4 Evidence

The diagnostics advisory committee (section 8) considered evidence on quantitative faecal immunochemical tests to assess people presenting to primary care who have symptoms but are at a low risk of colorectal cancer, from several sources. Full details of all the evidence are in the committee papers.

Clinical effectiveness

4.1 In total, 10 studies met the inclusion criteria for the systematic review. The studies were reported in 25 published papers and 2 unpublished manuscripts. Additional unpublished data were obtained for 2 of the published studies. Two of the included studies (Krivec et al. 2011; Thomas et al. 2016) were reported as conference abstracts only. Studies were included if they reported data for 1 of the intervention technologies in the scope and recruited people with lower abdominal symptoms who were being investigated for possible colorectal cancer. All included studies were appraised using the QUADAS‑2 tool if they reported diagnostic accuracy data and the PROBAST checklist if they also reported data for risk-prediction scores.

4.2 All of the included studies were diagnostic cohort studies; no randomised controlled trials or controlled clinical trials were identified. All 10 included studies were done in Europe, 1 of which was based in England (Thomas et al. 2016) and 3 in Scotland (Godber et al. 2016; McDonald et al. 2013; Mowat et al. 2015). Five of the studies had a high risk of bias. There were concerns about applicability for all of the included studies because none of them reported data that were specific to the population included in the scope of the assessment, that is, people with symptoms who are judged to be at low risk of colorectal cancer. Only 1 study (Mowat et al. 2015) was done in primary care.

4.3 The included studies reported data for the HM‑JACKarc, FOB Gold and OC Sensor assays only. No relevant data were found for the RIDASCREEN haemoglobin or the RIDASCREEN haemoglobin/haptoglobin assay. None of the included studies provided comparative accuracy data for the included technologies or made comparisons with guaiac-based faecal occult blood tests.

Diagnostic accuracy

4.4 The bivariate/hierarchical summary receiver operating characteristic (HSROC) model was used to calculate summary sensitivity and specificity estimates and to create HSROC curves for meta-analyses, which included 4 or more studies. For meta-analyses that included fewer than 4 studies, separate pooled estimates of sensitivity and specificity were calculated using random-effects logistic regression. Data were grouped by assay, target condition and the threshold used to determine a positive test.

OC Sensor test

4.5 Five studies reported data for the OC Sensor assay. One used the iO analyser (Mowat et al. 2015), 1 used the OC Sensor Diana analyser (McDonald et al. 2013), 2 used the MICRO desktop analyser (Rodriguez-Alonso et al. 2015; Terhaar sive Droste et al. 2011) and the fifth study did not report which analyser was used (Cubiella et al. 2014). All 5 studies reported diagnostic accuracy for colorectal cancer, although the prevalence of colorectal cancer ranged from 2.1% to 12.3%. Mowat et al. (2015) was the only study done in primary care. All studies reported the accuracy of a single faecal sample only and used varying thresholds to determine a positive test. A summary of the results is shown in table 1. Additional data from Terhaar sive Droste et al. (2011) were unpublished when this guidance was written and cannot be reported here.

Table 1 Accuracy of the OC Sensor for colorectal cancer

Study

Threshold

(micrograms Hb/g faeces)

Sensitivity %
(95% CI)

Specificity %
(95% CI)

Any detectable haemoglobin: 2 studies

Mowat et al. 2015

0

100 (87.7, 100)

43.4 (39.7, 47.1)

Rodriguez-Alonso et al. 2015

0

100 (88.4, 100)

43.3 (40.1, 46.4)

Summary estimate

100 (93.8, 100)

43.3 (40.9, 45.7)

10 micrograms Hb/g faeces: 4 studies (1 unpublished)

McDonald et al. 2012

≥10

100 (54.1, 100)

93.8 (90.3, 96.3)

Mowat et al. 2015

≥10

89.3 (71.8, 97.7)

79.1 (75.9, 82)

Rodriguez-Alonso et al. 2015

≥10

96.7 (82.8, 99.9)

79.9 (77.2, 82.3)

Summary estimate

92.1 (86.9, 95.3)

85.8 (78.3, 91.0)

15 micrograms Hb/g faeces or equivalent: 2 studies (1 unpublished)

Rodriguez-Alonso et al. 2015

≥15

96.7 (82.8, 99.9)

83.1 (80.6, 85.4)

Summary estimate

92.3 (86.6, 96.1)

86.9 (85.6, 88.1)

20 micrograms Hb/g faeces or equivalent: 3 studies (1 unpublished)

Cubiella et al. 2014

≥20

87.6 (79.0, 93.2)

77.4 (74.0, 80.4)

Rodriguez-Alonso et al 2015

≥20

93.3 (77.9, 99.2)

86.1 (83.8, 88.2)

Summary estimate

89.5 (84.9, 93.1)

86.6 (85.4, 87.7)

Abbreviations: CI, confidence interval; Hb, haemoglobin.

4.6 The external assessment group (EAG) considered that the optimal diagnostic threshold for colorectal cancer was either 10 or 15 micrograms of haemoglobin (Hb)/g faeces, but noted that most data were available for 10 micrograms Hb/g faeces. Test accuracy data from Mowat et al. (2015) and Rodriguez-Alonso et al. (2015) were used to illustrate diagnostic outcomes for a hypothetical cohort of 1,000 people, assuming a prevalence of colorectal cancer of 3.3%, and using thresholds of both 10 micrograms Hb/g faeces and any detectable haemoglobin (4 micrograms Hb/g faeces). The results are shown in table 2.

Table 2 Modelled outcomes for the OC Sensor test (colorectal cancer)

Threshold

10 micrograms Hb/g faeces

4 micrograms Hb/g faeces

Correct referrals for colonoscopy (true positives)

31

33

Incorrect referrals for colonoscopy (false positives)

198

548

Missed colorectal cancers (false negatives)

2

0

Colonoscopies correctly avoided (true negatives)

769

419

Abbreviation: Hb, haemoglobin.

4.7 Four studies (Cubiella et al. 2014; McDonald et al. 2012; Mowat et al. 2015; Rodriguez-Alonso et al. 2015) reported diagnostic accuracy for advanced neoplasia, which includes both colorectal cancer and high-risk adenoma. The definition of high-risk adenoma and the thresholds used varied between studies. Expanding the target condition reduced the sensitivity of the test, with summary sensitivity estimates of 62.9% (95% confidence interval [CI] 55.9% to 69.4%) at a threshold of 10 micrograms Hb/g, 63.9% (95% CI 58.2% to 69.2%) at a threshold of 20 micrograms Hb/g and 84.1% (95% CI 78.3% to 88.8%) at a threshold of any detectable haemoglobin. The sensitivity of the test was lower when the target condition was expanded to include other bowel pathologies. But data from studies that reported results for both colorectal cancer and high-risk adenoma suggested that many false-positive results for colorectal cancer could be from other bowel pathologies that may benefit from treatment.

4.8 Three studies reported diagnostic accuracy data for various non‑malignant or composite target conditions. McDonald et al. (2012) reported a sensitivity of 57.0% (95% CI 45.8% to 67.6%) and a specificity of 99% (95% CI 96.3% to 99.9%) for all colorectal cancers, high-risk adenomas and inflammatory bowel disease using a threshold of 10 micrograms Hb/g faeces. Mowat et al. (2015) used the same threshold and reported a sensitivity of 68.6% (95% CI 58.7% to 77.5%) and a specificity of 83.6% (95% CI 80.6% to 86.4%) for the same composite target condition. Additional data from Terhaar sive Droste et al. (2011) were unpublished at the time of writing so cannot be reported here.

HM-JACKarc system

4.9 Three studies reported accuracy data for the HM‑JACKarc automated system (Auge et al. 2016; Godber et al. 2016; Thomas et al. 2016). All 3 studies were done in outpatient clinics and used single faecal samples.

4.10 Two studies (Godber et al. 2016; Thomas et al. 2016) reported accuracy data for colorectal cancer. The prevalence of colorectal cancer was 2.2% in Godber et al. and 4.9% in Thomas et al. Godber et al. reported a sensitivity of 100% (95% CI 71.5% to 100%) and a specificity of 76.6% (95% CI 72.6% to 80.3%) at a threshold of 10 micrograms Hb/g faeces. Thomas et al. reported a sensitivity of 91.3% (95% CI 72.0% to 98.9%) and a specificity of 79.2% (95% CI 75.3% to 83.0%) at a threshold of 7 micrograms Hb/g faeces. Test accuracy data from Godber et al. were used to model outcomes for a hypothetical cohort of 1,000 people, assuming a prevalence of colorectal cancer of 2.2%. The results of this analysis are shown in table 3.

Table 3 Modelled outcomes for the HM-JACKarc assay (colorectal cancer)

Threshold of 10 micrograms of haemoglobin/g of faeces

Correct referrals for colonoscopy (true positives)

22

Incorrect referrals for colonoscopy (false positives)

229

Missed colorectal cancers (false negatives)

0

Colonoscopies correctly avoided (true negatives)

749

4.11 Two studies (Auge et al. 2016; Godber et al. 2016) reported data for a target condition of colorectal cancer and high-risk adenoma. Each study used a different definition of high-risk adenoma and reported different thresholds. The sensitivity estimates varied widely because of differences in the included populations. Godber et al. reported a sensitivity of 70.0% (95% CI 50.6% to 85.3%) and a specificity of 77.8% (95% CI 73.8% to 81.4%) at a threshold of 10 micrograms Hb/g faeces. Auge et al. reported a range of accuracy estimates with sensitivity ranging from 27.6% (95% CI 14.7% to 45.7%) at a threshold of 40 micrograms Hb/g faeces to 96.6% (95% CI 82.8% to 93.4%) at a threshold of any detectable haemoglobin. Specificity ranged from 10.6% (95% CI 6.9% to 15.9%) at a threshold of any detectable haemoglobin to 93.9% (95% CI 89.4% to 96.6%) at a threshold of 40 micrograms Hb/g faeces.

4.12 One study (Auge et al. 2016) also investigated the effect of multiple samples and sex on the accuracy of the HM‑JACKarc assay for detecting colorectal cancer and high-risk adenoma. The study had a prevalence of colorectal cancer of less than 1%. The authors reported that 100% sensitivity could be achieved by using a threshold of any detectable haemoglobin and using the highest value reported in 2 consecutive samples, but this reduced the specificity to 3.3%. Data were reported for single or multiple samples using a range of thresholds from any detectable haemoglobin to 40 micrograms Hb/g faeces. At thresholds above any detectable haemoglobin, using consecutive samples increased the test's sensitivity but this was still low at under 50% for all estimates.

4.13 Auge et al. (2016) also reported that sensitivity estimates at all thresholds were lower when the test was used in women than when used in men. Sensitivity estimates ranged from 8.3% at a threshold of 40 micrograms Hb/g faeces to 91.7% with any detectable haemoglobin for women, compared with a range of 41.2% at all thresholds above 20 micrograms Hb/g faeces to 100% with any detectable haemoglobin for men. Conversely, specificity estimates tended to be higher in women than in men.

4.14 Two studies (Godber et al. 2016; Thomas et al. 2016) reported accuracy data for various non‑malignant and composite target conditions. Godber et al. defined significant bowel disease as colorectal cancer, higher-risk adenoma, inflammatory bowel disease or colitis. They reported sensitivity and specificity estimates of 68.9% and 80.2% respectively at a threshold of 10 micrograms Hb/g faeces. Thomas et al. defined significant bowel disease as colorectal cancer, high-risk adenoma or inflammatory bowel disease. They reported sensitivity and specificity estimates of 72.1% and 80.6% respectively at a threshold of 7 micrograms Hb/g faeces.

FOB Gold assay

4.15 Two studies reported data for the FOB Gold assay. One was reported in a conference abstract only and used the Roche Modular P/917 analyser (Krivec et al. 2011). The other was unpublished at the time of writing and used the SENTiFIT270 analyser (Hospital Clinic de Barcelona 2015). Further data from Hospital Clinic de Barcelona are unpublished and cannot be reported here. Krivec et al. (2011) reported a sensitivity of 45.2% and a specificity of 92.3% for significant bowel disease (cancer, polyps or bleeding) using a threshold of 9.35 micrograms Hb/g faeces.

Test failures

4.16 Mowat et al. (2015) reported that fewer than 1% of samples were considered unsuitable for analysis using the OC Sensor test.

Test uptake

4.17 Four of the included studies reporting data for the OC Sensor reported test uptake (Cubiella et al. 2014; McDonald et al. 2013; Mowat et al. 2015; Rodriguez-Alonso et al. 2015), which ranged from 41% to 98%. Methods of inviting patients to take a test varied between studies.

4.18 Two of the included studies reporting data for the HM‑JACKarc reported test uptake. Godber et al. (2016) reported an uptake of 56% when collection devices and information were sent by post, whereas Thomas et al. (2016) reported an uptake of 66% when collection devices and information were provided at an outpatient appointment.

Management decisions

4.19 Mowat et al. (2015) reported that 11% of patients for whom a faecal immunochemical test sample was analysed were not referred to secondary care, 69% were referred for an endoscopy and 20% were referred to an outpatient clinic. However, decisions about the urgency of the referral were made before the test.

Prediction modelling studies

4.20 Two studies (Cubiella et al. 2016; Rodriguez-Alonso et al. 2015) reported data on using prediction models, which included results of faecal immunochemical tests. These studies were also appraised with the PROBAST tool. The studies were classified as having high concerns about the applicability of the included populations, and overall were rated as being at a high risk of bias.

4.21 Rodriguez-Alonso et al. (2015) did a multivariate analysis to identify independent predictors of colorectal cancer and advanced neoplasia. Faecal haemoglobin was measured using the OC Sensor assay. The model included age as a categorical variable. The following variables were identified as independent predictors of colorectal cancer:

  • male sex (odds ratio [OR] 2.39; 95% CI 1.039 to 5.519; p=0.041)

  • iron-deficiency anaemia (OR 2.99; 95% CI 1.27 to 7.03; p=0.012)

  • faecal haemoglobin (OR 86.60; 95% CI 11.70 to 64.16; p<0.001).

4.22 A pre-publication copy of a manuscript by Cubiella et al. (2016) reported the development and validation of a risk score known as the FAST score (faecal haemoglobin, age and sex test). Faecal haemoglobin was measured using the OC Sensor, OC‑Auto (an earlier version of the OC Sensor) and FOB Gold assays. The logistic regression model included age as a continuous variable, and sex and faecal haemoglobin as categorical variables. The results of the model suggested that a FAST score of 4.5 had a sensitivity of 89.3% (95% CI 84.1% to 93.0%) and a specificity of 82.3% (95% CI 81.1% to 83.5%) for colorectal cancer. To avoid missing any colorectal cancers, a lower FAST score threshold of 2.12 was needed. This gave a sensitivity of 100% (95% CI 97.7% to 100%) and a specificity of 19.8% (95% CI 18.6% to 21.1%).

Cost effectiveness

Review of economic evidence

4.23 Only 1 study was found that reported an economic analysis of using faecal immunochemical tests in people with symptoms; the economic analysis of faecal occult blood tests in NICE's guideline on suspected cancer. Faecal immunochemical tests were included in a scenario analysis in the guideline. Faecal occult blood tests were included in the base case, which showed that guaiac-based tests and barium enema were cost effective compared with colonoscopy at a maximum acceptable incremental cost-effectiveness ratio (ICER) of £20,000 per quality-adjusted life year (QALY) gained. In the scenario analysis, faecal immunochemical tests dominated (cost less and were more effective) barium enema and were cost effective at a maximum acceptable ICER of £20,000 per QALY gained.

Modelling approach

4.24 The EAG developed a de novo economic model to explore the cost effectiveness of using a quantitative faecal immunochemical test to guide referral of people who present to primary care with symptoms but have a low risk of colorectal cancer. The model took the perspective of the NHS and personal social services. In the base case it compared the use of 2 quantitative faecal immunochemical tests, the OC Sensor and HM‑JACKarc assays, with both guaiac-based faecal occult blood tests and no triage (that is, referral straight to colonoscopy). A watchful waiting strategy, which may currently be used in practice, was not included as a comparator because of variability in practice and a lack of data, but was incorporated into the guaiac-based faecal occult blood and faecal immunochemical testing strategies. The FOB Gold assay was not included in the base case because no data were available for the optimal threshold (10 micrograms Hb/g faeces) determined by the EAG in the clinical-effectiveness analyses. All costs and effects included in the model were discounted by 3.5%.

Model structure

4.25 The model had 3 parts. The first part was a decision tree with a 1‑year time horizon, which modelled the results of investigations for colorectal cancer (faecal immunochemical test, guaiac-based faecal occult blood test or no triage) for a cohort of patients, with symptoms, presenting to primary care. A positive faecal immunochemical test or guaiac-based faecal occult blood test resulted in referral for colonoscopy and a negative test resulted in watchful waiting, in which further investigations were done if a person's symptoms persisted. The decision tree was followed by 2 Markov state-transition models. One Markov model had a lifetime time horizon and a 1‑year cycle length and was used to estimate costs, life years and QALYs associated with the treatment and progression of colorectal cancer. The initial distribution of patients across the stages of disease at diagnosis was determined using data from the UK's National Cancer Intelligence Network. The other Markov model had a simple alive or dead structure and estimated life years and QALYs for people who did not have colorectal cancer, using UK life tables to model survival.

Model inputs

4.26 The model was populated with data from the clinical-effectiveness review, published literature and expert opinion. Diagnostic accuracy data were taken from the clinical-effectiveness review. The EAG concluded that a threshold of 10 micrograms Hb/g faeces with a single sample provided the optimal rule-out performance. That is, the threshold gave the maximum sensitivity and specificity, and had the lowest number of colorectal cancers missed. Data at this threshold were available for the HM‑JACKarc and OC Sensor assays. Data on the accuracy of guaiac-based faecal occult blood tests were taken from Gillberg et al. (2012), which was used in the economic model for NICE's guideline on suspected cancer. The accuracy estimates used in the base-case analysis are shown in table 4. The predictive values were calculated by the EAG assuming a prevalence of colorectal cancer of 1.5% to correspond with the prevalence assumed in NICE's guideline on suspected cancer.

Table 4 Diagnostic accuracy estimates used in the base-case model

Accuracy measure

OC Sensor assay

HM‑JACKarc assay

Guaiac-based faecal occult blood test

Sensitivity
(95% CI)

92.1%
(86.9% to 95.3%)

100%
(71.5% to 100%)

50%
(15.0% to 85.0%)

Specificity
(95% CI)

85.8%
(78.3% to 91.0%)

76.6%
(72.6% to 80.3%)

88%
(85.0% to 89.0%)

Positive predictive value

8.9%

6.1%

5.7%

Negative predictive value

99.8%

100%

99.1%

Abbreviation: CI, confidence interval.

Costs

4.27 Direct costs included in the model were test costs, cost of colonoscopy or CT colonography, adverse-event costs, CT scan costs, costs of first and follow-up investigations, cancer staging and treatment, drug costs, and GP and hospital visits. No costs were included in the Markov model used for outcomes for people without colorectal cancer. Costs were obtained from companies, published literature and routine sources of NHS costs. Test costs were calculated as average costs per test. The following test costs were used in the model:

  • OC Sensor: £4.53

  • HM‑JACKarc: £6.04

  • FOB Gold: £1.96

  • guaiac-based faecal occult blood test: £0.78 (rounded to 2 significant figures)

  • colonoscopy: £372

  • CT colonography: £136

  • CT scan: £116.

Health-related quality of life and quality-adjusted life year decrements

4.28 No disutilities for bleeding and perforation during colonoscopy were included in the model, because no evidence was found on quality-of-life effects in the literature and the events are often of short duration. The rates of adverse events from colonoscopy were assumed to be 0.26% for bleeding, 0.05% for perforation, and 0.0029% for death. Utilities associated with the different stages of colorectal cancer were taken from Ness et al. (1999) and sex- and age-related utilities for healthy patients were taken from Kind et al. (1999).

Base-case results

4.29 The following key assumptions were applied in the base-case analysis:

  • People who had a false-negative faecal immunochemical test or guaiac-based faecal occult blood test and whose symptoms persisted were diagnosed within 1 year if they survived.

  • The optimal threshold for the interventions was 10 micrograms Hb/g faeces.

  • People who had a delayed diagnosis had an increased probability of progressing to a more advanced cancer state.

  • Costs of laboratory staff were the same for both faecal immunochemical tests and guaiac-based faecal occult blood tests.

  • Testing had no long-term (after 1 year) effect on costs or QALYs in people without colorectal cancer.

  • Any differences in costs between the tests in patients without colorectal cancer occurred in year 1 only.

  • The prevalence of colorectal cancer was 1.5%.

  • The probabilities of adverse events during or after colonoscopy were as follows:

    • bleeding: 0.26%

    • bowel perforation: 0.05%

    • death: 0.0029%.

  • Only patients with negative test results whose symptoms did not persist did not have a colonoscopy.

  • The cost of a colonoscopy or CT colonography included a follow-up appointment with a gastroenterologist.

  • The adverse-event rates associated with CT colonography were the same as for colonoscopy.

  • A CT scan was done for all patients with colorectal cancer to stage the disease.

  • After year 15 in the colorectal cancer Markov model, colorectal-cancer-related mortality remains constant, but overall mortality increases because age-specific mortality is included from UK life tables.

4.30 The results of the base case are shown with the fully incremental probabilistic analysis in table 5 and the probabilistic pairwise comparisons in table 6. The ICERs for the deterministic analysis were slightly higher than those in the probabilistic analysis.

Table 5 base-case results – fully incremental probabilistic analysis

QALYs

Cost

Incremental QALYs

Incremental cost

ICER

gFOBT

18.6415

£230.49

OC Sensor assay

18.6439

£242.51

0.0024

£12.02

£5,039

No triage

18.6440

£500.60

Dominated by HM‑JACKarc

HM‑JACKarc assay

18.6444

£272.50

0.0005

£29.99

£61,619

Abbreviations: gFOBT, guaiac-based faecal occult blood test; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

Table 6 base-case results – probabilistic pairwise comparisons

Intervention

Comparator

Incremental QALYs

Incremental costs

ICER

HM‑JACKarc assay

gFOBT

0.0029

£42.01

£14,626

OC Sensor assay

0.0024

£12.02

£5,039

HM‑JACKarc assay

No triage

0.0004

−£228.10

Dominates

OC Sensor assay

−0.0001

−£258.09

£2,578,543a

a savings per QALY lost.

Abbreviations: gFOBT, guaiac-based faecal occult blood test; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life year.

4.31 The pairwise results suggest that both the OC Sensor and the HM‑JACKarc assays are cost effective compared with both guaiac-based faecal occult blood testing and no triage. The fully incremental probabilistic analysis suggests that the OC Sensor assay is cost effective. Despite dominating no triage, the HM‑JACKarc assay has a high ICER compared with the OC sensor because of the very small difference in QALYs and higher cost, which is accounted for by the test having more positive results and so a higher number of colonoscopies.

4.32 A breakdown of the costs and outcomes in the base case showed that the number of positive tests was highest for the HMJACK‑arc assay (245.36) and lowest for guaiac-based faecal occult blood testing (130.28). The OC Sensor assay had 153.50 positive tests. The increased number of positive tests increases the costs for faecal immunochemical tests because of the associated increase in colonoscopies. No colorectal cancer patients were missed with the HM‑JACKarc assay and so no delayed diagnosis occurred with this test. By comparison, 92% of colorectal cancers were detected by the OC Sensor assay and 50% with guaiac-based faecal occult blood tests.

4.33 The cost-effectiveness acceptability curves for all strategies show that at lower maximum acceptable ICERs, the tests associated with the lowest costs have the greatest probability of being cost effective, that is guaiac-based faecal occult blood testing and the OC Sensor assay. As the maximum acceptable ICER increases, the HM‑JACKarc assay and guaiac-based faecal occult blood testing have the greatest probability of being cost effective. Pairwise comparisons showed that, when compared with faecal immunochemical testing, no triage would be cost effective only when the maximum acceptable ICER is very high. There was more uncertainty about which strategy was the most cost effective when the faecal immunochemical tests were compared with guaiac-based faecal occult blood testing.

Analysis of alternative results

Test accuracy

4.34 The effect of changing assumptions about the accuracy of the tests was explored in several scenario analyses. Using alternative sources of accuracy data for guaiac-based faecal occult blood tests did not substantially alter the conclusions. The faecal immunochemical tests remained cost effective compared with guaiac-based faecal occult blood tests.

4.35 When a threshold of any detectable haemoglobin was considered for the OC Sensor test it resulted in an increased number of colonoscopy referrals and an ICER of £65,192 compared with guaiac-based faecal occult blood tests.

4.36 When a threshold of 20 micrograms Hb/g faeces was considered, and the FOB Gold assay was included in the analysis using a threshold of 20.5 micrograms Hb/g faeces, this resulted in an ICER of £4,725 per QALY gained for the FOB Gold assay compared with guaiac-based faecal occult blood tests. When compared with no triage, £950,152 was saved per QALY lost.

4.37 The FOB Gold assay was included in a scenario analysis with the base-case settings, but with a threshold of 6.8 micrograms Hb/g faeces, compared with 10 micrograms Hb/g faeces for the HM‑JACKarc and OC Sensor assays. The ICER for the FOB Gold compared with guaiac-based faecal occult blood testing was £15,720 per QALY gained, and compared with no triage was £2,273,829 saved per QALY lost.

Prevalence of colorectal cancer

4.38 Scenario analyses were done in which the prevalence of colorectal cancer was increased from 1.5% in the base case to 3% and 5.4%. Increasing the prevalence reduced the ICERs for the interventions compared with guaiac-based faecal occult blood testing. However, at 5.4% prevalence the ICER for the OC Sensor test compared with no triage became less cost effective, from £4,133,559 saved per QALY lost to £238,380 saved per QALY lost.

Test costs

4.39 A threshold analysis showed that for the ICER to remain below £30,000 per QALY gained the cost of the HM‑JACKarc test could be up to £32 more expensive than guaiac-based faecal occult blood tests. The OC Sensor assay could be up to £51 more expensive than guaiac-based faecal occult blood tests.

Initial or delayed diagnosis

4.40 In the base case, the following distribution of patients across the Dukes' stages was assumed in the colorectal cancer Markov model:

  • stage A: 13%

  • stage B: 37%

  • stage C: 36%

  • stage D: 14%.

4.41 When it was assumed that there were more patients in stages A and C (16% and 44% respectively) and fewer patients in stage B (25%) there was a slight loss of QALYs and reduction in costs for all strategies.

4.42 When it was assumed that there were more patients in stages A and D (19% and 15% respectively) and fewer in stages B and C (35% and 32% respectively) there was a slight gain in QALYs and an increase in costs for all strategies.

Colorectal cancer mortality and progression

4.43 When colorectal cancer progression was not considered in the model, the ICERs reduced from the base-case results for the HM‑JACKarc and the OC Sensor tests compared with guaiac-based faecal occult blood testing. The ICER for the OC Sensor compared with no triage became less cost effective at £163,305 saved per QALY lost.

Probability of symptoms persisting

4.44 When the probability of symptoms persisting after a negative test was doubled from the base case (65%), the interventions remained cost effective despite increased costs from increased colonoscopies and a slight reduction in QALYs. When the probability of symptoms persisting was halved from the base case (16.25%), the interventions remained cost effective with a slight QALY increase and reduction in costs.

Adverse events for colonoscopy

4.45 When a mortality rate of 0.0970% was considered for colonoscopy in a worst case scenario, strategies associated with a higher rate of referrals to colonoscopy (no triage and HM‑JACKarc assay) were dominated by guaiac-based faecal occult blood testing and the OC Sensor test respectively. When it was assumed that there are no adverse events associated with colonoscopy, no triage was dominated by the HM‑JACKarc assay because it provided an equivalent number of QALYs, but cost £227.30 less.

Probability of having CT colonography

4.46 When it was assumed that all referrals were to colonoscopy (compared with 88.3% in the base case), with no CT colonography, the cost of each of the testing strategies increased compared with the base case because of the increased cost of colonoscopy.

Probability of having a second index test

4.47 When it was assumed that 20% of patients, who remained symptomatic after a negative faecal immunochemical test or guaiac-based faecal occult blood test, had a second test, the cost of the faecal testing strategies increased, but not enough to affect the overall results.

  • National Institute for Health and Care Excellence (NICE)