Faecal immunochemical testing for adults with symptoms of colorectal cancer attending English primary care: a retrospective cohort study of 14 487 consecutive test requests

Faecal immunochemical testing (FIT) is recommended by the National Institute for Health and Care Excellence (NICE) to triage symptomatic primary care patients for further investigation of colorectal cancer.


| INTRODUC TI ON
Since July 2017 the faecal immunochemical test (FIT) has been recommended by the National Institute for Health and Care Excellence (NICE) DG30 guidelines 'to guide referral for suspected colorectal cancer in people without rectal bleeding who have unexplained symptoms but do not meet the criteria for a suspected cancer pathway'. 1 Guidance on faecal testing has generated significant debate. 2,3 Concerns have been raised about delayed cancer diagnosis due to false negatives and the potential to increase demand on already stretched endoscopy service due to false positives. [4][5][6] FIT is an immunoassay-based method that measures the globin component of human haemoglobin and its early degradation products. 7,8 FIT does not require dietary restriction, is specific to lower gastrointestinal (GI) cancers as upper GI enzymes degrade human globin, and is less affected by concomitant medication use than the guaiac method to detect faecal occult blood (gFOB). 7,9 Accordingly, three times fewer false-positive tests are reported when FIT is compared to gFOB in samples sent to the laboratory for symptomatic patients. 10 The adoption of FIT in primary care has been slow with notable variation in uptake and implementation across England, 5 low awareness of the NICE guidance 11 and low confidence in the accuracy of FIT amongst general practitioners (GPs). 12 The public report a preference for FIT over colonoscopy if no additional cancers are missed. 13 NICE DG30 recommends a quantitative FIT threshold of 10 µg Hb/g faeces should trigger referral for colonoscopy. 1 The NHS England Bowel Cancer Screening Programme (BCSP) currently uses a threshold of 120 µg Hb/g faeces. Colorectal cancer screening studies have consistently demonstrated improved performance of FIT compared to gFOB at a high analytical threshold. 9,14,15 However, NICE expressed concerns about the applicability of all ten of the studies used to underpin their DG30 recommendation: none reported data on patients with low-risk symptoms of colorectal cancer and only one study 16 was conducted in primary care. 1,17 Evidence has since been published in support the use of low-threshold FIT in symptomatic primary care patients. 5,18-25 A Danish study concluded that FIT may be used as a supplementary diagnostic test in individuals with non-alarm symptoms of colorectal cancer to diagnose serious bowel disease (colorectal cancer, high-risk adenoma and inflammatory bowel disease). 24 Authors of a Scottish study strongly recommended FIT should become integral to the assessment of all patients presenting to primary care with new bowel symptoms, to objectively determine the risk of underlying serious bowel disease. 20 Despite these important studies, data from symptomatic primary care patients prior to referral are lacking. 6 Unanswered questions remain: does FIT perform similarly in men and women, across age-groups and what is the optimal FIT threshold to detect colorectal cancer and significant lower GI disease? 6,26 The COVID-19 pandemic has introduced a new urgency to identify non-invasive approaches to triage for patients with symptoms of serious colorectal disease requiring further investigation. The large backlog of endoscopy created by the COVID-19 pandemic means that low cut-off FIT will be required to risk stratify patients referred with possible cancer symptoms into groups for urgent and less urgent endoscopy immediately after the pandemic has passed.
Understanding the performance of FIT at a range of thresholds in symptomatic patients is therefore a priority.
The Oxford University Hospitals Trust (OUH) adopted FIT prior to the DG30 guidance to comply with the 2015 NG12 NICE guidance for suspected cancer. 6 This coincided with a desire from the clinical laboratory to move away from gFOB. FIT was commissioned by Oxfordshire Clinical Commissioning Group (OCCG) as a direct access test for GPs in 2016. 5 We conducted a diagnostic accuracy study using linked electronic hospital records data to ascertain the diagnostic performance of FIT to detect serious bowel disease in the context of the NICE DG30 guidelines in England. We considered the diagnostic performance of FIT by age-group, gender and FIT threshold, and documented FIT negative cases of colorectal cancer. This study was registered as a service evaluation on the OUH Datix register (CSS-BIO-3 4730). We followed STARD reporting guidelines. 27 Leading up to the study period, the change in NICE guidance and the indications for FIT testing were communicated to GPs in Oxfordshire by email and newsletter from the OCCG. Samples were collected into standard collection pots by patients in primary care and analysed for FIT using the HM-JACKarc analyser (Hitachi Chemical Diagnostics Systems Co., Ltd) a method that has been independently evaluated with respect to analytical performance 16 and is recommended in the context of use for samples from primary care. 1 The method had a calibration range of 7-450 µg Hb/g faeces and immunoassay reproducibility, assessed across 12 months was between 4.5% and 8.7% when expressed as a percentage coefficient of variation. Sample preparation prior to analysis on the FIT instrument utilized the Extel Hemo-Auto MC device, a process which introduced additional variation, with overall analytical imprecision observed to be between 7.0% and 13.5% when specimens had been homogenised and sampled by laboratory staff. The selection of the faecal Hb concentration considered positive was made before the NICE recommendation to use 10 µg Hb/g faeces and was based on the methods lowest calibrant value of 7 µg Hb/g faeces and agreed with the OCCG based on initial method verification data. 10 Results were reported electronically to the requesting GP as either positive or negative. In selecting the approach to faecal sample handling we balanced two competing pre-analytical sources of error: the requirements to minimize sampling errors if undertaken by the patient, which may give rise to false negatives if the collection device was inadequately filled; and specimen degradation concerns if sampling is undertaken in the laboratory due to delays between specimen collection and stabilisation in the collection device buffer. We have highlighted the balancing of these risks in our contribution to the NICE DG30 FIT adoption resource. 1 Where more than one sample result was available for any individual patient, any positive result within those samples tested was considered a positive outcome on the basis that a single positive would trigger referral.

| ME THODS
To confirm the presence or absence of disease, OUHT clinical and diagnostic databases were searched for evidence of cellular pathology for up to 36 months following the FIT test for all patients.

| Clinical details and outcomes
FIT requests included a range of clinical features (symptoms, signs, test results) that gave rise to a concern about serious colorectal pathology. Patients commonly presented with combinations of clinical features (Table 1) (Table 1). Figure 1 shows the FIT values according to outcome group and gender.

| Trade-offs per 1000 patients tested
If all patients were referred and received definitive testing for colorectal cancer, using a FIT threshold of ≥10 µg Hb/g faeces would lead to seven patients without cancer referred for each person referred with cancer (  Table 6 details cancers diagnosed following a FIT result less than the current DG30 threshold of 10 µg Hb/g faeces. Nine cancers (7.8%, 9/115) were diagnosed within 6 months of a negative FIT (all within 3 months) Five of these diagnoses followed a change in bowel habit, four patients had anaemia, two had unexpected weight loss and two had abdominal pain. Seven of these cancers were located in the rectum or sigmoid colon, and three presented with obstruction.

| Sensitivity analysis
The sensitivity analysis investigating the effect of alternative followup periods of 3, 6 and 12 months showed no change in diagnostic performance for colorectal cancer detection using FIT ≥10 µg Hb/g faeces (Table 7).

| D ISCUSS I ON
We report a large in-context retrospective cohort study to ascertain the diagnostic performance of FIT used to investigate adult patients presenting to primary care in England with clinical features of colorectal disease. The sensitivity and specificity of FIT were both 91% for colorectal cancer using a FIT cut-off of ≥10 µg Hb/g faeces, the threshold currently recommended by NICE. If investigated further, one in ten patients with a positive FIT would be diagnosed with colorectal cancer. Using a FIT cut-off of ≥10 µg Hb/g faeces, one person with colorectal cancer would not be referred for further investigation for every one thousand patients tested. Selecting a higher threshold would mean that fewer patients without cancer would be eligible for referral for every person with cancer referred but a greater proportion of patients not referred for further investigation would have cancer. For serious colorectal disease, the specificity and PPV of FIT were higher and the sensitivity and NPV were lower than for colorectal cancer.

| Strengths and limitations
We present a large retrospective cohort study of FIT testing of English primary care patients to guide referral for colorectal cancer in the context of NICE DG30 guidelines. We ensured the population studied was relevant to DG30 by only including FIT requests originating in primary care following GP education on the use of FIT in this context. We have assumed that GPs have followed the guidance communicated to them about using FIT to triage 'low-risk' symptomatic patients. Although 'high-risk' symptoms qualifying for urgent colonoscopy were noted in the clinical details, such as weight loss or anaemia, it can be assumed that GPs assessed these cases to be lower risk and not to qualify for fast-track referral and that GPs required additional information to guide their management. This is consistent with previous literature recognising that symptoms and signs of disease form a 'symptom continuum' and that interpretation of alarm symptoms vary between GPs. 24 As a retrospective cohort study using routinely collected electronic hospital records data, the majority of patients did not have a gold standard investigation at the time of the FIT test to confirm whether serious colorectal disease was present or absent. All patients were therefore followed up for evidence of subsequent pathology in hospital clinical, laboratory, radiology, endoscopy and pathology database between 6 and 36 months after initial.
A potential criticism of this study is the veracity of the follow-up interval chosen. Whilst our primary analysis used 6 months of follow-up, previous studies have used a shorter minimum follow-up period of 3 months reporting similar results. 24 Changing the follow-up period to 3 and 12 months made no difference in our sensitivity analysis. If cancer was the cause of symptoms that prompted a FIT test then these symptoms will persist, evolve and worsen in the following months (even if the FIT test was to be a false negative). Premalignant lesions detected at colonoscopy triggered by a positive FIT would not have caused the presenting symptoms and are therefore likely to represent incidental findings, though their management will contribute to cancer prevention in the years to come. Preventing these cancers might therefore be regarded the result of opportunistic screening for asymptomatic lesions in symptomatic people. A longer follow-up period would therefore TA B L E 2 Area under the curve (AUC) for FIT for a diagnosis of cancer or serious colorectal disease by gender and age-group Specimen preparation is a critical step in the overall performance of the FIT method. 28,29 The methodological approach used in the present study involved collection of faeces by the patient into a standard sample pot which was then sampled into the FIT collection device by trained staff in the laboratory. This approach was adopted due to local concerns that patients may incorrectly use the collec-

ACK N OWLED G EM ENTS
We