Faecal immunochemical tests to triage patients with lower abdominal symptoms for suspected colorectal cancer referrals in primary care: a systematic review and cost-effectiveness analysis

Units for faecal immunochemical tests for haemoglobin

Different FITs use a variety of sampling methods, with variation in the mass of faeces collected and the volume and characteristics of the buffer used in the sampling device. These differences make the comparisons of test performance between FIT assay types difficult. FIT results can be expressed as Hb concentration in the sampling device buffer (ng Hb/ml buffer) or as Hb concentration by mass of faeces (µg Hb/g faeces). Initiatives aimed at standardising the units of measurement have resulted in recommendations, from the World Endoscopy Organization Colorectal Cancer Screening Committee’s Expert Working Group on ‘FIT for Screening’17 and the Guildford Medical Device Evaluation Centre,18 that manufacturers should adopt the use of ‘µg Hb/g faeces’.

OC-Sensor

OC-Sensor (Eiken Chemical Co./MAST Diagnostics, Tokyo, Japan) is a quantitative faecal immunochemical test. A sample is collected on a probe and inserted immediately into a unique specimen collection device that contains buffer. Analysis is fully automated using the OC-PLEDIA analyser or the OC-Sensor IO analyser – both quantitatively determine the concentrating of Hb present in faecal samples using polyclonal antibodies for human Hb and latex agglutination turbidimetry. 19,20 The OC-PLEDIA can process up to 320 samples per hour, with a capacity of 200 samples per run. The OC-Sensor IO analyser can process up to 88 samples per hour, with a maximum capacity of 20 samples per run.

HM-JACKarc system

The HM-JACKarc system (Kyowa Medex/Alpha Laboratories Ltd, Tokyo, Japan) is a fully automated quantitative faecal immunochemical test. A sample is obtained using the insertion of a probe attached to the cap of the specimen collection device, which is then inserted into a specialised collection tube containing buffer. The system picks up a small volume from the specimen collection devices and adds reagents, including latex reagent pre-coated with antibodies that are specific to the globin moiety of human Hb. Binding of the latex reagent to globin that is present in the faecal sample creates a complex that can be detected using turbidimetry. The system comprises an analyser, faecal sample collection devices (the Extel Hemo-auto MC A device; Kyowa Medex/Alpha Laboratories Ltd, Tokyo, Japan), latex agglutination reagents (Extel Hemo-Auto HS; Kyowa Medex/Alpha Laboratories Ltd) and buffer (Extel Hemo-auto; Kyowa Medex/Alpha Laboratories Ltd). The test has a measuring range of 7–400 µg Hb/g faeces. The HM-JACKarc analyser can process up to 200 samples per hour, with a maximum capacity of 80 samples per run. 21

FOB Gold

The FOB Gold system (Sentinel/Sysmex, Sentinel Diagnostics, Milan, Italy) is an automated quantitative faecal immunochemical test. Faecal samples are collected on probes, which are immersed immediately into solution within the specimen collection device. This ensures sample stability (14 days at 2–8 °C or 7 days at 15–30 °C). The devices are then placed into an automated analyser. A latex agglutination assay is used which is detected via turbidimetry. 22 The FOB Gold kit has CE (Conformité Européene)-marked applications for a range of clinical chemistry analysers, including those supplied by Roche, Siemens, Beckman Coulter and Abbott. The test has a measuring range of 10 ng/ml to the highest calibrator concentration used, and the instructions for use state that laboratories should establish their own population specific cut-off points. Test throughput is dependent on the analyser that is used to process samples.

RIDASCREEN haemoglobin/haptoglobin complex

The RIDASCREEN Hb test (R-Biopharm, Darmstadt, Germany) is a quantitative human Hb/Hp complex immunochemical test. Detection alone is automated. Samples are collected and kept in chilled storage media. Before analysis, the samples are diluted with extraction buffer and mixed. This can be done manually or using the DSX (DSX^®, DYNEX Technologies, Chantilly, VA, USA) automated enzyme-linked immunosorbent assay system. The test is run on a 96-well microtitre plate, which is pre-coated with polyclonal antibodies for human haptoglobin (Hp). The sample solution is applied, followed by a wash step and then application of monoclonal antibody for anti-Hb, which is conjugated to peroxidase. In the presence of a Hb/Hp complex, a sandwich complex forms between the polyclonal and monoclonal antibodies. After further washes, a substrate is added, which reacts with the peroxidase, creating a colour change that can be detected by a plate reader. The values produced by the plate reader are interpreted with the RIDA-SOFT Win.net software (R-Biopharm). The company recommends a cut-off value of > 2 µg/g to determine a positive sample. The test has a limit of detection of 0.42 µg/g. The company suggests that the determination of the Hb/Hp complexes has a diagnostic advantage: as the Hb/Hp complex is resistant to decomposition by acids or proteolytic enzymes, it will maintain in the faeces after long periods in the intestine. Thus, blood admixtures from larger intestinal polyps and colon carcinomas that are located higher up in the intestine can also be recorded with high sensitivity. 23 However, discussion with clinical experts at the scoping stage of this assessment has suggested that this method may also result in an increased number of FPs.

Potential advantages of faecal immunochemical testing

It has been suggested that faecal immunochemical testing may offer improved accuracy compared with guaiac faecal occult blood testing, particularly in relation to the rule-out of CRC. Although most studies do not provide evidence about the performance of the test in symptomatic populations, the idea that faecal immunochemical testing may be associated with improved diagnostic performance relative to guaiac faecal occult blood testing is supported by data from systematic reviews of studies that have been conducted in screening populations. 24,25 A meta-analysis25 of 18 studies demonstrated that faecal immunochemical testing (OC-Sensor) had a higher sensitivity (87% vs. 47%) with similar specificity (93% vs. 92%) to guaiac faecal occult blood testing (Hemoccult^® test, Beckman Coulter Inc., Brea, CA, USA) for screening for CRC. More recent studies comparing faecal immunochemical testing to guaiac faecal occult blood testing in screening populations have also reported increased sensitivity of faecal immunochemical testing for the detection of CRC of between 31.7% and 61.5%, relative to guaiac faecal occult blood testing, with no change in associated specificity. 26–28 A recent study in symptomatic and asymptomatic patients scheduled for diagnostic colonoscopy reported a smaller difference in sensitivity (14.7%). 29 The results of these studies indicate that faecal immunochemical testing may be associated with a decrease in the number of FN results and potentially missed CRC, relative to guaiac faecal occult blood testing, but not a reduction in FP results (inappropriate referrals).

This assessment systematically reviews the evidence about the performance of faecal immunochemical testing as a triage test for people presenting in primary care settings, with lower abdominal symptoms, who are at low risk for CRC according to the criteria defined in NG12,1 and for whom a referral for secondary care investigation for possible CRC is being considered. We have preferentially sought direct comparisons of faecal immunochemical testing and guaiac faecal occult blood testing, to inform comparative cost-effectiveness modelling; however, our assessment also included studies of the diagnostic accuracy of quantitative FIT assays alone (no comparison with gFOBT or other FIT). Where available, data were collected on the use of different faecal Hb concentration cut-off points and/or multiple sampling strategies in order to determine the best way to operationalise FIT use.

A meta-analysis of studies comparing faecal immunochemical testing and guaiac faecal occult blood testing reported that faecal immunochemical testing was associated with a small increase in participation in asymptomatic population-based screening (relative risk 1.16; 95% CI 1.03 to 1.30). 30 Initial reports from the NHS Bowel Cancer Screening Programme in England pilot of faecal immunochemical testing also indicate that faecal immunochemical testing may be associated with increased uptake compared with guaiac faecal occult blood testing (63.9% vs. 54.4% in 60-year-olds who were invited for screening for the first time). 31 We are not aware of any similar studies on testing uptake (or compliance) in symptomatic populations, and the extent to which the acceptability of FIT sample collection would be an issue for people with symptoms is unclear. In order to inform this question, we have collected all of the available data on the return rates for FIT sample collection devices that were issued to participants in the studies that were included in our systematic review.

Testing for occult blood in faeces in patients presenting to primary care settings

The NHS Bowel Cancer Screening Programme in England offers screening every 2 years to all men and women aged between 60 and 74 years. People aged > 74 years can request a screening kit by contacting the Programme. Screening is currently based on guaiac faecal occult blood testing, but FITs have been recommended by the UK Screening Committee for this purpose and a pilot evaluation has already been completed. FITs are currently recommended as the best non-invasive screening modality in all national and international recommendations. 32

According to NG12,1 patients should be referred for an appointment within 2 weeks if they have suspected CRC, defined as:

• aged ≥ 40 years with unexplained weight loss and abdominal pain, or

• aged ≥ 50 years with unexplained rectal bleeding, or

• aged ≥ 60 years with iron-deficiency anaemia or changes in their bowel habit

• tests showing occult blood in their faeces

• having a rectal or abdominal mass

• adults aged < 50 years with rectal bleeding and abdominal pain or change in bowel habit or weight loss or iron-deficiency anaemia.

According to NG12,1 testing for occult blood in faeces should be offered to adult patients who present with initial symptoms without rectal bleeding who are:

• aged ≥ 50 years with unexplained abdominal pain or weight loss

• aged < 60 years with changes in their bowel habit or iron-deficiency anaemia

• aged ≥ 60 years and having anaemia, even in the absence of iron deficiency.

Further testing following a positive test result for occult blood in faeces

Following a positive test result for occult blood in faeces, people in England are usually offered a colonoscopy within 2 weeks of referral to establish a diagnosis.

The 2011 NICE clinical guideline (CG131)33 states that patients should be advised that one or more investigations may be necessary to confirm or exclude a diagnosis of CRC. Colonoscopy is offered to patients without significant comorbidity to confirm a diagnosis of CRC; if a suspicious lesion is detected then a biopsy should be performed (unless contraindicated). For people with comorbidities, computed tomography colonography (CTC) can be offered as an alternative to colonoscopy.

The Scottish Intercollegiate Guidelines Network (SIGN) 201134 guidance for CRC (updated in 2016) states that patients aged > 40 years who present with new-onset, persistent or recurrent rectal bleeding should be referred for investigation. Review of the patient by a regional clinical genetics service is recommended for accurate risk assessment if family history of CRC is the principal indication for referral for investigation. GPs should perform an abdominal and rectal examination on all patients with symptoms that are indicative of CRC. A positive finding should expedite referral, but a negative rectal examination should not rule out the need to refer. All symptomatic patients should have a full blood count; in cases of anaemia, the presence of iron deficiency should be determined. When CRC is suspected clinically, the whole of the large bowel should be examined:

• Colonoscopy is recommended as a very sensitive method of diagnosing CRC, enabling biopsy and polypectomy.

• CTC can be used as a sensitive and safe alternative to colonoscopy.

Guidelines from clinical professional bodies follow a similar pattern: the Royal College of Radiologists recommends that symptomatic patients with suspected CRC should receive evaluation/diagnosis by imaging studies (colonoscopy, CTC or barium enema);35 the Association of Coloproctology of Great Britain and Ireland recommends that patients with higher-risk symptoms should be fast-tracked in special clinics or given urgent appointments in routine clinics. Patients so referred should be investigated with sigmoidoscopy (flexible or rigid) plus a high-quality, double-contrast barium enema, or colonoscopy or CTC. 36

Treatment of colorectal cancer

Following diagnosis and staging, CRC may be treated with surgery, chemotherapy and radiotherapy, or, in some cases, with biological agents such as cetuximab. Treatment is dependent on the stage of the cancer and is described in more detail in CG131. 33

Search strategy

Search strategies were based on intervention (FIT assays) and target condition (CRC), as recommended in the CRD guidance for undertaking reviews in health care37 and the Cochrane Handbook for Diagnostic Test Accuracy Reviews. 39

Candidate search terms were identified from target references, browsing database thesauri [e.g. MEDLINE medical subject heading (MeSH) and EMBASE Emtree] and from existing reviews that were identified during initial scoping searches. Strategy development involved an iterative approach, testing candidate text and indexing terms across a sample of bibliographic databases, aiming to reach a satisfactory balance of sensitivity and specificity. Search strategies were developed specifically for each database and the keywords associated with faecal immunochemical tests for occult blood were adapted according to the configuration of each database.

No restrictions on language, publication status or date of publication were applied. Searches took into account generic and other product names for the intervention. The main EMBASE strategy for each search was independently peer reviewed by a second Information Specialist, using the Canadian Agency for Drugs and Technologies in Health (CADTH) Peer Review checklist. 40 Identified references were downloaded in EndNote X6 software (Thomson Reuters, CA, USA) for further assessment and handling. References in retrieved articles were checked for additional studies. The final list of included papers was also checked on PubMed for retractions, errata and related citations. 41–44

The following databases were searched for relevant studies from database inception date to the most recent date available:

• MEDLINE (via Ovid): 1946 to March Week 3 2016

• MEDLINE In-Process & Other Non-Indexed Citations, and Daily Update (via Ovid): to 29 March 2016

• MEDLINE Epub Ahead of Print (via Ovid): 20 June 2016

• EMBASE (via Ovid): 1974 to 29 March 2016

• The Cochrane Library:

  • Cochrane Database of Systematic Reviews (CDSR) (via the internet): to Issue 3 of 12, March 2016

  • Cochrane Central Register of Controlled Trials (CENTRAL) (via the internet): to Issue 2 of 12, February 2016

  • Database of Abstracts of Reviews of Effects (DARE) (via the internet): to Issue 2 of 4, April 2015

  • Health Technology Assessment (HTA) database (via the internet): to Issue 1 of 4, January 2016

  • NHS Economic Evaluation Database (NHS EED) (via the internet): to Issue 2 of 4, April 2015

• International Network of Agencies for Health Technology Assessment (INAHTA) Publications (via the internet): to 30 March 2016: www.inahta.org/publications/

• National Institute for Health Research (NIHR) HTA programme (via the internet): to 30 March 2016

• Aggressive Research Intelligence Facility (ARIF) database (via the internet): to 30 March 2016: www.birmingham.ac.uk/research/activity/mds/projects/HaPS/PHEB/ARIF/index.aspx

• PROSPERO (International Prospective Register of Systematic Reviews) (via the internet): to 30 March 2016: www.crd.york.ac.uk/prospero/.

Completed and ongoing trials were identified by searches of the following resources:

• National Institutes of Health ClinicalTrials.gov: to 8 March 2016

  www.clinicaltrials.gov/

• EU Clinical Trials Register: to 8 March 2016

  www.clinicaltrialsregister.eu/ctr-search/search

• World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP): to 8 March 2016

  www.who.int/ictrp/en/.

Electronic searches were undertaken for abstracts and poster presentations from the following conferences:

• American Gastroenterological Association – Digestive Disease Week (DDW): 2011–15

• Annual Meeting of the American Association for Clinical Chemistry and Laboratory Medicine (AACC): 2011–15

• British Society of Gastroenterology (BSG) Annual Meeting: 2011–15

• EuroMedLab: IFCC-EFLM (International Federation of Clinical Chemistry and Laboratory Medicine/European Federation of Clinical Chemistry and Laboratory Medicine) European Congress of Clinical Chemistry and Laboratory Medicine: 2011–15

• United European Gastroenterology Week (UEGW): 2011–15.

Full search strategies are presented in Appendix 1.

Specialist members of the Assessment SubGroup and one additional clinical expert (CF) were contacted to seek additional studies.

Inclusion and exclusion criteria

Inclusion criteria for each of the clinical effectiveness questions are summarised in Table 2. Studies that fulfilled these criteria were eligible for inclusion in the review.

Inclusion screening and data extraction

Two reviewers (MW and SL) independently screened the titles and abstracts of all of the reports identified by searches and any discrepancies were discussed and resolved by consensus. Full copies of all of the studies that were deemed to be potentially relevant were obtained and the same two reviewers independently assessed these for inclusion; any disagreements were resolved by consensus. Details of the studies excluded at the full-paper screening stage are presented in Appendix 4.

When studies reported insufficient information (e.g. FIT assay not specified or incomplete accuracy data), the authors were contacted to request additional information. The authors of studies that included a mixed population (e.g. screening, surveillance and symptomatic patients) were contacted to request subgroup data for symptomatic patients. Authors were initially contacted by e-mail, which was followed up with a reminder e-mail after 2 weeks and, subsequently (where possible), by personal contact from a clinical specialist member of the Assessment SubGroup (RL).

Studies cited in the materials provided by the manufacturers of FIT assays were first checked against the project reference database, in Endnote; any studies that were not already identified by our searches were screened for inclusion following the process described above.

Data were extracted on the following: study details, inclusion and exclusion criteria, participant characteristics (demographic characteristics, presenting symptoms, other CRC risk factors), target condition {CRC, advanced neoplasia [high-risk adenoma (HRA) or CRC], other significant bowel disease outcomes (as reported)}, details of the FIT (manufacturer, analyser used, definition of a cut-off point, sampling procedure, detection method, etc.), details of any comparator test(s) (manufacturer, antibody, limit of quantitation, definition of cut-off point, sampling procedure, detection method, etc.), details of the reference standard, definitions of the target conditions, test performance outcome measures [numbers of true-positive (TP), FP, FN and true-negative (TN) test results] and proportion of study participants who returned a FIT sample (extracted as an indicator of acceptability). Data were extracted by one reviewer, using a piloted, standard data extraction form and checked by a second reviewer (MW and SL); any disagreements were resolved by consensus. Full data extraction tables are provided in Appendix 2.

Quality assessment

The methodological quality of included studies was assessed using QUADAS-2. 45

Note: Protocol change Studies that reported results for risk prediction scores that included faecal immunochemical testing, as well as test accuracy data for faecal immunochemical testing, were assessed using PROBAST (Prediction model study Risk Of Bias Assessment Tool)46 in addition to QUADAS-2.

Quality assessment was undertaken by one reviewer and checked by a second reviewer (MW and SL); any disagreements were resolved by consensus or discussion with a third reviewer.

The results of the quality assessments are summarised and presented in tables and graphs in the results of the systematic review, and are presented in full, by study, in Appendix 3.

Methods of analysis/synthesis

Sensitivity and specificity were calculated for each set of 2 × 2 data. The bivariate/hierarchical summary receiver operating characteristic (HSROC) model was used to estimate summary sensitivity and specificity with 95% CIs and prediction regions around the summary points, and to derive HSROC curves for meta-analyses involving four or more studies. 47–49 This approach allows for between-study heterogeneity in sensitivity and specificity, and for the trade-off (negative correlation) between sensitivity and specificity that is commonly seen in diagnostic meta-analyses. For meta-analyses with fewer than four studies we estimated separate pooled estimates of sensitivity and specificity, using random-effects logistic regression. 50 Heterogeneity was assessed visually using summary receiver operating characteristic (SROC) plots and statistically using the variance of logit (sensitivity) and logit (specificity), for which ‘logit’ indicates the logistic function: the smaller these values, the less heterogeneity between studies. Analyses were performed in Stata 10 (StataCorp LP, College Station, TX, USA), using the metandi command. For analyses that would not run in Stata we used Meta-DiSc (version 1.4, free source). 51

Studies were grouped by FIT assay type, target condition and threshold. We compared the accuracy of different FIT assays by tabulating summary estimates from analyses for commonly used thresholds. Stratified results tables receiver operating characteristic (ROC) space plots were used to illustrate the variation of test performance by threshold.

We used SROC plots to display summary estimates from analyses that included a minimum of four data points.

Overview of included studies

Details of the 10 included studies and their associated references are provided in Table 3. Additional data were supplied by the authors of two studies. 56,58 In the case of Terhaar sive Droste et al.,58 the authors provided overall data for symptomatic study from the master database, which holds data for all of their publications (e-mail from Sietze van Turenhout, University Medical Center, Amsterdam, to Marie Westwood, KSR Ltd, 12 June 2016, personal communication). The results section of this report cites studies using the primary publication and, where this is different, the publication (shown in bold text in Table 3) in which the referenced data were reported.

Five studies2,13,53,55,58 reported accuracy data for the OC-Sensor assay; one study52 used the IO analyser (Eiken Chemical Co.), one study13 used the OC-Sensor Diana automated immunoturbidimetric analyser (Eiken Chemical Co.), two studies53,58 used the MICRO desktop analyser (Eiken Chemical Co.) and one study55 did not report the analyser used. Three studies56,57,75 reported accuracy data for the HM-JACKarc automated system (Kyowa Medex, Tokyo, Japan). The remaining two studies reported accuracy data for the FOB Gold assay; one used the Roche Modular P/917 analyser (Roche Diagnostics Ltd, West Sussex, UK),54 and the other used the SENTiFIT 270 analyser (Sentinel Diagnostics, Milan, Italy) (Philippa Pinn, personal communication). There were no studies, using the RIDASCREEN Hb or the RIDASCREEN Hb/Hp complex assays, which met the inclusion criteria for this assessment. None of the included studies reported data comparing different FIT assays, or comparing one or more FIT assays with a gFOBT method. All of the studies included in our systematic review were diagnostic cohort studies that reported data on the diagnostic accuracy of faecal immunochemical testing for which the target condition was CRC or advanced neoplasia (defined as CRC or HRA). Five studies13,52,56,58,75 reported additional accuracy data for various non-malignant and composite target conditions.

Six of the diagnostic accuracy studies13,52,53,55,56,75 included in our systematic review also reported uptake rates for participants who were invited to provide a sample for faecal immunochemical testing.

No randomised controlled trials (RCTs) or controlled clinical trials (CCTs) were identified; no studies provided data on patient-relevant outcomes following faecal immunochemical testing compared with guaiac faecal occult blood testing or no faecal occult blood testing.

All 10 of the included studies2,13,52–58,75 were conducted in Europe: one in England,75 three in Scotland,13,52,56 three published studies53,55,57 and one unpublished study (Philippa Pinn, personal communication) in Spain, and one each in the Netherlands58 and Slovenia. 54 Three studies52,53,55 were publicly funded, five studies13,55–57,75 reported receiving some funding from manufacturers (including supply of test kits, reagents and analysers), one study54 did not report details of funding and the unpublished study was conducted at the request of the test manufacturer.

Full details of the characteristics of study participants, study inclusion and exclusion criteria, and FIT assay used and reference standard are reported in the data extraction tables presented in Appendix 2 (see Tables 65 and 66).

Study quality

All studies included in this systematic review were diagnostic cohort studies. The methodological quality of these studies was assessed using the QUADAS-2 tool45 (summarised in Table 4 and Figure 2). One of these studies53 and an additional report73 and unpublished paper (Callum Fraser, personal communication) linked to a second study55 reported the development and validation of risk prediction scores that included faecal immunochemical testing, in addition to test accuracy results. These studies were assessed using PROBAST (Table 5), as well as QUADAS-2. The full QUADAS-2 assessments for each study are provided in Appendix 3a and PROBAST assessment results are provided in Appendix 3b.

FIGURE 2.

Summary of QUADAS-2 results for studies of FIT assays.

Two studies54,75 were reported only as conference abstracts, with limited descriptions of methods.

Two studies55,57 were rated as ‘low’ risk of bias for all domains. (Confidential information has been removed.) The main potential sources of bias, across the included studies, concerned flow and timing, and application of the index test. Three studies13,52,75 were rated as ‘high’ risk of bias on the flow and timing domain because some patients who returned a sample for faecal immunochemical testing or who agreed to participate in the study were subsequently excluded from the analysis: Mowat et al. 201552 excluded 11% of participants who returned a FIT sample because they were not subsequently referred to secondary care or because the referral was cancelled; Thomas et al. 201675 excluded 12.5% of participants who returned a FIT sample (no reasons for exclusion were reported); McDonald et al. 201213 excluded 41% of people who originally agreed to participate in the study (38% did not return a FIT sample before endoscopy and 3% completed faecal immunochemical testing but not endoscopy). (Confidential information has been removed.)

All of the included studies were rated as having ‘high’ concerns about applicability with respect to participants. This was because no study reported data that were specific to the population that was defined in the scope for this assessment (people presenting, in primary care settings, with lower abdominal symptoms and who are at low risk for CRC according to the criteria defined in NG121). Only one study52 was conducted in a primary care setting, reporting that faecal immunochemical testing was ordered by GPs at the point of referral to secondary care. All of the studies included some participants who had symptoms that may be considered to be associated with a higher probability of CRC, and which are components of the criteria for 2-week referral as defined in NG121 (e.g. rectal bleeding). In addition, although all of the included studies were conducted in Europe, only four studies13,52,56,75 were conducted in the UK (one in England75 and three in Scotland13,52,56). Given that population studies have shown variation in faecal Hb concentrations, and hence potential variation in optimal thresholds for faecal immunochemical testing across different geographic location,88,89 this may limit the applicability of our findings to UK settings.

Diagnostic performance of the OC-Sensor faecal immunochemical test assay

Accuracy of OC-Sensor for the detection of colorectal cancer

Five studies13,52,53,55,58 reported data on the accuracy of OC-Sensor FIT using a single faecal sample, and thresholds ranging from any detectable Hb (the limit of detection for the assay is 452,53 to 4058 µg Hb/g faeces). Full test performance results for all thresholds evaluated in individual studies and summary estimates, where these were calculated, are provided in Table 6. The variation in test performance characteristics according to the threshold used is illustrated in a ROC space plot (Figure 3). Although test performance did not vary greatly with threshold, as might be expected, sensitivity estimates generally decreased and specificity estimates increased with increasing threshold. Specificity was low (< 50%) in both of the studies that used any detectable Hb to define a positive test. 52,53

TABLE 6 - Accuracy of OC-Sensor for the detection of CRC using a single faecal sample

The results of meta-analyses (Summary estimate) are reported in bold.

The limit of detection for the assay is 4 µg Hb/g faeces or 20 ng/ml buffer.

Calculated estimate.

Converted from ‘ng Hb/ml buffer’ using a multiplication factor of 0.2. 17,18

Sietze van Turenhout, personal communication; unpublished data provided in confidence.

Study	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
Any detectable Hb
Mowat 201552	0a	28	0	409	313	750	Reported	100 (87.7 to 100)b	43.4 (39.7 to 47.1)b
Rodríguez-Alonso 201553	0a	30	0	552	421	1003	Calculated	100 (88.4 to100)	43.3 (40.1 to 46.4)
Summary estimate								100 (93.8 to 100)	43.3 (40.9 to 45.7)
10 µg Hb/g faeces or equivalent
McDonald 201213	≥ 10c	6	0	17	257	280	Calculated	100 (54.1 to 100)	93.8 (90.3 to 96.3)
Mowat 201552	≥ 10	25	3	151	571	750	Reported	89.3 (71.8 to 97.7)b	79.1 (75.9 to 82)b
Rodríguez-Alonso 201553	≥ 10	29	1	196	777	1003	Calculated	96.7 (82.8 to 99.9)	79.9 (77.2 to 82.3)
dTerhaar sive Droste 201158	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)
Summary estimate								92.1 (86.9 to 95.3)	85.8 (78.3 to 91.0)
15 µg Hb/g faeces or equivalent
Rodríguez-Alonso 201553	≥ 15	29	1	164	809	1003	Calculated	96.7 (82.8 to 99.9)	83.1 (80.6 to 85.4)
dTerhaar sive Droste 201158	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)
Summary estimate								92.3 (86.6 to 96.1)	86.9 (85.6 to 88.1)
20 µg Hb/g faeces or equivalent
Cubiella 2014 (COLONPREDICT)55	≥ 20c	85	12	156	534	787	Calculated	87.6 (79.0 to 93.2)	77.4 (74.0 to 80.4)
Rodríguez-Alonso 201553	≥ 20	28	2	135	838	1003	Calculated	93.3 (77.9 to 99.2)	86.1 (83.8 to 88.2)
dTerhaar sive Droste 201158	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)
Summary estimate								89.5 (84.9 to 93.1)	86.6 (85.4 to 87.7)
Other thresholds
dTerhaar sive Droste 201158	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)
dTerhaar sive Droste 201158	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)	(Confidential information has been removed)

FIGURE 3.

(Confidential information has been removed.)

The optimal test performance (maximising both sensitivity and specificity) appeared to occur with thresholds of 10 or 15 µg Hb/g faeces, with most data being available for the 10 µg Hb/g faeces threshold. The summary estimates of sensitivity and specificity, using the 10 µg Hb/g faeces threshold, were 92.1% (95% CI 86.9% to 95.3%) and 85.8% (95% CI 78.3% to 91.0%), respectively, based on data from four studies,13,52,53,58 (see Table 6 and Figure 4). Mowat et al. 52 and Rodríguez-Alonso et al. 53 reported data for both the 10 µg Hb/g faeces threshold and the minimum threshold of any detectable Hb; the prevalence of CRC in these two studies52,53 was 3.3%. Using test performance data from these two studies,52,53 and a CRC prevalence estimate of 3.3%, to consider the outcome of testing for a hypothetical cohort of 1000 patients indicates that two CRCs would be missed using the 10 µg Hb/g faeces threshold and 198 unnecessary colonoscopies would be carried out (assuming that all patients with a positive FIT result receive colonoscopy and all of the colonoscopies that are conducted in patients without CRC are considered unnecessary); CRC would be correctly ruled out by faecal immunochemical testing, avoiding colonoscopy, in 769 people (Figure 5a). Alternatively, using a very low threshold (any detectable Hb) would result in no CRCs being missed, but would increase the number of ‘unnecessary’ colonoscopies to 548 and reduce the number of people in whom CRC would be correctly ruled out to 419 (Figure 5b). Please see subsequent sections for information on other significant bowel pathologies that may be detected by faecal immunochemical testing and hence may form part of the FP or ‘unnecessary colonoscopy’ population.

FIGURE 4.

(Confidential information has been removed.)

FIGURE 5.

Testing outcomes for a hypothetical cohort of 1000 patients using OC-Sensor at (a) the 10 µg Hb/g faeces threshold; and (b) any detectable Hb, for the target condition CRC.

Limited data from Cubiella et al. 55 indicated that the sensitivity of OC-Sensor (using a threshold of 20 µg Hb/g faeces) is higher for more advanced stages of CRC [American Joint Committee on Cancer (AJCC) stages II–IV] and is also higher for tumours of the distal colon than for tumours of the proximal colon or rectum (Table 7).

TABLE 7 - Sensitivity of OC-Sensor for determining the stage and location of CRC using a single faecal sample

Converted from ‘ng Hb/ml buffer’ using a multiplication factor of 0.2. 17,18

Study	Target condition	Subgroup	Threshold (µg Hb/g faeces)	TP	FN	Total n	Sensitivity % (95% CI)
CRC location
Cubiella 2014 (COLONPREDICT)55	Rectum	Participants with CRC	≥ 20a	26	4	30	86.7 (69.3 to 96.2)
	Distal colon	Participants with CRC	≥ 20a	40	4	44	90.9 (78.3 to 97.5)
	Proximal colon	Participants with CRC	≥ 20a	19	4	23	82.6 (61.2 to 95.0)
CRC stage (AJCC classification)
Cubiella 2014 (COLONPREDICT)55	0	Participants with CRC	≥ 20a	4	1	5	80 (28.4 to 99.5)
	I	Participants with CRC	≥ 20a	12	3	15	80 (51.9 to 95.7)
	II	Participants with CRC	≥ 20a	21	2	23	91.3 (72.0 to 98.9)
	III	Participants with CRC	≥ 20a	35	5	40	87.5 (73.2 to 95.8)
	IV	Participants with CRC	≥ 20a	12	1	13	92.3 (64.0 to 99.8)

Accuracy of OC-Sensor for the detection of advanced neoplasia (colorectal cancer or high-risk adenoma)

Four of the studies described in the previous section also reported data on the accuracy of OC-Sensor FIT using a single faecal sample, where the target condition was expanded to include CRC or HRA. 13,52,53,55 The thresholds assessed ranged from any detectable Hb (the limit of detection for the assay is 4 µg Hb/g faeces)52,53 to 20 µg Hb/g faeces. 53 Full test performance results for all thresholds evaluated in individual studies and summary estimates, where they were calculated, are provided in Table 8. The variation in test performance characteristics according to the threshold used is illustrated in a ROC space plot (Figure 6).

TABLE 8 - Accuracy of OC-Sensor for the detection of advanced neoplasia (CRC or HRA) using a single faecal sample

The limit of detection for the assay is 4 µg Hb/g faeces or 20 ng/ml buffer.

Calculated estimate.

Converted from ‘ng Hb/ml buffer’ using a multiplication factor of 0.2. 17,18

Study	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
Any detectable Hb
Mowat 201552	0a	61	7	376	306	750	Calculated	89.7 (79.9 to 95.8)b	44.9 (41.1 to 48.7)b
Rodríguez-Alonso 201553	0a	108	25	474	396	1003		81.2 (73.5 to 87.5)	45.5 (42.2 to 48.9)
Summary estimate								84.1 (78.3 to 88.8)	45.2 (42.7 to 47.7)
10 µg Hb/g faeces
McDonald 201213	≥ 10c	17	12	6	245	280	Calculated	58.6 (38.9 to 76.5)	97.6 (94.9 to 99.1)
Mowat 201552	≥ 10	45	23	131	551	750		66.2 (53.7 to 77.2)b	80.8 (77.6 to 83.7)b
Rodríguez-Alonso 201553	≥ 10	82	51	144	726	1003		61.7 (52.8 to 69.9)	83.4 (80.8 to 85.9)
Summary estimate								62.6 (56.0 to 68.9)	84.4 (82.7 to 86.1)
20 µg Hb/g faeces or equivalent
Cubiella 2014 (COLONPREDICT)55	≥ 20c	127	50	114	496	787	Calculated	71.8 (64.4 to 78.1)	81.3 (77.9 to 84.3)
Rodríguez-Alonso 201553	≥ 20	71	62	92	778	1003		53.4 (44.5 to 62.1)	89.4 (87.2 to 91.4)
Summary estimate								63.9 (58.2 to 69.2)	86.1 (84.2 to 87.8)
Other thresholds
Rodríguez-Alonso 201553	≥ 15	76	57	117	753	1003	Calculated	57.1 (48.3 to 65.7)	86.6 (84.1 to 88.7)

FIGURE 6.

Receiver operating characteristic space plot for the OC-Sensor assay using different thresholds for the target condition of advanced neoplasia (CRC or HRA).

For this expanded target condition, sensitivity estimates were low for all thresholds except any detectable Hb; the summary sensitivity and specificity estimates for this threshold were 84.1% (95% CI 78.3% to 88.8%) and 45.2% (95% CI 42.7% to 47.7%), respectively, based on data from the same two studies52,53 that evaluated this threshold for the detection of CRC.

Expanding the target condition from CRC only, to include CRC or HRA, resulted in an increase in prevalence from 3.3% to 11.5%. 52,53 Using test performance data from these two studies,52,53 and an estimate for the prevalence of advanced neoplasia of 11.5%, to consider the outcome of testing for a hypothetical cohort of 1000 patients indicates that 18 advanced neoplasias would be missed, even when using the minimum threshold of any detectable Hb. As data indicate that no CRCs would be missed at this threshold (see previous section), it may be assumed that the missed cases would all be HRAs. Using the any detectable Hb threshold, 485 unnecessary colonoscopies would be carried out (assuming that all patients with a positive FIT result receive colonoscopy and all of the colonoscopies that are conducted in patients without at least HRA are considered unnecessary); CRC and HRA would be correctly ruled out in 401 people (Figure 7a). If the 10 µg Hb/g faeces threshold were applied to the expanded target condition, for hypothetical cohort of 1000 patients, the number of missed cases would increase from 2 to 42 (two CRCs and 40 HRAs); using this threshold, 157 unnecessary colonoscopies would be carried out and CRC and HRA would be correctly ruled out in 729 people (Figure 7b).

FIGURE 7.

Testing outcomes for a hypothetical cohort of 1000 patients using OC-Sensor at (a) any detectable Hb, for the target condition of advanced neoplasia (CRC or HRA); and (b) the 10 µg Hb/g faeces threshold.

Diagnostic performance of the HM-JACKarc faecal immunochemical test assay

Accuracy of HM-JACKarc for the detection of colorectal cancer

Two studies56,75 reported data on the accuracy of HM-JACKarc FIT using a single faecal sample and thresholds of 10 µg Hb/g faeces56 and 7 µg Hb/g faeces. 75 Full test performance results are provided in Table 11. There was little variation in test performance between the 7- and 10 µg Hb/g faeces thresholds; the sensitivity estimates were 91.3% and 100%, respectively, and the corresponding specificity estimates were 76.6% and 79.2%. 56,75 As with the OC-Sensor FIT assay, the optimal test performance (maximising both sensitivity and specificity) appeared to occur with thresholds of 7 or 10 µg Hb/g faeces; none of the HM-JACKarc studies reported test performance characteristics for any detectable Hb. Using test performance data from the Godber et al. study56 – and a CRC prevalence estimate of 2.2% taken from the same study – to consider the outcome of testing for a hypothetical cohort of 1000 patients indicates that no CRCs would be missed using the 10 µg Hb/g faeces threshold, but 229 unnecessary colonoscopies would be carried out (assuming that all of the patients with a positive FIT result receive colonoscopy and that all colonoscopies conducted in patients without CRC are considered unnecessary). CRC would be correctly ruled out in 750 people (Figure 8). Please see subsequent sections for information on other significant bowel pathologies that may be detected by faecal immunochemical testing and hence may form part of the FP or ‘unnecessary colonoscopy’ population.

TABLE 11 - Accuracy of HM-JACKarc for the detection of CRC using a single faecal sample

Personal communication (e-mail from Ian Godber; Biochemistry Department, Monklands Hospital, NHS Lanarkshire, to Marie Westwood, KSR Ltd, 8 June 2016).

Calculated estimate.

Study	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
aGodber 201656	≥ 10	11	0	116	380	507	Reported	100 (71.5 to 100)b	76.6 (72.6 to 80.3)b
aThomas 201675	≥ 7	21	2	89	338	450	Calculated	91.3 (72.0 to 98.9)	79.2 (75.3 to 83)

FIGURE 8.

Testing outcomes for a hypothetical cohort of 1000 patients using HM-JACKarc: the 10 µg Hb/g faeces threshold, for the target condition of CRC.

Accuracy of HM-JACKarc for the detection of advanced neoplasia (colorectal cancer or high-risk adenoma)

Two studies56,57 reported data on the accuracy of HM-JACKarc FIT using a single faecal sample, for which the target condition was expanded to include CRC or HRA. Two studies56,57 assessed the diagnostic performance of the 10 µg Hb/g faeces threshold for different definitions of this expanded target condition (see Details of HM-JACKarc studies, above); these two studies56,57 reported very different estimates of sensitivity (70% and 34.5%, respectively). In addition to the difference in the definition of HRA, the Auge et al. study57 differed from Godber et al. study56 in that it included some patients who were undergoing colonoscopy for polyp surveillance and excluded people with GI bleeding or active rectal bleeding; the prevalence of CRC in the Auge et al. study57 was the lowest of any study included in this review (0.96%). 57 Auge et al. 57 reported test performance characteristics using a range of thresholds; sensitivity decreased and specificity increased with increasing threshold, and the ‘any detectable Hb’ threshold was associated with high sensitivity (96.6%) and very low specificity (10.6%). Full test performance results for all of the thresholds that were evaluated in individual studies are provided in Table 12. The variation in test performance characteristics according to the threshold used is illustrated in a ROC space plot (Figure 9).

TABLE 12 - Accuracy of HM-JACKarc for the detection of advanced neoplasia (CRC or high-/higher-risk adenoma) using a single faecal sample

The limit of detection for the assay is 0.6 µg Hb/g faeces or 0.6 ng/ml buffer.

Ian Godber, personal communication.

Calculated estimate.

Study	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
Any detectable Hb
Auge 201657	0a	28	1	160	19	208	Calculated	96.6 (82.8 to 93.4)	10.6 (6.9 to 15.9)
10 µg Hb/g faeces
Auge 201657	≥ 10	10	19	23	156	208	Calculated	34.5 (19.9 to 52.7)	87.2 (81.6 to 91.3)
bGodber 201656	≥ 10	21	9	106	371	507	Reported	70.0 (50.6 to 85.3)c	77.8 (73.8 to 81.4)c
Other thresholds
Auge 201657	≥ 20	9	20	13	166	208	Calculated	31 (17.3 to 49.2)	92.8 (88 to 95.7)
	≥ 30	9	20	12	167	208		31 (17.3 to 49.2)	93.3 (88.7 to 96.1)
	≥ 40	8	21	11	168	208		27.6 (14.7 to 45.7)	93.9 (89.4 to 96.6)

FIGURE 9.

Receiver operating characteristic space plot for the HM-JACKarc assay using different thresholds for the target condition of advanced neoplasia (CRC or HRA).

Based on the data from Godber et al. ,56 expanding the target condition from CRC only to include CRC or HRA resulted in an increase in prevalence from 2.2% to 5.9%. Using test performance data from this study – and an estimate for the prevalence of advanced neoplasia of 5.9% – to consider the outcome of testing for a hypothetical cohort of 1000 patients indicates that applying faecal immunochemical testing at the 10 µg Hb/g faeces threshold would result in 21 advanced neoplasias being missed. As data indicate that no CRCs would be missed at this threshold (see previous section), it may be assumed that the missed cases would all be higher-risk adenomas. Using the 10 µg Hb/g faeces threshold, 205 unnecessary colonoscopies would be carried out (assuming that all patients with a positive FIT result receive colonoscopy and that all of the colonoscopies that were conducted in patients without at least HRA are considered unnecessary); CRC and higher-risk adenoma would be correctly ruled out in 727 people (Figure 10).

FIGURE 10.

Testing outcomes for a hypothetical cohort of 1000 patients using HM-JACKarc: the 10 µg Hb/g faeces threshold for the target condition of advanced neoplasia (CRC or HRA).

Data from Auge et al. study57 indicated that, in this population (CRC prevalence < 1%), high sensitivity (good rule-out performance) could be achieved only by using the ‘any detectable Hb’ threshold; the sensitivity at this threshold was 96.6%. This study57 also compared the performance of double sampling with single sampling, and found that 100% sensitivity could be achieved by using the higher value from two consecutive samples and the ‘any detectable Hb’ threshold. The use of two consecutive samples increased sensitivity compared with single sampling, at all thresholds, but sensitivity remained low (< 50%) throughout. 57 Full results for single and double sampling at all of the thresholds that were evaluated are provided in Table 13.

TABLE 13 - Effects of multiple sampling on the accuracy of HM-JACKarc for the detection of advanced neoplasia (CRC or HRA)

The limit of detection for the assay is 0.6 µg Hb/g faeces or 0.6 ng/ml buffer.

Study	Sampling strategy	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
Auge 201657	First of two consecutive samples	0a	28	1	160	19	208	Calculated	96.6 (82.8 to 93.4)	10.6 (6.9 to 15.9)
	Higher of two consecutive samples	0a	29	0	173	6	208		100 (88.3 to 100)	3.3 (1.5 to 7.1)
	First of two consecutive samples	≥ 10	10	19	23	156	208		34.5 (19.9 to 52.7)	87.2 (81.6 to 91.3)
	Higher of two consecutive samples	≥ 10	12	17	37	142	208		41.4 (25.5 to 59.3)	79.4 (73 to 84.7)
	First of two consecutive samples	≥ 20	9	20	13	166	208		31 (17.3 to 49.2)	92.8 (88 to 95.7)
	Higher of two consecutive samples	≥ 20	10	19	26	153	208		34.5 (19.9 to 52.7)	85.6 (83.5 to 92.9)
	First of two consecutive samples	≥ 30	9	20	12	167	208		31 (17.3 to 49.2)	93.3 (88.7 to 96.1)
	Higher of two consecutive samples	≥ 30	10	19	25	154	208		34.5 (19.9 to 52.7)	86.1 (83.6 to 92.9)
	First of two consecutive samples	≥ 40	8	21	11	168	208		27.6 (14.7 to 45.7)	93.9 (89.4 to 96.6)
	Higher of two consecutive samples	≥ 40	10	19	21	158	208		34.5 (19.9 to 52.7)	88.3 (82.8 to 92.2)

Auge et al. 57 also reported lower sensitivity estimates, at all thresholds, when HM-JACKarc FIT was used, in women than in men. 57 Full results for test performance in men and women are provided in Table 14.

TABLE 14 - Effects of participant sex on the accuracy of HM-JACKarc for the detection of advanced neoplasia (CRC or HRA) using a single faecal sample

The limit of detection for the assay is 0.6 µg Hb/g faeces or 0.6 ng/ml buffer.

Study	Subgroup	Threshold (µg Hb/g faeces)	TP	FN	FP	TN	Total n	2 × 2 data	Sensitivity % (95% CI)	Specificity % (95% CI)
Auge 201657	Men	0a	17	0	69	6	92	Calculated	100 (81.6 to 100)	8 (3.7 to 16.4)
	Women	0a	11	1	91	13	116		91.7 (64.6 to 98.5)	12.4 (7.4 to 20)
	Men	≥ 10	8	9	10	65	92		47.1 (26.2 to 69)	86.7 (77.2 to 92.6)
	Women	≥ 10	2	10	13	91	116		16.7 (4.7 to 44.8)	87.6 (79.8 to 92.6)
	Men	≥ 20	7	10	6	69	92		41.2 (21.6 to 64)	92 (83.6 to 96.3)
	Women	≥ 20	2	10	7	97	116		16.7 (4.7 to 44.8)	93.3 (86.8 to 96.7)
	Men	≥ 30	7	10	6	69	92		41.2 (21.6 to 64)	92 (83.6 to 96.3)
	Women	≥ 30	2	10	6	98	116		16.7 (4.7 to 44.8)	94.3 (88 to 97.3)
	Men	≥ 40	7	10	6	69	92		41.2 (21.6 to 64)	92 (83.6 to 96.3)
	Women	≥ 40	1	11	5	99	116		8.3 (1.5 to 35.4)	95.2 (89.3 to 97.9)

Diagnostic performance of the FOB Gold faecal immunochemical test assay

Diagnostic performance of the RIDASCREEN faecal immunochemical test assay

No studies were identified that assessed the diagnostic performance of RIDASCREEN Hb or RIDASCREEN Hb/Hp complex in symptomatic patients.

Comparative diagnostic accuracy of different faecal immunochemical test assays and guaiac faecal occult blood testing

No studies were identified that directly compared the performance of different FIT assays, or that compared one or more FIT assays with a gFOBT method.

Selection of test strategies for inclusion in cost-effectiveness modelling

The selection of FIT accuracy data for use in cost-effectiveness modelling was based on the optimal threshold (maximum sensitivity and specificity) for each assay method and the threshold required to proved optimal rule-out performance (highest sensitivity and lowest number of cases missed). Because no studies were identified which directly compared the performance of one or more FIT assays with a gFOBT method, accuracy data for guaiac faecal occult blood testing used in the base case for cost-effectiveness modelling were those used in the NG126 model. The primary study from which these data were taken, Gilbert et al. ,91 was a retrospective study using a Swedish cancer registry, in which a diagnosis of CRC was classified as any diagnosed CRC within 2 years of guaiac faecal occult blood testing; it was unclear whether or not all of the patients who were included in the study had been symptomatic at the time of guaiac faecal occult blood testing, and the prevalence of CRC, in the subgroup of study participants used to provide data for the model was very low (0.36%). We therefore conducted scenario analyses using published estimates of gFOBT accuracy obtained from studies that were identified during the process of inclusion screening for our systematic review: Niv and Sperber92 and Bjerregaard et al. ;93 although guaiac faecal occult blood testing was not included in our systematic review, as an intervention, our searches included general terms for faecal occult blood testing and guaiac faecal occult blood testing. Both of these studies92,93 were conducted in symptomatic patients and the prevalence of CRC was 2.5% and 3.1%, respectively, making these studies closer to the population defined for this assessment.

Methods

Results

The total number of potentially relevant economic papers that were initially retrieved by searches was 923. In addition, 19 clinical guidelines relating to CRC were identified. Most of the economic studies that were identified were excluded because they related to screening for CRC in asymptomatic populations. None of the included studies directly assessed the decision problem of this diagnostic assessment.

The only included study was the one conducted to inform NG12. 1,6 This study1,6 is reviewed in detail below.

Diagnostic model structure

The model begins with a cohort of symptomatic patients, presenting in primary care in whom referral to secondary care for investigation of possible CRC is being considered. A patient in the cohort is offered one of the following choices: faecal immunochemical testing, guaiac faecal occult blood testing or no triage testing at all (referral straight to colonoscopy). A positive FIT or gFOBT results in referral to colonoscopy, whereas a negative test results in a watchful waiting strategy, in which a repeated test or further investigation can be performed when symptoms persist. Patients who had a FN gFOBT or FIT, and whose symptoms persisted, were assumed to receive a colonoscopy and thus be diagnosed within 1 year should they survive. Given the delay in diagnosis, it was also assumed that they would have an increased probability of progressing to a worse cancer state (i.e. these patients enter the CRC Markov model in a worse health state). This logically implies that no patients whose symptoms do not persist will have CRC, that is, all FIT-/gFOBT-negative patients who become asymptomatic do not have the target condition. These patients plus those found not to have CRC by colonoscopy were further modelled with a simple alive or dead Markov model, as described below (see Healthy population Markov model). Colonoscopy is considered to be the gold standard investigation for the diagnosis of CRC owing to its ability to visualise the entire colon and allow biopsies to be performed. 6 CTC is an alternative investigation of choice for people who are unfit for colonoscopy, but does not include biopsy. 6 For patients who test positive for CRC after colonoscopy or CTC, it was assumed that contrast-enhanced CT of the chest, abdomen and pelvis was performed, to determine the stage of disease. A time frame of 1 year was assumed for the diagnostic model. A schematic representation of the decision-analytic model is provided in Figure 12.

FIGURE 12.

Diagnostic decision-analytic model.

Colorectal cancer Markov model structure

The disease natural history used in the model was consistent with the existing UK-based screening economic models and divided the disease states by Dukes’ grading (i.e. stage A – tumour confined to the mucosa; stage B – tumour infiltrating through muscle; stage C – lymph node metastases present; and stage D – distant metastases). 111 The structure of the Markov model was similar to the one used in NG12,6 where cancer stages were defined based on the Dukes’ grading system for CRC and then mapped into the health states of the Markov model. The initial distribution of patients with CRC through the model’s health states was determined by the probability of being in a certain Dukes’ stage. After the initial distribution of patients in the CRC model was determined, it was assumed that patients may stay in their current health state, progress to the health state representing the next worst stage of the condition or die (from CRC or another cause). A schematic representation of the CRC model is provided in Figure 13. A lifetime horizon with a 1-year cycle length captures the probability of progression for treated and untreated patients.

FIGURE 13.

Colorectal cancer Markov model.

Healthy population Markov model

Colonoscopy is the reference standard used to determine the accuracy of quantitative FIT assays and guaiac faecal occult blood testing, for which CRC is the target condition. Thus, it was assumed that patients with a negative colonoscopy result do not have the target condition. It was also assumed that patients with the target condition will remain symptomatic; it therefore follows that all patients with a FN FIT or gFOBT result will remain symptomatic. All patients with a negative test result who become asymptomatic were assumed not to have the target condition.

Patients who do not have the target condition were further modelled with a simple alive or dead Markov model, as depicted in Figure 14. Survival estimates for this model were based on UK life tables. 98 Hence, patients entering this model can either die of any cause (including CRC, as a negative test result or a negative colonoscopy does not imply that these patients will never develop CRC) or stay in the alive health state. LYs and QALYs are accounted, but costs were not calculated in this model. This approach essentially implies two main assumptions for the difference in outcomes between intervention and comparator for those without CRC:

1. for life expectancy, this is caused only by a difference in mortality due to colonoscopy/CTC

2. for cost, this is due only to the difference in cost of guaiac faecal occult blood testing and colonoscopy/CTC.

FIGURE 14.

Healthy population Markov model.

This approach assumes that testing has no long-term (after 1 year) effect on costs or QALYs in disease-negative people and is consistent with our approach taken in previous NICE diagnostic assessments. 112 We have assumed that, in patients without CRC, faecal occult blood testing would not significantly delay diagnosis of the underlying cause of presenting symptoms and hence would not incur any extra cost or effect on mortality.

Model parameters

This section describes the input parameters used in the diagnostic model and in the Markov models and how their values were estimated for the base case.

Overview of main model assumptions

Table 43 summarises the main assumptions made in our economic model.

Scenario analyses

Scenario analyses were performed to explore the impact on costs and QALYs of using different assumptions on diagnostic accuracy of faecal immunochemical testing/guaiac faecal occult blood testing, prevalence, test costs, initial and delayed CRC diagnosis, probability of CRC progression, probability of remaining symptomatic after a negative test result, adverse events (including mortality) attributable to colonoscopy, probability of being referred to colonoscopy (vs. CTC) and probability of receiving a second FIT/gFOBT.

Base-case analysis

The base-case lifetime results, reported as cost per QALY gained (ICER) per patient and strategy, are summarised in Table 45. From these results, it is clear that the difference in QALYs between all of the strategies is minimal and that the no triage strategy (referral straight to colonoscopy) is the most expensive one.

Note that, where the question is whether or not to recommend faecal immunochemical testing instead of guaiac faecal occult blood testing, then the pairwise comparisons of the different FIT assays against gFOBT assays may be more informative than the full incremental analysis shown above. The base-case ICERs for the two interventions against guaiac faecal occult blood testing are presented in Table 46. In both cases, the ICER was < £30,000. Therefore, both interventions are deemed cost-effective compared with guaiac faecal occult blood testing. Table 46 also shows the results obtained when the comparator is no triage (referral straight to colonoscopy). The HM-JACKarc strategy dominates no triage (referral straight to colonoscopy); it results in more QALYs and it is cost saving. It should be noted that, the ICER obtained when the OC-Sensor was compared with no triage (referral straight to colonoscopy) is extremely high (£4,133,559) but that this results from both negative incremental costs and incremental QALYs (i.e. the ICER is in the south-west quadrant of the cost-effectiveness plane); in this case, the cost savings outweigh the loss in QALYs and, therefore, the OC-Sensor strategy is more cost-effective than no triage (referral straight to colonoscopy).

Finally, Table 47 shows the breakdown of the total costs and total QALYs for all of the strategies that were included in the analysis. As some values were very small (e.g. number of adverse events or deaths), the number of patients per outcome of the diagnostic model is multiplied by 1000. Costs and QALYs are shown per patient, for consistency with Table 45 above. Note that number of patients with the disease might be lower than the prevalent CRC population. This is simply because in the model we assumed three possible outcomes of colonoscopy: positive, negative and death. Therefore, those dead, who eventually had CRC, were not counted as having the disease. The number of positive tests was the largest for HM-JACKarc (245.36), followed by OC-Sensor (153.50) and gFOBT (130.28). In the model, all of these patients were assumed to be referred for further investigation with colonoscopy or CTC. Following further investigation, the number of patients with CRC who had a positive FIT was 14.99 for HM-JACKarc, 13.80 for OC-Sensor and 7.49 for gFOBT. Thus, all patients with CRC tested positive with HM-JACKarc, compared with approximately 92% with OC-Sensor and 50% with gFOBT. The immediate consequence of this was that diagnosis of CRC was delayed in no patients tested with HM-JACKarc, in 8% of patients tested with OC-Sensor and in 50% of patients with with gFOBT. The apparent advantage of FIT with respect to gFOBT in detecting patients with CRC earlier comes at the cost of performing more colonoscopies, as the total number of colonoscopies performed was 490.62 for HM-JACKarc, 428.61 for OC-Sensor and 412.95 for gFOBT. This also implies that total costs and risks associated with colonoscopy are higher for FIT than for gFOBT. Because of this, and because performing a FIT was estimated to have a higher cost than gFOBT, the HM-JACKarc diagnostic strategy was more expensive (£255.27) than the OC-Sensor (£225.16) and gFOBT (£214.17) strategies. All testing strategies were clearly cheaper than no triage (£484.19).

Differences in QALYs between the four strategies were minimal. For HM-JACKarc, OC-Sensor and no triage (referral straight to colonoscopy), the total QALYs gained were 0.9048 per patient for the first year (diagnostic phase). For gFOBT this was 0.9044. In fact, there was a difference in QALYS between HM-JACKarc, OC-Sensor and no triage (referral straight to colonoscopy) observed at the fifth decimal place. QALYs were the highest for HM-JACKarc, followed by no triage (referral straight to colonoscopy) and OC-Sensor. This was because for HM-JACKarc no delayed diagnosis was observed. Thus, the difference in QALYs between HM-JACKarc and no triage (referral straight to colonoscopy) was due only to the number of colonoscopies. This was smaller for HM-JACKarc and, therefore, fewer deaths occurred with this strategy. When compared with OC-Sensor, both no triage (referral straight to colonoscopy) and HM-JACKarc strategies, resulted in more colonoscopies being performed, implying more deaths with the last two strategies. However, for OC-Sensor, delayed diagnosis was observed, and this seemed to ‘outweigh’ the gain in QALYs due to fewer deaths.

Note, finally, that the differences in costs and QALYs observed in the two Markov models between the four strategies were also minimal. The healthy population Markov model was used to account for lifetime QALYs of patients who did not have the target condition. The number of QALYs accrued in this model was simply determined by the initial number of patients entering in the model. This was the highest for guaiac faecal occult blood testing, as this was the strategy leading to the fewest colonoscopies and, therefore, the fewest deaths. Hence, the total number of QALYs estimated by the healthy population Markov model was 17.6971 for gFOBT and OC-Sensor (higher than gFOBT in the fifth decimal place), 17.6970 for HM-JACKarc and 17.6968 for no triage (referral straight to colonoscopy). Likewise, the total costs and QALYs estimated by the CRC Markov model were determined only by the initial distribution of patients per Dukes’ stage entering in the model. Note that all of the four strategies correctly diagnosed all CRC alive patients. Thus, although the total number of patients entering the model was approximately the same (prevalence minus those who died) for all of the strategies, differences in staging were observed, due to delayed diagnosis, with the gFOBT and OC-Sensor strategies. Because of this, estimated costs (£19.48) and QALYs (0.0222) were higher for no triage (referral straight to colonoscopy) and HM-JACKarc, as for these two strategies no delayed diagnosis was observed. Compared with gFOBT and OC-Sensor, these patients with CRC were healthier and, therefore, lived longer, but also incurred more costs. For gFOBT, diagnosis of CRC was delayed in 50% of patients, the highest among the four strategies compared. Thus, costs and QALYs estimated by the CRC Markov model for gFOBT were the lowest (£18.10 and 0.0203, respectively). Finally, for OC-Sensor, 8% delayed diagnosis was observed. Therefore, costs (£19.26) and QALYs (0.0219) estimated by the CRC Markov model for this strategy were similar to (but slightly lower than) those estimated for HM-JACKarc and no triage (referral straight to colonoscopy).

Sensitivity analysis

The cost-effectiveness results from the PSA are very similar to the deterministic results except that, in this simulation, the ICERs calculated in the PSA were lower than those shown in Table 45. These results are presented in Table 48.

The probabilistic ICERs for the two interventions compared with every comparator are shown in Table 49. These results are also similar to those presented in Table 46 for the deterministic case, and so are the conclusions drawn.

The scatterplot of the PSA outcomes in the cost-effectiveness plane in Figure 15 shows that the uncertainty around QALYs is similar for all of the strategies, but the uncertainty around costs is clearly larger (and with higher cost values) for the no triage (referral straight to colonoscopy) strategy. The CEACs for each strategy are shown in Figure 16. It can be observed that at low values of the ICER threshold, the probability of being cost-effective is higher for those strategies with lower costs (i.e. gFOBT and OC-Sensor). As the ICER threshold increases, the CEAC for HM-JACKarc increases, whereas those for gFOBT and OC-Sensor decrease. Both CEACs for HM-JACKarc and gFOBT seem to converge to 0.5, but for all values of the ICER threshold considered in this analysis the cost-effectiveness probability for gFOBT was the highest. In particular, at the ICER threshold equal £30,000, the cost-effectiveness probabilities for gFOBT, HM-JACKarc, OC-Sensor and no triage (referral straight to colonoscopy) were 0.47, 0.41, 0.11 and 0.01, respectively.

FIGURE 15.

Cost-effectiveness plane with PSA outcomes for all of the strategies in the base-case scenario.

FIGURE 16.

Cost-effectiveness acceptability curves for all of the strategies in the base-case scenario.

Pairwise comparisons of the two different FIT assays against the two comparators included in this diagnostic assessment (gFOBT and no triage) were also explored in the PSA. The results are presented in Figure 17 (cost-effectiveness planes) and Figure 18 (CEACs). When the comparator was no triage (referral straight to colonoscopy), all PSA outcomes were located in the southern quadrants of the cost-effectiveness plane, indicating cost savings. They seem to be evenly scattered over the two quadrants. Thus, for approximately half of the PSA outcomes the cost savings due to faecal immunochemical testing do not outweigh the loss in QALYs and, therefore, the FIT strategies are no more cost-effective than no triage (referral straight to colonoscopy). However, as most of the PSA outcomes are located very close to the y-axis (where incremental QALYs are zero), no triage (referral straight to colonoscopy) will be cost-effective compared with HM-JACKarc, or OC-Sensor, only for very large values of the ICER threshold. In fact, this was not observed for any of the values of the ICER threshold that were considered in the CEACs shown in Figure 18. When the comparator was gFOBT, the PSA outcomes were mostly located in the northern quadrants of the cost-effectiveness plane (given that faecal immunochemical testing is more expensive). When the intervention was the OC-Sensor, we observed PSA outcomes scattered over the four quadrants of the cost-effectiveness plane, indicating that there is large uncertainty about which strategy is the most cost-effective. Note, finally, that the probabilistic ICERs for both HM-JACKarc compared with gFOBT (£14,626) and OC-Sensor compared with gFOBT (£5039) were estimated to be in the north-east quadrant of the cost-effectiveness plane, with an ICER of < £30,000, meaning that on average the FIT strategies are cost-effective compared with gFOBT. However, this is not readily apparent when we observed the CEACs in Figure 18, as these cross each other at values of the ICER threshold that are much larger than the probabilistic ICERs. This can be explained by the asymmetry of the distribution of the PSA outcomes on the cost-effectiveness plane. Nevertheless, given the small difference in QALYs between all of the strategies observed in the results, the conclusions from the PSA regarding which strategy is the most cost-effective are expected to be sensitive to changes in the accuracy estimates. This is explored in several scenarios within the next section.

FIGURE 17.

Cost-effectiveness plane with PSA outcomes for intervention vs. comparator in the base-case scenario. (a) Cost-effectiveness plane – HM-JACKarc vs. no test; (b) cost-effectiveness plane – HM-JACKarc vs. gFOBT; (c) cost-effectiveness plane – OC-Sensor vs. no test; and (d) cost-effectiveness plane – OC-Sensor vs. gFOBT.

FIGURE 18.

Cost-effectiveness acceptability curves for intervention vs. comparator in the base-case scenario. (a) Acceptability curves (HM-JACKarc vs. no test); (b) acceptability curves (HM-JACKarc vs. gFOBT); (c) acceptability curves (OC-Sensor vs. no test); and (d) acceptability curves (OC-Sensor vs. gFOBT).

Scenario analysis

In the scenarios presented below, only the most significant differences with respect to the base-case scenario are discussed. Complete results are provided in Appendix 11.

Diagnostic accuracy of faecal immunochemical test/guaiac faecal occult blood test

Accuracy scenario I: guaiac faecal occult blood test accuracy estimates based on the Niv and Sperber study

In this scenario analysis, we considered the accuracy estimates for guaiac faecal occult blood testing reported by Niv and Sperber,92 as these are considered more representative of the population of this diagnostic assessment than those reported in NG12. 6 The only difference with respect to the base-case results can be observed in the gFOBT estimates. In this case, the total QALYs and total costs estimated per patient were 18.6227 and £277.54, respectively. Thus, in this scenario, guaiac faecal occult blood testing was dominated by both FIT strategies. Probabilistic results show that the main difference with respect to the base-case scenario was that guaiac faecal occult blood testing was now clearly less cost-effective. This can be seen in the CEACs plot for all of the strategies in Figure 19, where guaiac faecal occult blood testing is the most cost-effective strategy only when the ICER threshold ranges between (approximately) £12,750 and £24,500 (whereas in the base-case gFOBT was the most cost-effective strategy for the whole range of ceiling ratios that were considered in the analysis). The CEACs for each intervention compared with guaiac faecal occult blood testing separately are presented in Figure 20. These show that both interventions are more cost-effective than guaiac faecal occult blood testing for all of the ceiling ratio values that were considered in this analysis.

FIGURE 19.

Cost-effectiveness acceptability curves for all of the strategies: alternative gFOBT accuracy estimates.

FIGURE 20.

Cost-effectiveness acceptability curves for intervention vs. gFOBT, alternative gFOBT accuracy estimates. (a) acceptability curves – HM-JACKarc vs. gFOBT; and (b) acceptability curves – OC-Sensor vs. gFOBT.

Accuracy scenario II: guaiac faecal occult blood test accuracy estimates based on the Bjerregaard et al. study

The accuracy estimates for guaiac faecal occult blood testing reported in Bjerregaard et al. 93 were considered for this scenario, as these are also regarded as more representative of the population of this diagnostic assessment than those in NG12. 6 In this scenario, the total QALYs and total costs estimated per gFOBT patient were 18.6230 and £258.97, respectively. Thus, guaiac faecal occult blood testing was dominated by the OC-Sensor. The ICER reported when the intervention was HM-JACKarc was £13,482. Therefore, at the common value of the ceiling ratio of £30,000, HM-JACKarc was deemed cost-effective compared with guaiac faecal occult blood testing. All of the CEACs were similar to those in the scenario for which the gFOBT accuracy estimates were obtained from Niv and Sperber92 and, therefore, they are not shown here. The main difference with respect to the previous scenario is that guaiac faecal occult blood testing is now the most cost-effective strategy when the ICER threshold ranges between (approximately) £12,250 and £23,500. The CEACs for each intervention compared with guaiac faecal occult blood testing are also similar, but the cost-effectiveness probability for guaiac faecal occult blood testing is now higher.

Accuracy scenario III: faecal immunochemical test threshold any detectable haemoglobin (OC-Sensor only)

In this scenario, a threshold of any detectable Hb was assumed for faecal immunochemical testing. The idea of this scenario is to look at the effects of using a threshold at the lowest level that faecal immunochemical testing is able to detect. For this threshold, our systematic review obtained data only for the OC-Sensor. The total QALYs and total costs estimated per patient for the OC-Sensor strategy were 18.6240 and £375.40, respectively. For gFOBT and no triage (referral straight to colonoscopy) these were as in the base-case scenario. In this case, no triage (referral straight to colonoscopy) was dominated by the OC-Sensor, and the ICER obtained when the OC-Sensor was compared with gFOBT was £65,192. Thus, at the common value of the ceiling ratio of £30,000, the OC-Sensor was not cost-effective compared with gFOBT. This is illustrated in the CEAC presented in Figure 21.

FIGURE 21.

Cost-effectiveness acceptability curves for OC-Sensor vs. gFOBT: scenario – any detectable Hb.

Accuracy scenario IV: faecal immunochemical test threshold of ≥ 20 µg Hb/g faeces (OC-Sensor and FOB Gold)

In this scenario, we assumed a threshold of ≥ 20 µg Hb/g faeces. For this threshold, our systematic review obtained data for the OC-Sensor assay and also for the FOB Gold assay. Although the reliability of the accuracy data for the latter was questionable, as explained in detail in Chapter 3 (see Diagnostic performance of the FOB Gold faecal immunochemical test assay), we decided to include this FIT assay in this scenario. Deterministic results are shown in Table 50. The no triage (referral straight to colonoscopy) strategy was the most expensive, but also the strategy with the most QALYs. However, as in previous scenarios, little difference in QALYs was observed between any of the strategies.

TABLE 50 - Lifetime results for all of the strategies in the scenario of FIT threshold of ≥ 20 µg Hb/g faeces (deterministic)

	QALYs	Cost (£)	Incremental QALYs	Incremental cost (£)	ICER (£)
gFOBT	18.6219	232.28
FOB Gold	18.6237	240.62	0.0018	8343	4725
OC-Sensor	18.6238	241.83	0.0001	1210	12,576
No triage (referral straight to colonoscopy)	18.6239	503.67	0.0002	261,839	1,449,585

The ICERs for the two interventions in comparison with guaiac faecal occult blood testing are shown in Table 51. In both cases, the ICER was very low (around £5000). Therefore, the two interventions are cost-effective compared with gFOBT. Table 51 also shows the results obtained when the comparator is no triage (referral straight to colonoscopy). In both cases, the ICERs obtained were very high, but these result from both negative incremental costs and incremental QALYs. In this case, the cost savings outweigh the loss in QALYs and thus the FIT strategies are more cost-effective than no triage (referral straight to colonoscopy).

TABLE 51 - Lifetime results for intervention vs. comparator: FIT threshold of ≥ 20 µg Hb/g faeces (deterministic)

	Incremental QALYs	Incremental cost (£)	ICER (£)
FOB Gold vs. gFOBT	0.0018	8.34	4725
FOB Gold vs. no triage (referral straight to colonoscopy)	–0.0003	–263.05	950,102
OC-Sensor vs. gFOBT	0.0019	9.55	5131
OC-Sensor vs. no triage (referral straight to colonoscopy)	–0.0002	–261.84	1,449,585

The CEACs of all of the strategies presented in Figure 22 show that guaiac faecal occult blood testing is the strategy with the highest probability of being cost-effective for all of the values of the ceiling ratio considered in this analysis. In particular, at an ICER threshold equal to 30,000, this probability was approximately 0.5. The probability of being cost-effective was similar for both interventions, but slightly higher for the FOB Gold. Compared with no triage (referral straight to colonoscopy), both interventions have a higher probability of being cost-effective for all of the values of the ICER threshold considered in the analysis (CEACs not shown). CEACs for each intervention compared with gFOBT are shown in Figure 23. We can observe that, at lower values of the ICER threshold, the gFOBT has a higher probability of being cost-effective because the interventions are more expensive. However, as the ICER threshold increases, the probability of the two interventions being cost-effective also increases and becomes slightly higher than that for gFOBT, although in both cases the probability is close to 0.5.

FIGURE 22.

Cost-effectiveness acceptability curves for all of the strategies: FIT threshold of ≥ 20 µg Hb/g faeces.

FIGURE 23.

Cost-effectiveness acceptability curves for intervention vs. gFOBT: FIT threshold of ≥ 20 µg Hb/g faeces. (a) Acceptability curves – FOB Gold vs. gFOBT; and (b) acceptability curves – OC-Sensor vs. gFOBT.

Accuracy scenario V: threshold of ≥ 10 µg Hb/g faeces (base case) including FOB Gold

In this scenario, we included the FOB Gold assay as well as the interventions and comparators that were considered in the base case. We assumed a threshold of 6.8 µg Hb/g faeces for the FOB Gold test, as this was the closest available value to the base-case threshold of ≥ 10 µg Hb/g faeces, and because for this threshold the FOB Gold test showed performance characteristics (sensitivity and specificity) similar to those for the other FITs at the base-case threshold. In the full incremental deterministic analysis, the FOB Gold assay was dominated by the OC-Sensor, and, as no triage (referral straight to colonoscopy) was dominated by HM-JACKarc, the results of this scenario reduce to those in the base case presented in Table 45. The ICER of FOB Gold compared with gFOBT was £15,720, and compared with no triage (referral straight to colonoscopy) was £2,273,829 (in the south-west quadrant of the cost-effectiveness plane). Thus, in both cases, FOB Gold is cost-effective given the common threshold ICER of £30,000. The CEACs for each strategy are shown in Figure 24. It can be observed that the FOB Gold strategy has a low probability of being cost-effective. This is not surprising, as the full incremental analysis results showed that FOB Gold was dominated by the OC-Sensor. The rest of the CEACs are similar to those obtained in the base-case scenario (see Figure 16).

FIGURE 24.

Cost-effectiveness acceptability curves for all of the strategies in the scenario including FOB Gold.

The CEACs in Figure 25 show that the probability of FOB Gold being cost-effective compared with no triage (referral straight to colonoscopy) is high even at large values of the ceiling ratio of the ICER threshold. However, gFOBT clearly more cost-effective than FOB Gold at lower values of the ICER threshold. As the ceiling ratio increases, the probability of being cost-effective for both strategies seems to converge to 0.5.

FIGURE 25.

Cost-effectiveness acceptability curves for FOB Gold vs. comparator in the scenario including FOB Gold. (a) Acceptability curves – FOB Gold vs. no test; and (b) acceptability curves – FOB Gold vs. gFOBT.

Colorectal cancer mortality and progression

The impact of not considering CRC progression explicitly in the model was explored in this scenario. The full incremental results in Table 57 were similar to those in the base case, with a slight gain in QALYs for all of the strategies and an increase in costs.

TABLE 57 - Lifetime results for all of the strategies: scenario – without CRC progression (deterministic)

NA, not applicable.

	QALYs	Cost (£)	Incremental QALYs	Incremental cost (£)	ICER (£)
gFOBT	18.701	263.92	NA	NA	NA
OC-Sensor	18.711	279.10	0.0101	15.18	1508
No triage (referral straight to colonoscopy)	18.713	538.93	Dominated by HM-JACKarc
HM-JACKarc	18.713	310.01	0.0019	30.90	16,528

When the interventions were compared with gFOBT and no triage (referral straight to colonoscopy) separately (Table 58), we observed, as in previous scenarios, that HM-JACKarc dominated no triage (referral straight to colonoscopy); the ICER for OC-Sensor compared with no triage (referral straight to colonoscopy) was large but in the south-west quadrant of the cost-effectiveness plane (hence, OC-Sensor was cost-effective); and the ICERs obtained when the comparator was gFOBT were both < £30,000 (and, therefore, both were cost-effective).

TABLE 58 - Lifetime results for intervention vs. comparator: scenario – without CRC progression (deterministic)

	Incremental QALYs	Incremental cost (£)	ICER (£)
HM-JACKarc vs. gFOBT	0.0119	46.09	3859
HM-JACKarc vs. no triage (referral straight to colonoscopy)	0.0003	–228.92	–820,947
OC-Sensor vs. gFOBT	0.0101	15.18	1508
OC-Sensor vs. no triage (referral straight to colonoscopy)	–0.0016	–259.83	163,305

The main difference with respect to the base-case PSA is that results are now more favourable to the interventions, as can be observed in the CEACs for all of the strategies in Figure 26 and the CEACs for the interventions against gFOBT in Figure 27.

FIGURE 26.

Cost-effectiveness acceptability curves for all of the strategies: scenario – without CRC progression.

FIGURE 27.

Cost-effectiveness acceptability curves for intervention against gFOBT in the scenario without CRC progression. (a) Acceptability curves – HM-JACKarc vs. gFOBT; and (b) acceptability curves – OC-Sensor vs. gFOBT.

Clinical effectiveness

All of the studies included in the systematic review were diagnostic cohort studies reporting accuracy data. Eight published studies13,52,53,55–58,75 and one unpublished study (Hospital Clinic de Barcelona 2015; Philippa Pinn, personal communication) assessed accuracy of a quantitative FIT method for the target condition of CRC or for the composite target condition of advanced neoplasia (CRC or HRA); the remaining study,54 which was published as a conference abstract, reported accuracy data for only the composite target condition ‘significant bowel disease’ (defined as CRC polyps or bleeding). Five studies13,52,53,55,58 reported accuracy data for the OC-Sensor assay. Three studies56,57,75 reported accuracy data for the HM-JACKarc automated system. One study,54 published as an abstract, and one unpublished study (Hospital Clinic de Barcelona 2015) reported accuracy data for the FOB Gold assay. No studies using the RIDASCREEN Hb or the RIDASCREEN Hb/Hp complex assays met the inclusion criteria for this assessment. Two studies of OC-Sensor53,55 also developed risk scores for CRC or advanced neoplasia that included faecal immunochemical testing; in the case of the COLONPREDICT study,55 risk score development was reported in a separate publication,73 namelya manuscript, which was provided ahead of publication (Callum Fraser, personal communication). None of the included studies reported data comparing different FIT assays, or comparing one or more FIT assays with a gFOBT method. No RCTs or CCTs were identified, and no studies provided data on patient-relevant outcomes following faecal immunochemical testing compared with gFOBT or no faecal occult blood testing.

No study reported data that were specific to the population defined in the scope for this assessment (people presenting, in primary care settings, with lower abdominal symptoms who are at low risk for CRC according to the criteria defined in NG121). Only one study52 was conducted in a primary care setting, and this study reported that faecal immunochemical testing was ordered by GPs at the point of referral to secondary care. In addition, all of the studies included some participants who had symptoms that may be considered to be associated with a higher probability of CRC and that are components of the criteria for 2-week referral as defined in NG121 (e.g. rectal bleeding).

When faecal immunochemical testing was based on a single faecal sample and a threshold of 10 µg Hb/g faeces was applied, sensitivity estimates indicated that a negative test using either OC-Sensor and HM-JACKarc may be considered adequate to rule out CRC; the summary estimate of sensitivity for C-Sensor was 92.1% (95% CI 86.9% to 95.3%), based on four studies, and the only study of HM-JACKarc to assess the 10 µg Hb/g faeces cut-off point reported a sensitivity of 100% (95% CI 71.5% to 100%). 56 The corresponding specificity estimates were 85.8% (95% CI 78.3% to 91.0%) and 76.6% (95% CI 72.6% to 80.3%), respectively. (Confidential information has been removed.) When a hypothetical cohort of 1000 patients is considered, assuming prevalence estimates derived from the included studies, the estimated number of CRC cases that would be missed using each of the three FIT assays (OC-Sensor and HM-JACKarc 10 µg Hb/g faeces threshold and (confidential information has been removed) is 2, 0 and (confidential information has been removed), respectively; under the same assumptions, the number of ‘unnecessary’ colonoscopies carried out on people without CRC would be 198, 229 and (confidential information has been removed), respectively (see Chapter 3, Results of the assessment of clinical effectiveness assessment, see Figures 5a, 9 and 11). Reducing the threshold to ‘any detectable Hb’ produced summary estimates of sensitivity and specificity, for OC-Sensor, of 100% (95% CI 87.7% to 100%) and 43.4% (95% CI 39.7% to 47.1%), based on two studies. For the hypothetical cohort of 1000 patients, this lowering of the threshold would result in no cases of CRC being missed, but would increase the number of ‘unnecessary’ colonoscopies from 198 to 548 (see Figure 5b). No study of HM-JACKarc assessed the any detectable Hb threshold for CRC. (Confidential information has been removed.)

No study reported data on the effects of patient age, sex or presenting symptoms, or on the effects of multiple sampling, on the accuracy of faecal immunochemical testing for the target condition of CRC. However, the results of multivariable regression analyses indicate that participant age73 and sex,53,73 but not presenting symptoms (NICE and SIGN 2-week referral criteria),53 as well as faecal Hb concentration,53,73 are independent predictors of CRC. Validation of the FAST score (Cubiella et al. 2016;127 Callum Fraser, personal communication), which incorporates age, sex and faecal Hb concentration, has suggested that a score threshold of 2.12 (range of score not reported) can be used to reliably rule out CRC (sensitivity 100%, 95% CI 97.7% to 100%); however, the specificity of the score is very low (19.8%, 95% CI 18.6 to 21.1%) at this threshold.

If a lower diagnostic threshold is applied, that is, the target condition includes HRA as well as CRC (advanced neoplasia), then the rule-out performance of all FIT methods is reduced. For faecal immunochemical testing based on a single faecal sample and a threshold of 10 µg Hb/g faeces, the sensitivity estimates indicated that neither a negative OC-Sensor nor a negative HM-JACKarc FIT would be likely to be considered adequate to rule out CRC; the summary estimate of sensitivity for OC-Sensor was 62.9% (95% CI 55.9% to 69.4%), based on three studies, and the estimate of sensitivity for HM-JACKarc was 70.0% (95% CI 50.6% to 85.3%), based on one study. 56 The corresponding specificity estimates were 84.6% (95% CI 82.8% to 86.2%) and 77.8% (95% CI 73.8% to 81.4%), respectively. One additional study57 reported sensitivity and specificity estimates for HM-JACKarc at the 10 µg Hb/g faeces threshold and any detectable Hb thresholds; however, this study57 differed from others in the systematic review in that it included some patients who were undergoing colonoscopy for polyp surveillance and excluded people with GI bleeding or active rectal bleeding; the prevalence of CRC in this study was the lowest of any included study (0.96%). 57 When a hypothetical cohort of 1000 patients is considered, assuming that prevalence estimates have been derived from the included studies, the estimated number of HRA cases that would be missed using either of the two FIT assays (OC-Sensor and HM-JACKarc at the 10 µg Hb/g faeces threshold) is 40 and 22, respectively; under the same assumptions the number of ‘unnecessary’ colonoscopies carried out on people without CRC would be 157 and 205, respectively (see Chapter 3, Results of the assessment of clinical effectiveness assessment, Figures 7a and 10). (Confidential information has been removed.) Using the OC-Sensor assay and a threshold of any detectable Hb increased the summary estimate of sensitivity to 84.1% (95% CI 78.3% to 88.8%), with a corresponding specificity estimate of 45.2% (95% CI 42.7% to 47.7%). For the hypothetical cohort of 1000 patients, this lower threshold would still result in 18 cases of HRA being missed and 485 ‘unnecessary’ colonoscopies being conducted in people without HRA or CRC (see Figure 7a).

Only the Auge et al. study57 reported any data on the effects of patient characteristics (sex) and number of faecal samples tested on the accuracy of faecal immunochemical testing for the target condition of advanced neoplasia. However, the results of multivariable regression analyses indicate that participant age53,73 and sex,53,73 but not presenting symptoms (NICE and SIGN 2-week referral criteria),53 as well as faecal Hb concentration,53,73 are independent predictors of advanced neoplasia. Validation of two risk scores based on the analyses (both incorporating age, sex Hb concentration) gave increased estimates of sensitivity and specificity. One study reported sensitivity and specificity estimates of 88.1% (95% CI 74.3% to 96.0%) and 63.3% (95% CI 57.4% to 69.0%), for a threshold of ‘5’ on a score with a scale of 0–11. 53 The FAST score (Cubiella et al. 2016;127 Callum Fraser, personal communication) threshold of 2.12 (range of score not reported) gave sensitivity and specificity estimates of 96.7% (95% CI 94.9% to 98.0%) and 21.5% (95% CI 20.1% to 22.9%). These data suggest that risk scores may have the potential to provide a more reliable rule-out method than faecal immunochemical testing alone at lower thresholds of disease.

No studies were identified which assessed the diagnostic performance of RIDASCREEN Hb or RIDASCREEN Hb/Hp complex in symptomatic patients, and no studies were identified which directly compared the performance of different FIT assays, or which compared one or more FIT assays with a gFOBT method.

The selection of FIT accuracy data for use in cost-effectiveness modelling was based on the optimal threshold (maximum sensitivity and specificity) for each assay method and the threshold required to provide optimal rule-out performance (highest sensitivity and lowest number of cases missed). Because no studies were identified which directly compared the performance of one or more FIT assays with a gFOBT method, accuracy data for gFOBT that were used in the base case for cost-effectiveness modelling were those used in the NG12 model. 6 The primary study from which these data were taken, Gillberg et al. ,91 was a retrospective study using a Swedish cancer registry, in which a diagnosis of CRC was classified as any diagnosed CRC within 2 years of guaiac faecal occult blood testing. It was unclear whether or not all of the patients included in the study had been symptomatic at the time of guaiac faecal occult blood testing, and the prevalence of CRC, in the subgroup of study participants used to provide accuracy data for the model, was very low (0.36%); the prevalence of CRC for the whole population of the Gillberg et al. study91 (1.8%) was closer to the prevalence estimate that was used in the NG12 model6 (1.5%). Because of these problems with the Gillberg et al. study91 data, we chose to conduct scenario analyses using published estimates of gFOBT accuracy in symptomatic patients obtained from studies that were identified during the process of inclusion screening for our systematic review: the Niv and Sperber92 and Bjerregaard et al. 93 studies. Both of these studies were conducted in symptomatic patients and the prevalence of CRC was 2.5% and 3.1%, respectively, making these studies closer to the population defined for this assessment.

Cost-effectiveness

We assessed the cost-effectiveness of using quantitative faecal immunochemical tests for Hb (faecal immunochemical testing) as a triage test for people presenting, in primary care settings, with lower abdominal symptoms who are at low risk for CRC according to the criteria defined in NG12. 1 Our review of economic analysis found no papers or reports assessing specifically this decision problem. The most relevant study was the one conducted in NG12,6 which estimated the cost-effectiveness of five different investigations (i.e. faecal occult blood testing, barium enema, flexible sigmoidoscopy, CTC or colonoscopy) for suspected CRC ordered in primary care for patients aged 40 years and over with a change in bowel habit as the main symptom. This study was used to guide a part of the development of our diagnostic assessment model.

Diagnostic accuracy of faecal immunochemical testing was captured in the model using data on sensitivity and specificity for the detection of CRC using a single faecal sample obtained from our systematic review. For the base-case scenario, we considered a threshold of 10 µg Hb/g faeces or equivalent. For this threshold, our systematic review obtained data only for OC-Sensor and HM-JACKarc. The comparators chosen for this diagnostic assessment were gFOBTs and no triage (referral straight to colonoscopy). For gFOBT, we considered the sensitivity and specificity estimates used in NG121 (base-case scenario) and in two papers identified in our systematic review (scenario analyses) that were thought to represent a better match for the population that was specified for this diagnostic assessment. Additionally, the FOB Gold FIT system was included as an intervention in scenario analyses, despite the limitations of the accuracy data found in our systematic review, as described in Chapter 3.

The base-case cost-effectiveness results suggested that the difference in QALYs between all of the strategies included in this assessment is minimal and that the ‘no triage’ strategy (referral to colonoscopy) is the most expensive. Overall, faecal immunochemical testing was cost-effective when compared with no triage. This was either because the latter was dominated (less effective and more costly) or because the ICERs obtained were high, but in the south-west quadrant of the cost-effectiveness plane, that is, faecal immunochemical testing was slightly less effective, but cheaper, than no triage. In this case the cost savings could be said to ‘outweigh’ the slight loss in QALYs. When the comparator was gFOBT, the cost-effectiveness results showed that faecal immunochemical testing was more effective and more costly than gFOBT, but the ICERs obtained were below the common threshold ICER of £30,000, and thus the interventions can also be considered cost-effective. However, it should be noted that the PSA results showed that this finding should be interpreted with caution. Because the PSA outcomes were scattered over the four quadrants of the cost-effectiveness plane in an asymmetrical fashion, gFOBT had in general a high probability of being cost-effective (especially at low values of the ICER threshold) and sometimes gFOBT was the strategy with the highest probability of being cost-effective.

The results of the different scenario analyses did not differ substantively from the base-case results. The scenarios for which the accuracy estimates for gFOBT were based on the studies by Niv and Sperber92 and Bjerregaard et al. 93 (those that are considered more representative of the population of this diagnostic assessment) were more favourable than the base-case scenario with regard to the interventions. In only two scenarios would FIT strategies be considered to be not cost-effective because the ICER exceeded the £30,000 threshold. The highest ICER was obtained when OC-Sensor was compared with gFOBT when a threshold of any detectable Hb was assumed for this FIT assay (£65,192). This was expected, as reducing the threshold for FIT results in the test being less effective in avoiding colonoscopies, that is, this threshold is associated with the highest number of FPs. When HM-JACKarc was compared with gFOBT in the scenario with high mortality due to colonoscopy, the ICER was £45,271.

Clinical effectiveness

Extensive literature searches were conducted in an attempt to maximise retrieval of relevant studies. These included electronic searches of a variety of bibliographic databases, as well as screening of clinical trials registers and conference abstracts to identify unpublished studies. Because of the known difficulties in identifying test accuracy studies using study design-related search terms,128 search strategies were developed to maximise sensitivity at the expense of reduced specificity. Thus, large numbers of citations were identified and screened, relatively few of which met the inclusion criteria of the review.

The possibility of publication bias remains a potential problem for all systematic reviews. Considerations may differ for systematic reviews of test accuracy studies. It is relatively simple to define a positive result for studies of treatment, for example a significant difference between the treatment and control groups, which favours treatment. This is not the case for test accuracy studies, which measure agreement between index test and reference standard. It would seem likely that studies finding greater agreement (high estimates of sensitivity and specificity) will be published more often; however, the relative priorities given to sensitivity and specificity estimates may vary depending on the intended application of the test. In addition, test accuracy data are often collected as part of routine clinical practice, or by retrospective review of records; test accuracy studies are not subject to the formal registration procedures applied to RCTs and are therefore more easily discarded when results appear unfavourable. The extent to which publication bias occurs in studies of test accuracy remains unclear; however, simulation studies have indicated that the effect of publication bias on meta-analytic estimates of test accuracy is minimal. 129 Formal assessment of publication bias in systematic reviews of test accuracy studies remains problematic and reliability is limited. 39 We did not undertake a statistical assessment of publication bias in this review. However, our search strategy included a variety of routes to identify unpublished studies and resulted in the inclusion of a number of conference abstracts.

Despite our extensive searches, no studies were identified that assessed the diagnostic performance of RIDASCREEN Hb or RIDASCREEN Hb/Hp complex in symptomatic patients. We identified one article that assessed the performance of Hb/Hp complex in faeces for detecting CRC; the reported sensitivity and specificity estimates were 77% and 95%, using a threshold of 2 µg/g faeces. 130 The test manufacturer stated, of this study, that: ‘My understanding is that all of the patients were symptomatic’ (e-mail from Andrea Lennerz, R-Biopharm AG, Germany, via Rebecca Albrow, NICE, to Marie Westwood, KSR Ltd, 27 May 2016, personal communication). However, this study did not meet the inclusion criteria for our systematic review, as it used an early developmental immunoluminometric assay for Hb/Hp complex and not an assay that is currently commercially available or in use in a NHS laboratory.

No studies were identified which directly compared the performance of different FIT assays, or which compared one or more FIT assays to a gFOBT method; therefore, all of the data included in this assessment refer to the clinical effectiveness of individual FIT methods and not to their comparative effectiveness.

Although gFOBT was not one of the interventions included in our systematic review, it was considered to be a comparator for cost-effectiveness modelling. In order to inform cost-effectiveness modelling, our searches therefore included general terms for faecal occult blood testing and guaiac faecal occult blood testing. Diagnostic accuracy studies of gFOBT conducted in symptomatic patients, retrieved by these searches, were identified during the systematic review process. However, we do not consider that we have conducted a full systematic review of guaiac faecal occult blood testing in symptomatic patients, as our searches did not include terms for gFOBT product names and our inclusion criteria specified only those studies of gFOBT that included a comparison to one or more FIT methods.

Clear inclusion criteria were specified in the protocol for this review, a copy of which is available online (www.nice.org.uk/guidance/GID-DG10005/documents/final-protocol). The eligibility of studies for inclusion is therefore transparent. In addition, we have provided specific reasons for exclusion for all of the studies which were considered potentially relevant at initial citation screening and were subsequently excluded on assessment of the full publication (see Appendix 4). The review process followed recommended methods to minimise the potential for error and/or bias;37 studies were independently screened for inclusion by two reviewers, and data extraction and quality assessment were undertaken by one reviewer and checked by a second (MW and SL). Any disagreements were resolved by consensus.

All of the studies included in this review were assessed for risk of bias and applicability using the QUADAS-2 tool,45 which is recommended by the Cochrane Collaboration. 39 QUADAS-2 is structured into four key domains: participant selection, index test, reference standard and the flow of patients through the study (including timing of tests). Each domain is rated for risk of bias (low, high or unclear); the participant selection, index test and reference standard domain are also separately rated for concerns regarding the applicability of the study to the review question (low, high or unclear). The results of the QUADAS-2 assessment are reported, in full, for all of the included studies in Appendix 3 and are summarised in Chapter 3 (see Study quality). Those studies that reported development of risk scores, in addition to test accuracy data, were also assessed using the PROBAST tool. 46 PROBAST has been designed to assess both the risk of bias and concerns regarding applicability of a study that evaluates (develops and/or validates) a multivariable diagnostic or prognostic prediction model. It has a domain-based structure, similar to that of QUADAS-2, and is intended to be used for the assessment of primary studies that are included in a systematic review. PROBAST is not yet published, but has been used with the consent of the steering group, of which the lead author of this assessment report is a member.

This assessment includes information on the development and validation of risk scores that incorporate faecal Hb levels measured by faecal immunochemical testing, as well as test accuracy data for faecal immunochemical testing alone. Available data on the performance characteristics of such scores have been included in order to inform the question of whether or not the performance characteristics of faecal immunochemical testing can be improved by using it as part of a simple scoring system that includes other information that is readily available to medical practitioners.

All of the studies included in our systematic review were conducted in Europe; however, only four studies13,52,56,75 were conducted in the UK (one in England75 and three in Scotland13,52,56). Given that population studies have shown variation in faecal Hb concentrations, and hence potential variation in optimal thresholds for faecal immunochemical testing, across different geographic location,88,89 this may limit the applicability of our findings to UK settings. In addition, the definitions of HRA used in UK studies are generally based on number and size of lesions, as per BSG guidelines,90 whereas those conducted in mainland Europe have used definitions that also include morphology. However, it should also be noted that validation data for the FAST score (Callum Fraser, personal communication) indicated that there was no significant difference in the sensitivity of the tool (for either CRC or advanced neoplasia) between Scotland and Spain, the two countries where the majority of studies in our systematic review were conducted.

Although the sample sizes of studies included in our systematic review were generally large for diagnostic accuracy studies (median n = 474, range 83–2058), it should be noted that study samples may not be representative of all of those people presenting in primary care with symptoms as specified in NG12. 1 Only three of the studies13,52,56 included in our systematic review explicitly reported the number of people who were invited to participate, and proportions agreeing were relatively low (64%,13 56%56 and 48%52), implying that study samples may be self-selecting for more motivated patients.

A further limitation of the study populations is that no study reported data that were specific to the population that was defined in the scope for this assessment (people presenting, in primary care settings, with lower abdominal symptoms who are at low risk for CRC according to the criteria defined in NG121). Only one study52 was conducted in a primary care setting, and this study reported that faecal immunochemical testing was ordered by GPs at the point of referral to secondary care. Although the primary care study might be considered the most applicable to this setting, because it is the only study to allow a change in referral decision based on FIT result, this design also gives rise to a risk of bias, as 11% of participants who returned a FIT sample were excluded from the analysis because they were not subsequently referred to secondary care or the referral was cancelled. 52 All of the studies included some participants who had symptoms that may be considered to be associated with a higher probability of CRC and which are components of the criteria for 2-week referral as defined in NG121 (e.g. rectal bleeding). The prevalence of CRC and advanced neoplasia is likely to differ with different presenting symptoms and with the health-care setting in which testing is undertaken (primary care vs. secondary care); the median prevalence of CRC reported in the studies included in our systematic review was 3.0% (range 1.0–12.3%) compared with the estimate of 1.5% for the relevant symptomatic group used in NG12. 6 Although it is known that the prevalence of the target condition can affect test performance characteristics,7 there are insufficient data to adequately assess these effects for the use of faecal immunochemical testing in symptomatic patients. The results of our systematic review do not suggest significant differences in test performance between the primary care study52 and other included studies that used the same FIT assay, threshold(s) and target condition (see Chapter 3, Diagnostic performance of the OC-Sensor faecal immunochemical test assay). Similarly, validation data for the FAST score (Callum Fraser, personal communication) indicated that there was no significant difference in the sensitivity or specificity estimates for the score between primary and secondary care settings.

Systematic reviews conducted previously have assessed the test characteristics of various FIT assays in screening settings8,131 and compared the performance of faecal immunochemical testing for CRC screening with that of gFOBT. 24,25 However, given the potential for target condition prevalence to affect test performance characteristics,7 it is important to determine the diagnostic accuracy of faecal immunochemical testing in the population of interest; we are not aware of any previous systematic review on this topic. We identified one large systematic review132 that assessed the value of symptoms and additional diagnostic tests for CRC, used in symptomatic patients in primary care. However, the searches for this review were completed in 2008 and identified only three studies of quantitative FIT assays. Our searches identified all of the three studies:133–135 one study133 was excluded from this assessment because it included a mixed population of symptomatic, screening and high-risk patients and no separate data for the symptomatic subgroup, and the remaining two studies134,135 were excluded because they used development stage Hb/Hp complex tests, which are not currently available in the NHS. Additionally, Williams et al. 136 have recently published a systematic review of risk prediction models for CRC or HRA in people with symptoms. This review136 included 15 risk models, none of which included faecal Hb measured by faecal immunochemical testing (or any other method) as a variable. Our systematic review is the first to assess the performance of faecal immunochemical testing, as a triage test for CRC, in people with symptoms and to consider the potential utility of applying faecal immunochemical testing as part of a simple risk score.

Cost-effectiveness

A major strength of our analysis is that it is the first study to assess the cost-effectiveness of faecal immunochemical testing as a triage test for symptomatic low-risk people presenting in primary care settings, based on a systematic review of the accuracy of faecal immunochemical testing. The input parameters used to model the accuracy of faecal immunochemical testing in the diagnostic part of the cost-effectiveness model (decision tree) were informed by a comprehensive high-quality systematic literature review on the clinical effectiveness of several FIT assays. For consistency, we sourced costs (other than those for faecal immunochemical testing and gFOBT) and CRC natural progression parameters from the current clinical guideline NG12. 6 These parameters were updated to their most recent values when deemed necessary. Although many of the input parameters of our cost-effectiveness model are the same as those described in NG12,6 the ones pertaining to faecal immunochemical testing differed because we performed cost-effectiveness analyses for specific FIT assays. In NG12,6 faecal immunochemical testing was included only as a single intervention and only in a scenario analysis.

A further strength of our study is that it includes detailed data on resource use and equipment costs for faecal immunochemical testing. Data on resource use were obtained from expert opinion (e-mail from Callum Fraser, NHS Tayside, to Marie Westwood, KSR Ltd, 19 July 2016, personal communication). This was important because figures on resource use for the population considered in this diagnostic assessment are scarce in the literature; we could not find any published study. Data on equipment costs were obtained from the relevant companies and are presented in a detailed and transparent way in this diagnostic assessment. This was also important because, in the papers found, which assessed the costs of faecal immunochemical testing, it was not always clear how the total costs were estimated from resource use and equipment costs.

Finally, we have considered a large variety of scenarios to explore the uncertainties around the assumptions that were made in the cost-effectiveness model and, in order to quantify these uncertainties in a statistical way, we performed PSAs for all of them.

The secondary care aspects of our model are a simplification of what actually occurs in practice. The ‘no triage’ comparator arm would be more accurately described as ‘no FIT/gFOBT’, as the secondary care consultation with a gastroenterologist clearly forms part of the triage before colonoscopy. The model structure follows a similar approach to that used in NG12,6 in that it assumes that all of the patients with a positive FIT/gFOBT and all of the patients in the ‘no FIT/gFOBT’ arm receive colonoscopy. However, the alternative would involve trying to estimate the proportion of patients whom the gastroenterologist would refer for colonoscopy without faecal immunochemical testing/gFOBT (this proportion would apply to the ‘no FIT/gFOBT’ arm). We would also need to estimate the effect that the positive FIT/gFOBT result has on the gastroenterologist’s probability of requesting colonoscopy (numbers referred from the FIT/gFOBT-positive pathway). We are not aware of any published data to inform these parameters, and asking clinicians for estimates of their own performance (the accuracy of their referral decisions) has a high risk of bias.

There is a potential problem with the population, which is based on NG12,6 in that it is likely to be heterogeneous, such that many patients might better be treated with a watchful waiting strategy rather than being referred immediately to secondary care. The model assumes that all of the patients who present with the symptoms specified in NG126 would be treated in the same way by the GP. Hence, we cannot capture the role of the GP’s clinical judgement in ruling out other possible explanations for the presenting symptoms for which faecal immunochemical testing/gFOBT would not be helpful. In practice, this is may be less of a problem than it appears, as the evidence from our systematic review is in a clinically appropriate population (those for whom the GP is considering referral to secondary care or for whom a referral decision has already been made). The prevalence of CRC in this population is also uncertain; NG126 relied on expert opinion to provide an estimate, which was 1.5%. In the absence of any data that were specific to this population, we also chose this value for the base case, although we tested the effect of using prevalence as observed in the accuracy studies from our systematic review.

There is a potential for patients with a FIT FP test for CRC to benefit from referral to secondary care/colonoscopy for diagnosis and treatment of conditions other than CRC (see Chapter 5, Clinical effectiveness). This is not captured in our model, which is, therefore, likely to underestimate the cost-effectiveness of faecal immunochemical testing.

The limitations described for clinical effectiveness regarding the lack, or the poor quality, of accuracy data for the different FIT assays are also applicable to the assessment of cost-effectiveness. Accuracy data were not available for all of the relevant FIT methods and there were no studies that compared more than one FIT method or that compared faecal immunochemical testing with gFOBT.

We lacked data to inform the parts of the diagnostic model that followed a FIT or gFOBT negative result. Therefore, we assumed that those who have CRC would remain symptomatic and would be diagnosed using colonoscopy or CTC. However, we did not know how many non-patients with CRC would also remain symptomatic; hence, we sought expert opinion. We also had to seek expert opinion on the number of patients who would get a second FIT/gFOBT, as well as the time to diagnosis of CRC. We did not seek expert opinion on the accuracy of a second FIT/gFOBT, even though we had no data, because we believe that this is something that cannot be reliably estimated by a clinician. We therefore sent out a questionnaire to 10 specialists, five of whom returned it, although not all of the questions were completely answered. We also conducted scenario analyses on the proportion who were symptomatic and time to diagnosis.

In the absence of data for this population, the probability of undergoing colonoscopy, as opposed to CTC, was estimated to be 83% by the experts (four answers). However, this value seems to be low compared with the screening population. Logan et al. 126 found that, for approximately 98% of the screening patients with an abnormal test result, the first investigation was colonoscopy. However, the ratio of colonoscopy to CTC seems to vary by centre and ranges from 0.3% to 9.1%;137 therefore, this parameter was varied in the scenario analysis.

In the base-case scenario, we assumed that adverse event rates for CTC were the same as those for colonoscopy. However, Burling et al. 19 reported that perforation rates for colonoscopy are four times higher than for CTC. No death events were reported and the proportion of patients experiencing bleeding after examination was not described. Thus, it is likely that our assumption would result in an overestimation of the total number of adverse events. Nevertheless, as there is uncertainty around the appropriate estimates for these adverse events, we decided to keep this assumption and interpret it as a worst-case scenario regarding the total number of adverse events. We explored the impact of this assumption on the cost-effectiveness results by performing additional scenario analyses.

Cost estimates for faecal immunochemical testing and gFOBT reported in the literature vary significantly. In particular, the costs of faecal immunochemical testing found in the literature ranged from £1.96 to £10.23 per test, and from (confidential information has been removed) to £9.57 per gFOBT. The estimated costs for the OC-Sensor and the HM-JACKarc FIT assays used in this diagnostic assessment were £4.53 and £6.04, respectively. Although the cost estimates used in this diagnostic assessment are properly reported, we acknowledge that there is uncertainty around the underlying assumptions leading to them. For example, it can be seen that, in Table 32, information for certain costs categories is missing. This may lead to an underestimation of the cost estimates for faecal immunochemical testing. On the other hand, the manufacturers indicated that the prices provided were also subject to discount, which could imply a lower cost in practice. Ideally, training and staff costs should have been included for faecal immunochemical testing and gFOBT. However, information about training and staff costs is scarce and, when available, relates to dedicated service provision for the national screening programme. For this reason, we decided not to include these costs in our health economic model. Nevertheless, as FIT assays use automated immunoassay platforms, which are standard in clinical laboratories, we believe that training costs would be minimal. To overcome these limitations, we performed threshold analyses on the cost differences between tests.

Our diagnostic model did not include reduction in quality of life (disutilities) for the adverse events that are associated with colonoscopy. Although no evidence was found on the effects of bleeding and perforation on quality of life, some effects on quality of life might be expected. Nevertheless, as these events are often of short duration, the effects on quality of life can be assumed negligible. Kapidzic et al. 138 also reported a slight reduction in quality of life for participants in a screening programme who had a positive result. This was due to the anxiety of knowing that they tested positive. However, these anxiety episodes can also be assumed to be of short duration, with minimal effects on overall quality of life.

As mentioned in Chapter 3 (see Methods of analysis/synthesis), the sensitivity and specificity point estimates for the OC-Sensor assay were derived using the bivariate model, which accounts for the correlation between sensitivity and specificity. However, in the PSA, sensitivity and specificity were sampled independently from the beta distributions presented in Table 23. Thus, the PSA does not account for the correlation between sensitivity and specificity. This was a pragmatic choice, as insufficient data were available to allow an analysis accounting for correlation for the HM-JACKarc or the FOB Gold assays.

The effect of a delayed CRC diagnosis was assumed to be mediated through the probability of progression within Dukes’ states, for undiagnosed patients, given a delay in the base case of 6 months. This was because the proportion of patients per Dukes’ stage for 6 months seemed to match reasonably well with the idea that FN patients would remain symptomatic and thus be picked up well within 1 year. Estimates for delayed diagnosis were also obtained from clinical experts, although only three of the experts who received our questionnaire answered this question; expert estimates were even lower, with two of them estimating a delay of approximately 2–3 months, whereas the third expert estimated this at between 1 and 2 months. A delay of 1 year also seems to be too long because it would result in 45% at Dukes’ stage D. This was deemed too high, given that the initial proportion of patients at stage D was 14%. The only published source that we found reporting delayed diagnosis was the Meester et al. study. 139 This study used a microsimulation model to calculate the effect of delay to colonoscopy on an average-risk population cohort in the USA who underwent annual FIT screening (from the age of 50 to 75 years), with follow-up colonoscopy examinations for individuals with positive results in the following 12 months. For the scenario of diagnostic follow-up at 12 months from a positive FIT, diagnosed cancers shifted towards more advanced stages, but still with only 8% in stage D. Regardless of the precise delay and its effect on progression, the cost-effectiveness results did not change significantly assuming 6 months’ or 1 year’s delay and so we decided not to run an additional scenario for a smaller delay. Note, however, that, as the delay in diagnosis diminishes, the differences in QALYs between the strategies decrease as well. In a hypothetical scenario with no delay, the difference in QALYs will be caused only by mortality. In this situation, the strategy with the least number of colonoscopies (gFOBT) is likely to dominate all of the others.

Colorectal cancer mortality was modelled using observational data by CRC stage at the time of diagnosis from the NCIN,98 as used in NG12. 6 Patients who survive might eventually progress, thus increasing mortality. It was not clear the extent to which the effect of progression was included. We chose to include a separate estimate of the rate of progression as in NG12. 6 However, the progression probabilities, which were obtained from Tappenden et al. ,99 were for undiagnosed patients with CRC. This is an obvious limitation, as diagnosed patients are considered in the model following colonoscopy. We also did not model separately the effect of treatment on patients with CRC mortality or cost, assuming that the estimates of mortality and cost were from all diagnosed patients with CRC, including those who had been treated and indeed ‘cured’. This does mean that the effect of newer treatments and thus the effect of delay to diagnosis cannot be taken into account.

Despite the limitations of the model for CRC progression described above, it should be emphasised that the lifetime cost-effectiveness results of this diagnostic assessment are completely determined by the results of the diagnostic part of the economic model. Thus, any difference between the strategies observed in the lifetime results was already captured in the diagnostic phase. Beyond the diagnostic model, the differences between the strategies were very small because these were caused only by the number of CRC-diagnosed patients and the distribution of these between Dukes’ stages. Given that death (due to colonoscopy) in the diagnostic phase is almost negligible, we could roughly estimate these as 15 patients with CRC from a hypothetical cohort of 1000 (as the estimated prevalence of the disease is 1.5%). Therefore, any difference in the lifetime results between the strategies will be caused by only these 15 patients. Hence, all of the limitations and uncertainties of the model for CRC progression will have a minimal impact on the model results of this diagnostic assessment. If the disease prevalence, the life expectancy or the costs and benefits of the treatments and surveillance of the population with the disease are likely to make a difference in the lifetime cost-effectiveness results then all of the limitations of the CRC model should be explored with extra caution.

Finally, we would like to emphasise the importance of specificity on the cost-effectiveness results. This can be partly explained, for example, with simple algebra. As described in Chapter 4 (see Model parameters), the probability that FIT/gFOBT is positive can be calculated as (sensitivity × prevalence) + [(1 – specificity) × (1 – prevalence)]. The first part of this equation (sensitivity × prevalence) is the probability of a TP, that is, of sending patients who will be diagnosed with CRC. The second part [(1 – specificity) × (1 – prevalence)] is the probability of a FP, that is, of wrongly sending healthy patients for colonoscopy. It is clear that in a situation for which the prevalence is low, the weight of specificity on this equation is larger than that of sensitivity, that is, any change in specificity will have a greater effect on the probability of FPs than any change in sensitivity will have on the probability of TPs. In this diagnostic assessment, the prevalence of CRC in the base-case population was assumed to be 1.5%. This means that the probability of a FP can vary from 0 to 0.985, whereas the probability of a TP can vary only from 0 to 0.015. The question then is what are the consequences, in terms of cost and QALYs, of either not missing TPs or avoiding FPs. People without CRC might be subjected to unnecessary colonoscopies, incur unnecessary cost and experience unnecessary adverse events (including death). On the other hand, patients with CRC might receive an earlier diagnosis, resulting in an increased number of QALYs. In fact, there was very little difference in QALYs between the FIT/gFOBT and the ‘no triage test’, which indicates little difference between avoiding FPs or not missing TPs in terms of health effects. Nevertheless, it appears that, in our model, improved specificity does outweigh improved sensitivity. This was clearly shown by the fact that OC-Sensor was almost identical to ‘no triage test’ in terms of QALYs, but cheaper, whereas HM-JACKarc, with its higher sensitivity and lower specificity, produced fewer QALYs and was more costly than OC-Sensor. This effect was also observed in the scenario analyses using the estimates of gFOBT accuracy from Niv and Sperber92 and Bjerregaard et al. ,93 where, with respect to the base-case scenario, sensitivity increased more than specificity decreased (approximately 20–25% vs. 9–15%). The net effect of these changes was that the cost-effectiveness results for gFOBT were worse than those obtained in the base-case scenario.

Clinical effectiveness

There remain a number of areas of uncertainty in relation to the performance characteristics of faecal immunochemical testing in specific patient subgroups and test combinations.

Population data indicate that concentrations of faecal Hb vary with age and sex, being higher in men and the elderly. 88,140 Further, a recent study conducted in Scotland found that faecal Hb concentrations also increased with increasing levels of deprivation (measured using the Scottish Index of Multiple Deprivation) and that this trend remained after controlling for age and sex. 141 Thus, at any cut-off concentration, more men, more older people and more people in high deprivation groups are likely to have a positive result on faecal immunochemical testing. The performance of faecal immunochemical testing in these subgroups, within the symptomatic population, is therefore an important consideration for this assessment; optimal thresholds may differ between subgroups and may also differ depending on the geographic location in which the test is being applied. The baseline characteristics of our included studies (where reported) indicated that the average age of participants ranged from 59 to 67 years (total range was 16–94 years) and the percentage of males ranged from 40.4% to 56%. However, only one of the studies included in our systematic review compared the accuracy of a FIT assay (HM-JACKarc) in men and women,57 and no study compared different age groups. The study reporting separate data for men and women found that, at all thresholds, the observed sensitivity of HM-JACKarc for advanced neoplasia was higher in men than in women and the observed specificity was similar for men and women. 57 This indicates that, at any given threshold, more women than men with CRC or HRA may be missed by using a FIT as a triage test to determine referral to secondary care. However, it should be noted that this study differed from others in the systematic review in that it included some patients who were undergoing colonoscopy for polyp surveillance and excluded people with GI bleeding or active rectal bleeding; the prevalence of CRC in this study was the lowest of any included study (0.96%). 57 More data are needed to adequately assess whether or not there are clinically relevant differences in test performance between men and women, and these data are needed for all FIT assays. We did not identify any studies that compared the accuracy of faecal immunochemical testing in different age groups, or with varying levels of deprivation. Validation data for the FAST score (Callum Fraser, personal communication) indicated that there were no significant differences in the sensitivity of this tool between men and women and between patients aged < 50 years and those who were ≥ 50 years old.

The effects on FIT performance of using multiple faecal samples per patient remain unclear. One published study57 and one unpublished study, Hospital Clinic de Barcelona 2015 (Philippa Pinn, personal communication), included in our systematic review, compared single sampling with double sampling, and both asked participants to collect two consecutive faecal samples. The Auge et al. study57 used HM-JACKarc and reported that sensitivity for advanced neoplasia was increased (at all thresholds) when the highest value from two consecutive FIT samples was used, compared with using only the first sample; in this study, FIT results were discordant in 39.2% of participants. The Hospital Clinic de Barcelona 2015 study (confidential information has been removed). There is currently insufficient information about intra-individual variation in faecal Hb concentration over time to determine the clinical utility of multiple sampling.

The scope of this assessment did not include evaluation of the performance characteristics of faecal immunochemical testing when used in combination with other biomarkers. However, we identified one study142 that did not meet the inclusion criteria for our systematic review because it used an obsolete FIT assay, but that reported data on the performance characteristics of faecal immunochemical testing in combination with faecal calprotectin, M2-PK or both (for which a positive result was defined as at least one test being positive) for the target conditions of CRC and HRA, as well as data on the performance characteristics of faecal immunochemical testing alone. Faecal calprotectin is an inflammatory marker, whereas M2-PK is a key enzyme in tumour metabolism. 142 This study142 found that, in all cases, the addition of at least one further test to faecal immunochemical testing resulted in markedly increased sensitivity and decreased specificity. The sensitivity and specificity estimates for faecal immunochemical testing alone and CRC were 61.7% (95% CI 47.4% to 74.2%) and 88.8% (95% CI 84.1% to 92.3%), respectively; for the combination of faecal immunochemical testing and faecal calprotectin these estimates were 90.9% (95% CI 78.8% to 96.4%) and 35.9% (95% CI 29.7% to 42.6%), respectively; for faecal immunochemical testing and M2-PK, sensitivity and specificity were 91.5% (95% CI 80.1% to 96.6%) and 57.1% (95% CI 50.6% to 63.2%), respectively, and, for all of the three markers, they were 95.7% (85.7% to 98.8%) and 24.1% (18.8% to 30.2%), respectively. 142 Although all of the sensitivity estimates were generally lower, this pattern was repeated when the target condition was advanced neoplasia. 142 However, the FIT threshold in this study (20 µg Hb/g faeces) was higher than that which the results of our systematic review indicate is likely to be the optimal threshold (10 µg Hb/g faeces or a lower threshold). A second study143 of accuracy for the target condition of advanced neoplasia, which did not meet the inclusion criteria for this assessment because it used a qualitative FIT method, also found that combining faecal calprotectin with faecal immunochemical testing (where a positive result was defined as either or both tests positive) resulted in increased sensitivity and decreased specificity [92% (95% CI 82% to 97%) and 49% (95% CI 43% to 54%)] compared with faecal immunochemical testing alone [74% (95% CI 62% to 83%) and 82% (95% CI 78% to 86%)]. 143 The effectiveness of combining other biomarkers with quantitative faecal immunochemical testing (at the threshold at which faecal immunochemical testing is likely to be used in practice) remains unclear. We did not identify any data about the effects of adding faecal calprotectin (or any other biomarker) to risk scores that include faecal immunochemical testing. A study provided ahead of publication (e-mail from Karel Moons, University Medical Center, Utrecht, the Netherlands, to Marie Westwood, KSR Ltd, 20 June 2016, personal communication) reported the development of a multivariable diagnostic model for significant colorectal disease (defined as CRC, adenoma of > 10 mm, IBD or diverticulitis). This model included age, symptoms, digital rectal examination, and point-of-care FITs and faecal calprotectin tests. The authors concluded that this model may avoid approximately 30% of colonoscopy referrals from primary care at the cost of delaying around 6% of diagnoses (mostly HRA or diverticulitis); whether faecal calprotectin adds value, over and above point-of-care FIT was uncertain. Given the trade-off between ease of use/simplicity and diagnostic performance, the clinical value of including additional variables (e.g. symptoms and further diagnostic tests) in risk scores for CRC and/or other significant bowel disease is likely to require further investigation.

There is uncertainty about downstream consequences of using faecal immunochemical testing to triage symptomatic patients in primary care. It can be seen from the findings of our systematic review (see, in Chapter 3, the sections relating to the diagnostic performance of the OC-Sensor, HM-JACKarc, FOB Gold and RIDASCREEN FIT assays) that triage using faecal immunochemical testing at thresholds around 10 µg Hb/g faeces has the potential to correctly rule out CRC and avoid colonoscopy in approximately 75% of symptomatic patients and that this estimate does not appear to vary greatly between FIT assays. Further, the relatively high proportion of FIT FPs observed when the target condition is CRC may be mitigated by the detection of other bowel pathologies in these patients; we estimate that between 22.5% and 93% of patients with a positive FIT and no CRC will have other significant bowel pathologies, depending largely on how many, and which, diagnoses are included in the target condition.

The full potential benefits of faecal immunochemical testing in symptomatic patients, including those relating to diagnoses other than CRC, remain unclear. This issue may be particularly important in younger patients, in whom the prevalence of CRC is lowest and other diagnoses are more likely. Our systematic review identified a recently published protocol for a cluster randomised trial (NCT02308384) of a clinical education intervention (provision of a starting package of faecal immunochemical testing with clinical instructions, including guidance on interpretation of results) in general practices in Denmark, which may partially address these issues. 144 The stated aim of this trial is to assess the diagnostic value and clinical implications of using faecal immunochemical testing, in general practice, for patients presenting with non-alarm symptoms of CRC (‘low risk, but not no risk’). This population is defined as patients aged ≥ 30 years with symptoms and signs of CRC, but who do not meet the Danish Cancer Patient Pathway referral criteria; typical indications are given as change in bowel habit, abdominal pain, anaemia or decrease in Hb of > 10%, non-specific symptoms (weight loss, fatigue, loss of appetite). The specified outcome measures for clinical impact are age- and sex-standardised number and rate of urgent referrals for CRC; age- and sex-standardised number and rate of colonoscopies; all colonoscopy findings by ICD-10 code; age- and sex-standardised number of CRCs diagnosed; stage distribution of CRC diagnosed (I–IV). This trial has the potential to inform the clinical effectiveness component of this assessment, but will not address all potentially relevant applications of faecal immunochemical testing, as it is designed to assess faecal immunochemical testing as a rule-in test, with all patients who have a positive result (≥ 50 µg/l Hb, equivalent to 10 µg Hb/g faeces using OC-Sensor) receiving urgent colonoscopy referral; interpretation guidance in the protocol states that a negative FIT should not exclude CRC.

Finally, the acceptability of faecal immunochemical testing, as indicated by reported sample return rates, varied widely between the studies included in our systematic review. There is evidence that faecal immunochemical testing is more acceptable, when used as a screening test/method, than gFOBT; the use of faecal immunochemical testing in screening programmes results in increased uptake compared with gFOBT. 145 However, these data are unlikely to be transferable to symptomatic patients, for whom factors such as the setting in which the test is requested, and advice and information received directly from clinicians, will affect uptake rates. It therefore remains unclear whether or not the method of assessing faecal occult blood has any effect on uptake in patients who are symptomatic.

Cost-effectiveness

The main uncertainties described for the review of clinical effectiveness also apply to the assessment of cost-effectiveness. These are caused by a lack of clinical effectiveness data for the performance characteristics of faecal immunochemical testing in specific patient subgroups and for the effects of using multiple faecal samples. Once these uncertainties are resolved, the cost-effectiveness of faecal immunochemical testing could be assessed for different subgroups of patients (e.g. elderly) and faecal immunochemical testing using multiple faecal samples per patient could be included as an intervention in our cost-effectiveness model.

MEDLINE (via Ovid): 1946 to March Week 3 2016

Date searched: 30 March 2016.

Records found: 3198.

1. ((immunochem$ or immuno-chem$ or immunohistochem$ or immuno-histochem$ or immunol$ or immunochromatographic or immuno-chromatographic or immunoassay or immuno assay) adj4 (f?ecal or f?eces or stool or stools)).ti,ab,ot,hw. (910)

2. iFOBT.ti,ab,ot,hw. (83)

3. 1 or 2 (932)

4. F?ecal h?emoglobin.ti,ab,ot,hw. (94)

5. H?emoccult.ti,ab,ot,hw. (669)

6. FOBT.ti,ab,ot,hw. (1013)

7. (guaiac$ or gFOBT).ti,ab,ot,hw. (3444)

8. Guaiac/ (319)

9. or/4-8 (4898)

10. (f?ecal or f?eces or stool or stools).ti,ab,ot,hw. (246,345)

11. occult blood/ or occult blood.ti,ab,ot,hw. (6307)

12. (test$ or measur$ or screen$ or exam$).ti,ab,ot,hw. (6,577,832)

13. 10 and 11 and 12 (3172)

14. 3 or 9 or 13 (7254)

15. exp colorectal neoplasms/ (162,552)

16. exp cecal neoplasms/ (4892)

17. ((colorect$ or rectal$ or rectum$ or colon$ or sigma$ or sigmo$ or rectosigm$ or bowel$ or anal or anus) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot,hw. (199,488)

18. CRC.ti,ab,ot. (13,848)

19. ((cecum or cecal or caecum or caecal or il?eoc?ecal or il?eoc?ecum) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (1846)

20. (large intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (1675)

21. (lower intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (26)

22. 15 or 16 or 17 or 18 or 19 or 20 or 21 (208,172)

23. 14 and 22 (3206)

24. (FOB gold$ or FOBgold$).ti,ab. (11)

25. (jack-arc$ or jackarc$ or HM-JACKarc$).ti,ab. (0)

26. (RIDASCREEN$ H?emo$ or RIDASCREEN$ Hapto$).ti,ab. (1)

27. (OC Sensor$ or OC-Sensor$ or OC Pledia$ or OC-Pledia$).ti,ab. (38)

28. or/24-27 (43)

29. 23 or 28 (3208)

30. exp animals/ not (exp animals/ and humans/) (4,205,567)

31. 29 not 30 (3198)

MEDLINE In-Process & Other Non-Indexed Citations, MEDLINE Daily Update (via Ovid): March 29, 2016

Date searched: 30 March 2016.

Records found: 255.

1. ((immunochem$ or immuno-chem$ or immunohistochem$ or immuno-histochem$ or immunol$ or immunochromatographic or immuno-chromatographic or immunoassay or immuno assay) adj4 (f?ecal or f?eces or stool or stools)).ti,ab,ot,hw. (134)

2. iFOBT.ti,ab,ot,hw. (9)

3. 1 or 2 (138)

4. F?ecal h?emoglobin.ti,ab,ot,hw. (16)

5. H?emoccult.ti,ab,ot,hw. (23)

6. FOBT.ti,ab,ot,hw. (72)

7. (guaiac$ or gFOBT).ti,ab,ot,hw. (321)

8. Guaiac/ (0)

9. or/4-8 (418)

10. (f?ecal or f?eces or stool or stools).ti,ab,ot,hw. (18,710)

11. occult blood/ or occult blood.ti,ab,ot,hw. (302)

12. (test$ or measur$ or screen$ or exam$).ti,ab,ot,hw. (671,105)

13. 10 and 11 and 12 (202)

14. 3 or 9 or 13 (632)

15. exp colorectal neoplasms/ (220)

16. exp cecal neoplasms/ (1)

17. ((colorect$ or rectal$ or rectum$ or colon$ or sigma$ or sigmo$ or rectosigm$ or bowel$ or anal or anus) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot,hw. (14,744)

18. CRC.ti,ab,ot. (2522)

19. ((cecum or cecal or caecum or caecal or il?eoc?ecal or il?eoc?ecum) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (197)

20. (large intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (40)

21. (lower intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (0)

22. 15 or 16 or 17 or 18 or 19 or 20 or 21 (15,079)

23. 14 and 22 (254)

24. (FOB gold$ or FOBgold$).ti,ab. (2)

25. (jack-arc$ or jackarc$ or HM-JACKarc$).ti,ab. (3)

26. (RIDASCREEN$ H?emo$ or RIDASCREEN$ Hapto$).ti,ab. (0)

27. (OC Sensor$ or OC-Sensor$ or OC Pledia$ or OC-Pledia$).ti,ab. (3)

28. or/24-27 (6)

29. 23 or 28 (255)

30. exp animals/ not (exp animals/ and humans/) (3648)

31. 29 not 30 (255)

MEDLINE Epub Ahead of Print (via Ovid): June 20, 2016

Date searched: 21 June 2016.

Records found: 80.

1. ((immunochem$ or immuno-chem$ or immunohistochem$ or immuno-histochem$ or immunol$ or immunochromatographic or immuno-chromatographic or immunoassay or immuno assay) adj4 (f?ecal or f?eces or stool or stools)).ti,ab,ot,hw. (36)

2. iFOBT.ti,ab,ot,hw. (3)

3. 1 or 2 (37)

4. F?ecal h?emoglobin.ti,ab,ot,hw. (4)

5. H?emoccult.ti,ab,ot,hw. (1)

6. FOBT.ti,ab,ot,hw. (24)

7. (guaiac$ or gFOBT).ti,ab,ot,hw. (77)

8. Guaiac/ (0)

9. or/4-8 (102)

10. (f?ecal or f?eces or stool or stools).ti,ab,ot,hw. (1652)

11. occult blood/ or occult blood.ti,ab,ot,hw. (73)

12. (test$ or measur$ or screen$ or exam$).ti,ab,ot,hw. (131,690)

13. 10 and 11 and 12 (65)

14. 3 or 9 or 13 (156)

15. 15 exp colorectal neoplasms/ (0)

16. exp cecal neoplasms/ (0)

17. ((colorect$ or rectal$ or rectum$ or colon$ or sigma$ or sigmo$ or rectosigm$ or bowel$ or anal or anus) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot,hw. (3630)

18. CRC.ti,ab,ot. (799)

19. ((cecum or cecal or caecum or caecal or il?eoc?ecal or il?eoc?ecum) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (16)

20. (large intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (12)

21. (lower intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (0)

22. 15 or 16 or 17 or 18 or 19 or 20 or 21 (3677)

23. 14 and 22 (80)

24. (FOB gold$ or FOBgold$).ti,ab. (1)

25. (jack-arc$ or jackarc$ or HM-JACKarc$).ti,ab. (0)

26. (RIDASCREEN$ H?emo$ or RIDASCREEN$ Hapto$).ti,ab. (0)

27. (OC Sensor$ or OC-Sensor$ or OC Pledia$ or OC-Pledia$).ti,ab. (5)

28. or/24-27 (5)

29. 23 or 28 (80)

30. exp animals/ not (exp animals/ and humans/) (0)

31. 29 not 30 (80)

EMBASE (via Ovid): 1974 to 2016 March 29

Date searched: 30 March 2016.

Records found: 5255.

1. Fecal Immunochemical Test/ [EMTREE candidate term 13.1.16] (135)

2. ((immunochem$ or immuno-chem$ or immunohistochem$ or immuno-histochem$ or immunol$ or immunochromatographic or immuno-chromatographic or immunoassay or immuno assay) adj4 (f?ecal or f?eces or stool or stools)).ti,ab,ot,hw. (1630)

3. iFOBT.ti,ab,ot,hw. (164)

4. 1 or 2 or 3 (1672)

5. F?ecal h?emoglobin.ti,ab,ot,hw. (155)

6. H?emoccult.ti,ab,ot,hw. (866)

7. FOBT.ti,ab,ot,hw. (1857)

8. (guaiac$ or gFOBT).ti,ab,ot,hw. (5159)

9. Guaiac/ (696)

10. or/5-9 (7538)

11. (f?ecal or f?eces or stool or stools).ti,ab,ot,hw. (347,542)

12. occult blood/ or occult blood.ti,ab,ot,hw. (11,184)

13. (test$ or measur$ or screen$ or exam$).ti,ab,ot,hw. (9,112,793)

14. 11 and 12 and 13 (5121)

15. 4 or 10 or 14 (11,582)

16. exp colon tumor/ (239,813)

17. exp rectum tumor/ (182,401)

18. exp colon cancer/ (191,557)

19. exp rectum cancer/ (148,462)

20. ((colorect$ or rectal$ or rectum$ or colon$ or sigma$ or sigmo$ or rectosigm$ or bowel$ or anal or anus) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot,hw. (318,031)

21. CRC.ti,ab,ot. (27,205)

22. ((cecum or cecal or caecum or caecal or il?eoc?ecal or il?eoc?ecum) adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (2649)

23. (large intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (1929)

24. (lower intestin$ adj3 (cancer$ or neoplas$ or oncolog$ or malignan$ or tumo?r$ or carcinoma$ or adenocarcinoma$ or sarcoma$ or adenom$ or lesion$)).ti,ab,ot. (29)

25. or/16-24 (329,839)

26. 15 and 25 (5264)

27. (FOB gold$ or FOBgold$).ti,ab. (27)

28. (jack-arc$ or jackarc$ or HM-JACKarc$).ti,ab. (7)

29. (RIDASCREEN$ H?emo$ or RIDASCREEN$ Hapto$).ti,ab. (2)

30. (OC Sensor$ or OC-Sensor$ or OC Pledia$ or OC-Pledia$).ti,ab. (151)

31. or/27-30 (169)

32. 26 or 31 (5274)

33. animal/ (1,733,374)

34. animal experiment/ (1,919,080)

35. (rat or rats or mouse or mice or murine or rodent or rodents or hamster or hamsters or pig or pigs or porcine or rabbit or rabbits or animal or animals or dogs or dog or cats or cow or bovine or sheep or ovine or monkey or monkeys).ti,ab,ot,hw. (6,178,883)

36. or/33-35 (6,178,883)

37. exp human/ (16,972,709)

38. human experiment/ (350,478)

39. or/37-38 (16,974,155)

40. 36 not (36 and 39) (4,862,933)

41. 32 not 40 (5255)

The Cochrane Library (Wiley) (via the internet)

www.cochranelibrary.com/

Date searched: 30 March 2016.

Records found:

• CDSR: Issue 3 of 12, March 2016: 25 records

• DARE: Issue 2 of 4, April 2015: 34 records

• HTA Database: Issue 1 of 4, January 2016: 27 records

• NHS EED: Issue 2 of 4, April 2015: 103 records

• CENTRAL: Issue 2 of 12, February 2016: 446 records.

#1 (immunochem* or “immuno-chem*” or immunohistochem* or “immuno-histochem*” or immunol*) near/4 (f*ecal or f*eces) (237)

#2 iFOBT (14)

#3 “F*ecal h*emoglobin” (15)

#4 H*emoccult (143)

#5 FOBT (247)

#6 gFOBT or guaiac (132)

#7 [mh ^guaiac] (30)

#8 f*ecal or f*eces or stool or stools (9516)

#9 MeSH descriptor: [Occult Blood] this term only (480)

#10 “occult blood” (907)

#11 #9 or #10 (907)

#12 test* or measur* or screen* or exam* (430,115)

#13 #8 and #11 and #12 (683)

#14 #1 or #2 or #3 or #4 or #5 or #6 or #7 or #13 (931)

#15 MeSH descriptor: [Colorectal Neoplasms] explode all trees (6053)

#16 MeSH descriptor: [Cecal Neoplasms] explode all trees (9)

#17 (colorec* or rectal* or rectum* or colon* or sigma* or sigmo* or rectosigm* or bowel* or anal or anus) near/3 (cancer* or neoplas* or oncolog* or malignan* or tumo*r* or carcinoma* or adenocarcinoma* or sarcoma* or adenom* or lesion*) (12,274)

#18 CRC (1334)

#19 (cecum or cecal or caecum or caecal or il*eoc*ecal or il*eoc*ecum) near/3 (cancer* or neoplas* or oncolog* or malignan* or tumo*r* or carcinoma* or adenocarcinoma* or sarcoma* or adenom* or lesion*) (99)

#20 “large intestin*” near/3 (cancer* or neoplas* or oncolog* or malignan* or tumo*r* or carcinoma* or adenocarcinoma* or sarcoma* or adenom* or lesion*) (54)

#21 “lower intestin*” near/3 (cancer* or neoplas* or oncolog* or malignan* or tumo*r* or carcinoma* or adenocarcinoma* or sarcoma* or adenom* or lesion*) (0)

#22 #15 or #16 or #17 or #18 or #19 or #20 or #21 (12627)

#23 #14 and #22 (638)

#24 “FOB gold*” or “FOBgold*” (2)

#25 “jack-arc*” or “jackarc*” or “HM-JACKarc*” (0)

#26 “RIDASCREEN* Haemo*” or “RIDASCREEN* Hemo*” or “RIDASCREEN* Hapto*” (0)

#27 “OC Sensor*” or “OC-Sensor*” or “OC Pledia*” or “OC-Pledia*” (16)

#28 #24 or #25 or #26 or #27 (16)

#29 #23 or #28 (638) (CDSR: 25; DARE: 34; HTA: 27; NHS EED: 103; CENTRAL: 446)

International Network of Agencies for Health Technology Assessment Publications

www.inahta.org/publications/

Date searched: 30 March 2016.

Records found: 8.

National Institute for Health Research Health Technology Assessment programme

www.nets.nihr.ac.uk/projects?collection=netscc&meta_P_sand=Project

Date searched: 30 March 2016.

Records found: 1.

Aggressive Research Intelligence Facility Reviews Database

www.birmingham.ac.uk/research/activity/mds/projects/HaPS/PHEB/ARIF/databases/index.aspx

Date searched: 30 March 2016.

Records found: 17.

PROSPERO (via internet)

www.crd.york.ac.uk/prospero/

Date searched: 30 March 2016.

Records found: 9.

ClinicalTrials.gov (via internet)

https://clinicaltrials.gov/

Date searched: 8 March 2016.

Records found: 51.

European Union Clinical Trials Register (via internet)

www.clinicaltrialsregister.eu/ctr-search/search

Date searched: 8 March 2016.

Records found: 1.

International Clinical Trials Registry Platform (via internet)

http://apps.who.int/trialsearch/

Date searched: 8 March 2016.

Records found: 29.

American Gastroenterological Association – Digestive Disease Week 2011–15

Date searched: 13 April 2016.

Records found: 92.

www.gastrojournal.org/content/ddw_abstracts

2011 DDW Abstract Supplement

www.gastrojournal.org/issue/S0016-5085%2811%29X6001-8

2012 DDW Abstract Supplement

www.gastrojournal.org/issue/S0016-5085%2812%29X6001-3

2013 DDW Abstract Supplement

www.gastrojournal.org/issue/S0016-5085%2813%29X6001-9

2014 DDW Abstract Supplement