{55,56,58,62–71,74,75,77,78} and 16 studies^{55,56,58,62–71,75,77,78} for sensitivity and specificity, respectively). The summary values from bivariate meta-analysis for sensitivity and specificity of NBI and FICE for all characterisations in the whole colon did not reach 0.90 (i.e. 90%) in either case. Limiting the analysis to high-confidence characterisations of polyps in the whole colon increased the summary sensitivity and specificity values from bivariate meta-analysis; for i-scan (two studies^77,79), both values were > 0.90; whereas for NBI (11 studies^{55–57,59–65,77}) only the summary value for sensitivity was > 0.90. As mentioned above, none of the FICE studies analysed outcomes for high-confidence assessments of diminutive polyps. As with the NPV results, the higher sensitivity and specificity values seen for i-scan might be explained by the endoscopists in the two i-scan studies^77,79 being experienced endoscopists working in specialist and academic centres. Therefore, we conducted a post hoc analysis restricting the meta-analysis to high-confidence characterisations in the whole colon obtained from studies that reported the endoscopists had prior experience with NBI (four studies^59,60,62,77). The summary sensitivity and specificity results from this post hoc analysis of NBI were almost identical to those obtained from all the NBI studies.

Some NBI and i-scan studies provided data on characterisations of polyps in the rectosigmoid colon, but no evidence was available for FICE. For all characterisations of polyps (regardless of confidence level) in the rectosigmoid colon, the NBI (four studies^54,55,58,63) and i-scan (two studies^81,82) results were similar to those obtained from the whole colon. Limiting the analysis to high-confidence characterisations of polyps in the rectosigmoid colon increased the summary sensitivity and specificity values from bivariate meta-analysis of NBI, and the study estimates from i-scan were also higher (meta-analysis was not possible for i-scan). A post hoc analysis restricting the NBI meta-analysis to high-confidence characterisations in the rectosigmoid colon obtained from studies that reported the endoscopists had prior experience with NBI (two studies^54,62) increased the summary sensitivity and specificity values further. However, there was no evidence for i-scan because the single study⁸² that reported on high-confidence characterisations in the rectosigmoid colon did not report on whether or not the endoscopist had prior experience using i-scan.

Overall, there is evidence showing that, in general, sensitivity and specificity estimates increase when only high-confidence characterisations of polyps are considered rather than all characterisations (i.e. not on the basis of high confidence). It is worth reiterating that the level of confidence with which polyp classifications are made is subjective and is likely to vary between endoscopists. Some endoscopists may refer to the relevant classification system to make a confident polyp characterisation. The studies included in our systematic review did not explicitly state how confidence was achieved. This creates possible uncertainty in the interpretation of diagnostic accuracy based on high-confidence characterisations.

We also generated SROC curves to explore the effect of endoscopist experience with NBI on sensitivity and specificity when characterising polyps in the whole colon. This confirmed that endoscopists with prior experience of using NBI to characterise diminutive colorectal polyps achieve higher sensitivity and specificity than endoscopists with no prior experience of using NBI to characterise diminutive colorectal polyps (other than any training that they undertook at the start of the study). It was not possible to discern this effect when comparing the post hoc meta-analysis of high-confidence characterisations in the whole colon made by endoscopists with prior experience of NBI with the meta-analysis of all high-confidence characterisations in the whole colon. This may be because, three studies in the pool of 11 NBI studies^{55–57,59–65,77} providing data on high-confidence characterisations in the whole colon included endoscopists with a mix of prior experience^56,57,65 and two did not report on prior experience^63,64 with NBI, which would probably have masked any difference between NBI-experienced (four studies^59,60,62,77) and NBI-naive endoscopists (two studies^55,61).

Finally, a post hoc bivariate meta-analysis pooling together all the available evidence for high-confidence characterisations of polyps in the whole colon was undertaken and yielded a sensitivity of 0.92 (95% CI 0.87 to 0.95) and a specificity of 0.83 (95% CI 0.78 to 0.87). There were differing opinions among the clinical experts we consulted regarding whether or not it was appropriate to pool evidence from different VCE technologies. The technologies have the same aim (to enhance surface vessel patterns), but achieve this either by filtering the light source (NBI) or by using digital post-processing software to convert white-light images such that they appear like narrow-band images (i-scan and FICE). This post hoc analysis should therefore be treated as illustrative because of the uncertainty regarding whether or not a class-effect can be assumed and also because the available evidence is predominantly from NBI (11 studies^{55–57,59–65,77}) with only two i-scan studies^77,79 and none for FICE.

In terms of the other outcomes of interest in this review, none of the studies measured HRQoL, anxiety, number of outpatient appointments or telephone consultations, incidence of colorectal cancer or mortality. Only three^57,75,77 of the NBI studies and one⁷⁷ of the i-scan studies reported AEs (e.g. complications of polypectomy such as bleeding). All studies reported that there were none. Thus, there are only limited data available on AEs in this review. This is an outcome that future studies should consider measuring. A few of the NBI studies reported on the number of polyps that would be resected and discarded if a resect and discard type of management strategy had been in place.^68,70 Given the limited evidence available, it is challenging to determine the number of polyps that would be designated to be left in place, the number of polyps that would be designated to be resected and discarded and the number of polyps that would be designated for resection and histopathological examination. Likewise, only limited data were available on the length of time to perform the colonoscopy, which means that no firm estimates can be made of the additional time it would take during colonoscopy to make real-time assessments of polyp histopathology.

Table 66 summarises the results of the studies included in this review in relation to the two PIVI requirements that new technologies for the real-time endoscopic assessment of the histopathology of diminutive colorectal polyps should meet, before a resect and discard strategy could be applied in practice. To reiterate, the criteria specify that, for colorectal polyps ≤ 5 mm in size to be resected and discarded without histopathological assessment, the endoscopic technology (when used with high confidence) should have a ≥ 90% agreement in assignment of post-polypectomy surveillance intervals when compared with decisions based on histopathology assessment of all identified polyps. The criteria also specify that, in order for a technology to be used to guide the decision to leave suspected rectosigmoid hyperplastic polyps ≤ 5 mm in size in place (without resection), the technology should provide ≥ 90% NPV (when used with high confidence) for adenomatous histopathology (see Chapter 1, Care pathway). Not all the studies that assessed surveillance intervals evaluated these in accordance with the PIVI criteria. We have therefore included here the results only of those studies that clearly calculated concordance of surveillance intervals between VCE and histopathology in line with the PIVI requirements. Neither of the two FICE studies that measured surveillance intervals used the PIVI requirements to do this.^83,84 None of the FICE studies examined the NPV for high-confidence assessments in the rectosigmoid colon either. This means that this review did not identify any evidence that enables us to assess how FICE meets the PIVI requirements.

TABLE 66

Summary of the review’s results in relation to the PIVI criteria

As Table 66 shows, all but one⁷⁶ of the NBI and i-scan studies that measured surveillance interval assignment in line with the PIVI criteria^{55,57,58,61–64,67,76,79,82} found a concordance of ≥ 90% between NBI or i-scan and histopathology and thus met this criterion of the PIVI statement (Ladabaum and colleagues⁵⁸ achieved this for only one of the two guidelines used to determine surveillance interval). Most studies did not provide a CI, but where this was reported the lower limit fell below 90% in two of six cases. All the NBI and i-scan studies that measured the NPV of high-confidence assessments of diminutive polyps in the rectosigmoid colon found a ≥ 90% NPV, and thus met the second criterion of the PIVI statement. However, NPV and surveillance interval results for i-scan were provided by only one and two studies, respectively, and so the evidence in relation to how i-scan meets the PIVI requirements is limited. Our findings suggest that, on the whole, NBI appears to meet the PIVI criteria, supporting the use of NBI to carry out a resect and discard strategy in practice. We note that, in general, when there were discrepancies between the surveillance intervals set following NBI and histopathology, NBI surveillance intervals tended to be shorter than they would have been with histopathology (i.e. patients are seen again sooner).

It should be noted that our assessment here of the findings of the studies included in this review against the PIVI criteria does not take into account the settings of these studies (i.e. whether they were carried out in specialist, academic settings or routine practice). This could impact on whether or not VCE technologies meet the PIVI criteria. The DISCARD 2 study,¹⁴⁸ which is a large, multicentre prospective UK study, concluded that NBI cannot be recommended for use in routine clinical practice, as when it is used by non-experts in this setting it does not result in a high enough concordance rate with histopathology for determining surveillance intervals. This study was not included in our systematic review as it did not meet the inclusion criteria as a result of only 22% of the colonoscopies being conducted using HD equipment. In this respect it differs from the studies included in this review and the decision problem for this assessment. It is possible that without HD equipment, diagnostic accuracy and appropriate allocation of surveillance intervals may be lower than that achieved when HD equipment is used.

The results of our systematic review have some similarities to those of previous systematic reviews of VCE for characterising colorectal polyps, notwithstanding certain differences between reviews in scope and study inclusion criteria.^42–44,149

For example, the ASGE Technology Committee conducted a systematic review to examine whether NBI, i-scan or FICE met the PIVI performance thresholds and, therefore, whether or not the evidence supported a ‘diagnosis-and-leave’ approach (ASGE Technology Committee, 2015, p. 1).¹⁴⁹ Literature searches were done on a number of standard health research databases, up to May 2014 (thus the search is around 2 years older than our literature search). For NBI the review included 19 studies giving estimates of NPVs and 10 studies giving estimates of agreement in post-polypectomy surveillance intervals. For i-scan there were eight studies of NPVs and one study of agreement in post-polypectomy surveillance intervals. For FICE there were eight NPV studies and two studies of agreement in post-polypectomy surveillance intervals. The majority of the studies used HD endoscopy systems, and some allowed use of magnification (in contrast to our systematic review).

In the ASGE systematic review¹⁴⁹ the pooled random-effects NPV for studies in which an optical characterisation of diminutive polyps with NBI was made with high confidence was 93% (95% CI 90% to 96%). This increased to 95% (95% CI 92% to 98%) when high-confidence characterisations were made by endoscopists experienced in optical assessment of colorectal polyps. In our systematic review the majority of NBI studies reported NPVs for high-confidence assessments of > 78%, with five studies reporting NPVs of ≥ 90%^{20,55,57,64,77} (though note that the lower limit of the 95% CI fell below 90% in the majority of studies). The agreement in assignment of post-polypectomy surveillance intervals based on optical characterisation of diminutive colorectal polyps with high confidence using NBI was 91% (95% CI 88% to 95%). For i-scan there was no pooled NPV estimate given for high-confidence predictions. The overall pooled random-effects NPV (any level of confidence prediction) was 84% (95% CI 76% to 91%). A subgroup analysis based on endoscopist experience in performing and interpreting optical biopsies of colorectal polyps reported a pooled random-effects NPV of 96% (95% CI 94% to 98%) for experienced endoscopists compared with a pooled random-effects NPV of 72% (95% CI 69% to 76%) for novice endoscopists. As discussed earlier, our systematic review also found that diagnostic accuracy (in terms of sensitivity and specificity) increased in studies (of NBI) involving experienced endoscopists compared with those with less experience. The one i-scan study included in the ASGE review,¹⁴⁹ which compared surveillance intervals based on optical assessment with histopathology, reported an agreement level of 69.5% (95% CI 63% to 75%), thus not meeting the PIVI threshold. The overall pooled random-effects NPV for FICE was 80% (95% CI 76% to 85%). This estimate did not improve when restricted to studies of endoscopists experienced in use of optical assessment of colorectal polyps.

Another systematic review, reported by Wanders and colleagues,⁴² assessed the diagnostic performance of VCE. This review assessed the sensitivity, specificity and NPV of NBI, FICE and i-scan for optical diagnosis of colonic polyps (in addition to autofluorescence imaging and confocal laser endomicroscopy, which are not within the scope of our systematic review). Key research databases were searched up to January 2013 (thus 3 years older than our systematic review). The inclusion criteria were broader than our review, permitting studies of diminutive and larger polyps, studies of real time as well as post-procedure image-based VCE, studies with or without magnification and studies with standard or HD endoscopy systems. However, subgroup analyses were presented based on these criteria, allowing a comparison more aligned to the scope of our systematic review to be made. Pooled bivariate meta-analysis sensitivity for the subgroup of five NBI studies with diminutive polyps where the prediction was made with high confidence was 87% (95% CI 78% to 93%) and corresponding pooled specificity was 85% (95% CI 74% to 92%). These estimates are reported to have been assessed in the context of the PIVI statement, which implies they are based on characterisations of polyps in the rectosigmoid colon. If this is the case then the corresponding NBI pooled sensitivity and specificity estimates for polyps characterised with high confidence in the rectosigmoid colon in our bivariate meta-analysis are 87% (95% CI 80% to 92%) and 95% (95% CI 87% to 98%), respectively (n = four studies). Thus, our estimates are similar in terms of sensitivity but not specificity. A pooled NPV of 83% (95% CI 75% to 88%) was reported for NBI, restricted to real-time studies (n = 35), but not further restricted in terms of diminutive polyps in the rectosigmoid colon based on high-confidence decisions (i.e. in accordance with the PIVI statement) or in terms of the definition status of the endoscopy systems used (standard or high) or magnification status (with or without). The authors suggest that studies of only rectosigmoid colon NPVs are likely to show a good diagnostic performance, as the prevalence of non-neoplastic polyps is increased in the rectosigmoid. For FICE, bivariate sensitivity and specificity are reported for diminutive polyps, though not stated to be for any particular confidence level (four studies). The estimates were 84% (95% CI 73% to 94%) and 87% (95% CI 79% to 94%), respectively, similar to our results (see Table 65). Owing to the lack of suitable studies, no diagnostic accuracy estimates were presented for diminutive polyps characterised with i-scan.

Also of note was that, in the review by Wanders and colleagues,⁴² sensitivity and specificity did not differ (statistically) significantly according to whether the EXERA or LUCERA NBI system was used. Even though only the LUCERA system is available for use in the UK, the inclusion criteria for our systematic review, based on the National Institute for Health and Care Excellence scope, allowed studies of both of these systems to be included. (Note that 16 of the NBI studies used EXERA, five used LUCERA and three did not report which system was used – see Table 5.) We did not plan to conduct a formal subgroup analyses based on type of NBI system.

Cost-effectiveness

A systematic search of the literature found two economic evaluations^112,113 of VCE compared with histopathology. Both studies compared the resect and discard strategy with current practice of submitting all polyps to histopathology. The evaluations were published in the USA. The studies found that there were cost savings for the resect and discard group between US$25 and US$174 per person.

A study by Olympus, the manufacturer of NBI, described a budget impact analysis of NBI in NHS England. The decision tree model has a time horizon of 7 years, and in each year there is a cohort of patients who undergo endoscopy. The study found that NBI offered cost savings of £141M over 7 years.

We developed an independent cost-effectiveness model comparing NBI, FICE and i-scan with histopathology. The base-case analysis uses a VCE strategy in a bowel screening population where diminutive polyps in the whole colon are optically characterised. The model uses estimates of diagnostic accuracy from our meta-analysis for diminutive polyps characterised with high confidence in the whole colon. The results from our economic model suggest that VCE is cost saving compared with histopathology, with a mean saving of between £73 and £87 per person over their lifetime. The QALYs are similar between the technologies with a very small increase in QALYs with NBI and i-scan compared with histopathology of between 0.0005 and 0.0007 QALYs per person, whereas FICE is associated with 0.0001 fewer QALYs per person than histopathology. VCE technologies have a cost saving of about £50 per polyp resection avoided compared with histopathology. The model estimates that the correct surveillance interval would be given to 95% of patients with NBI, 94% of patients with FICE and 97% of patients with i-scan. Results are most sensitive to the pathology cost, the probability of perforation with polypectomy and the proportion of patients who die from perforation. PSAs were conducted for pairwise and incremental comparisons for histopathology with VCE technologies. The probabilistic ICERs were similar to the base-case deterministic ICERs. At a willingness-to-pay threshold of £20,000 and £30,000, i-scan was most cost-effective in 95% and 33% of simulations, respectively.

Analyses were also conducted for a surveillance population of patients who had previously had one or more adenomas detected at an earlier colonoscopy and a symptomatic patient population that had been referred for colonoscopy with symptoms suggestive of colorectal cancer. These populations had a lower risk of adenomas than the screening population. All VCE technologies were less expensive and more effective than histopathology for the surveillance population and symptomatic population analyses.

Analyses were conducted for a DISCARD strategy in which diminutive polyps in the rectosigmoid colon are optically characterised. These analyses used the diagnostic accuracy from our meta-analysis for diminutive polyps characterised with high confidence in the rectosigmoid colon (see Figure 16). All VCE technologies were less expensive and more effective than histopathology. There were smaller differences in costs and QALYs between VCE and histopathology for this analysis than for the base-case analysis.

The base-case results show that the VCE technologies are associated with cost savings compared with histopathology and small gains in QALYs. Given the large number of colonoscopies performed every year, the potential cost savings for the NHS are substantial. The cost savings are a result of a reduction in the number of polypectomies performed (with a consequent reduction of adverse events from bleeding and perforation) and polyps sent for histopathological examination. Our base-case analysis estimated that there would be around 40% fewer polypectomies performed and this would result in between 3% and 15% of adenomas left in situ with VCE and ≥ 90% fewer hyperplastic polyps resected. The model estimates that VCE would lead to incorrect surveillance intervals for between 3% and 6% of patients. The QALY gains are attributable to the reduction in adverse events, such as perforation. The QALY losses are as a result of the long-term consequences of not resecting adenomas and patients receiving incorrect surveillance intervals.

The base-case analyses indicate that the cost-effectiveness of histopathology compared with VCE varies according to the VCE technology. The differences in cost-effectiveness between the VCE technology are largely attributable to the differences in the diagnostic sensitivity of the technologies, with our meta-analysis calculating sensitivity for i-scan of 0.96 and for FICE of 0.814. We urge caution when comparing between the results of different VCE technologies, given the differences in the diagnostic accuracy studies for these technologies in our meta-analyses.

Strengths of the assessment

The strengths of this assessment include that we carried out the systematic review and economic analysis independent of competing interests, and the methods we used were prespecified in a published protocol. We sought feedback from our Expert Advisory Group on the draft protocol and incorporated its comments into the final version. The protocol was published on the National Institute for Health and Care Excellence website and was discussed by experts in the topic area recruited by National Institute for Health and Care Excellence (specialist members of the Appraisal Committee). The protocol was also published on the PROSPERO prospective register of systematic reviews website.

We critically appraised all of the diagnostic test accuracy studies included in the review using recognised criteria^38,39 to assess potential risks of bias and to assess the generalisability of the results. Our Expert Advisory Group commented on the protocol and a draft of this report, and we also sought specialist methodological input from the NIHR Complex Reviews Support Unit to conduct this assessment.

Our economic model is in line with current BSG¹⁰⁹ and ESGE³¹ guidelines, unlike other models that have examined VCE. Hassan and colleagues¹¹² assumed that all patients undergoing screening would have a repeat colonoscopy at 10 years, which is not the recommended surveillance interval in BSG or ESGE guidelines. In Kessler and colleagues,¹¹³ the polyp groups used are inconsistent with both guidelines. Kessler and colleagues¹¹³ divide patients into four groups by the types of polyps that patients have, whereas guidelines divide patients into risk groups by the number of adenomas that they have. Solon and colleagues¹¹⁷ did not examine surveillance intervals, so their study is not representative of UK practice.

Our model uses the SBCS model to generate long-term outcomes. The SBCS model was developed for the NHS Bowel Cancer Screening Programme.¹²² Using long-term outcomes from the SBCS model allows guidance to be consistent across NHS evidence streams.

In line with National Institute for Health and Care Excellence methodological guidance,¹¹⁹ we derived as much of our evidence from systematic searches as feasible. The diagnostic accuracy data were obtained from a robust systematic review and meta-analysis using appropriate bivariate meta-analysis techniques, where possible.⁴¹ Cost data were derived from appropriate NHS sources, and quality-of-life data were derived from EQ-5D and expressed in QALYs as the primary measure of benefit. Additionally, we conducted a wide variety of sensitivity analyses to explore uncertainty.

Limitations of the assessment

The evidence base for this assessment was particularly limited for FICE and to a lesser extent for i-scan. This limits the conclusions we can draw about the diagnostic accuracy of these technologies for assessing diminutive colorectal polyps in real time. None of the FICE studies we identified assessed surveillance intervals nor NPV in relation to the PIVI criteria, which meant that there was no evidence available to assess how use of FICE meets the PIVI requirements for a resect and discard strategy to be adopted using this technology in practice. Most of the studies included in this review evaluated NBI, but there was heterogeneity in the NBI studies in terms of the original purpose of the studies, country and settings, likely prevalence of adenomas (which can then impact NPV estimates), polyp classification systems used and experience of endoscopists. This makes it difficult to determine the diagnostic accuracy of NBI and to provide clear implications for practice. However, despite this heterogeneity, NBI appears to meet the PIVI requirements (with the caveat that, when reported, the lower limit of 95% CIs was sometimes below the 90% PIVI threshold), supporting its use for a resect and discard strategy in practice.

One limitation of this review is that we did not formally investigate the impact study setting has on diagnostic accuracy estimates. Some research has shown that studies conducted in academic or specialist centres tend to find better diagnostic accuracy outcomes than those conducted in generalist settings or community practice.¹⁴⁸ It is not possible to determine from this review how accurate NBI is for the real-time diagnosis of diminutive polyps when used in different settings. We also did not formally investigate the impact of the classification system used for characterising polyps in the studies. There was much variation in the reporting of the classification schemes used, which would have introduced uncertainty in assembling subgroups. Expert clinical advice suggested that diagnostic performance is unlikely to vary according to different schemes, as some of the classification schemes are derived from others (e.g. the NBI International Colorectal Endoscopic classification²⁰ is based on the Kudo scheme²² among other schemes). Caution is also advised in the interpretation of our subgroup analysis based on endoscopist’s experience with VCE, as there was variation between studies in how experience was measured and also there were small numbers of studies in the subgroups.

In order to construct an economic model for histopathology compared with VCE technology, it was necessary to make several assumptions. First, it is not reported in the studies identified how the sensitivity and specificity for individual polyps relates to the surveillance intervals for patients. Although some studies in the systematic review of diagnostic accuracy examined correct assignment of surveillance intervals, the data from these studies were insufficient to incorporate in the model. Therefore, we assumed that diagnostic accuracy data for individual polyps were applicable to the entire patient, and assigned patients into risk categories a priori using data from Raju and colleagues.¹³² When comparing our modelled outcomes with those found in the systematic review of diagnostic accuracy studies, the model’s correct prediction of surveillance intervals was similar to that found in the systematic review (see Chapter 4 for details). Furthermore, we assumed that the prevalence of adenomas was constant across risk groups with adenomas to predict the number of polyps that patients have. It may be that patients in different risk groups have different ratios of adenomas to polyps. If patients with low-risk adenomas have a higher number of polyps per adenoma than patients in the higher-risk categories, this would adversely affect the cost-effectiveness of histopathology compared with VCE, as more hyperplastic polyps would be resected and sent to histopathology.

The long-term cost and QALY outcomes derived from the SBCS model were estimated assuming use of standard colonoscopy for any follow-up surveillance. These long-term costs and QALY outcomes do not therefore show the true extent of the future colonoscopies; for example, we would expect there to be future cost savings for VCE in any future colonoscopies. It was not feasible to include our decision tree within the SBCS model. However, we included a sensitivity analysis to investigate the likely impact of including VCE, which had only a small effect on the model results. This was because the majority of patients were low risk (i.e. few of them would have repeat colonoscopy).

The economic analysis includes only diminutive polyps. Although the decision problem focuses on diminutive polyps, some people with diminutive polyps will also have larger polyps (falling into the ‘small’ and ‘large’ categories). We attempted to incorporate large and small polyps using data from studies identified in the systematic review and meta-analysis as well as targeted searches, but there were insufficient data to allow coherent analysis of larger polyps. In practice, large polyps would be assessed using only histopathology, and the effect would be an increase in the number of patients with intermediate- and high-risk adenoma (i.e. shorter surveillance intervals), and a decrease in the number of polyps characterised as adenomas in intermediate- and high-risk patients. It is this last feature of the analysis that made assessing large polyps infeasible as no data were available that indicated the number of polyps found in patients with large polyps at intermediate or high risk. Additionally, no information could be identified on what proportion of patients in the intermediate risk category had two or fewer adenomas with one adenoma being large. Including small polyps would affect only the proportion of patients assessed using only histopathology. Surveillance intervals for small polyps are identical to diminutive polyps.

Furthermore, the model does not differentiate between the type of polyp such as depressed polyps or sessile serrated polyps. No diagnostic accuracy data were identified, specifically, for either type of polyp. Additionally, sessile serrated polyps are rare and no diagnostic accuracy data were available for diminutive sessile serrated polyps from our systematic review of diagnostic studies (see Chapter 4). These polyps may be more likely to be given a low-confidence assessment, in which case they would therefore undergo histopathology.

In the absence of data on adverse events for diminutive polyps, we have used adverse event rates observed for all polyps. However, this overestimates the number of adverse events as adverse events for diminutive polyps are rarer than for larger polyps. Indeed, comments from our clinical advisors suggest that diminutive polyps are very unlikely to result in perforation. We have varied the adverse event rate in sensitivity analyses (see Table 50), where the lower estimate for adverse events for perforation and bleeding was set to zero. With these changes to the adverse event rates, the results are similar to reported in our base-case analyses.

Another uncertainty is the variation in diagnostic accuracy of VCE that would occur as a result of polyps that are unable to be successfully retrieved for histopathological analysis (e.g. as a result of fragmentation). We have noted earlier in this report (see Chapter 1, Description of the diagnostic technologies under assessment) that histopathology, despite being the accepted reference standard, is imperfect. Evidence shows that polyp retrieval failure increases significantly with smaller polyps, particularly those that are diminutive, even when resected by experienced colonoscopists. Lost polyps would be classified as adenomas, even though many would be hyperplastic. A retrospective analysis of 4383 polyps resected from 1495 patients undergoing colonoscopy in the Bowel Cancer Screening Programme reported a polyp retrieval failure rate of 6.1%. In our systematic review estimates of polyps not successfully resected for histopathological analysis, where reported, ranged from 0.5% (Basford and colleagues⁷⁹) to 13% (DISCARD³), though in most studies estimates were < 5%. The effect of this is to reduce the diagnostic accuracy of histopathology relative to that of VCE.³ We note that some polyps resected using the VCE strategy would also be sent to histopathology. We have not been able to incorporate this uncertainty into our economic analysis as a result of a lack of data to inform how this would affect all of the relevant input parameters. It may lead to a small reduction in the cost of histopathological assessment because of fewer polyps being sent to the laboratory.

The data on recurrence rates post polypectomy in the SBCS model have several limitations. The transition probabilities reported in Table 45 are not age dependent; however, the transition probabilities used in the model are age dependent. The study populations do not reflect the English bowel cancer screening population, are quite small in size, do not use the BSG surveillance guidelines to categorise adenomas, and report highly varying recurrence rates. The SBCS data on recurrence rates for people classified as intermediate risk or high risk and undergoing 1- or 3-yearly surveillance have not been updated with more recent data from the NHS Cancer Screening Programme.

The full uncertainty around the model results have not been explored in the PSA, as the long-term outcome parameters have not been varied. These data were not available from the SBCS model.