Abstract
Objective To develop and validate a model using available information at the time of Faecal Immunochemical testing (FIT) in primary care to improve selection of symptomatic patients for colorectal cancer (CRC) investigations.
Design Population based cohort study. Setting All adults ≥ 18 years of age referred to Nottingham University Hospitals NHS Trust between 2018 and 2022 with symptoms of suspected CRC who had a FIT.
Participants The derivation cohort (Nov/2017-Nov/2021) included 34,435 patients with FIT results who had 533 (1.5%) CRCs at 1-year. The validation analysis included 34,231 patients with first FITs in the derivation cohort with 516 (1.5%) cancers, and 16,735 patients with first FITs in the validation cohort with 206 (1.2%) cancers.
Main outcome measures Predicted 1-year CRC diagnosis using Cox proportional hazards modelling with selected multiple fractional polynomial transformations for age, faecal haemoglobin concentration (f-Hb) value, mean corpuscular volume (MCV), platelet count and sex. In the internal-external validation we calculated discrimination and calibration to assess performance and estimated net benefit values across a range of CRC risk thresholds to assess clinical utility.
Results In the survival model multiple fractional polynomial transformations were selected for age, f-Hb and platelet count, with MCV included as a linear variable and sex as a binary variable. Haemoglobin was not selected. At a CRC risk threshold of 0.6% (equivalent to f-Hb=10 µgHb/g (µg/g)) overall performance of the validated model across age strata using Harrell’s C index was ≥ 0.91% (overall C-statistic 93%, 95% CI 92%-95%) with acceptable calibration. Using this model would yield similar numbers of detected and missed cancers but require 20% fewer investigations than a f-Hb ≥10 µg/g strategy. For approximately 100,000 people per year with symptoms of suspected CRC, we predict it might save >10,000 colonoscopies with no evidence that more cancers would be missed if we used our model to triage investigations compared to using FIT at the currently recommend level for referral.
Conclusions Including age, sex, MCV, platelets and f-Hb in a survival analysis model to predict the risk of CRC yields greater diagnostic utility than a simple binary cut off f-Hb≥10 µg/g. Enacting model-based triage of a symptomatic CRC pathway may decrease the burden on endoscopy whilst maintaining diagnostic accuracy. Further targeted validation of this approach is required in external populations with symptoms of possible CRC.
Transparency statement The lead author and manuscript’s guarantor (CJC) affirms that the manuscript is an honest, accurate and transparent account of the study being reported; than no important aspects of the study have been omitted: and that any discrepancies from the study as originally planned have been explained.
Role of the funding source This project was funded by the National Institute for Health and Care Research (NIHR) [Health Technology Assessment (HTA) Programme (Project number 133852); awarded to CJR, WH & LS] and will be published in full in the HTA journal. Further information is available at: [https://fundingawards.nihr.ac.uk/award/NIHR133852]. The views expressed are those of the authors and not necessarily those of the NIHR or Department of Health and Social Care and sponsored by Nottingham University Hospitals NHS Trust. The funder and sponsor had no role in the study design, in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. We confirm the independence of researchers from funders and that all authors, external and internal, had full access to all of the statistical reports and tables in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. SERB was supported by an NIHR Advanced Fellowship while undertaking this work (NIHR301666) and received additional support from the Higgins family. BDN was supported by a National Institute of Health Research Academic Clinical Lectureship and a CRUK Research Careers Committee Postdoctoral Fellowship (RCCPDF\100005).
Ethics approval statement HRA and Health and Care Research Wales (HCRW) approval was given for this study - IRAS project ID: 312362; Protocol number: 22ON007; REC reference: 22/HRA/2125; Sponsor: Nottingham University Hospitals NHS Trust.
Introduction
Colorectal cancer (CRC) is the 4th most common cancer in the United Kingdom and is the 2nd largest cause of cancer death1. Bowel cancer screening programmes reduce mortality but takes a long time to have an effect. Currently, screening only accounts for approximately 10% of CRC diagnoses – the majority occur through symptomatic patients being referred with a suspicion of cancer through a variety of pathways2,3. Additionally, 20% of CRCs present as an emergency4. The “risk threshold” for urgent referrals for investigation was set at a 3% positive predictive value of cancer by the English National Institute for Health and Care Excellence (NICE), i.e. those referred by primary care on a cancer diagnostic pathway should have a risk of a specific cancer of 3% or more5.
Faecal Immunochemical Testing for haemoglobin (Hb) (FIT) identifies haemoglobin in faeces as an indicator of possible CRC. This is the approach used in the English asymptomatic population-based bowel cancer screening programme (at a higher threshold than used in symptomatic patients). FIT for symptomatic patients was recommended by NICE in 2023 with a FIT cut-off of ≥10 µgHb/g (µg/g) set for this purpose6. However, current demand for colonoscopy or computed tomography colonography (CT-colonography) capacity in the UK and many other countries far outstrips capcity7,8. This imbalance between demand for investigation at f-Hb ≥10 µg/g and supply of colonoscopy or CT-colonography means that investigations will be delayed for some people at higher risk of CRC whilst many normal investigations are being performed. This situation has been worsened by the COVID-19 pandemic9.
Recent evidence from Nottingham suggests that stratifying by age and the presence or absence of anaemia could identify those people with a FIT ≥10 µg/g at a low CRC risk, well below the defined 3%, who do not need investigation10. While stratified approaches could work, an alternative approach, is to use a clinical prediction model to estimate risk of CRC at an individual level to tailor investigation. Such an approach should maintain diagnostic performance whilst decreasing the burden on diagnostic services by reducing the number of colonoscopies and/or CTC’s performed. Information from the patient i.e. f-Hb level, age, sex and blood indices could be used to inform whom to investigate based on their predicted risk of CRC. Such predictions could inform all stakeholders (patients, general practitioners, policy makers) as to who could either safely avoid investigation or have it routinely (and be reassured that the risk of CRC is low) whilst prioritising those with the highest risks of CRC for urgent colonic investigations. A recent systematic review highlighted the potential merits of this approach using f-Hb; however, it concluded that models to date had been developed with poor methodology and few externally validated models11.
Our aim was to develop and validate a clinically useful prediction model to estimate 1-year risk of CRC using all people in the referral population for Nottingham University Hospitals NHS Trust who had completed a FIT in primary care. We planned to compare a model-based strategy to the NICE f-HB cut-off of 10 µg/g and a risk of referral threshold of 3% (f-Hb ≥ 40 µg/g and others including a 1% and 2% risk) akin to the NICE early cancer diagnosis recommendations and assess diagnostic utility using a net benefit approach6.
Methods
Nottingham Rapid Colorectal Cancer Diagnosis Pathway (NRCCD)
In Nottingham since November 2016, a locally commissioned pathway allows FIT to be requested by general practitioners as a triage tool for all symptomatic patients referred with suspected CRC, except those with rectal bleeding or a palpable rectal mass, as described elsewhere. From November 2021 general practitioners were also able to request the test for those with rectal bleeding. In addition, a Full Blood Count (FBC) blood test was mandated for all CRC referrals irrespective of symptoms or age.
The study is reported consistent with the TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) guidelines12. The study was undertaken as part of the COLOFIT programme of work seeking to establish the optimal role of FIT in the clinical pathway.
Study setting
The study was conducted at Nottingham University Hospitals (NUH) NHS Trust, using data for all primary care requested FIT results, processed within pathology services at Nottingham University Hospitals NHS Trust (NUH) among13:
Adults (≥ 18 years of age)
Patients within Nottinghamshire registered at a General Practice that would refer to Nottingham University Hospitals (Nottingham City and South Nottingham Integrated Care Partnerships)
From 01/Nov/2017 until 31/Nov/2021 for a derivation cohort.
From 01/Dec/2021 until 31/Nov/2022 for a validation cohort.
All other patients were excluded. FIT requests and results reporting was electronic. FIT dispatch and return were postal from the laboratory and were required for all referred patients to NUH. The kits were distributed and analysed according to manufacturer’s protocols by our accredited FIT laboratory using an OC-Sensor™ platform (Eiken Chemical Co., Tokyo, Japan) as previously described14.
Data Management
The variables of interest were extracted and linked by patients’ unique identification numbers using Microsoft SQL Server from our Trust’s Enterprise Data Warehouse as previously described10. The data were anonymised before being accessed by the researchers, so the researchers had no access to identifiable patient level data: no patient level data left NUH NHS Trust. The anonymous data for analysis were analysed on a secure SQL server within NUH that only the analytical team could access for analysis (CC, JW)10.
Outcomes
CRC was defined from linked Infoflex (Civica) data where all cancers diagnosed at NUH NHS Trust are recorded. Fact and date of death were obtained from the NHS personal demographics service and underlying cause of death (coded with ICD-10) from https://www.hed.nhs.uk/Info/. Patients were followed up for one year for CRC diagnosis or death.
Exposures
Each individual had their first recorded FIT (index FIT) identified and were subsequently linked to all the required datasets within NUH NHS Trust’s Enterprise Data Warehouse as previously described10. This included age (at date of FIT) using year of birth, sex (defined as male/female), and recorded ethnicity (categorised as White, Black, Asian, Other and unrecorded). Measures of ferritin, iron, transferrin saturation and faecal calprotectin were extracted. Blood test results included haemoglobin/mean corpuscular volume (MCV)and platelets as thrombocytosis has been shown to be associated with undiagnosed colorectal cancer15. Missing blood tests values were assumed to be missing at random and imputed separately in the derivation and validation cohorts using two level multiple imputation-chained equations by predictive mean matching, with a random intercept for each patient and time from test fitted as a within patient gradient. Ten imputed datasets were used with up to 10 iterations per dataset. Adequate mixing of imputed values was assessed visually with plots. Additional predictors in the imputation model included age, sex, CRC, and death. We were not able to disaggregate by sex and gender as only sex is recorded electronically.
Statistical analysis
Predictive model building
We constructed a multivariate regression model using F-HB result, age, sex, and haemoglobin, platelets and MCV test results as potential predictors with the outcome of CRC diagnosis within one year of the FIT. A Cox proportional hazards survival model was selected as the primary model to account for censoring from non-CRC deaths. A multivariable selection algorithm was used to select fractional polynomial transformations for f-Hb, haemoglobin, MCV, platelets and age. This used backward elimination with weighted likelihood ratio testing across the stacked imputed datasets whilst keeping the familywise error rate at p = 0.0516. Individual level weights were calculated as the smallest proportion of non-missing data across the imputed values divided by the number of imputed datasets following Morris et al16. All pairwise interactions between age, f-Hb, platelets and MCV were tested using generalised likelihood ratio tests incorporating all the transformed components for each predictor17.
We also developed a logistic regression model with the binary outcome of CRC at one year ignoring censoring, using the same model selection approach as for the Cox model. This was to assess whether the Cox survival analysis was influenced by a change in time to diagnosis of cancer from the FIT rather than predicting the risk of cancer itself. It is presented in the supplementary material.
All analyses were carried out using R16 within R Studio.
Model targeted validation
We initiated an external validation (using data from a different geographical location) as planned in the statistical analysis plan, but initial work showed poor calibration in the external dataset chosen. We judged this to be potentially related to pre-analytical (sample collection) and population differences. Whilst work is on-going to understand the impact of these factors on model calibration in the external dataset, we validated the model in the target population in Nottingham18. The equation developed in the derivation cohort was applied to the Nottingham validation cohort of patients having a FIT with the imputed blood test values18. At the time of analysis we only had access to one year of test results for validation subsequent to the derivation cohort. Therefore, for the following performance measures only first FIT per patient excluding repeat tests in subsequent years was used so that patients were not included in both the validation and derivation samples.
Concordance was then measured using Harrel’s C-statistic across the imputed datasets and pooled using Rubin’s rules. Concordance was stratified by age and ethnicity to assess for potential inequalities in the performance of the model. The calibration of the models was assessed by plotting the observed 1-year Kaplan Meier survival probability against the expected 1-year survival predicted by the fitted model, stratified by the linear predictor in deciles of patients with CRC (to avoid strata with few events). The sensitivity, specificity, true positive, false positive, true negative and false negative values and rates were calculated for different test thresholds for the logistic and survival models in the validation and derivation cohorts. To account for censoring in the survival models we calculated false positive (FPV) and true negative values (TNV) as the observed CRC free survival in patients above and below each selected threshold respectively, using Kaplan Meier estimates similar to the approach described by Vickers et al19. The false positive and true positive rates were then calculated by multiplying FPV and (1-FPV) respectively by the proportion of patients above the selected threshold. True and false negative rates were calculated as the difference between the false and true positive rates and the overall observed CRC survival and risk respectively.
These estimates were then used to calculate the net benefit and extrapolated (by multiplying by 100,000) to indicate the number of potential missed cancers and reduction in colonoscopies per 100,000 FITs. These performance estimates were compared to a f-Hb only model with a threshold of ≥10 µg/g and ≥40 µg/g (f-Hb ≥10 µg/g as a currently recommended threshold and f-Hb ≥ 40 µg/g as equivalent to the 3% risk cut off recommended urgent cancer referrals), and an intermediate model with f-Hb, age and sex including fractional polynomial transformations for f-Hb and age. 95% confidence intervals were calculated with bootstrapping, resampling each of the imputed datasets with replacement (n = 1000) and pooling the subsequent estimates to calculate the 2.5% and 97.5% quantiles20.
Data sharing statement
This work uses data that has been provided by patients and collected by the NHS as part of their care and support. Under the Data Protection Impact Assessment approval for this work (DPIA reference: IG0889) we are unable to share the original data outside Nottingham University Hospitals NHS Trust.
Ethics approval statement
HRA and Health and Care Research Wales (HCRW) approval was given for this study - IRAS project ID: 312362; Protocol number: 22ON007; REC reference: 22/HRA/2125; Sponsor: Nottingham University Hospitals NHS Trust.
Patient and public involvement
PPI involvement in development of the COLOFIT work was extensive with review by several PPI panels and named PPI representation on the grant application.
Funding
This project was funded by the National Institute for Health and Care Research (NIHR) [Health Technology Assessment (HTA) Programme (Project number 133852)) and will be published in full in the HTA journal. Further information is available at: [https://fundingawards.nihr.ac.uk/award/NIHR133852]. The views expressed are those of the authors and not necessarily those of the NIHR or Department of Health and Social Care.
Results
Study population
For the derivation cohort 34,435 patients had 37,216 FIT results recorded between 01/Nov/2017 and 31/Nov/2021, after excluding 2,558 repeat tests within 12 months. Between 1/Dec/2021 and 31/Nov/2022 there were 21,012 patients; however, after excluding 1948 repeat tests within 12 months there were 20,234 FITs recorded for the validation cohort. Summary characteristics are shown in table 1, with similar age and sex distributions in the two cohorts. 533 CRC diagnoses (1.5%) were made within one year of FIT in the derivation cohort, and 214 (1.1%) colorectal cancer (or CRC) diagnoses within the validation cohort.
After taking the first FIT test to calculate the performance characteristics in the validation analysis, there were 34,231 patients with first FITs in the derivation cohort with 516 cancers, and 16,735 patients with first FITs in the validation cohort with 206 cancers.
Table 1 summarises the cohorts, with missing haemoglobin (9.5%), MCV (9.7%), and platelet count (9.8%) blood values within the year prior and 14 days post FIT in the derivation cohort, increasing to 13% for the validation cohort (table 1). Measures of ferritin, iron, transferrin saturation and faecal calprotectin were missing in 20-90% of patients and were therefore not used in model building. There was adequate mixing of the imputed values of haemoglobin, MCV, and platelet count after the first couple of iterations within both the derivation and validation cohort (supplementary figures S1 & S2).
Predictor Model building
Cox Proportional Hazards Survival Model
In the survival model multiple fractional polynomial transformations were selected for age, f-Hb, and platelet count (figures S3-S5), with MCV included as a linear variable and sex as a binary variable. Haemoglobin was not selected. The fitted model is shown in table S1, with the fitted equation predicting one-year survival from CRC in table S2. There was minimal evidence for interactions between the transformed covariates (all p > 0.1, generalised likelihood ratio tests, table S3). The concordance for the model in the derivation cohort pooled between the imputed datasets was C = 0.937 (0.916-0.957).
Performance in the validation cohort
Cox Proportional Hazards Survival Model
Stratified C-statistic performance
Concordance, as measured by a pooled C-statistic, remained highest in younger patients across the derivation and validation cohorts, with no clear drop in performance in the validation cohort (table 2). Similarly, the performance did not drop within the ethnicities, although there was greater uncertainty in the smaller strata for ethnicities other than White.
Calibration
Figure S6 shows that although there was some reduction in the calibration of the model in the validation cohort compared to the derivation it remained acceptable and did not need recalibrating, allowing for the increased variability from smaller numbers.
Performance
A f-Hb cut-off of 10 or greater was equivalent to a 1-year cancer risk of 0.64% in the derivation cohort. We combined this threshold with the model predicted cancer risk thresholds of 1%, 2% and 3% to calculate the positive predictive value, negative predictive value, sensitivity and specificity of the Cox model compared to a f-Hb ≥10 and f-Hb ≥40 cut off in the derivation and validation cohort (table 3). The Cox model with blood tests had a 1% to 2% increase in the positive predictive value with a similar negative predictive value compared to a binary f-Hb cut off. This was reflected in the improvement in specificity at the expense of sensitivity using the Cox model with blood tests compared to binary f-Hb cut offs and a Cox model with only f-Hb, age and sex using similar fractional polynomial transformations to those in the full model (supplementary table S6).
Net benefit analysis
Figure 2 shows the net benefit plots for the derivation and validation models comparing the balance between true positives and false positives, weighted for the different cancer threshold probabilities that can trigger referral to secondary care. This shows that at all thresholds there was a net benefit using the Cox model with blood tests compared to f-Hb only. Extrapolating true and false positive and negative rates to 100,000 FITs in the validation cohort showed that using the Cox model with blood tests reduced the number of normal colonoscopies needed by 1,729 colonoscopy tests (95% CI 1458 to 2007) compared to ≥ 40 f-Hb cut off, and to 11,787 colonoscopy tests (95% CI 11,240 to 12,326) compared to ≥ 10 f-Hb cut off, an 18-40% decrease. There was no significant predicted change in missed cancers (+8 compared to ≥ 40 f-Hb cut off (95% CI –43 to 63) and +7 compared to ≥ 10 f-Hb cut off (95% CI –6 to 27). When compared to the Cox model with only f-Hb, age, and sex, the addition of blood tests similarly reduced the number of colonoscopies that would be required (Table 4).
Discussion
We show that incorporating simple blood test results, age and sex with f-Hb into a clinical risk prediction model for patients with symptoms of possible CRC may improve the diagnostic pathway. Our models, which included f-Hb, age, sex, MCV and platelet count, increased the positive predictive value for CRC compared to the f-Hb cut off of ≥10 µg/g currently recommended by NICE and others, with minimal change in the negative predictive value6,21,22. The corresponding reduction in false positive rate could lead to fewer referrals and colonic investigations. During validation, the false negative rate increased resulting in more missed cancers, but to a smaller extent than the reduction in false positives. Compared to f-Hb ≥10 µg/g the Cox model, if implemented in a population of 100,000 people having FITs, might reduce the number of referrals and tests by >10,000 with no evidence that more cancers would be missed. Using a f-Hb ≥10 µg/g was associated with a cancer risk of 0.64% which is considerably below the current recommended urgent referral guidelines of 3%5. At risk thresholds up to a 5% risk of CRC our model showed similar improvements over the equivalent binary f-Hb cut offs or f-Hb modelled as a continuous variable. These findings require further external validation to assess generalisability to other populations and health care settings and the value of these trade-offs require robust health economic assessment18. This work is ongoing as part of the COLOFIT programme.
The strengths of this study are the large population-based cohort of patients with symptomatic FIT testing over a period of 6 years. All relevant routine blood tests and CRC diagnoses were recorded within electronic health records, including complete follow up for death and its cause. Our choice of modelling the covariates as continuous transformations avoids the pitfalls of selective cut offs and crude thresholds. This allows a more personalised approach to individual risk that is more informative and potentially more useful for the health care system when prioritising who to investigate in diagnostic CRC pathways. We have incorporated the effect of missing data within our models and undertaken a targeted validation within a separate cohort of more recent patients within Nottingham. There are some limitations, however. We were unable to utilise some indices for anaemia such as ferritin due to the high proportion of missing data for these individual measures. We had smaller numbers for our validation cohort and did not include repeat tests. We plan to continue to reassess the calibration and performance of the model as data accrues and validate the model in external datasets within different referral regions to assess how transferrable the model is14. We identified fewer cases of CRC in the validation cohort. This may reflect changes in the tested population over time, such as the inclusion of those with rectal bleeding and wider use of the test which has occurred since the pandemic. Nonetheless, the population within the validation cohort represents a ‘pan-risk’ group of patients where CRC is a possibility, where other factors to discriminate those at risk of CRC are needed. Despite this population change, the internal-external validation showed the performance of the model did not decrease over time in the same Nottinghamshire population: indeed, performance was slightly superior in the validation cohort, which is also worthy of further study. We have modelled the risk of CRC but other diagnoses are made in patients urgently referred for possible cancer, so further work will be required to detail the potential benefits of, for example, diagnosing inflammatory bowel disease and polyps.
We also did several sensitivity analysis and comparisons between Cox and logistic modelling. When we modelled f-Hb as a continuous variable, transformed, and added either age or age and sex none of these models performed as well as the more complex model we developed. When comparing Cox and logistic modelling techniques we found the differences were minor in terms of performance but calibration in validation was superior for the Cox model.
There have been some prior attempts to utilise blood tests to improve the performance of FIT. In a study of 16,604 FITs in a patient cohort in Oxford three models were developed combining f-Hb, age and sex. The authors found no additional benefit from the use of blood tests; however, the cohort described derived from patients fulfilling the DG30 criteria (i.e. of lower risk of cancer), FITs were analysed on the HM JACK-arc rather than the OC Sensor analyser, and, importantly, FITs were undertaken using faecal samples collected into universal stool collection pots and transported to the laboratory before being transferred in to stabilising buffer23. Haemoglobin degradation is likely to have occurred during this time24. The COLONPREDICT study included f-Hb, age, sex, rectal bleeding, benign anorectal lesions, rectal mass, serum carcinoembryonic antigen, blood haemoglobin, colonoscopy in the last 10 years, and treatment with aspirin25. We were unable to include family history in the model developed in Nottingham as this was not available from the data used (and is not routinely captured in the NHS). The FAST score did not incorporate blood tests into the risk prediction model, instead using f-Hb level, age and sex26. A further study modelling the risk of CRC using logistic regression in a population in Scotland concluded no benefit to the use of a risk score in those with a low FIT test however it suggested that modelled risk might allow the raising of the current threshold and thus reduce endoscopy demand27. A recent systematic review of the performance of f-Hb-based risk prediction models identified 22 studies combining FIT with one or more variable to predict the risk of CRC or advanced colonic polyps11. The review found that 10 studies reported development of a model, whilst four reported validation of models and three presented both derivation and validation11. The models were developed in modestly sized cohorts and were considered methodologically poor with a lack of validation. None presented a net-benefit analysis.
While the output of the Cox and logistic models require a calculator for computation, the aim of the study was to identify the best fitting model which yielded net benefit, regardless of complexity. Equations are implementable within current NHS IT systems like how renal function is automatically computed nationally. Other examples of successful implementation of model-based decision making in primary care, include the QRisk algorithms28. The advantage of this approach is that it allows calculation of individualised risk prediction using available results and demographics such that a tailored approach could be considered i.e. we could determine at a given level of risk of CRC what f-Hb result (incorporating age, sex and FBC) should trigger referral for further investigation. Table 5 shows some clinical scenarios of how the predicted risk varies for a woman with a f-Hb of 40 µg/g depending on her full blood count results and age.
Implementation of such an approach requires a re-evaluation of the guidance on prioritising investigation for CRC. At present the aspiration in the English NHS is to refer people being assessed in primary care with a risk of cancer of 3% for further rapid investigation29. In CRC this threshold is de facto much lower, as the current f-Hb threshold of ≥10 µg/g represents a risk of CRC of 1% or lower. Stakeholder consensus would be required to decide what the level of CRC risk should be. This would determine the diagnostic yield and health service burden all stakeholders are willing and able to accept and deliver. For example, if a 3% risk of CRC were to be implemented this would be equivalent to enacting a f-Hb threshold of ≥40 µg/g10. This would reduce the number of normal colonoscopies required substantially but at a cost of more missed cancers.
Our model at both a 1% or 3% threshold of CRC risk would offer additional improvements in false negative and false positive rates compared to the equivalent f-Hb ≥10 µg/g or f-Hb ≥40 µg/g thresholds. In addition, if the level of f-Hb in the national Bowel Cancer Screening Programme (BCSP) is reduced alongside the expansion of the age inclusion criteria, a currently unfeasible increase in capacity will be required to deliver all the extra colonoscopies. Enacting model-based triage in symptomatic patients suspected of CRC should free up colonoscopy resource to allow expansion of the BCSP, pending further externally validation of our model in other populations.
In conclusion, enacting a model-based triage of a symptomatic CRC pathway could decrease the burden on endoscopy whilst maintaining diagnostic accuracy as targeted validation of our model suggested that using the model may lead to a similar proportion of cancers detected whilst reducing the number of colonoscopies performed compared to the equivalent binary f-Hb cut offs. The current f-Hb cut off of 10 µg/g or greater is equivalent to an individual CRC risk less than one percent, resulting in many false positives and therefore colonoscopies that is arguably unsustainable within the health system.
Data Availability
This work uses data that has been provided by patients and collected by the NHS as part of their care and support. Under the Data Protection Impact Assessment approval for this work (DPIA reference: IG0889) we are unable to share the original data outside Nottingham University Hospitals NHS Trust.
Footnotes
Postal address and affiliation: Gastrointestinal and Liver Theme, National Institute for Health Research (NIHR) Nottingham Biomedical Research Centre (BRC), Nottingham University Hospitals NHS Trust and the University of Nottingham, School of Medicine, Queen’s Medical Centre, Nottingham, United Kingdom
Contributor and guarantor information The guarantor (CJC) accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Authorship statement All authors were part of the conceptualisation of the work and have reviewed and interpreted the findings. JW, DJH and CJC wrote the applications for permission to access the data, the protocol for the study and gained HRA approval. JJ carried out data management with CJC, and CJC carried out the analysis. DJH, JW and CJC wrote the first draft of the manuscript, and all authors were involved in the writing, reviewing and editing drafts of the paper and approving the manuscript for submission. CJC is guarantor.
Copyright/license for publication The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, a worldwide licence to the Publishers and its licensees in perpetuity, in all forms, formats and media (whether known now or created in the future), to i) publish, reproduce, distribute, display and store the Contribution, ii) translate the Contribution into other languages, create adaptations, reprints, include within collections and create summaries, extracts and/or, abstracts of the Contribution, iii) create any other derivative work(s) based on the Contribution, iv) to exploit all subsidiary rights in the Contribution, v) the inclusion of electronic links from the Contribution to third party material where-ever it may be located; and, vi) licence any third party to do any or all of the above
Competing interests Mixed competing interests: “All authors have completed the ICMJE uniform disclosure form at http://www.icmje.org/disclosure-of-interest/ and declare: no support from any organisation for the submitted work; CJR has received grant funding from ARC medical, Norgine. Medtronic, 3D Matrix solutions and Olympus medical. He was an expert witness for ARC medical and Olympus medical.”; no other relationships or activities that could appear to have influenced the submitted work.”
We have added in an additional new reference published on the 29th of February on line in Colorectal Diseases. This has added an additional line of text to the manuscript in the discussion.
References
