Virtual chromoendoscopy to assess colorectal polyps during colonoscopy

4.4 Seventeen studies reported on the sensitivity of NBI and 16 studies reported on the specificity of NBI for characterisations of polyps made with any level of confidence. The sensitivity ranged from 0.55 to 0.97 and the specificity ranged from 0.62 to 0.95. Bivariate meta-analysis of the 16 studies reporting on both sensitivity and specificity produced summary values of 0.88 (95% confidence interval [CI] 0.83 to 0.92) for sensitivity and 0.81 (95% CI 0.75 to 0.85) for specificity.

4.5 The sensitivity and specificity of NBI was higher for polyps diagnosed with high confidence, compared with those diagnosed with any level of confidence (that is, those assessed with low and high confidence). Eleven studies reported on the sensitivity and specificity of NBI for assessing polyps that were characterised with high confidence. Bivariate meta-analysis produced summary values of 0.91 (95% CI 0.85 to 0.95) for sensitivity and 0.82 (95% CI 0.76 to 0.87) for specificity.

4.6 A post-hoc bivariate meta-analysis was run for high-confidence characterisations, which only included studies with endoscopists who were experienced in using NBI (4 studies). The analysis produced summary values of 0.92 (95% CI 0.89 to 0.94) for sensitivity and 0.82 (95% CI 0.72 to 0.89) for specificity. Compared with the analysis for endoscopists with different levels of experience, the point estimate for sensitivity increased slightly from 0.91 to 0.92 and the specificity did not change. The confidence interval for sensitivity narrowed for experienced endoscopists compared with that for endoscopists with a variety of experience. The confidence interval for specificity for experienced endoscopists widened (0.72 to 0.89) compared with endoscopists with different levels of experience (0.76 to 0.87).

4.7 Sixteen studies reported on the negative predictive value of NBI for characterising diminutive polyps in the whole colon, made with any level of confidence. The negative predictive value ranged from 43% to 96%. The lower bound of the 95% confidence interval fell below 90% in all studies, apart from Patel et al. (2016).

4.8 Thirteen studies reported on the negative predictive value for high-confidence characterisations of polyps in the whole colon. The negative predictive value was higher for characterisations made with high confidence compared with those made with all levels of confidence. The range was 48% to 98%. When reported, the lower bound of the 95% confidence interval fell below 90% in all but 2 studies.

4.9 One study looked at the difference between the negative predictive value of characterisations done by specialists in colonoscopy and general endoscopists. The study found that specialists achieved a higher negative predictive value (90.9%; CI 70.8 to 98.9) than generalists (71.4%; 95% CI 47.8 to 88.8). However, the difference was not statistically significant.

4.10 Four studies reported on the sensitivity and specificity of NBI for assessing polyps in the rectosigmoid colon with high confidence and 3 studies reported data for assessing polyps in the rectosigmoid colon with any level of confidence. Bivariate meta-analysis for characterisations made with any level of confidence produced summary values of 0.85 (95% CI 0.75 to 0.91) for sensitivity and 0.87 (95% CI 0.74 to 0.94) for specificity. For characterisations made with high confidence, summary values were 0.87 (95% CI 0.80 to 0.92) for sensitivity and 0.95 (95% CI 0.87 to 0.98) for specificity.

4.11 A post-hoc bivariate meta-analysis was run for the 2 studies that included endoscopists who were experienced in using NBI. For high-confidence characterisations, it produced summary values of 0.90 (95% CI 0.71 to 0.97) for sensitivity and 0.98 (95% CI 0.91 to 1.00) for specificity. When compared with the bivariate analysis for endoscopists with different levels of experience, the point estimate for sensitivity increased from 0.87 to 0.90 and the point estimate for specificity increased from 0.95 to 0.98. The confidence interval for sensitivity widened for experienced endoscopists (0.71 to 0.97) compared with that for endoscopists with different levels of experience (0.80 to 0.92). The confidence interval for specificity narrowed slightly for experienced endoscopists (0.91 to 1.00) compared with that for endoscopists with different levels of experience (0.87 to 0.98).

4.12 Thirteen studies reported on the agreement between surveillance intervals set when using NBI compared with those set by histopathology; agreement ranged from 84% to 99%.

4.14 All 3 studies reported the sensitivity and specificity of FICE for characterising polyps in any part of the colon. The sensitivity ranged from 0.74 to 0.88 and the specificity ranged from 0.82 to 0.88. Bivariate meta-analysis using all 3 studies produced summary values of 0.81 (95% CI 0.73 to 0.88) for sensitivity and 0.85 (95% CI 0.79 to 0.90) for specificity. The negative predictive values ranged from 70% to 84%.

4.16 Two studies reported on high-confidence characterisations of polyps in the whole colon. Bivariate meta-analysis produced summary values of 0.96 (95% CI 0.92 to 0.98) for sensitivity and 0.91 (95% CI 0.84 to 0.95) for specificity.

4.17 Two studies reported that the negative predictive value of i‑scan for detecting colorectal polyps in the whole colon was above 90%. But, the lower bound of the confidence interval for both studies was below 90%.

4.18 Two studies reported that the negative predictive value of i‑scan for detecting colorectal polyps in the distal or rectosigmoid colon was above 90%. But, the lower bounds of the confidence interval were below 90%.

4.21 The decision tree compared the virtual chromoendoscopy strategies with a histopathology strategy. It had 4 main arms, 1 for each test that was assessed: NBI, FICE, i‑scan and standard endoscopy with histopathology. The comparator arm of the decision tree assumed that all polyps are resected and sent to histopathology and everyone is given the correct surveillance interval.

4.22 Firstly, the cohort was divided into 4 risk categories based on the number of adenomas that they have:

4.23 The model then calculated the proportion of patients in each category expected to have a correct surveillance interval assigned and the proportions expected to have an incorrect surveillance interval assigned.

4.24 With a virtual chromoendoscopy strategy, the following errors could lead to an incorrect surveillance interval (too long or too short) being assigned in the model:

4.25 The ScHARR bowel cancer screening (SBCS) model was designed to assess the cost effectiveness of different screening strategies for colorectal cancer for a lifetime time horizon. The model simulated the progression of colorectal cancer in people who are eligible for the bowel cancer screening programme in England.

4.26 The population in the base-case analysis was people taking part in the bowel cancer screening programme who had been referred for colonoscopy. Patients were included if they had at least 1 diminutive polyp (5 mm or less), and were excluded if they had 1 or more non-diminutive polyps (more than 5 mm). In addition, scenario analyses looked at:

4.27 Two different diagnostic strategies were explored in the economic analyses, the virtual chromoendoscopy strategy (used in the base case) and the DISCARD strategy (Detect, InSpect, ChAracterise, Resect, and Discard; used in some scenario analyses). The criteria common to both strategies were that diminutive polyps:

4.28 The characteristic unique to the virtual chromoendoscopy strategy was that diminutive polyps, in the whole of the colon, diagnosed with high confidence as hyperplastic are left in place.

4.29 The characteristics unique to the DISCARD strategy were that diminutive polyps:

4.30 The model inputs were taken from various sources, including routine sources of cost data, published literature, and the clinical-effectiveness review and meta-analyses.

4.31 The prevalence of adenomas was estimated for 3 populations: the screening population (base case), the surveillance population (scenario analysis) and the symptomatic population (scenario analysis). For the base-case analysis on the screening population, the prevalence of adenomas was taken from a published study by Raju et al. (2013) that retrospectively analysed data from a US colon cancer screening programme. The distributions of adenomas and the data sources for each population are reported in table 1.

4.32 Data on diagnostic accuracy were taken from the clinical-effectiveness review and meta-analysis for NBI, FICE and i‑scan, as shown in table 2. Data were used for polyps in the whole colon that were characterised with high confidence in the base-case analysis for NBI and i‑scan. Data were used for polyps in the whole colon that were characterised with any level of confidence in the base-case analysis for FICE. It was assumed that the proportion of low-confidence characterisations was the same for all 3 technologies, and was calculated using data from 12 NBI studies, because data were not available for FICE and i‑scan. The comparator, histopathology, was assumed to be 100% accurate.

Abbreviations: CI, confidence interval; EAG, external assessment group; FICE, flexible spectral imaging colour enhancement; NBI, Narrow Band Imaging.

4.33 The probabilities of adverse events occurring during colonoscopy were assumed to be 0.003 for hospitalisation for bleeding with polypectomy, 0.003 for perforation with polypectomy, and 0.052 for death of patients with perforation during polypectomy. These values were taken from published values used in the SBCS model.

4.34 For the base-case analysis, the costs of colonoscopy, polypectomy, adverse events and histopathology were taken from the NHS reference costs for 2014/15 (see table 3). Training costs were assumed to be £14.72 per patient, based on the assumption that endoscopists complete 150 endoscopies per year and that training costs are equivalent to 2 days of pay (£1,104) per year.

4.35 The cost of upgrading equipment was not included in the model. It was assumed that most hospitals already had equipment with virtual-chromoendoscopy-enabled technology in place, and hospitals that do not have this equipment will get it in the future as part of standard procurement. Therefore, the base-case analysis assumes that the cost of maintaining and purchasing equipment is included in the Healthcare Resource Group (HRG) cost of colonoscopy.

4.36 Health-related quality of life was calculated in the SBCS model. The base-case analysis used utility values taken from a study by Ara and Brazier (2011). The model assumes a utility of 0.697 for people with cancer and a utility of 0.798 for people without cancer.

4.37 A scenario analysis was done using utility values from a study identified by the EAG through a targeted search (Farkkila et al. 2013). For the scenario analysis, it was assumed that the utility for people with cancer was 0.761 and for people without cancer was 0.798.

4.38 No disutility values for adverse events during polypectomy, such as bowel perforation and bleeding, were found. Therefore, the values were taken from studies that reported on similar events. A QALY loss of 0.006 was taken from Dorian et al. (2014) for the disutility of a major gastrointestinal bleed and a QALY loss of 0.010 was taken from Ara and Brazier (2011) for the disutility of bowel perforation.

4.39 The costs and QALYs for the end points of the decision tree were calculated by running the SBCS model with a cohort of patients aged 65.

4.40 The following changes were made to the SBCS model for this assessment:

4.46 The EAG did 12 scenario analyses, and a further 2 scenario analyses were done as an addendum to the assessment report. Fewer scenario analyses were done for FICE, because data were unavailable. Results of the scenario analyses show that NBI and i‑scan were dominant in all scenario analyses when compared with histopathology.

4.47 When FICE was compared with histopathology, it was cost effective in all scenario analyses. FICE was cheaper and more effective than histopathology and therefore was dominant when:

4.48 When alternative utility values were used from Farkkila et al. (2013), FICE was cheaper and slightly less effective compared with histopathology and had an ICER of £1,273,941 saved per QALY lost.

4.49 When diagnostic accuracy data were used from studies that reported data for endoscopists experienced in using NBI for the whole colon and the rectosigmoid colon, the results were similar to the base-case analyses for virtual chromoendoscopy and NBI dominated histopathology.

4.50 The effect of using virtual chromoendoscopy (NBI) for surveillance was explored and found to be small; it was estimated to increase cost savings by £20 and increase QALYs gained by 0.0003.

4.51 The EAG produced an addendum with 2 scenario analyses on adverse events. The first analysis varied the rate of perforation during colonoscopy using ratios from the data in Rutter et al. (2014), and found that cost savings for all 3 technologies decreased slightly in relative and absolute terms, and the QALYs decreased slightly in absolute terms, whereas the relative change was large (see table 5). NBI and i‑scan still dominated histopathology and the ICER for FICE increased to £126,229 saved per QALY lost. The second analysis included the risk of an adverse event happening during all colonoscopies, as well as for colonoscopies with polypectomy. This analysis also used data from Rutter et al. and found that cost savings for all 3 technologies decreased slightly in relative and absolute terms, and the QALYs decreased slightly in absolute terms, whereas the relative change was large (see table 6). NBI and i‑scan still dominated histopathology and the ICER for FICE increased to £342,438 saved per QALY lost.

Abbreviations: FICE, flexible spectral imaging colour enhancement; Inc, incremental; NBI, Narrow Band Imaging; QALY, quality-adjusted life year.

Abbreviations: FICE, flexible spectral imaging colour enhancement; inc, incremental; NBI, narrow band imaging; QALY, quality-adjusted life year.

4.52 The one-way deterministic sensitivity analyses found that the parameters with the most influence on the cost effectiveness of the tests were pathology cost, the probability of perforation with polypectomy, and the proportion of patients who die from perforation. All one-way sensitivity analyses showed that NBI, FICE and i‑scan were cost effective compared with histopathology at a maximum acceptable ICER of £30,000 per QALY gained.

4.53 The EAG did a probabilistic sensitivity analysis by varying the base-case inputs for the decision tree. The analysis was done by running the model 5,000 times. Each time it was run, the inputs were varied according to the distribution of the input.

4.54 The probabilistic sensitivity analysis found that i‑scan was more likely to be cost effective than NBI and FICE. At a maximum acceptable ICER of £20,000 per QALY gained, i‑scan was cost effective in 85.2% of the analyses, and at a maximum acceptable ICER of £30,000 per QALY gained i‑scan was cost effective in 99.5% of the analyses.