Incorporating faecal haemoglobin measurement using the faecal immunochemical test (FIT) in the referral, triage and prioritisation pathway for patients with colorectal symptoms

Waiting times for colonoscopies are long and risk harm due to the delayed diagnosis of serious gastrointestinal diseases, including colorectal cancer (CRC). We have reported that the New Zealand Ministry of Health, now Manatū Hauora, referral criteria for direct access outpatient colonoscopy or computed tomography colonography (CTC),¹ hereafter the direct access criteria, have a low specificity for CRC.² This low specificity, together with high colorectal symptom burden in the general population contributes to high referral rates and low yield from investigation. Improving access to colorectal investigations for New Zealanders who have significant bowel disease is an immediate priority for gastroenterology and surgical services and should be undertaken to increase the detection of significant diseases, reduce time to diagnosis, and reduce the number of investigations performed with no significant finding.

Data indicates that incorporating faecal haemoglobin (FHb) measurement by using the faecal immunochemical test (FIT) into the new patient investigation pathway may help to satisfy these goals.^3,4 FIT has been extensively investigated in symptomatic populations and has been successfully incorporated into new patient symptomatic pathways,⁵ and used to re-prioritise cases waiting for colonoscopy following pandemic related delays.⁶ Indeed, the British Society of Gastroenterology (BSG) and the Association of Coloproctology of Great Britain and Ireland (ACPGBI) recommend the implementation of FIT as a diagnostic tool for all patients with symptoms or signs of a suspected CRC diagnosis, other than those with an anal or rectal mass or anal ulceration.⁷ Identifying combined FIT/symptom thresholds to direct referral and investigation of cases requires both a reliable estimate of the prior risk of disease for common clinical presentations, and knowledge of the diagnostic accuracy of FIT at the proposed FHb thresholds, neither of which is known absolutely.

We aimed to derive a colorectal symptom pathway incorporating standard referral (direct access criteria), and FIT data, to guide referral, triage, and prioritisation of cases.

Methods

An overview of the study design is shown in Figure 1.

View Figures 1-4, Tables 1-4.

Diagnostic accuracy of a single rule out FIT for colorectal cancer

Search strategy and exclusions are summarised in Figure 2. Studies were included if they reported the diagnostic accuracy of a single rule out FIT (at threshold >10mcg/g or at a threshold of >4mcg/g or lower) for colorectal cancer in a cohort of patients with unexplained colorectal symptoms. Studies reported as full papers, with a complete, patient level dataset (sufficient to allow sensitivity and specificity to be calculated), were considered for inclusion, including those reporting outcomes based on either clinical follow-up (>6 months) or colonic investigation. A quality assessment tool for diagnostic accuracy studies (QUADAS-2) was used to facilitate the assessment of study quality. Studies from the same population were allowed when referral dates did not overlap. Studies were excluded if they did not meet inclusion criteria and furthermore if they included surveillance cases, or where the mode of investigation or follow-up period was deemed insufficient to diagnose incident colorectal cancer. Database search, literature review, quality assessment, decision to include or exclude, and data extraction was made by one author (James Falvey). Patient level data was manually extracted from each included study and grouped according to FIT threshold for meta-analysis. FHb thresholds for analysis were at >10mcg haemoglobin per gram of stool (mcg/g), and at the limit of detection of the test (LoD) (any threshold <4mcg/g). Meta analyses were undertaken using a random effects model due to heterogeneity in study design, and performed using MedCalc for Windows, version 19.4 (MedCalc Software, Ostend, Belgium). The pragmatic approach to the LoD was taken to avoid overestimating sensitivity for CRC, reflect variation in assay sensitivity, and to remain consistent with prior methodology.⁸

The prior risk of CRC according to the direct access criteria has been reported previously (2018 dataset).² Briefly, a retrospective cohort study was performed which collected referral, demographic and outcome data for all first primary care referrals for direct access colorectal investigations made to Canterbury district health board (now Te Whatu Ora – Waitaha Canterbury) using a dedicated electronic referral form (eform) in the year 2018. The eform includes a free text section for clinical history and tick boxes that allow the case history to be summarised with respect to the direct access criteria. General practitioners have access to additional guidance regarding the investigation and referral of cases through an online resource (Community HealthPathways). Faecal occult blood (FOB) testing is not included in the direct access criteria or included as a required field in the eform. Cases were followed for a median of 33 months. One hundred and twenty-eight CRC cases were detected among 3,200 referrals. For the purposes of this study, referrals for patients with suspected Inflammatory bowel disease (IBD) were not included (2 CRC among 214 referrals for suspected IBD [0.9%]).

Likelihood ratios (LR) derived from summary accuracy data were used to calculate disease prevalence for clinical groups following FIT and were unadjusted. Simple proportions were converted to odds (and vice versa) as required and are presented as percentages or number needed to investigate (NNI) or decline (NND) to detect or miss one CRC. Detection of high-risk adenoma is estimated based on the prevalence of advanced polyps in the 2018 dataset, and using LRs derived from published data for advanced adenomas.⁹¬ FIT positivity rates at each threshold were calculated as follows: n=([C – cbP]/[c{{a}} – c{{b}}])/P, where n is the proportion of cases with a test result at or above the threshold, C is the total number of cases of CRC in a population (P), and c{{a}} and c{{b}} are the prevalence of CRC for cases with test results above and below the threshold, respectively. To determine the investigative resource requirements of the pathway, the secondary care decision aid has been followed with additional assumptions for categories requiring triagers’ discretion as follows:

• iron deficiency anaemia (IDA) 80% colonoscopy and 20% CTC.

• rectal bleeding (RB) 40–49 years/detectable FHb <10mcg/g, 50% colonoscopy and 50% CTC.

• RB <39 years/FHb >10mcg/g, 100% colonoscopy.

• altered bowel habit (ABH) >50 years/detectable FHb <10mcg/g, 50% colonoscopy and 50% CTC.

• other presentations FHb >10mcg/g, 100% colonoscopy, detectable FHb <10mcg/g, 50% CTC and 50% decline.

Modelling does not make allowance for any change in primary care referral practice, or the effect of expanding access criteria in the proposed pathway (to younger patients with rectal bleeding, or due to lowering age thresholds for Māori and Pacific people). The proportion of cases meeting criteria for colonoscopy, but who in usual clinical practice would be offered an alternative mode of investigation due to age-related frailty or the presence of significant comorbidity has not been estimated for either the current proposal or the direct access criteria, or the ACPGBI/BSG guideline for urgent colonoscopy. Outcome estimates of sensitivity, specificity, NNI, and NND were determined for each of these criteria. The upper 95% CI of negative likelihood ratios (NLR) derived from summary data were used to determine the worst-case missed cancer rates for rule out thresholds. Ninety-five percent CI were calculated by the binomial exact method.

CRC incidence by novel symptom criteria was determined from the 2018 dataset by mapping the direct access criteria to the novel criteria. Symptom thresholds for referral are unchanged from the direct access criteria (e.g., altered bowel habit [ABH] refers to looser and/or more frequent stools, and unexplained rectal bleeding [RB] refers to cases where benign anal causes have been treated or excluded).

Results

Meta-analysis

Eighteen studies were included in meta-analysis. The study characteristics, quality assessment, and diagnostic accuracy of included studies are shown in Table 1. One study was excluded as it did not report FIT accuracy data at a threshold consistent with our analysis.²⁵ Of those studies excluded on methodological or other grounds despite apparently meeting inclusion criteria, one provided insufficient information regarding the reference standard and had too short a period of clinical follow-up,²⁶ one was deemed at high risk of bias in both case selection and case follow-up,²⁷ four had incomplete investigation or follow-up of cases (usually FIT negative),^28–31 one did not contain patient level data,³² and one was excluded due to multiple samples counted as a single positive if any gave an above threshold result.³³ Forrest plots for the primary analyses are shown in Figure 3. Summary sensitivity and specificity of FIT at FHb threshold >10mcg/g were 89.0% (95%CI 87.0–90.9%)(I² 33.14%) and 80.1% (95%CI 77.7–82.4%)(I² 98.2%) respectively, and at the LoD were 95.7% (95%CI 93.2–97.7%)(I² 58.84%) and 60.5% (95%CI 53.8–67.0%)(I² 99.4%), respectively. Correspondingly, the NLR of FIT for CRC at thresholds of >10mcg/g and at LoD were 0.14 (95%CI 0.12–0.16) and 0.07 (95%CI 0.04–0.11), respectively. There was significant heterogeneity between studies. The source of this was investigated by subgroup analysis according to the following study characteristics: cohort date (pre vs not pre-2017), retrospective vs prospective data collection, recruitment location (primary or secondary care), analyser (HM-jack arc, OC-sensor, other/unknown), colorectal cancer prevalence (>3% vs <3%), and reference standard (colonoscopy only,^{9,10,12,15,16} any colonic investigation,^{3,11,14,17,19,24} follow-up^{4,13,18,20–23})(see Table 1 for study characteristics). Significant heterogeneity was still identified within the subgroups for both sensitivity and specificity, and the estimates between subgroups did not differ significantly; however, the limited sample size limits the robustness with which these effects can be explored. Outliers were sought with respect to study design, prevalence, sensitivity, and specificity; however, exclusion of individual studies did not significantly influence results.

Canterbury colorectal symptom pathway

The proposed pathway is summarised in Figure 4. For Symptom/FHb categories where the risk of CRC is low, triagers will use discretion in determining the most appropriate outcome (e.g., CTC, flexible sigmoidoscopy, outpatient review, or a further period of observation in primary care) based on age, case presentation, co-morbidity, and local resource availability. The outcome in the secondary care decision aid (Figure 4b) for such categories is denoted ‘triagers discretion’, and for simplicity in modelling, it is assumed that all cases will undergo either colonoscopy or CTC (see methods). Estimated CRC prevalence by clinical category, age, and FHb threshold are shown in Table 2. Pathway sensitivity and specificity are shown in Table 3, along with those for the direct access criteria, and those of the ACPGBI/BSG guidance for urgent colonoscopy. Within the limits of the analysis, the sensitivity for advanced polyps is estimated to fall from 84.2% for the direct access criteria to 70.4% for the Canterbury pathway.

Discussion

Our study demonstrates how incorporating FIT in the investigation, referral, and prioritisation of patients with colorectal symptoms may both improve sensitivity for colorectal cancer, while simultaneously reducing the number of investigations performed. Indeed, including FHb measurement to guide patient care was previously an established strategy in New Zealand. In Canterbury, between 2010–2017, a qualitative FIT was incorporated in the colorectal symptom assessment and referral pathway and provided the strongest single predictor for colorectal cancer diagnosis, above anaemia and rectal bleeding.³⁴ Thereafter, while Canterbury moved away from FHb and adopted the direct access criteria (based on the 2005 NICE guideline, CG27), the United Kingdom sought to increase sensitivity for CRC (NICE NG12) by incorporating a rule-in guaiac based FOB (gFOB) for primary care patients with low risk symptoms (CRC risk of <3%).³⁵ In the UK the low specificity of gFOB led to increased demand for colorectal investigation and a higher NNI to detect one cancer,³⁶ and this was addressed by replacing gFOB with FIT >10mcg/g (NICE DG30) in 2017.³³ The discriminatory value of a quantitative FIT for CRC, and its validity beyond population screening and low risk symptoms to high risk scenarios such as rectal bleeding have subsequently been confirmed.^11,12,37,38 This reflects a broader concept, that the discriminating power of FIT for CRC is determinate (notwithstanding variation in tumour size, location, biology, and stool sampling method), while its clinical utility varies by FHb threshold and the pre-test probability of disease.

A major strength of this study is that our conclusions are based on actual cancer rates in our referral population, and a current estimate of FIT accuracy. Furthermore, because we are primarily concerned with optimising the clinical pathway sensitivity and understanding the risk of declining investigation, we have estimated the worst-case scenario miss rates for each clinical presentation using the upper 95^th CI of the NLR calculated from our summary sensitivity and specificity, which encompass the least favourable estimate of the diagnostic accuracy of rule out FIT found in contemporary meta-analysis.^8,39,40 Table 4 shows that this is important because there was significant heterogeneity between studies included in meta-analysis, and this impacts on the accuracy with which the diagnostic accuracy of FIT can be estimated.

We estimate the missed cancer rate for the Canterbury pathway to be 2.9%, compared with 9.5% for the direct access criteria (Table 3). In our proposal, cases meeting symptom threshold who have FHb >10mcg/g undergo investigation with colonoscopy. However, because almost 10% of CRC are missed at this threshold, we further recommend that all cases with detectable FHb <10mcg/g be referred and investigated appropriately given case characteristics and resource availability. Although the CRC rate of those with detectable FHb<10mcg/g is just 1.47%, we see benefit in a pathway that provides definitive care at the first contact, reducing the risk of frequent repeat testing (and associated false positive tests) and the inequity that is likely to result from such an approach. A higher rule out threshold of >10mcg/g would reduce colonoscopy volumes further; however, the pathway as it stands has the potential to lower demand for colonoscopy by 47%. More restrictive criteria may not only delay diagnosis for some cases of colorectal cancer, but also reduce sensitivity of the pathway for other significant colorectal disease.^9,41–43

Simplifying the clinical categories when compared with the direct access criteria is justified on several grounds. Foremost, the distinction between urgent and non-urgent categories in the direct access criteria appears arbitrary with some non-urgent categories having greater risk of CRC than others afforded urgent investigation.² Thereafter, as New Zealand data have repeatedly shown that CRC risk falls in the order IDA>RB>ABH,^2,34,44 and because FIT discriminates CRC risk with greater power than any of these,¹² it follows that all cases be stratified by FIT, and usability of the resulting criteria is enhanced by simplifying the clinical component.

We see benefit in amending the access criteria in several other ways. Lowering age thresholds for investigating Māori and Pacific people by 10 years reflects disease risk and survival outcomes for these peoples,⁴⁵,46and aims to align the rate of investigation of these peoples with the higher rate of investigation of NZ Europeans found in the 2018 dataset.² The change also aligns the colorectal pathway with the Canterbury upper gastrointestinal pathway, which has lower age thresholds for at risk populations, and is consistent with the recent extension of age criteria for Māori and Pacific people in the National Bowel Screening Program (NBSP). Meanwhile, recommending FHb measurement for patients below the age of 50 years who have unexplained RB addresses concern regarding the increasing incidence of CRC in the young,⁴⁷ and brings order to the current ad hoc approach for this patient group, the higher rule out threshold (>10mcg/g) being justified by the low prior risk.

Primary sector engagement indicates a strong preference for FHb testing in primary care supported by comprehensive education and online resources (in Canterbury via Community HealthPathways) for both test interpretation and to guide primary care-based management of colorectal symptoms. FIT request and interpretation in primary care has several benefits: promoting decision making by a single physician aware of the entire patient history, optimising sample return through explanation of the investigative process and the immediate provision of standardised collection device and requisition form for all faecal tests, and allowing follow-up in primary care for cases not returning samples using community-based staff. Furthermore, FHb testing in primary care avoids unnecessary case referral, saving time in primary and secondary care, and facilitating rapid clinical decision making.

There are few data regarding accuracy of FIT for most non-malignant colorectal conditions other than IBD and high-risk adenomas. FIT has good diagnostic accuracy for colonic IBD,⁴² and is likely to have utility in the diagnosis of other bleeding pathologies such as drug induced, ischaemic, or diverticular colitis, colonic angiodysplastic bleeding and ulcerative conditions such as stercoral ulceration or rectal ulcer syndrome. However, there is no expectation that FIT would be useful in the diagnosis of microscopic colitis or other non-bleeding pathologies, and FIT is unreliable in the diagnosis of proximal gastrointestinal bleeding and small bowel Crohn’s disease.^48,49 To ensure that a FIT based colorectal investigation pathway does not compromise the diagnosis of either malignant, or non-malignant colorectal disease, we have started with the lowest rule out threshold and propose to develop the pathway iteratively in response to prospective data.

FIT is currently being used in Canterbury to re-prioritise cases awaiting non-urgent new patient colonoscopy, and outcome data from the project will be reported in due course. Subsequently, the safe implementation of a pathway for new patients is dependent on robust primary sector engagement, education, and strong governance. Several future scenarios are conceivable. Future data may show that FIT has greater accuracy than estimated in the present study. In that situation, it would be appropriate to adopt a higher rule out threshold, retaining excellent sensitivity for CRC with lower rates of investigation. The current modelling assumes no change in GP referral practice; however, it is likely that GPs will have a lower threshold for investigating with FIT than they currently have for referring for invasive investigation. To maintain a high yield from invasive investigation in this situation, it may be necessary to increase the FIT threshold at which cases are accepted. Effort should be made to avoid this scenario, and the extreme case of surrogate screening, by emphasising the importance of symptom threshold for testing, as the higher the rule out FIT threshold is set, the less reassurance an individual symptomatic patient will receive from a ‘negative’ test. Reassuringly, where data are available, FIT testing rates have been shown to stabilise over time, suggesting that surrogate screening is unlikely to be widespread.^13,33

Comparison of the current proposal with the new UK guideline is pertinent. Neither pathway recommends FIT in cases with anorectal lesions, and both recommend FIT in primary care. Thereafter, the pathways diverge with the ACPGBI/BSG recommending urgent referral for symptomatic cases with FHb >10mcg/g, while the Canterbury pathway follows a graduated approach, both to urgency and mode of investigation dependent on FHb concentration and case presentation. The ACPGBI/BSG delegates decisions regarding referral of cases with FHb<10mcg to the discretion of the referring doctor, while the Canterbury pathway anticipates accepting cases for non-urgent investigation when the FHb is detectable above the LoD. Referral of cases below this threshold, made due to enduring concern, would be judged on their merit. To derive future iterations of the pathway, and ultimately a national solution for FIT in symptomatic cases, the merits of resource distribution between various facets of colonoscopy activity must be considered, including for symptomatic, screening, and surveillance cases. According to our dataset, the number needed to investigate to detect one CRC in the current direct access criteria is 21 (90.5% sensitive), compared with 12 for the current proposal (97% sensitive), 14 in the NBSP (assuming 7% CRC detection at colonoscopy), and 6 in the latest UK guideline (sensitivity 90%). Further discussion on this point is important but beyond the current work.

Much is beyond the scope of this study. We have neither performed economic analysis, nor detailed the complex processes required to ensure patient engagement and equity of outcome for population groups. Neither have we sought to resolve all clinical scenarios. For example, given the increasing risk of CRC with age, is there an upper age threshold beyond which all cases should be investigated irrespective of FHb result? How long should a persistently symptomatic patient be observed and managed in primary care before repeat FHb testing, and how should a repeat FIT result be interpreted? Neither have we fully addressed the investigation of colorectal symptoms in younger cases where IBD is the more common diagnosis, nor how to approach a case at risk of both CRC and IBD.

Improving access to colonoscopy for patients at risk of serious disease is an immediate concern for New Zealand. Formally incorporating FHb measurement into the assessment, referral, and prioritisation of colorectal symptoms appears achievable and should enable a high sensitivity for colorectal cancer, while also reducing the number of colonoscopies performed with no significant finding. This will expedite the investigation of those at higher risk, as colonoscopies can be undertaken more rapidly in this group. Robust primary and secondary sector education, community collaboration, development of strategies to ensure equity, research, prospective data gathering, analysis and feedback, are all essential for the initial and future success of the pathway.