Role of the faecal immunochemical test in patients with risk-stratified suspected colorectal cancer symptoms: A systematic review and meta-analysis to inform the ACPGBI/BSG guidelines.

Background: The UK National Institute for Health and Care Excellence (NICE), recommended in 2017 the use of the faecal immunochemical test (FIT) to guide investigations in patients presenting with NICE-defined low-risk symptoms suspicious for colorectal cancer (CRC). At that time, NICE did not recommend FIT use for high-risk symptoms. This is the first systematic review to evaluate the diagnostic accuracy of FIT in NICE-defined high and low-risk symptoms and was designed to inform the joint ACPGBI/BSG guidelines.
Methods: We performed a systematic literature review and meta-analysis. PROSPERO registration number CRD42021224674. Medline and EMBASE databases were searched from inception to 31st March 2022. We included studies recruiting adult patients presenting with suspected CRC symptoms in whom FIT was performed and diagnostic accuracy data for CRC detection could be derived at a limit of detection (LoD) and/or 10 µg haemoglobin/gram faeces threshold in four commonly used analysers. FIT performance was assessed for high-risk, low-risk and individual symptoms where possible. Bivariate meta-analysis was performed where study numbers allowed. Findings: Thirty-one studies (79566 patients) met inclusion criteria. At 10 µg/g, for "all symptoms" (n = 35,945) sensitivity and specificity were 91.0% (95% CI: 88.9, 92.7) and 75.2% (95% CI: 69.6, 80.1); for "high-risk" symptoms (n = 18,264), 88.7% (95% CI: 84.4, 92.0) and 78.5% (95% CI: 73.0, 83.2); and for "low-risk" symptoms (n = 2161), 88.7% (95% CI: 78.1, 95.3) and 88.5% (95% CI: 87.1, 89.9), respectively. At LoD, for "all symptoms" (n = 26,056) sensitivity and specificity were 94.7% (95% CI: 90.5, 97.1) and 66.5% (95% CI: 58.7, 73.6); for "high-risk" symptoms (n = 16,768), 92.8% (95% CI: 86.4, 96.3) and 70.3% (95% CI: 66.5, 73.8); and for "low-risk" symptoms (n = 2082), 94.7% (95% CI: 85.4, 98.9) and 71.9% (95% CI: 69.9, 73.9), respectively. Summary estimates were similar across different analysers. Interpretation: FIT sensitivity for CRC detection is maximised at the LoD; its performance is similar in high and low-risk symptoms, and across different analysers where a common threshold is used. FIT performance for CRC detection is adequate and transferrable to clinical diagnostic pathways. Funding: This review was part-funded by NHS England awarded to RM Partners. RB and RC were funded by research fellowships awarded by Croydon University Hospital.

Evidence before this study

Faecal immunochemical tests (FIT) are used by the National Institute for Health and Care Excellence (NICE) to guide referral for investigation of patients with low-risk symptoms, but to date there have been no recommendations on the use of FIT in high-risk symptoms. Previous meta-analyses were hampered by a low number of studies and heterogeneity with mixed cohorts including patients in screening populations and in some cases CRC/polyp surveillance populations or by using a mixture of reference standards which may introduce verification bias.

We performed a systematic review searching MEDLINE and EMBASE databases from inception to 31st March 2022, using the terms listed in appendix A. Thirty-one studies met inclusion criteria. For an “all symptoms” analysis at 10 µg/g threshold (n=35,945) the sensitivity and specificity were 91.0% (95% CI: 88.9, 92.7) and 75.2% (95% CI: 69.6, 80.1). For “high-ripk“ symptoms (n=18,264), the sensitivity and specificity were 88.7% (95% CI: 84.4, 92.0) and 78.5% (95% CI: 73.0, 83.2). For “low-risk“ symptoms (n=2161), the sensitivity and specificity were 88.7% (95% CI: 78.1, 95.3) and 88.5% (95% CI: 87.1, 89.9). As might be expected, reducing the FIT threshold to the limit of detection gains a marginal increase in sensitivity with concurrent decrease in specificity.

Added value of this study

This is the first systematic review to assess the effect of “high-risk” and “low-risk” symptom criteria on the diagnostic accuracy of FIT for CRC detection. It also assesses the diagnostic accuracy of FIT for individual symptoms. To minimise the effect of verification bias, FIT utility was evaluated separately in true diagnostic accuracy studies with cohorts receiving full colonic imaging as the reference standard and studies of FIT in clinical diagnostic pathways with cohorts receiving mixed reference standards, and then compared.

Implications of all the available evidence

FIT performance for CRC detection is similar in “high risk” and “low risk” symptom clusters as well as rectal bleeding, change in bowel habit and iron deficiency anaemia. The current definitions of “high-risk” and “low-risk” symptoms for CRC are no longer needed in the FIT era. FIT performance was also similar in both diagnostic studies and clinical pathways, therefore can be used safely as an initial triage for all patients presenting with new symptoms suspicious for CRC. No clinically significant difference exists in the diagnostic performance of the two most-commonly used FIT analysers.

Alt-text: Unlabelled box

Introduction

Colorectal cancer is the second most common cause of cancer death in the UK, accounting for 10% of cancer-related mortality. Despite the introduction of the national bowel cancer screening programme in 2008, only 9.8% of cases are diagnosed via this pathway; nearly all other patients are diagnosed because of bowel symptoms.

In England, the National Institute for Health and Care Excellence (NICE) employs symptom-based criteria to guide urgent referral for suspected colorectal cancer (CRC) (Table 1). The original 2005 NICE guidance included only high-risk symptoms for CRC. This was superseded in 2015 by the much-expanded NG12 guidance that also included medium and low risk symptoms. In 2017, NICE introduced DG30 guidance recommending use of FIT in low-risk symptoms in primary care to guide referral for further investigation. This guidance did not recommend the use of FIT in patients with high-risk symptoms, who continue to be referred with increasing numbers, creating a significant demand for diagnostic services. The COVID-19 pandemic further added to the backlog of patients awaiting endoscopy which forced many UK clinical services to adopt FIT into their clinical pathways to meet local needs, by reducing referrals and balancing demand for endoscopy.

Table 1

Symptoms stratified according to cancer risk according to NICE CG27, NG12 and DG30 guidelines.

Symptoms	2005 NICE Guidance (CG27)	2015 NICE Guidance (NG12)	2017 NICE Guidance (DG30)	Risk of cancer
Rectal bleeding for 6 weeks (>60 years) Rectal bleeding + diarrhoea for 6 weeks (>40 years) Change in bowel habit for 6 weeks (>60 years) Mass (any age) Iron deficiency anaemia	REFER	REFER	REFER	High: >5%
Abdominal pain AND weight loss (>40 years)Rectal bleeding (>50 years)Rectal bleeding + (Iron deficiency anaemia/change in bowel habit/weight loss, <50 years)ron deficiency anaemia (>60 years)Change in bowel habit (>60 years)		REFER	REFER	Medium: 3–5%
Abdominal pain OR weight loss (>50 years) Change in bowel habit (<60 years) Iron deficiency anaemia (<60 years) Anaemia, Non-Iron deficient (>60 years)		Test with FOBT before referral	Test with FIT before referral	Low: 1–3%
Other symptoms			FIT, if no rectal bleeding	Low: <1%

NICE= National Institute for Health and Care Excellence. FOBT= Guaiac faecal occult blood test. FIT= Faecal immunochemical test.

This systematic review was designed to inform the joint guidelines of The Association of Coloproctology of Great Britain and Ireland (ACPGBI) and British Society of Gastroenterology (BSG) on the role of FIT in symptomatic patients due for release in 2022. The analysis aimed to assess the diagnostic accuracy of quantitative FIT for CRC detection in purely symptomatic patient cohorts at the NICE DG30-recommended threshold of 10 µg haemoglobin per gram faeces (hereafter µg/g) and at the limit of detection (LoD) of available assays taking account of the reference standard used in the included studies; the recruitment setting (primary or secondary care); and the symptom clusters including high and low-risk symptoms.

Methods

We used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to structure and report our systematic review. The review was registered with the International Prospective Register of Systematic Reviews (PROSPERO), registration number CRD42021224674.

Data sources and search strategy

We used a combination of search terms related to FIT (appendix A) and included all studies identified. Databases searched were MEDLINE and EMBASE via OVID from inception to 31 March 2022. In addition, references in included studies and systematic reviews were checked for further studies, and we also contacted several authors. Search results were combined, and duplicates removed using EndNote reference software.

Study selection and inclusion criteria

Figure 1 illustrates the flow through the study selection process. Three authors (RB, RC and NDS) independently screened titles with disagreement resolved via discussion or through consultation with a fourth author (MA). A similar process was followed for abstract screening. Final eligibility was determined by review of full text, with papers included if they met the following criteria:

Figure 1

PRISMA diagram showing selection of articles for inclusion in the review.

Population, setting and study design

We included cohort studies performed on adult patients consulting a physician with symptoms suggestive of CRC in whom quantitative FIT was performed as part of their work-up. Studies recruiting in both primary and secondary care were included. Studies reporting on mixed cohorts including screening or follow-up populations were excluded.

Index test

We included data from studies that evaluated the diagnostic accuracy of quantitative FIT for CRC using the FOB Gold (Sentinel Diagnostics, Italy), HM-JACKarc (Hitachi Chemical Diagnostics Systems, Tokyo, Japan), OC-Sensor (Eiken Chemical, Japan) and QuikRead Go (Aidian Oy, Espoo, Finland) FIT assays. All included studies used single-sample FIT.

Reference test

We included diagnostic accuracy studies for FIT that used full colonic imaging with either colonoscopy or CT colonography (CTC) as the reference test for the diagnosis of CRC. We also included pragmatically designed studies using FIT within symptomatic pathways where registry follow-up formed part of the reference standard for some or all FIT negative patients. We grouped studies by reference standard and performed subsequent meta-analyses accordingly. The first group (hereafter referred to as tier 1) included studies where at least 90% of the cohort underwent full colonic imaging with either colonoscopy or CTC as the reference test (colonoscopy and CTC having equivalent sensitivity for CRC detection). The second group (hereafter referred to as tier 2) included studies with mixed reference standards including plain CT and flexible sigmoidoscopy, which reflects clinical practice, as well as studies with a minimum of 3 months’ registry follow-up.

Data extraction

Two reviewers (RB and RC) independently extracted data which were compared and cross-checked by the other and queries clarified with the senior author (MA). The extracted data included: study design; publication year; geographical location; patient numbers; recruitment setting (primary or secondary care, or both); assay(s) used; thresholds employed (LoD or 10 µg/g); reference standard used; and presenting symptom cluster as stratified by NICE (NG12 hereafter high-risk, DG30 hereafter low-risk, or unstratified) or individual symptoms of rectal bleeding, iron deficiency anaemia and change in bowel habit.

Data were extracted to two-by-two tables either from absolute numbers of true-positive, false-negative, true-negative, and false-positive observations, or derived from reported sensitivity, specificity, positive and negative predictive value data.

Endpoints

The primary aim was to assess the diagnostic accuracy of quantitative FIT for CRC detection in purely symptomatic patients at the NICE DG30 recommended threshold of 10 µg/g and the LoD for each analyser. Secondary goals were to assess the utility of FIT for symptom clusters: high-risk; low-risk; iron-deficiency anaemia, change in bowel habit and rectal bleeding.

Risk of bias assessment

Studies were assessed for potential risks of bias and applicability independently by three authors (RB, RC and NDS) using the Quality Assessment of Diagnostic Accuracy Studies 2 tool (QUADAS-2). An example screening tool and summary table of assessments are included in appendix B.

The extent to which publication bias occurs in studies of test accuracy is uncertain, however, simulation studies have indicated that the effect of publication bias on meta-analytic estimates of test accuracy is minimal. Formal assessment of publication bias in systematic reviews of test accuracy studies remains problematic and reliability is limited. We did not undertake a statistical assessment of publication bias; however, our search strategy included a variety of routes to identify unpublished studies and resulted in the inclusion of several conference abstracts.

Statistical analysis

Meta-analyses were performed grouping studies by reference standard (tier 1 or tier 2) and threshold for a negative FIT result (LoD or 10 µg/g). The LoD of the HM-JACKarc and FOB Gold assays is <2 µg/g,, OC-Sensor <4 µg/g and QuikRead go <10 µg/g. Studies using the HM-JACKarc and OC-Sensor at these thresholds were combined within the LoD analyses. Subgroup analyses were performed to investigate the potential effects of analyser, threshold, presenting symptom cluster and recruitment location, on estimates of test accuracy.

Where four or more studies were available, bivariate random-effects analysis was performed to give summary estimates of sensitivity and specificity using STATA 13. Where this was not possible, random-effects meta-analysis was performed using Meta DiSc 1.4. Hierarchical summary receiver operating characteristic curves (HSROCs) were constructed using a bivariate model, based on Reitsma et al, and graphically depicted summary operating points and 95% confidence regions. HSROC analyses were performed using version 0.5.10 of package mada in R version 4.1.0.

Role of the funding source

The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

Literature search, study characteristics and symptom clusters

The literature search yielded 10,447 discrete titles once results from Medline and EMBASE were combined with duplicates removed (Figure 1). Title screening removed 9751 irrelevant titles. For the remaining 696 titles, abstracts were reviewed and screened to 221 full-text articles. Full-text articles which were screened and excluded, with reasons for exclusion are listed in appendix C.

Thirty-one articles (n = 79,566) met inclusion criteria and had extractable data. Of these 16 (n = 35,945) were eligible for inclusion in tier 1 and 11 (n = 43,621) in tier 2. Four studies reported data on a specific symptom derived from a previously included cohort and were used for evaluation of FIT in those symptoms only, to avoid “double counting”.

Nine studies (n = 31,190) reported data on patients presenting with high-risk symptoms and 6 studies (n = 6842) on low-risk symptoms. Four studies (n = 1050) yielded data for iron-deficiency anaemia and 3 studies each yielded data on FIT utility in change in bowel habit (n = 11,211) and rectal bleeding (n = 3665).

The median CRC prevalence within included studies was 3.7% (range 1.1–16.0%).

Table 2 shows details of geographical study setting; recruitment location (primary or secondary care); reference tests; study cancer prevalence; presenting symptoms; and individual study sensitivity, specificity, positive and negative predictive values for CRC by FIT threshold, for all included studies.

Table 2

Study characteristics stratified by reference standard.

Study	Setting	Region	Reference test(s)	Study cancer prevalence (%)	Analyser	Presenting symptoms	n patients	Threshold (µg/g)	Sensitivity % (95% CI)	Specificity % (95% CI)	PPV % (95% CI)	NPV % (95% CI)
Tier 1 - >90% Colonoscopy or CT Colonography
Chapman, 202115	Primary care	UK (Eng), single centre	Colonoscopy	5.2	OC-Sensor	High-risk	732	>10	89.5 (75.9, 95.8)	74.1 (70.7, 77.2)	15.9 (11.6, 21.4)	99.2 (98, 99.7)
								>4 (LoD)	97.4 (86.5, 99.5)	74.7 (71.1, 78.1)	19.8 (14.7, 26.1)	99.8 (98.7, 100)
					HM-JACKarc			>10	84.2 (69.6, 92.6)	78 (74.7, 80.9)	17.3 (12.5, 23.4)	98.9 (97.6, 99.5)
								>4	92.1 (79.2, 97.3)	70 (66.5, 73.3)	14.4 (10.5, 19.4)	99.4 (98.2, 99.8)
D'Souza, 202016	Secondary care	UK (Eng), single centre	Colonoscopy	4.0	HM-JACKarc	High-risk	160	>10	87.5 (52.9, 97.8)	84.2 (77.6, 89.2)	22.6 (11.4, 39.8)	99.2 (95.7, 99.9)
								>2 (LoD)	100 (67.6, 100)	71.1 (63.4, 77.7)	15.4 (8, 27.5)	100 (96.6, 100)
						Low-risk	138	>10	100 (51, 100)	93.3 (87.7, 96.4)	30.8 (12.7, 57.6)	100 (97, 100)
								>2 (LoD)	100 (51, 100)	82.1 (74.7, 87.7)	14.3 (5.7, 31.5)	100 (96.6, 100)
						Unstratified	298	>10	91.7 (64.6, 98.5)	88.7 (84.5, 91.8)	25 (14.6, 39.4)	99.6 (97.8, 99.9)
								>2 (LoD)	100 (75.7, 100)	76.2 (71, 80.8)	15 (8.8, 24.4)	100 (98.3, 100)
D'Souza (2), 202017	Secondary care	UK (Eng), multicentre	Colonoscopy	3.3	HM-JACKarc	High-risk	7194	>10	92.2 (88.3, 94.9)	82.3 (81.4, 83.2)	16.2 (14.4, 18.2)	99.7 (99.5, 99.8)
								>2 (LoD)	97.7 (95, 98.9)	63 (61.8, 64.1)	8.9 (7.9, 10)	99.9 (99.7, 99.9)
						Low-risk	1944	>10	86.8 (75.2, 93.5)	88 (86.5, 89.4)	16.9 (12.9, 21.8)	99.6 (99.1, 99.8)
								>2 (LoD)	94.3 (84.6, 98.1)	71.2 (69.1, 73.2)	8.4 (6.4, 10.9)	99.8 (99.3, 99.9)
						Unstratified	9822	>10	90.9 (87.3, 93.5)	83.5 (82.8, 84.3)	16.1 (14.5, 17.8)	99.6 (99.5, 99.7)
								>2 (LoD)	97 (94.5, 98.3)	64.9 (63.9, 65.8)	8.7 (7.9, 9.7)	99.8 (99.7, 99.9)
Farrugia, 202018	Secondary care	UK (Eng), single centre	Colonoscopy, CT Colonography	6.2	HM-JACKarc	High-risk	519	>10	84.8 (69.1, 93.3)	81.3 (77.6, 84.5)	23.5 (16.8, 31.9)	98.8 (97.1, 99.5)
						Low-risk	79		100 (56.6, 100)	91.9 (83.4, 96.2)	45.5 (21.3, 72)	100 (94.7, 100)
						Unstratified	612		86.8 (72.7, 94.2)	82.2 (78.9, 85.1)	24.4 (18, 32.3)	99 (97.6, 99.6)
Godber, 201619	Secondary care	UK (Sco), single centre	Colonoscopy	2.3	HM-JACKarc	Unstratified	484	>10	100 (74.1, 100)	76.5 (72.5, 80.1)	9 (5.1, 15.4)	100 (98.9, 100)
Hererro, 201820	Secondary care	Spain, multicentre	Colonoscopy	13.6	OC-Sensor	Unstratified	1572	>10	93.5 (89.3, 96.1)	63.4 (60.8, 65.9)	28.7 (25.5, 32.2)	98.4 (97.3, 99)
*Khasawneh, 202021	Primary care	UK (Eng)	CT Colonography	1.2	OC-Sensor	CIBH	5818	>10	88.9 (79.6, 94.3)	80.8 (79.7, 81.8)	5.5 (4.3, 6.9)	99.8 (99.7, 99.9)
								>4 (LoD)	91.7 (83, 96.1)	69.7 (68.5, 70.9)	3.7 (2.9, 4.6)	99.9 (99.7, 99.9)
Laszlo, 202133	Primary and secondary care	UK (Eng), multicentre	Colonoscopy, CT Colonography, CT, Flexible Sigmoidoscopy	2.5	OC-Sensor	High-risk	3596	>10	83.3 (74.3, 89.6)	80.1 (78.8, 81.4)	9.7 (7.8, 12)	99.5 (99.1, 99.7)
								>4 (LoD)	87.8 (79.4, 93)	73.1 (71.6, 74.5)	7.7 (6.2, 9.5)	99.6 (99.2, 99.8)
								>6	86.7 (78.1, 92.2)	76.1 (74.7, 77.5)	8.5 (6.9, 10.5)	99.6 (99.2, 99.7)
McSorley, 202022	Primary care	UK (Sco), multicentre	Colonoscopy	5.5	HM-JACKarc	High-risk	4841	>10	94.7 (91.4, 96.8)	47 (45.6, 48.5)	9.4 (8.4, 10.6)	99.4 (98.9, 99.6)
*Morales-Arraez, 201823	Unclear	Spain, single centre	Colonoscopy	11.4	OC-Sensor	IDA	245	>10	92.9 (77.4, 98)	57.1 (50.5, 63.5)	21.8 (15.4, 30.1)	98.4 (94.4, 99.6)
Mowat, 201624	Primary care	UK (Sco), single centre	Colonoscopy	3.7	OC-Sensor	Unstratified	750	>10	89.3 (72.8, 96.3)	79.1 (76, 81.9)	14.2 (9.8, 20.1)	99.5 (98.5, 99.8)
								>4 (LoD)	100 (87.9, 100)	43.4 (39.8, 47)	6.4 (4.5, 9.1)	100 (98.8, 100)
Navarro, 202025	Secondary care	Spain, single centre	Colonoscopy	5.0	FOB Gold	Unstratified	727	>10	94.4 (81.9, 98.5)	75.1 (71.8, 78.2)	16.5 (12.1, 22.2)	99.6 (98.6, 99.9)
Rodriguez-Alonso, 201526	Secondary care	Spain, single centre	Colonoscopy	3.0	OC-Sensor	Unstratified	1003	>10	96.7 (83.3, 99.4)	79.9 (77.2, 82.3)	12.9 (9.1, 17.9)	99.9 (99.3, 100)
Schwettmann58	Secondary care	Norway, single centre	Colonoscopy	16.0	FOB Gold	Unstratified	163	>10	96.2 (81.1, 99.3)	51.8 (43.5, 60.0)	27.5 (23.9, 31.5)	98.6 (91.1, 99.8)
Tsapournas, 202027	Secondary care	Sweden, multicentre	Colonoscopy	5.3	QuikRead go	Unstratified	242	>10	92.3 (66.7, 98.6)	77.3 (71.4, 82.2)	18.8 (11.1, 30)	99.4 (96.9, 99.9)
						Rectal bleeding	60		100 (61, 100)	74.1 (61.1, 83.9)	30 (14.5, 51.9)	100 (91.2, 100)
Turvill, 202139	Secondary care	UK (Eng), multicentre	Colonoscopy, CT Colonography, CT, Flexible Sigmoidoscopy	3.0	HM-JACKarc	Unstratified	5040	>10	87.4 (81.2, 91.8)	80.9 (79.8, 82)	12.4 (10.5, 14.5)	99.5 (99.3, 99.7)
								>2 (LoD)	92.7 (87.4, 95.9)	60.7 (59.4, 62.1)	6.8 (5.8, 8)	99.6 (99.3, 99.8)
Tier 2 – other reference standards
Ayling, 201928	Secondary care	UK (Eng), single centre	Colonoscopy, CT	3.9	OC-Sensor	High-risk	178	>10	66.7 (30, 90.3)	95.4 (90.4, 97.9)	40 (16.8, 68.7)	98.4 (94.4, 99.6)
						IDA	137		71.4 (35.9, 91.8)	95.9 (91.8, 98)	41.7 (19.3, 68)	98.8 (95.7, 99.7)
Bailey J, 202129	Primary care	UK (Eng)	Registry F/up	1.7	OC-Sensor	High-risk	13042	>10	92.1 (87.8, 94.9)	81.7 (81, 82.4)	8.2 (7.2, 9.3)	99.8 (99.7, 99.9)
								>4 (LoD)	96.5 (93.2, 98.2)	69.5 (68.7, 70.3)	5.3 (4.7, 6)	99.9 (99.8, 100)
Bailey S, 202130	Primary care	UK (Eng)	Registry F/up	1.1	HM-JACKarc	Low-risk	3890	>10	84.3 (72, 91.8)	85 (83.9, 86.1)	7 (5.2, 9.2)	99.8 (99.5, 99.9)
Juul, 201831	Primary care	Denmark	Colonoscopy, Registry F/up	1.6	OC-Sensor	Unstratified	3462	>10	94.4 (84.9, 98.1)	85.7 (84.4, 86.8)	9.4 (7.3, 12.2)	99.9 (99.7, 100)
Khan, 202032	Secondary care	UK (Eng), single centre	Colonoscopy, CT Colonography, CT	4.8	HM-JACKarc	High-risk	928	>10	86.7 (73.8, 93.7)	83.5 (80.9, 85.8)	21.1 (15.8, 27.5)	99.2 (98.2, 99.6)
Maclean, 202134	Secondary care	UK (Eng), single centre	Colonoscopy, CT Colonography, Flexible Sigmoidoscopy	2.5	QuikRead go	Low-risk	553	>10	92.9 (68.5, 98.7)	70.1 (66.1, 73.8)	7.5 (4.4, 12.4)	99.7 (98.5, 100)
Mowat, 202135	Primary care	UK (Sco), single centre	Colonoscopy, CT Colonography, CT, Flexible Sigmoidoscopy, Registry follow-up	2.0	HM-JACKarc	Unstratified	5381	>10	86.7 (78.9, 91.9)	79.4 (78.3, 80.5)	7.7 (6.3, 9.4)	99.7 (99.4, 99.8)
								>2 (LoD)	97.1 (91.9, 99)	49.5 (48.1, 50.8)	3.7 (3, 4.5)	99.9 (99.7, 100)
								>7	88.6 (81.1, 93.3)	75.9 (74.7, 77)	6.8 (5.6, 8.3)	99.7 (99.5, 99.8)
Nicholson, 201836	Primary care	UK (Eng)	Colonoscopy, CTC, Registry F/up	7.0	HM-JACKarc	Low-risk	238	>10	85.7 (48.7, 97.4)	90.5 (86, 93.6)	21.4 (10.2, 39.5)	99.5 (97.4, 99.9)
								>7	85.7 (48.7, 97.4)	89.2 (84.5, 92.6)	19.4 (9.2, 36.3)	99.5 (97.3, 99.9)
Nicholson, 202037	Primary care	UK (Eng)	Registry F/up	1.1	HM-JACKarc	Unstratified	9896	>10	90.5 (83.4, 94.7)	91.3 (90.8, 91.9)	10.1 (8.3, 12.2)	99.9 (99.8, 99.9)
								>7	91.4 (84.5, 95.4)	89.8 (89.2, 90.4)	8.7 (7.2, 10.6)	99.9 (99.8, 99.9)
Pin Vieto, 202038	Primary care	Spain, multicentre	Registry F/up	1.4	OC-Sensor	Unstratified	5623	>10	80.2 (70.3, 87.5)	84.1 (83.1, 85)	6.9 (5.4, 8.7)	99.7 (99.4, 99.8)
						CIBH	1144		93.3 (70.2, 98.8)	82.2 (79.9, 84.3)	6.5 (3.9, 10.6)	99.9 (99.4, 100)
Widlak, 201740	Secondary care	UK (Eng), single centre	Colonoscopy, CT Colonography, CT, Flexible Sigmoidoscopy	5.8	HM-JACKarc	Unstratified	430	>7	88.0 (70, 95.8)	93.1 (90.2, 95.2)	44 (31.2, 57.7)	99.2 (97.7, 99.7)
Studies with specific symptom analyses derived from other included cohorts
Cunin, 202041	Secondary care	UK (Eng), single centre	Colonoscopy, CT Colonography, CT	5.2	HM-JACKarc	IDA	189	>10	80 (58.4, 91.9)	81.7 (75.1, 86.8)	34 (22.2, 48.3)	97.2 (93, 98.9)
D'Souza, 202142	Secondary care	UK (Eng), multicentre	Colonoscopy	3.3	HM-JACKarc	IDA	479	>10	100 (89.6, 100)	81.6 (77.8, 84.9)	28.7 (21.2, 37.5)	100 (99, 100)
						CIBH	4249		82.7 (73.1, 89.4)	87.5 (86.5, 88.5)	11.4 (9.1, 14.2)	99.6 (99.4, 99.8)
								>2 (LoD)	91.4 (83.2, 95.8)	68.4 (67, 69.8)	5.3 (4.3, 6.6)	99.8 (99.5, 99.9)
Digby, 202043	Primary care	UK (Sco), single centre	Colonoscopy	5.6	OC-Sensor	Rectal bleeding	462	>10	96.2 (81.1, 99.3)	38.3 (33.9, 42.9)	8.5 (5.8, 12.3)	99.4 (96.7, 99.9)
Hicks, 202144	Secondary care	UK (Eng), multicentre	Colonoscopy	3.3	HM-JACKarc	Rectal bleeding	3143	>10	96.6 (92.3, 98.5)	76.6 (75, 78.1)	16.8 (14.5, 19.5)	99.8 (99.5, 99.9)

CI: confidence interval; CIBH: change in bowel habit; IDA: iron-deficiency anaemia; LoD: limit of detection; Unstratified: Symptoms not stratified by “high-risk” or “low-risk” definitions; NG12: NICE guideline 12 (Suspected cancer: recognition and referral) – “high-risk symptoms”; DG30: NICE diagnostics guidance 30 (Quantitative faecal immunochemical tests to guide referral for colorectal cancer in primary care) – “low-risk symptoms”; * published as an abstract.

QUADAS-2 assessment

Appendix B shows the quality assessment of the 31 included studies using the QUADAS-2 instrument. Nine studies were assessed as having low risk of bias and applicability concerns across all domains. Ten studies were assessed as having high risk of bias in the patient selection domain. These concerns principally centred around either non-consecutive recruitment,,,,,, (for both tier 1 and tier 2 studies), or discretionary referral for investigation in the event of ongoing clinical concern for tier 2 studies.,,,, There were no significant differences seen in FIT sensitivity when comparing those studies assessed as being at high risk of bias with those at low risk of bias, however two studies, showed a moderately lower specificity at a threshold of 10 µg/g (47.0% (95% CI: 45.6, 48.5) and 57.1% (95% CI: 50.3, 60.5) respectively. Seven tier 2 studies were assessed as being at risk of bias in the reference standard domain, owing to the use of multiple reference tests or registry follow-up. Ten studies had applicability concern in the patient selection domain, often due to concerns regarding exclusions within the study population (Tiers 1 and 2).,,

Tier 1 analysis

Sixteen studies were included in tier 1: 6 using OC-Sensor; 6 using HM-JACKarc; 1 comparing both analysers; 1 using QuikRead go and 2 using FOB Gold (Table 3).

Table 3

Accuracy of FIT, tier 1 reference standard (≥90% of participants received colonoscopy or CTC), comparing individual and combined assays, symptom clusters and study setting: Summary estimates (95% CI).

Presenting symptoms	Analyser	Threshold (µg/g)	Setting	Number of studies (references)	n patients in analysis	Sensitivity (%)	Specificity (%)
Symptom groups
All (High-risk, low-risk or unstratified)	Pooled analysers	LoD(>4 OC-Sensor>2 HM-JACKarc)	Any	7 (15, 16, 17, 21, 24, 33, 39)a	26056	94.7 (90.5, 97.1)b	66.5 (58.7, 73.6)b
			Primary care	3 (15, 21, 24)	7300	94.9 (89.8, 97.9)c	67.5 (66.4, 68.6)c
			Secondary care	3 (16, 17, 39)	15160	95.7 (93.5, 97.3)c	63.7 (62.9, 64.5)c
		>10	Any	16 (15c, 16, 17, 18, 19, 20, 21, 33, 22, 23, 24, 25, 26, 27, 39, 58)	35945	91.0 (88.9, 92.7)b	75.2 (69.6, 80.1)b
				16 (15d, 16, 17, 18, 19, 20, 21, 33, 22, 23, 24, 25, 26, 27, 39, 58)	35945	91.2 (89.2, 92.8)b	75.0 (69.4, 79.8)b
			Primary care	4 (15c, 21, 22, 24)	12141	90.1 (83.9, 94.1)b	72.6 (58.6, 83.3)b
				4 (15d, 21, 22, 24)	12141	91.1 (85.7, 94.5)b	71.6 (57.7, 82.2)b
			Secondary care	10 (16, 17, 18, 19, 20, 25, 26, 27, 39, 58)	19963	91.6 (89.2, 93.6)b	77.2 (71.1, 82.3)b
	OC-Sensor	>4a	Any	4 (15, 21, 24, 33)	10896	95.0 (80.7, 98.9)b	65.8 (53.2, 76.5)b
			Primary care	3 (15, 21, 24)	7300	94.9 (89.8, 97.9)c	67.5 (66.4, 68.6)c
			Secondary care	0
		>6	Any	1 (33)d	3596	86.7 (77.9, 92.9)	76.1 (74.7, 77.5)
			Primary care	0
			Secondary care	0
		>10	Any	7 (15, 20, 21, 23, 24, 26, 33)	13716	90.2 (86.2, 93.1)b	74.5 (68.1, 79.9)b
			Primary care	3 (15, 21, 24)	7300	89.1 (82.7, 93.8)c	79.9 (79.0, 80.9)c
			Secondary care	2 (20, 26)	2575	93.9 (90.1, 96.5)c	70.3 (68.4, 72.1)c
	HM-JACKarc	>2a	Any	3 (16, 17, 39)e	15160	95.7 (93.5, 97.3)c	63.7 (62.9, 64.5)c
			Primary care	0
			Secondary care	3 (16, 17, 39)	15160	95.7 (93.5, 97.3)c	63.7 (62.9, 64.5)c
		>4	Any	1 (15)d,f	732	92.1 (78.6, 98.3)	70.0 (66.5, 73.4)
			Primary care	1 (15)d	732	92.1 (78.6, 98.3)	70.0 (66.5, 73.4)
			Secondary care	0
		>10	Any	7 (15, 16, 17, 18, 19, 22, 39)	21829	90.6 (87.6, 92.9)b	78.2 (69.2, 85.2)b
			Primary care	2 (15, 22)	5573	93.4 (90.0, 95.9)c	51.1 (49.8, 52.5)c
			Secondary care	5 (16, 17, 18, 19, 39)	16256	89.7 (86.4, 92.3)b	82.4 (79.2, 85.2)b
	FOB Gold	>10	Any	2 (25, 58)e	890	95.2 (86.5, 99.0)c	71.3 (68.0, 74.3) c
			Primary care	0
			Secondary care	2 (25, 58)	890	95.2 (86.5, 99.0)c	71.3 (68.0, 74.3) c
	QuikRead go	>10	Any	1 (27)e	242	92.3 (64.0, 99.8)	77.3 (71.3, 82.6)
			Primary care	0
			Secondary care	1 (27)	242	92.3 (64.0, 99.8)	77.3 (71.3, 82.6)
High-risk	Pooled analysers	LoD(>4 OC-Sensor>2 HM-JACKarc)	Any	4 (16, 17, 21, 33)a	16768	92.8 (86.4, 96.3)b	70.3 (66.5, 73.8)b
			Primary care	2 (15, 21)	6550	93.6 (87.3, 97.4)c	70.2 (69.1, 71.3)c
			Secondary care	2 (16, 17)	7534	97.7 (95.1, 99.2)c	63.1 (62.0, 64.3)c
		>10	Any	7 (15c, 16, 17, 18, 21, 23, 33)	18264	88.7 (84.4, 92.0)b	78.5 (73.0, 83.2)b
				7 (15d, 16, 17, 18, 21, 23 33)	18264	89.3 (85.2, 92.4)b	78.0 (72.2, 82.9)b
			Primary care	2 (15c, 21)	6550	87.3 (79.6, 92.9)e	80.5 (79.5, 81.4)e
				2 (15d, 21)	6550	89.1 (81.7, 94.2)e	80.0 (79.0, 81.0)e
			Secondary care	3 (16, 17, 18)	7873	91.3 (87.5, 94.2)c	82.3 (81.4, 83.2)c
	OC-Sensor	>4a	Any	3 (15, 21, 33)	10146	91.0 (86.1, 94.6)c	71.2 (70.3, 72.1)c
			Primary care	2 (15, 21)	6550	93.6 (87.3, 97.4)c	70.2 (69.1, 71.3)c
			Secondary care	0
		>6	Any	1 (33)	3596	86.7 (77.9, 92.9)	76.1 (74.7, 77.5)
			Primary care	0
			Secondary care	0
		>10	Any	4 (15, 21, 23, 33)	10391	88.7 (78.8, 94.3)b	74.2 (65.0, 81.7)b
			Primary care	2 (15, 21)	6550	89.1 (81.7, 94.2)c	80.0 (79.0, 81.0)c
			Secondary care	0
	HM-JACKarc	>2a	Any	2 (16, 17)e	7534	97.7 (95.1, 99.2)c	63.1 (62.0, 64.3)c
			Primary care	0
			Secondary care	2 (16, 17)	7534	97.7 (95.1, 99.2)c	63.1 (62.0, 64.3)c
		>4	Any	1 (15)f	732	92.1 (78.6, 98.3)	70.0 (66.5, 73.4)
			Primary care	1 (15)	732	92.1 (78.6, 98.3)	70.0 (66.5, 73.4)
			Secondary care	0
		>10	Any	4 (15, 16, 17, 18)	8605	89.0 (82.5, 93.3)b	81.1 (79.1, 82.9)b
			Primary care	1 (15)	732	92.1 (78.6, 98.3)	70.0 (66.5, 73.4)
			Secondary care	3 (16, 17, 18)	7873	91.3 (87.5, 94.2)c	82.3 (81.4, 83.2)c
Low-risk	HM-JACKarc	>2a	Any	2 (16, 17)e	2082	94.7 (85.4, 98.9)c	71.9 (69.9, 73.9)c
			Primary care	0
			Secondary care	2 (16, 17)	2082	94.7 (85.4, 98.9)c	71.9 (69.9, 73.9)c
		>10	Any	3 (16, 17, 18)e	2161	88.7 (78.1, 95.3)c	88.5 (87.1, 89.9)c
			Primary care	0
			Secondary care	3 (16, 17, 18)	2161	88.7 (78.1, 95.3)c	88.5 (87.1, 89.9)c
Individual symptoms
CIBH	OC-Sensor	>4a	Primary care	1 (21)	5818	91.7 (82.7, 96.9)	69.7 (68.5, 70.9)
		>10		1 (21)	5818	88.9 (79.3, 95.1)	80.8 (79.7, 81.8)
	HM-JACKarc	>2a	Secondary care	1 (42)	4249	91.4 (83.0, 96.5)	68.4 (67.0, 69.8)
		>10		1 (42)	4249	82.7 (72.7, 90.2)	87.5 (86.5, 88.5)
	Pooled analysers	LoD(>4 OC-Sensor>2 HM-JACKarc)	Any	2 (21, 42)	10067	91.5 (85.9, 95.4)	69.1 (68.2, 70.1)
		>10		2 (21, 42)	10067	85.6 (79.0, 90.8)	83.6 (82.9, 84.3)
IDA	OC-Sensor	>10	Unclear	1 (23)	245	92.9 (76.5, 99.1)	57.1 (50.3, 63.8)
	HM-JACKarc		Secondary care	1 (42)	479	100 (89.4, 100)	81.6 (77.7, 85.1)
	Pooled analysers		Any	2 (23, 42)	724	96.7 (88.7, 99.6)	73.6 (70.1, 76.9)
Rectal bleeding	OC-Sensor	>10	Primary care	1 (43)	462	96.2 (80.4, 99.9)	38.3 (33.7, 43.0)
	HM-JACKarc		Secondary care	1 (44)	3143	96.6 (92.2, 98.9)	76.6 (75.0, 78.1)
	QuikRead go			1 (27)	60	100 (54.1, 100)	74.1 (60.3, 85.0)
	Pooled analysers		Any	3 (27, 43, 44)	3665	96.6 (92.8, 98.8)	71.7 (70.2, 73.2)

LoD for the assay.

Bivariate Meta-analysis (STATA 13).

Random effects meta-analysis (Meta DiSc 1.4).

All studies conducted in patients presenting with high-risk symptoms.

All studies conducted in secondary care settings.

All studies conducted in primary care settings.

gOnly data for the HM-JACK assay for Chapman 2021 included to avoid double counting.

hOnly data for the OC-Sensor assay for Chapman 2021 included to avoid double counting.

CIBH: change in bowel habit; CI: confidence interval; DG: diagnostic guidance; IDA: iron deficiency anaemia; LoD: limit of detection.

FIT performance by presenting symptom and study setting were explored by pooling all studies using a common threshold (LoD for the assay or >10 µg/g) across analysers. At >10 µg/g threshold, for the ‘all symptoms’ analysis (16 studies, n = 35,945) the summary estimates of sensitivity and specificity were 91.0% (95% CI: 88.9, 92.7) and 75.2% (95% CI: 69.6, 80.1). For studies reporting on high-risk symptoms (7 studies, n = 18,264), the summary estimates of sensitivity and specificity were 88.7% (95% CI: 84.4, 92.0) and 78.5% (95% CI: 73.0, 83.2), and for studies reporting on low-risk symptoms (3 studies, n = 2161), the summary sensitivity and specificity estimates were 88.7% (95% CI:78.1, 95.3) and 88.5% (95% CI: 87.1, 89.9). At the LoD, for the ‘all symptoms’ analysis (7 studies, n = 26,056) the summary estimates of sensitivity and specificity were 94.7% (95% CI: 90.5, 97.1) and 66.5% (95% CI: 58.7, 73.6); for high-risk symptoms (4 studies, n = 16,768), 92.8% (95% CI: 86.4, 96.3) and 70.3% (95% CI: 66.5, 73.8), and for low risk symptoms (2 studies, n = 2082), 94.7% (95% CI: 85.4, 98.9) and 71.9% (95% CI: 69.9, 73.9), respectively. Figure 2 shows HSROC curves comparing FIT performance at LoD by symptom cluster (high-risk, low-risk, or unstratified) for any analyser and studies conducted in any setting. Figure 3 shows the same comparison at 10 µg/g.

Figure 2

HSROC curves for pooled assays at an LoD cutoff for patients with unstratified presenting symptoms, high-risk symptoms or low-risk symptoms, where studies were conducted in any setting (primary care, secondary care, both or unclear). 95% confidence region for each summary estimate is represented by the dashed-line curve.

Figure 3

HSROC curves for pooled assays at a cutoff of 10 µg/g for patients with unstratified presenting symptoms, high-risk symptoms or low-risk symptoms, where studies were conducted in any setting (primary care, secondary care, both or unclear). Curves generated using only OC-Sensor data for Chapman 2021 study to avoid double counting. 95% confidence region for each summary estimate is represented by the dashed-line curve.

For individual assay analyses (Table 3), the summary estimates of sensitivity at >10 µg/g were 90.2% (95% CI: 86.2, 93.1) for OC-Sensor (7 studies, n = 13,716), 90.6% (95% CI: 87.6, 92.9) for HM-JACKarc (7 studies, n = 21,829), 95.2% (95% CI: 86.5, 99.0) for FOB Gold (2 studies, n = 890) and 92.3% (95% CI: 64.0, 99.8) for QuikRead go (1 study, n = 242), and corresponding specificity estimates were 74.5% (95% CI: 68.1, 79.9) for OC-Sensor, 78.2% (95% CI: 69.2, 85.2) for HM-JACKarc, 71.3% (95% CI: 68.0, 74.3) for FOB Gold and 77.3% (95% CI: 71.3, 82.6) for QuikRead go. Using the LoD for the assay (4 µg/g for OC-Sensor and 2 µg/g for HM-JACKarc), the summary estimates of sensitivity were 95.0% (95% CI: 80.7, 98.9) for OC-Sensor (4 studies, n = 10,896) and 95.7% (95% CI: 93.5, 97.3) for HM-JACKarc (3 studies, n = 15,160) and the corresponding specificity estimates were 65.8% (95% CI: 53.2, 76.5) and 63.7% (95% CI: 62.9, 64.5); there were no data on the accuracy of FOB Gold using an LoD threshold. The LoD for QuikRead go is 10 µg/g and thus data for this analyser were described with others at a threshold of 10 µg/g.

A comparison of the HSROC curves for OC-Sensor, HM-JACKarc and pooled analysers, across all symptom groups and in any setting at various thresholds are in Appendix D.

For individual symptoms (Table 3), pooled estimates were produced for all analysers, all settings at the 10 µg/g threshold. A summary estimate was also possible for change in bowel habit at the LoD threshold. For change in bowel habit at the LoD threshold (2 studies, n = 10,067), the summary sensitivity was 91.5% (95% CI: 85.9, 95.4) and specificity 69.1% (95% CI: 68.2, 70.1), and at the threshold of 10 µg/g (2 studies, n = 10,067), 85.6% (95% CI: 79.0, 90.8) and 83.6% (95% CI: 82.9, 84.3), respectively. For iron deficiency anaemia, the summary sensitivity at a threshold of 10 µg/g (2 studies, n = 724) was 96.7% (95% CI: 88.7, 99.6) and specificity 73.6% (95% CI: 70.1, 76.9). For rectal bleeding, the summary sensitivity at a threshold of 10 µg/g (3 studies, n = 3665) was 96.6% (95% CI: 92.8, 98.8) and specificity 71.7% (95% CI: 70.2, 73.2).

Comparison of settings (primary versus secondary) care was not always possible and the numbers of studies in these analyses were generally small; there was no clear pattern of difference in test performance by setting (Tables 3 and 4).

Table 4

Accuracy of FIT, tier 2 reference standard (Mixed reference tests and registry follow up), comparing individual and combined assays, symptom clusters and study setting: Summary estimates (95% CI).

Presenting symptoms	Analyser	Threshold (µg/g)	Setting	Number of studies (references)	n patients in analysis	Sensitivity (%)	Specificity (%)
Symptom groups
All (High-risk, low-risk or unstratified)	Pooled analysers	LoD(>4 OC-Sensor>2 HM-JACKarc)	Any	2 (29, 35)b	18,423	96.7 (94.1, 98.3)e	63.7 (63.0, 64.4)e
			Primary care	2 (29, 35)	18,423	96.7 (94.1, 98.3)e	63.7 (63.0, 64.4)e
			Secondary care	0
		>10	Any	10 (28, 29, 30, 31, 32, 34, 35, 36, 37, 38)	43,191	88.2 (84.3, 91.2)d	85.7 (81.0, 89.3)d
			Primary care	7 (29, 30, 31, 35, 36, 37, 38)	41,532	88.8 (84.6, 92.0)d	85.7 (82.3, 88.5)d
			Secondary care	3 (28, 32, 34)	1659	86.4 (75.7, 93.6)e	80.3 (78.2, 82.2)e
	OC-Sensor	>4a	Any	1 (29)b,c	13,042	96.5 (93.2, 98.5)	69.5 (68.7, 70.3)
			Primary care	1 (29)c	13,042	96.5 (93.2, 98.5)	69.5 (68.7, 70.3)
			Secondary care	0
		>10	Any	4 (28, 29, 31, 38)	22,305	86.6 (73.9, 93.6)d	87.5 (80.0, 92.4)d
			Primary care	3 (29, 31, 38)	22,127	89.8 (86.2, 92.7)e	82.9 (82.4, 83.4)e
			Secondary care	1 (28)c	178	71.4 (29.0, 96.3)	95.9 (91.7, 98.3)
	HM-JACKarc	>2a	Any	1 (35)b	5381	97.1 (91.9, 99.4)	49.5 (48.1, 50.8)
			Primary care	1 (35)	5381	97.1 (91.9, 99.4)	49.5 (48.1, 50.8)
			Secondary care	0
		>7	Any	4 (35, 36, 37, 40)	15,945	90.0 (85.0, 93.5)d	88.0 (81.0, 92.6)d
			Primary care	3 (35, 36, 37)	15,515	89.9 (85.1, 93.5)e	85.0 (84.4, 85.5)e
			Secondary care	1 (40)	430	88.0 (68.8, 97.5)	93.1 (90.2, 95.4)
		>10	Any	5 (30, 32, 35, 36, 37)	20,333	87.8 (83.3, 91.2)d	86.4 (81.9, 89.9)d
			Primary care	4 (30, 35, 36, 37)	19,405	88.1 (83.1, 91.7)d	87.1 (81.8, 91.0)d
			Secondary care	1 (32)c	928	86.7 (73.2, 94.9)	83.5 (80.8, 85.9)
	QuikRead go	>10	Any	1 (34)f,g	553	92.9 (66.1. 99.8)	70.1 (66.1, 74.0)
			Primary care	0
			Secondary care	1 (34)g	553	92.9 (66.1. 99.8)	70.1 (66.1, 74.0)
High-risk	Pooled analysers	>10	Any	3 (28, 29, 32)	14,148	90.7 (86.6, 93.8)c	82.0 (81.3, 82.6)c
			Primary care	1 (29)	13,042	92.1 (87.8, 94.9)	81.7 (81.0, 82.4)
			Secondary care	2 (28, 32)	1065	84.3 (71.4, 93.0)e	85.0 (82.7, 87.2)e
	OC-Sensor	>4a	Any	1 (29)b	13,042	96.5 (93.2, 98.5)	69.5 (68.7, 70.3)
			Primary care	1 (29)	13,042	96.5 (93.2, 98.5)	69.5 (68.7, 70.3)
			Secondary care	0
		>10	Any	2 (28, 29)	13,220	91.4 (87.1, 94.7)e	81.8 (81.2, 82.5)e
			Primary care	1 (29)	13,042	92.1 (87.8, 95.2)	81.7 (81.0, 82.4)
			Secondary care	1 (28)	178	71.4 (29.0, 96.3)	95.9 (91.7, 98.3)
	HM-JACKarc	>7	Any	0
			Primary care	0
			Secondary care	0
		>10	Any	1 (32) f	928	86.7 (73.2, 94.9)	83.5 (80.8, 85.9)
			Primary care	0
			Secondary care	1 (32)	928	86.7 (73.2, 94.9)	83.5 (80.8, 85.9)
Low-risk	Pooled analysers	>10	Any	3 (30, 34, 36)	4681	86.1 (75.9, 93.1)e	83.6 (82.5, 84.6)e
	HM-JACKarc	>7	Any	1 (36)b	238	85.7 (48.7, 97.4)	89.2 (84.5, 92.6)
			Primary care	1 (36)	238	85.7 (48.7, 97.4)	89.2 (84.5, 92.6)
			Secondary care	0
		>10	Any	2 (30, 36)b	4128	84.5 (72.6, 92.7)e	85.3 (84.2, 86.4)e
			Primary care	2 (30, 36)	4128	84.5 (72.6, 92.7)e	85.3 (84.2, 86.4)e
			Secondary care	0
	QuikRead go	>10	Any	1 (34)f	553	92.9 (66.1. 99.8)	70.1 (66.1, 74.0)
			Primary care	0
			Secondary care	1 (34)	553	92.9 (66.1. 99.8)	70.1 (66.1, 74.0)
Individual symptoms
CIBH	OC-Sensor	>10	Primary care	1 (38)	1144	93.3 (68.1, 99.8)	82.2 (79.8, 84.4)
IDA			Secondary care	1 (28)	137	66.7 (22.3, 95.7)	95.4 (90.3, 98.3)
	HM-JACKarc			1 (41)	189	80.0 (56.3, 94.3)	81.7 (75.0, 87.2)
Rectal bleeding	OC-Sensor		Primary care	1 (43)	462	96.2 (80.4, 99.9)	38.3 (33.7, 43.0)

LoD for the assay.

All studies conducted in primary care settings.

All studies conducted in patients presenting with high-risk symptoms.

Bivariate Meta-analysis (STATA 13).

Random effects meta-analysis (Meta DiSc 1.4).

All studies conducted in secondary care settings.

All studies conducted in patients presenting with low-risk symptoms.

CIBH: change in bowel habit; CI: confidence interval; IDA: iron deficiency anaemia; LoD: limit of detection.

Tier 2 analysis

Eleven studies used alternative reference standards and were included in tier 2: 4 using OC-Sensor; 6 using HM-JACKarc and 1 using QuikRead go (Table 4). There were no tier 2 studies using the FOB Gold assay.

As observed for tier 1 studies, overall analyses of studies conducted across any setting and for the all-symptom groups showed similar summary estimates of sensitivity and specificity across different analysers when comparing a common threshold. Further pooling was undertaken combining data for different assays where a common threshold (LoD for the Assay or 10 µg/g) was used to explore the potential effects of presenting symptoms and study setting on estimates of test accuracy.

Comparison between tier 1 and tier 2 studies

Where sufficient data allowed, a comparison was made between the performance characteristics of FIT based on a full colonic imaging reference standard (tier 1) and those with mixed reference standard (tier 2), (Tables 3 and 4). The performance characteristics of FIT for CRC detection, using a LoD threshold, were similar in Tier 1 and 2 studies. The overall summary sensitivity for all analysers and all presenting symptoms, were 94.7% (95% CI: 90.5, 97.1) and 96.7% (95% CI: 94.1, 98.3) for tier 1 and tier 2, respectively and the corresponding summary specificity were 66.5% (95% CI: 58.7, 73.6) and 63.7% (95% CI: 63.0, 64.4). At a threshold of 10 µg/g, there is more variation with the overall summary sensitivity for all analysers and all presenting symptoms being 91.0% (95% CI: 88.9, 92.7) for tier 1 and 88.2% (95% CI: 84.3, 91.2) for tier 2 with the corresponding specificity being 75.2% (95% CI: 69.6, 80.1) and 85.7% (95% CI: 81.0, 89.3) respectively. Figure 4 shows HSROC curves comparing FIT performance, at LoD for tier 1 and tier 2 studies and Figure 5 provides the same comparison at 10 µg/g.

Figure 4

HSROC curves for pooled assays at an LoD cutoff, for tier 1 studies and for tier 2 studies, for patients with unstratified presenting symptoms where studies were conducted in any setting (primary care, secondary care, both or unclear). 95% confidence region for each summary estimate is represented by the dashed-line curve.

Figure 5

HSROC curves for pooled assays at cutoff of 10 µg/g, for tier 1 studies and for tier 2 studies, for patients with unstratified presenting symptoms where studies were conducted in any setting (primary care, secondary care, both or unclear). 95% confidence region for each summary estimate is represented by the dashed-line curve.

Discussion

This meta-analysis was undertaken to inform the joint ACPCBI/BSG guidelines on the use of FIT in symptomatic patients and indicates that FIT has a high sensitivity well above 90% for CRC irrespective of the presenting symptom(s), particularly when used at a threshold at or near to the LoD for the assay.

Summary of principal findings

Performance of FIT in NICE-defined and individual symptoms

This is the first review to investigate the potential effect of presenting symptoms cluster, according to NICE definitions of “high-risk” and “low-risk”, on the diagnostic accuracy of FIT. Within the limitations of the available data (in the context of the relatively recent introduction of DG30 guidelines for low-risk symptoms and hence lack of longitudinal data), the sensitivity of FIT for CRC detection is unaffected by the definition of “symptomatic patients” used. This suggests that current definitions of “high-risk” and “low-risk” symptoms for CRC are no longer required in the FIT era, and that FIT can be used for all symptomatic patients when CRC is suspected to triage their need and urgency for investigation.

Drawing conclusions regarding the diagnostic accuracy of FIT for individual symptoms is more challenging particularly given that patients commonly present with multiple symptoms. Whilst many studies give a breakdown of numbers of patients presenting with individual symptoms, relatively few provided analysable data by symptom (change in bowel habit, rectal bleeding or iron-deficiency anaemia). Within these limitations the summary sensitivity and specificity were broadly similar in patients presenting with rectal bleeding, change in bowel habit and iron-deficiency anaemia, and were comparable to the “all symptom”, high-risk, low-risk symptoms clusters. Therefore, there is no reason to treat these symptoms differently or exclude them from FIT testing.

Comparisons between analysers

We did not find any clinically significant difference in the performance of FIT for detection of CRC between currently available analysers. One study, included in our review, directly compared the performance of HM-JACKarc and OC-Sensor in a cohort of 732 patients and reported the sensitivity of OC-Sensor was marginally higher than HM-JACKarc at low thresholds of 4 µg/g and 10 µg/. However, this study compared FIT performance at 4 µg/g, which is the LoD of OC-Sensor, but above the LoD for HM-JACKarc.

Performance of FIT in formal diagnostic accuracy studies and clinical pathways

The performance of FIT for CRC detection at the LoD were similar in both Tier 1 and 2 studies. At a threshold of 10 µg/g, there is more variation with a lower overall summary specificity for all analysers and all presenting symptoms for Tier 1 compared to Tier 2 studies. This appears to be driven by the low specificity in the McSorley study at 47% (95% CI: 45.6, 48.5) compared with the other studies (Table 2). In this study primary care physicians were not blinded to FIT result and indeed given guidance reassuring them of the low risk of CRC with a negative FIT result. Consequently, “FIT positivity” rate amongst those referred was 55% which is more than double that seen in other diagnostic accuracy studies. This referral bias may have led to the drop in specificity observed. Within the limitations described comparisons of the performance characteristics of FIT estimated from tier 1 studies and from tier 2 studies suggest that FIT performance for CRC detection is adequate and transferrable to clinical diagnostic pathways for CRC.

Comparison with existing literature

Earlier meta-analyses,,, were hampered by a low number of studies and heterogeneity with mixed cohorts including patients in screening populations and in some cases CRC/polyp surveillance populations. Two meta-analyses in 2021,, included studies with different reference standards (ie registry or clinical follow-up cohorts and colonic investigation cohorts). This approach may introduce verification bias and prevents comparison of the performance characteristics of FIT, as has been done in our review. Despite these methodological differences and variation in the data, the results have been consistently similar which supports the robustness of use of FIT in symptomatic patients for detection of CRC.

Potential impact on referral rates

Figure 6 demonstrates the potential impact of the use of FIT applied to a hypothetical cohort of 1000 patients presenting to primary care using data from the tier 1 all analyser, all- symptoms analyses, at LoD and 10 µg/g thresholds, applied to a CRC prevalence rate of 3.3% as observed in 2019 England national data. Using a LoD threshold, 335 patients would return a faecal haemoglobin (f-Hb) above the LoD threshold or “FIT positive” and ideally would undergo full colonic imaging with colonoscopy or CT colonography. Of these, 31 would be correctly identified as having CRC. 2 patients with CRC would not be detected (FIT negative or undetectable f-Hb) and 643 patients would be correctly identified by FIT as not having CRC. Applying a threshold of 10 µg/g, 258 would be “FIT positive” and of these 30 CRC would be identified. 3 CRCs would not be detected and 739 would have been correctly identified by FIT as not having CRC. According to this analysis, use of a threshold of 10 µg/g rather than the LoD has the potential to reduce colonic investigations by 23%, with the caveat of a marginally higher rate of “FIT negative” or undetected cancers.

Figure 6

FIT and colonoscopy outcomes comparing a LoD threshold with a threshold of 10 µg/g for a hypothetical 1000 patients with a CRC prevalence of 3.3%.

Strengths and limitations

We used the same thorough search strategy for the meta-analysis that informed NICE DG30 guidelines; the search strategy was based on terms for the test and target condition and did not include any study design filters. We excluded diagnostic case-control or ‘two-gate’ studies, as this study design has been found to produce inflated estimates of test accuracy, compared with those derived from diagnostic cohort studies., Our primary analysis considered diagnostic accuracy studies where more than 90% of patients studied had full colonic imaging, the ‘gold standard’ for diagnosis of CRC, (Tier 1). However, we also included pragmatic studies that reflect clinical practice, where not all patients were suitable to undergo full colonic imaging, but (clinically) justified other investigations such as CT scan and flexible sigmoidoscopy, and also included studies with Registry follow-up for some or all FIT negative patients (Tier 2). The minimum follow-up period was chosen following review of the Nicholson study which showed no significant difference in sensitivity after a follow-up period of 3, 6 or 12 months, with FIT negative cancers presenting within first 3 months.

The inclusion of these studies allowed comparisons between the diagnostic performance of FIT when the ‘gold standard’ method is used to determine the presence or absence of CRC, and the diagnostic performance of FIT when the presence or absence of CRC is determined as it would be in clinical practice. We did not pool data across the two categories of reference standard, whether for individual symptoms or for wider populations because, where a different reference standard is used, this essentially explores the performance of FIT to detect a different definition of the target condition.

A potential limitation of this study is the strong geographical weighting to United Kingdom-based studies (23/31 studies). The other eight included studies recruited cohorts from other European nations (5 Spanish, 1 Danish, 1 Norwegian and 1 Swedish). This geographical bias could potentially limit the generalisability of our study although we have not seen significant variation in FIT performance in these studies.

Current uncertainties

There is currently a relative paucity of diagnostic accuracy data for the QuikRead go (2 studies) and FOB Gold analysers (2 studies). Meta-analysis for these was only possible for the overall all symptoms, all analysers analysis at a threshold of 10 µg/g. No data were available for FOB Gold at the LoD threshold.

The broadening of the symptom definition within NG12 guidelines diluted their PPV and thus risks over-burdening diagnostic services. Two studies, demonstrated a reduction in PPV from 7.5% to 3.7% and 8.5% to 3.5% respectively following the introduction of NG12 with a concurrent increase in referrals. A large Danish prospective cohort study of 37,455 patients reported the PPV for CRC for the symptoms of abdominal pain, change in stool frequency, change in stool texture and rectal bleeding were 0.3%; 0.4%; 0.2% and 0.6% respectively. For studies included in our review, the use of FIT at a threshold of 10µg/g increased the PPV for CRC for change in bowel habit to between 5.5% and 11.4%,,; rectal bleeding between 8.5% and 30%,, and iron-deficiency anaemia between 21.8% and 41.7%.,,, There is a current lack of data for included individual symptoms at a LoD threshold and at any threshold for other symptoms such as abdominal pain and weight loss.

Whilst we are relatively confident that FIT has sufficient operational sensitivity (irrespective of population or other variables), practical considerations about capacity for investigations as well as unnecessary alarm and investigations for false positive patients might suggest that work is still needed to refine and optimise the criteria used to select patients for testing to increase FIT specificity, at these low thresholds. Multivariable prediction modelling studies may be useful, in this context, to assess the independent predictive value of a “positive” FIT result, in the context of individual symptoms and clinical risk factors. Prediction modelling studies should consider the trade-off between the potential for improved predictive performance and ease of use (the extent to which the components of any risk score developed are readily available to and easily used by clinicians). Furthermore, despite the high sensitivity of FIT for CRC detection, there are still a small number of cancers that will not be detected, and it is therefore essential that appropriate “safety-netting” be in place to refer patients with persistent symptoms and a “negative” FIT. Research into means of optimising FIT sensitivity by repeat testing, or sampling technique may further reduce false negative results.

Summary and conclusions

There is evidence to suggest that FIT can be used at a threshold of 10 µg/g or the LoD as an initial test to triage patients when CRC is suspected irrespective of the presenting symptom cluster to determine the need and urgency for investigations. Within the limitations of the available data, the sensitivity of FIT assays at these thresholds appears to be unaffected by assay (OC-Sensor or HM-JACKarc), study setting or definition of “symptomatic patients”. Although the sensitivity is maximised at LoD, the specificity is relatively low. At a threshold of 10 µg/g, the specificity improves with a slightly lower sensitivity. Clinical services should consider the trade-off between the impact on diagnostic services and potential missed cancer rates when deciding on the most appropriate FIT threshold to use in their clinical setting.

Contributors

Muti Abulafi: Supervision, Conceptualisation, Methodology, Investigation, Writing – Original Draft, Review and Editing, Project Administration, Funding Acquisition. Richard Booth: Investigation, Writing – Original Draft, Review and Editing, Visualisation. Rachel Carten: Investigation, Writing – Review and Editing. Nigel D'Souza: Investigation, Writing – Review and Editing. Marie Westwood: Methodology, Analysis, Review and Editing. Jos Kleijnen: Methodology, Analysis, Review and Editing.

Data sharing statement

We obtained permission from Dr Kai Saw and Professor Ian Bissett, Faculty of Medical and Health Sciences, University of Auckland, New Zealand to use the unique R data code to generate the HSROC curves.

Declaration of interests

We declare no competing interests.