Combining faecal immunochemical testing with blood test results for colorectal cancer risk stratification: a consecutive cohort of 16,604 patients presenting to primary care

Three approaches were investigated to optimise FIT.

These cut-offs were based on the assays Limit of Detection (2 μg Hb/g faeces) and the cut-off recommended by NICE for primary care triage in 2017 (10 μg Hb/g faeces) and align with existing research [2, 9].

A test was considered positive if patients fell above the cut-off value for FIT (2 or 10 μg Hb/g faeces) and had an abnormal blood test result. The threshold for abnormal blood tests were pre-specified based on standard clinical practice [19].

Logistic regression was used to generate predicted probabilities of colorectal cancer. Backward stepwise selection was used to select covariates. Because serum ferritin and CRP were only available for a subset of cases, stepwise selection was conducted on an imputed dataset with 10 replicates using predictive mean matching. In models where CRP or serum ferritin were retained, coefficients for each variable in the imputed and complete case datasets were compared, and if similar, the model results from the complete case dataset were reported. The three modelling approaches are defined below.

The restricted cubic spline function for FIT was specified to have knots at 2, 10, 50, and 100 μg Hb/g faeces. Four knots were selected to yield a model with at least 20 events per variable, to minimise optimism bias [20]. Ninety-five percent confidence intervals were estimated using the Wilson Score method [21]. The positive predictive value (PPV) and negative predictive value (NPV) were additionally expressed as the number of positive FITs to detect one cancer (number needed to scope) and the cancer miss rate per 10,000 negative tests. To permit a comparison of model performance, the probability cut-off to determine a positive result was selected to match the sensitivity of the FIT alone at a cut-off of 10 μg Hb/g faeces.

A total of 16,604 of 18,656 available FITs (89%) were included in the study. Included patients were representative of the overall sample (Table 1). Study subjects had a median age of 61 and were 58% female. One-hundred thirty-nine (139) cancers were diagnosed within 6 months of the FIT test (0.8%). Patients who were diagnosed with cancer were older (median age 72) and more likely to be male (60%), to have a FIT ≥10 μg Hb/g faeces, and/or to have abnormal blood tests (Table 1, Fig. 2, Additional file 1: Table S1).

For 90% of included patients, the free text in the electronic FIT request mapped onto at least one of the pre-specified 11 clinical indication categories. The most common of these was change in bowel habit (40%), then anaemia (26%) and abdominal pain (18%, Table 1). The most common clinical indications in people with cancer were anaemia (35%), change in bowel habit (32%), blood in stool (16%), and abdominal pain (17%).

Low haemoglobin was the most common abnormal blood test result (31% of all patients, 52% of those with a subsequent cancer diagnosis, Table 1) followed by low MCH (16% and 34%, respectively).

In this large cohort of patients tested with FIT in primary care, neither age, nor blood test results remained strong enough predictors of colorectal cancer to improve on the performance of FIT. While the number of false positives could be reduced by taking into account blood tests, the large associated increase in false negatives outweighed the benefit. In addition, there was no evidence to suggest that using clinical indication as a rule-out or rule-in factor would improve the efficiency of FIT triage. The lack of an apparent age-effect after taking into account FIT suggests that age-specific thresholds for FIT positivity would not improve test performance.

This is the largest cohort of primary care patients tested with FIT in the UK. The sample comprises patients reflecting true clinical practice which involves uptake of the DG30 guideline over time and clinical judgement. The study also comprises tests prior to and during the COVID-19 pandemic and suggests that the performance of FIT in primary care has remained stable throughout. Centralised FIT and blood testing at the the OUH Clinical Biochemistry Laboratory allowed for highly complete assessment of FIT and blood test values. We accessed the referral text to explore the performance of FIT in strata defined by GP-reported symptoms. The prevalence of and type of symptoms reported may have differed if we had accessed primary care records or asked patients to report their symptoms directly [23]. A 6-month follow-up period was used for the primary analysis to optimise the number of cancers included, but in sensitivity analyses that aligned with national guideline-setting, a 12-month follow-up showed similar results.

With respect to limitations, the gold standard would have been to have every patient undergo a colonoscopy after FIT. Due to the observational nature of this study, we instead used hospital-based records to determine outcomes after FIT. This may have resulted in some underestimation of disease. However, by linking multiple local data sources for patients tested in a single central laboratory in a clearly defined geographical catchment area we increased the likelihood that serious disease diagnosed during the study period was captured. It is plausible that some patients may have been lost to follow-up by moving out of area or by dying, but this is unlikely to have had a meaningful impact on our findings, as we used a short follow-up period and there is minimal patient movement between localities, particularly during investigation. An alternative approach would have been linkage to Public Health England’s National Cancer Registration and Analysis Service (NCRAS). While this may have provided a more complete accounting of outcomes, it would have restricted the timeliness of our analyses as cancer registry data is currently available only up until the end of 2018. Taking into account the follow-up period, this would have limited us to fewer than 4000 FITs for inclusion.

Another potential limitation is that the sample was restricted to individuals for whom blood test results were available within a 90-day window surrounding FIT, but this excluded only 10% of the sample. As no predictive model was identified that performed better than FIT alone, neither internal validation nor optimism correction were pursued.

Recent studies have reported FIT to be an effective tool to triage for “high-risk” patients referred for definitive investigation [9, 13]. Our findings contribute to a growing literature suggesting that FIT also performs well in the “lower risk” primary care setting [6, 7, 16, 24, 25]. This is the one of few studies to formally and systematically evaluate blood tests in addition to FIT in symptomatic patients, and one of few to analyse FIT supplemented with other variables. The f-Hb, age, and sex test score (FAST) was not superior to FIT alone in the primary care setting [12]. COLONPREDICT included FIT, age, sex, rectal bleeding, benign anorectal lesions, rectal mass, serum carcinoembryonic antigen, blood haemoglobin, colonoscopy in the last 10 years, and treatment with aspirin. COLONPREDICT was derived and validated in a higher-risk referred population [26] and at a threshold equivalent to approximately 90% sensitivity (f-Hb≥20 μg Hb/g faeces and COLONPREDICT score≥5.6) had a specificity of 78.7% compared to 69.6% in FIT alone [27].

A UK-based study of whether demographic, lifestyle (e.g., smoking, physical activity), or clinical factors (family history, symptoms) could add to the predictive value of FIT found that only family history of polyps showed a significant association once FIT was taken into account [28]. In the current study, family history was not retained in stepwise models; however, the indicator was based on referral notes whereas in the aforementioned study, patients were prospectively asked about family history.

FIT is a sensitive and specific test and as such can serve as a valuable rule-in and rule-out test for patients presenting to primary care. However, it remains worthwhile to investigate strategies to further enhance the sensitivity and specificity of FIT and guide prioritisation of FIT-positive patients for immediate colonoscopy. Risk stratification tools that have been developed in the screening setting incorporating polygenic risk scores [29‐31], urinary volatile organic compounds [10, 32], and circulating and/or faecal tumour DNA [33] could be explored to complement FIT for triage of primary care patients.

Currently, there is limited evidence to support the use of repeat FIT testing to select initially FIT-negative patients for referral or to reassure about non-referral. Based on exploratory analyses, 1113 patients in our study had second FITs also meeting inclusion criteria. In that group, there were 6 cancers, and no false negatives at a threshold of 2 or 10. Of the 1007 patients without cancer, forty-four (4.3%) patients had initially negative FITs followed by false-positive FITs at a threshold of 10, and 942 (93.5%) participants without cancer had two negative FITs. Further research is needed to inform the timing, interpretation, and utility of repeat FIT testing for triage in symptomatic patients attending primary care.

Both FIT and the “gold standard” colonoscopy result in false negatives [34]. In this study, no practical rules using blood tests or clinical indication to reduce false negatives were apparent. To reduce the likelihood of false FIT-negative results, future research may benefit from an agnostic approach to building the prediction model. For example, new predictive markers could be discovered by applying machine learning models to large, representative databases of electronic health records [35].