Early detection of colorectal neoplasia: application of a blood-based serological protein test on subjects undergoing population-based screening

Abstract

Background

Blood-based biomarkers used for colorectal cancer screening need to be developed and validated in appropriate screening populations. We aimed to develop a cancer-associated protein biomarker test for the detection of colorectal cancer in a screening population.

Methods

Participants from the Danish Colorectal Cancer Screening Program were recruited. Blood samples were collected prior to colonoscopy. The cohort was divided into training and validation sets. We present the results of model development using the training set. Age, sex, and the serological proteins CEA, hsCRP, TIMP-1, Pepsinogen-2, HE4, CyFra21-1, Galectin-3, ferritin and B2M were used to develop a signature test to discriminate between participants with colorectal cancer versus all other findings at colonoscopy.

Results

The training set included 4048 FIT-positive participants of whom 242 had a colorectal cancer. The final model for discriminating colorectal cancer versus all other findings at colonoscopy had an AUC of 0.70 (95% CI: 0.66–0.74) and included age, sex, CEA, hsCRP, HE4 and ferritin.

Conclusion

The performance of the biomarker signature in this FIT-positive screening population did not reflect the positive performance of biomarker signatures seen in symptomatic populations. Additional biomarkers are needed if the serological biomarkers are to be used as a frontline screening test.

Background

Colorectal cancer (CRC) is the third leading cause of cancer and accounts for approximately one-tenth of cancer cases and cancer-related deaths globally [1]. While the incidence and mortality from CRC are increasing in low- and middle-income countries, a decreasing trend is observed in several high-income countries [2]. Overall, the risk of being diagnosed with CRC is associated with increasing age [3], but emerging results have shown, however, that the risk of young-onset CRC is increasing in some high-income countries [4–6]. In addition, the disease is a leading cause of cancer-related mortality among adults younger than 50 years of age in the USA [7].

Survival from CRC is highly dependent on the stage at diagnosis; screening leads to more early-stage diagnoses, thereby improving survival from CRC [8, 9]. Identification and removal of adenomas generally prevents long-term mortality from colorectal cancer [10]. Furthermore, evidence suggests that organised screening programmes reduce the incidence of colorectal cancer [11].

The number of subjects undergoing CRC screening will most likely increase in coming years as a consequence of both increased demand in low- and middle-income countries [12] and an extended age interval for screening recommendations in high-income countries, for instance 45-85 years of age in USA [13]. Direct colonoscopy is the “gold” standard for early detection of CRC, but high costs and limited capacity makes direct colonoscopy unrealistic and infeasible for general population-based screening [14, 15]. Most nationwide colorectal screening programmes use the faecal immunochemical test (FIT) with compliance rates above the minimum acceptable rate of 45% but below the desirable rate of 65% [16–18].

Test accuracy and compliance are important determinants of the cancer detection rate in a screening population. The test concept and its ease to compliance is an important determinant for the overall compliance of subjects invited to participate in CRC screening [19]. Blood-based tests appear to be preferred compared to faeces-based tests, and such screening tests may increase compliance compared to FIT [19, 20]. If a blood-based test is to be used as a frontline CRC screening test it requires comparable diagnostic accuracy to FIT in an appropriate screening population.

Blood-based, cancer-associated biomarker tests using multiple circulating proteins have shown promising proof-of-concept results as a test for early detection of CRC [21]. However, almost all studies have been performed in symptomatic populations, performed in limited and often insufficiently sized studies, or without proper internal or external validation of the findings [21]. We have previously presented the results of a combination of eight blood-based proteins as a detection test for CRC with an AUC of 0.84 derived from a large prospective cohort of symptomatic patients undergoing diagnostic colonoscopy [22]. A blood-based biomarker protein test intended for use as a frontline screening test for CRC needs, however, to be developed and validated in a sufficiently sized appropriate screening population.

The aim of this study was to develop a blood-based, cancer-associated protein biomarker test for the detection of CRC in a large screening population [23]. Specifically, the focus of the test was to discriminate in an asymptomatic, FIT-positive (FIT + ) screening population at colonoscopy:

1. CRC from all other findings.

2. CRC, high-risk adenomas (HRA), and medium-risk adenomas (MRA) versus all other findings.

3. CRC versus clean colorectum.

The performance of the developed model was assessed post hoc head-to-head with the FIT in a cohort representative of a true screening cohort.

Methods

This cohort study is reported according to the TRIPOD statement [24] and presents the results of model development of a blood-based, cancer-associated protein biomarker test for detection of CRC in a large FIT-positive screening population. Secondly, we present the performance of the developed model in a true screening population and compare the results with the performance of FIT.

Study design and participants

The present study represents results from FIT + and FIT- participants from Part 1 of the Endoscopy III protocol, which was approved by the Ethics Committee at the Capital Region of Denmark (H-4-2013-050) and the Danish Data Protection Agency (2007-58-0015/HVH-2013-022) [23].

Part 1 of the Endoscopy III project consists of FIT + (faecal haemoglobin ≥100 ng/mL; OC-Sensor, Eiken Pharmaceuticals, Japan) and FIT- (haemoglobin <100 ng/mL) participants from the nationwide Danish Colorectal Cancer Screening Program. After a 4-year run-in period, the screening programme offers biennial FIT and subsequent colonoscopy for all FIT + subjects; screening is offered to all residents between 50 and 74 years of age. Participation in the Danish Colorectal Cancer Screening Program, subsequent work-up, and treatment are free of charge. The FIT− participants were randomly chosen from the database of the Danish Colorectal Cancer Screening Program and offered to participate in the protocol. A total of 5143 FIT− subjects were included between April 2014 and March 2016. FIT + participants of the screening programme undergoing screening colonoscopies at nine hospitals (Amager, Bispebjerg, Herlev, Herning, Hillerød, Horsens, Hvidovre, Randers and Viborg hospitals) were eligible for inclusion in the FIT + subgroup. In addition to FIT + result and invitation to the subsequent colonoscopy, the participants received information about the study protocol. On the day of colonoscopy participants received verbal information and signed an informed consent form. Prior to colonoscopy, an interview was performed, and blood samples were collected. A total of 8415 FIT + participants were included between April 2014 and August 2016. Until the end of 2017 the Danish National Screening for colorectal cancer was prevalent screening and the included subjects in our study were all from prevalent screening.

The cohort of FIT + (N = 8415) as well as the FIT− (N = 5143) participants from Part 1 of the Endoscopy III project were divided into training and validation sets, where the validation set included subjects from the end of the study, counting backwards until 200 consecutive CRC FIT + cases were identified, and then assigning subjects from the start of the study up until this cut point to the training set. The actual date of the cut point separated all included FIT + and FIT− subjects into the training and validation sets, respectively. This design was reviewed with The Notified Body in Ireland (representative of the European Union) and agreed that the samples from the training set will be analysed and results locked before analysing the samples from the validation set. The present study reports the results of model development using the training set.

Demographics and medical history were obtained by interview and through the patients’ electronic medical files. Findings and interventions at colonoscopy and subsequent histopathology of any removed tissue were entered into a Web-based database together with demographics and medical history. The database was continuously audited to ensure accurate data capture. The identification of interval CRC in the FIT− group was done retrieving data from The Danish Pathology Register based on SNOMED codes [25]. The Danish Pathology Register is nationwide and mandatory by law to register all cancer cases. An interval CRC was defined as a CRC registered up to 2 years from the date of inclusion in the current study.

The database was locked prior to the analysis of the blood samples. The recording of demographics, medical history and findings at colonoscopy were thereby blinded from the results of the analysed blood samples.

Blood samples

Blood samples (90 mL) for serum, K2 EDTA (Ethylenediaminetetraacetic acid) plasma and buffy-coats were collected after bowel preparation, but before sedation and subsequent colonoscopy. According to a validated standard operating procedure, the samples were collected and centrifuged at 3000×g for 10 min at +21 °C; K2 EDTA plasma was centrifuged twice with a pipetting step in between. All samples were stored at −80 °C within two hours of collection; the water-cooled freezers were placed in a locked storage room and were under constant electronic surveillance for continuous temperature monitoring. Trial specimens had a unique 8-digit barcode number and storage was managed by FreezerWorks (Seattle, WA, USA), a PC-based storage management system. Shipments from participating Danish hospitals to the study repository centre at Hvidovre Hospital and subsequent shipment to the various participating national and international laboratories were on dry ice in secured transport boxes.

Outcomes

The presence of CRC and colorectal adenomas for the FIT + participants were assessed at colonoscopy and subsequent histopathology reports of identified lesions. CRC was defined as a positive histopathology report of carcinoma in the colon or rectum (biopsy at colonoscopy or specimens examined after subsequent colorectal surgery). HRAs were defined as lesions ≥20 mm, the total number of lesions ≥5, or resection by piece-meal technique. MRAs were defined as lesions sized 10–20 mm, a total of 3–4 lesions independent of size, tubulovillous or villous lesion, or high-grade neoplasia. LRAs were defined as lesions <10 mm, the total number of lesions <3, tubular lesions, or low-grade dysplasia. Clean colorectum was defined as no CRC or adenomas at colonoscopy.

For the FIT− participants the presence of CRC was defined by a positive histopathology report of carcinoma in the colon or rectum within 2 years from the date of inclusion in the study.

Predictors

Age, sex and the following blood-based, cancer-associated proteins were chosen as possible predictors based on the results of previous studies [22]: CEA (carcinoembryonic antigen), hsCRP (high-sensitivity C-reactive protein), TIMP-1, Pepsinogen-2, HE4 (human epididymis secretory protein 4), CyFra21-1, Galectin-3, ferritin and B2M (β₂ microglobulin). EDTA plasma samples were shipped to the Abbott Center of Excellence at Amsterdam UMC (Amsterdam, the Netherlands), where the above-mentioned blood-based proteins were determined using the ARCHITECT i2000^® Automated platform (Abbott Laboratories Inc., Abbott Park, IL, USA).

Statistics

The sample size of Part 1 of the Endoscopy III protocol was planned to include 8000 FIT + participants [23]. The sample size of the current study was thereby predefined based on the recommendation that the cohort being split into a training set and a validation set in a consecutive order so that the last 200 CRC cases were included in the validation set.

Univariable logistic regression was performed for all predictors and multivariable logistic regression was used for model development and validated using residual diagnostics. Outcomes were included as binary dependent variables. All biomarkers were log-transformed (log base 2) and included as continuous explanatory variables. Age was divided by 10 and included as a continuous explanatory variable. Sex was included as a binary explanatory variable. All biomarkers were included in the multivariable analyses and the final models were identified using stepwise backward selection and tenfold cross-validation, with biomarkers significant at the 5% level, as well as age and sex. Results of the logistic regression analyses are presented as odds ratios and 95% confidence intervals, and sensitivity estimates at 30, 70, 80 and 90% specificity. The area under the receiver-operating characteristic curve (AUC) was used as a measure of discrimination.

The outcome of CRC versus no CRC was analysed post hoc for the FIT + and FIT− subjects weighing the analysis to reflect the distribution of FIT + and FIT− in the Danish screening population, the rate was 6.89% FIT + in the period of inclusion to this study. The results of the FIT, the developed model in the FIT + cohort, as well as a fitted model for the entire cohort are presented. The latter analysis also includes linear splines in order to achieve a good fit to the data.

All available data were used. Patients with missing data regarding outcomes or predictors were excluded from analyses and no imputations were performed. A locked web-based database was used. Analyses were performed using SAS (v9.4) and R (version 3.6.3).

Results

In total, 4048 FIT + participants of the Danish Colorectal Cancer Screening Program undergoing colonoscopy were included in the training set (Fig. 1). A total of three subjects in the training set had missing data for one or more biomarkers, and 4,045 were analysed for model development. Of these, 2433 (60%) were males with a median age of 66 years (IQR: 58–72) as compared to females with a median age of 64 years (IQR: 56–71). The training set comprised 242 (6%) participants with CRC, 548 (14%) with HRA, 812 (20%) with MRA, 704 (17%) with LRA and 1742 (43%) with clean colorectum. The distribution of the 242 participants with CRC was 112 (46%) with UICC (Union for International Cancer Control) Stage I, including 58 with T1 lesions, 48 (20%) with UICC Stage II, 65 (27%) with UICC Stage III, and 17 (7%) with UICC Stage IV. All subjects underwent elective resection except for eight patients with severe disseminated UICC Stage IV disease, where resection was not an option. For the post hoc analyses, a total of 2582 FIT− participants were included in the training set of whom three subjects were subsequently diagnosed with colorectal cancer within 2 years from the date of inclusion in the study.

Fig. 1.

Flow of the study.

The univariable logistic regression analyses of age, sex and the blood-based biomarkers are presented in Table 1 with regards to discrimination between CRC and all other findings at colonoscopy; CRC, HRA and MRA versus all other findings at colonoscopy; and CRC versus clean colorectum at colonoscopy. Of the included biomarkers, only CEA, CyFra21-1, hsCRP and age were significantly associated with all three outcomes, while B2M, ferritin, HE4, Pepsinogen-2 and TIMP-1 were significantly associated with some of the outcomes. The significant association between Galectin-3 and any of the three outcomes was not demonstrated. Limited discriminatory power was shown for all of the blood-based biomarkers with AUCs ranging from 0.49 to 0.60. None of the individual protein biomarkers demonstrated better discrimination than age alone.

Table 1.

Univariable analysis of predictors.

	Colorectal cancer versus all others			Colorectal cancer, high-risk adenomas and medium-risk adenomas versus all others			Colorectal cancer versus clean colon
Predictors	Odds ratio (95% CI)	P value	AUC	Odds ratio (95% CI)	P value	AUC	Odds ratio (95% CI)	P value	AUC
CEA	1.34 (1.20; 1.50)	<0.0001	0.57	1.14 (1.08; 1.21)	<0.0001	0.53	1.45 (1.29; 1.64)	<0.0001	0.60
TIMP-1	1.31 (0.89; 1.91)	0.1662	0.52	1.20 (0.99; 1.46)	0.0645	0.52	1.66 (1.12; 2.45)	0.0115	0.55
Ferritin	0.77 (0.70; 0.85)	<0.0001	0.59	1.03 (0.98; 1.08)	0.2034	0.51	0.84 (0.76; 0.93)	0.0004	0.59
CyFra21-1	1.34 (1.15; 1.57)	0.0002	0.57	1.09 (1.00; 1.18)	0.0451	0.52	1.43 (1.22; 1.68)	<0.0001	0.58
hsCRP	1.16 (1.05; 1.27)	0.0030	0.55	1.05 (1.00; 1.10)	0.0396	0.52	1.18 (1.07; 1.30)	0.0080	0.56
Galectin-3	1.06 (0.77; 1.44)	0.7277	0.51	0.97 (0.83; 1.13)	0.6908	0.50	1.12 (0.80; 1.56)	0.5067	0.52
Pepsinogen-2	0.92 (0.77; 1.08)	0.3057	0.52	1.13 (1.05; 1.23)	0.0024	0.53	1.06 (0.89; 1.26)	0.5097	0.51
B2M	1.18 (0.85; 1.66)	0.3252	0.53	1.11 (0.94; 1.32)	0.2202	0.52	1.44 (1.02; 2.04)	0.0410	0.55
HE4	0.99 (0.79; 1.24)	0.9242	0.49	1.23 (1.11; 1.37)	<0.0001	0.55	1.24 (0.99; 1.55)	0.0625	0.55
Age	1.67 (1.40; 1.99)	<0.0001	0.61	1.41 (1.31; 1.53)	<0.0001	0.58	2.02 (1.69; 2.41)	<0.0001	0.66
Female sex	0.91 (0.80; 1.05)	0.1966	0.52	0.76 (0.71; 0.81)	<0.0001	0.57	0.75 (0.65; 0.86)	0.0001	0.57

AUC area under the receiver-operating characteristics curve, CEA carcinoembryonic antigen, TIMP-1 TIMP metallopeptidase inhibitor 1, hsCRP high-sensitivity C-reactive protein, B2M β₂ microglobulin, HE4 human epididymis secretory protein 4.

All blood-based biomarkers are log-transformed using log base 2. Age was divided by 10.

The multivariable model for discriminating CRC from all other findings at colonoscopy included sex, age, CEA, ferritin, hsCRP and HE4 (Table 2). HE4 was not significantly associated with the outcome in the univariable analysis and the point estimate of OR = 0.99 together with an AUC of 0.49 indicated it as a poor predictor. However, in the multivariable model, the performance of HE4 was significantly increased. The OR point estimates of CEA, ferritin and hsCRP in the multivariable model did not change substantially from the univariable models. The AUC of the model was 0.70 (95% CI: 0.66–0.74) (Fig. 2).

Table 2.

Multivariable model for discriminating colorectal cancer from all other findings at colonoscopy.

Predictor	Odds ratio (95% CI)	P value	AUC	Sensitivity at 30% specificity	Sensitivity at 70% specificity	Sensitivity at 80% specificity	Sensitivity at 90% specificity
Female sex	0.80 (0.69; 0.92)	0.002	0.70 (0.66; 0.74)	0.86	0.40	0.29	0.18
Age	2.00 (1.64; 2.43)	<0.001
CEA	1.45 (1.28; 1.64)	<0.001
Ferritin	0.71 (0.64; 0.78)	<0.001
hsCRP	1.21 (1.10; 1.35)	<0.001
HE4	0.421 (0.31; 0.57)	<0.001

AUC area under the receiver-operating characteristic curve, CEA carcinoembryonic antigen, hsCRP high-sensitivity C-reactive protein, HE4 human epididymis secretory protein 4.

All blood-based biomarkers are log-transformed using log base 2. Age was divided by 10.

Fig. 2. Receiver-operating characteristic curves for multivariable models.

The receiver-operating characteristic curves for the multivariable models discriminating colorectal cancer (CRC) from all other findings at colonoscopy (remaining); colorectal cancer, high-risk adenomas, and medium-risk adenomas (CRC + HRA + MRA) versus all other findings at colonoscopy (remaining); and colorectal cancer (CRC) versus clean colon (clean) at colonoscopy. CRC vs remaining CRC + HRA + MRA vs remaining. CRC vs clean Flip a coin.

Only CEA and B2M together with sex and age were included in the model for discriminating between CRC, HRA and MRA versus all other findings at colonoscopy (Table 3). The OR point estimate for CEA did not change substantially from the univariable to the multivariable model. B2M was not significantly associated with the outcome in the univariable analysis and in the multivariable model, the OR point estimate was reversed from 1.11 to 0.75. The AUC of the model was 0.61 (95% CI: 0.59–0.63) (Fig. 2).

Table 3.

Multivariable model for discriminating between colorectal cancer, high-risk adenomas and medium-risk adenomas versus all other findings at colonoscopy.

Predictor	Odds ratio (95% CI)	P value	AUC	Sensitivity at 30% specificity	Sensitivity at 70% specificity	Sensitivity at 80% specificity	Sensitivity at 90% specificity
Female sex	0.76 (0.71; 0.81)	<0.001	0.61 (0.59; 0.63)	0.82	0.45	0.31	0.17
Age	1.44 (1.31; 1.57)	<0.001
CEA	1.14 (1.07; 1.21)	<0.001
B2M	0.745 (0.62; 0.91)	0.003

AUC area under the receiver-operating characteristic curve, CEA carcinoembryonic antigen; B2M β₂ microglobulin.

All blood-based biomarkers are log-transformed using log base 2. Age was divided by 10.

The model for discriminating CRC versus clean colorectum had an AUC of 0.73 (95% CI: 0.69–0.77) and included age, sex, CEA, ferritin, hsCRP, and HE4 (Fig. 2 and Table 4). Again, the OR point estimates of CEA, ferritin, hsCRP and age did not differ from the univariable analyses. The OR point estimate of HE4 was reversed from 1.24 in the univariable analysis to 0.45 in the multivariable model.

Table 4.

Multivariable model for discriminating colorectal cancer versus clean colon at colonoscopy.

Predictor	Odds ratio (95% CI)	P value	AUC	Sensitivity at 30% specificity	Sensitivity at 70% specificity	Sensitivity at 80% specificity	Sensitivity at 90% specificity
Female sex	0.67 (0.57; 0.78)	<0.001	0.73 (0.69; 0.77)	0.91	0.56	0.37	0.23
Age	2.30 (1.87; 2.82)	<0.001
CEA	1.54 (1.34; 1.77)	<0.001
Ferritin	0.74 (0.67; 0.82)	<0.001
hsCRP	1.20 (1.08; 1.34)	0.001
HE4	0.45 (0.33; 0.62)	<0.001

AUC area under the receiver-operating characteristic curve, CEA carcinoembryonic antigen, hsCRP high-sensitivity C-reactive protein, HE4 human epididymis secretory protein 4.

All blood-based biomarkers are log-transformed using log base 2. Age was divided by 10.

CyFra21-1 was not included in any of the multivariable models despite being significantly associated with all the outcomes in the univariable analyses.

Analysis of the FIT concentration for the CRC outcome in all subjects weighted according to the distribution of FIT− and FIT + in the Danish screening population showed a sensitivity of 78%, a specificity of 94%, a PPV of 5.9%, and NPV of 99.9%. The AUC of this model was 0.89. Using the model with protein levels as well as age and sex developed on the FIT + subjects, the sensitivity was 26% at 94% specificity, with PPV of 2.1%, NPV of 99.6% and AUC of 0.74. Fitting the proteins to the entire dataset results in an AUC = 0.78 with a sensitivity of 29% at 94% specificity, and PPV of 2.5% and NPV of 99.6%.

Discussion

At present, most research of blood-based, cancer-associated biomarkers for early detection of colorectal neoplasia has relied on subjects with symptoms or even patients with a diagnosis of the CRC [21, 22, 26–28]. Although promising, biomarkers for screening need validation in screening studies or at least in studies performed with the inclusion of subjects undergoing current population screening [21].

The results of the present study were based on subjects undergoing a Danish nationwide population screening with FIT (faecal immunochemical test) to identify those who should be recommended a follow-up colonoscopy due to a FIT + result. The estimated model for discrimination between CRC and all others had an AUC of 0.70, and the model for discriminating between CRC, HRA, and MRA versus all others at colonoscopy had an AUC of 0.61. Therefore, the discriminatory power of all models was far from supporting the previous promising results achieved by the same protein biomarkers analysed in blood samples from symptomatic subjects [22]. A biomarker test used in screened subjects with a FIT + test need to show similar discrimination of colorectal lesions as symptomatic subjects, i.e., 75–80% sensitivity at 90–95% specificity [29, 30]. The goal of 80% sensitivity for detecting CRC in the present study resulted in only a 30% specificity, whereas 80% specificity was associated with a limited 29% sensitivity. When evaluating the performance of the biomarker model in a cohort representative of a true screening population the performance was clearly inferior to FIT. Consequently, the results need further discrimination improvements before consideration as a potential frontline test for CRC screening. These proof-of-concept results show promise in a screening population and evaluation with additional biomarkers may enhance algorithm performance.

Previous studies of protein biomarkers for early detection of colorectal neoplasia have indicated similar discrepancies between results generated from subjects with symptoms or diagnosed disease versus subjects that participated in CRC screening [31, 32]. This discrepancy issue may be associated with certain differences between symptomatic and non-symptomatic subjects [31, 32]. In particular, the only FDA approved blood-based biomarker for CRC screening, the mSept9 gene methylation test, had a reported sensitivity of 64% at 88% specificity in an asymptomatic average-risk population [33], but the same test had much higher sensitivity, 80% at 83% specificity, for subjects that had been diagnosed with CRC [34].

Indeed, it is well known that screening leads to a different distribution of the numbers of CRCs in each specific stage. Among symptomatic patients with CRC the distribution in our previous study [22] were Stage I: 10%, Stage II: 34%, Stage III: 37% and Stage IV: 18%. The distribution of subjects by stage in the present study was Stage I: 46%, Stage II: 20%, Stage III: 27% and Stage IV: 7%. In addition, among the 112 patients in this study with Stage I cancers, 52% had T1 lesions, which are rare among symptomatic patients. Therefore, the significant differences of results from symptomatic and screened CRC patients may be explained in part by the substantial difference in the stage distribution since the sensitivity in the detection of Stage I cancers are significantly lower than for Stages II–IV [22, 33]. In addition, the higher proportion of screened FIT + subjects with T1 lesions may underline the limited ability of the chosen protein biomarkers to discriminate between subjects with bowel neoplasia and subjects with clean colorectum. Finally, the difference between those being identified by symptoms and those being identified by screening directed colonoscopy is substantiated by results showing stage-dependent better outcomes among screened patients [35, 36].

From the cancer-associated biomarkers chosen from our previous study of symptomatic patients [22] CEA, CyFra21-1 and hsCRP were identified as having significant discrimination for all three clinical outcomes. Amongst the remaining protein biomarkers, B2M, ferritin, HE4, Pepsinogen-2 and TIMP-1 were significantly associated with some of the three outcomes, while Galectin-3 did not play any significant role. Subsequent multivariable analyses demonstrated that the various chosen biomarkers resulted in combinations of proteins, which were different depending by outcomes. Moreover, some proteins changed from increased for some combinations, while they were decreased in other combinations. Such changes within the combinations of the proteins were also observed in the multivariable analyses of the previous study with symptomatic subjects [22]. At present, such variability cannot be explained in detail, but may be based on various covariate interactions that are more determinant among screening population subjects with low levels of the various protein biomarkers vs. among symptomatic subjects, where possible interactions may be less prominent due to high protein biomarker levels.

It is an important question as to whether the limited discrimination of the present study based on blood samples from subjects undergoing population-based screening may be relevant in future screening. At least three potential possibilities might be investigated further: first, the compliance to FIT testing is 60–65%, leaving 35–40% of the population unscreened. Many subjects do not attend the screening programmes due to faeces aversion, social barriers, co-morbidity, and specifically fear of discomfort at bowel preparation required for screening, and subsequent colonoscopy in the event of a positive FIT result [37, 38]. Therefore, potential future screening options may be considered as alternative tests for non-compliant subjects; blood-based screening may be one such option [19]. Although the present results on protein biomarkers are insufficient for use in population-based screening, due to the limited ability to detect adenoma and Stage I lesions and particularly T1 lesions, the non-compliant will be provided an option to be screened and malignant lesions ≥Stage II may generally be identified. Subjects without detection of neoplastic lesions will in some countries be offered annual or biannual screening, which presumably detect subjects with ≥Stage II malignant lesions. Secondly: focus on protein biomarkers is the initial step to identify blood-based, screening relevant biomarkers. At present, few studies have presented results on combinations of various blood-based biomarkers, including proteins and ctDNA in a multi-analyte test [39, 40]. Based on the current research, future multi-analyte tests may be a combination that includes proteins, ctDNA methylations/mutations/fragmentations, miRNA, nucleosomes, histone modifications, and/or metabolomes, but such analyses are still not performed in studies based on CRC screening. Thirdly: the FIT cut-offs that determine who should be offered subsequent colonoscopy have been increased [14], or even set at very high cut-offs in order to mitigate the capacity constraints of follow-up colonoscopy in some countries that have implemented population screening. Recently, it was indicated that a Triage test that combines age of the subject, FIT result, and blood-based biomarker levels/presence may reduce colonoscopy rates by ~30% [15, 41]. Therefore, a focus on research of the Triage concept may lead to improved selection of those subjects that need follow-up colonoscopy and at the same time reduce the numbers of unnecessary colonoscopies significantly. Although there may be a plethora of various biomarkers to include in such Triage tests, the proteins, as presented in the present study may play a significant role.

Blood-based samples included in the present study were collected after bowel preparation, but before sedation and subsequent colonoscopy. Optimally, the blood collection of blood-based biomarkers for early detection or screening for colorectal neoplasia needs to be collected before the initiation of bowel preparation. Results from a recent study comparing protein biomarker levels in samples collected before and after bowel preparation showed that the level of one of the determining biomarkers in this study CyFra21-1, was decreased by almost 30% by bowel preparation [42]. Therefore, the result of the present study may have been impacted by the blood collection timing.

In conclusion, the selected serological proteins cannot be used independently as a frontline screening test for CRC but serve as a strong initial proof-of-concept. In addition to the development of a blood-based screening test, future research should examine the differences in biomarkers between symptomatic and asymptomatic populations and the ability of blood-based markers for detection of T1 tumours and advanced adenomas. Most importantly, the future development and validation of blood-based biomarkers for CRC screening should be performed in screening populations, and serious consideration should be made to combine additional biomarkers and other factors in order to improve detection accuracy.

Author contributions

HJN conceived the study and provided the funding. LNJ, JWH, MRM, JV, LM, AK, BA, MR, SG, GJD and IJC helped design the study. LNJ, JWH, MRM, JV, SB, LM, AK, PI, LF, GJD, FM, BA and MR acquired the data. JK, IJC and HJN interpreted the results. JK wrote the draft. All authors revised the manuscript, approved the final version and agreed to be fully accountable for all aspects of the work.

Funding

This study was financially supported by: The Andersen-Isted Fund, The Augustinus Foundation, The Vilhelm Bang Fund, The Beckett Fund, The Inger Bonnén Fund, The Hans and Nora Buchard Fund, CEO Jens Bærentsen, The Walter and O. Kristiane Christensen Fund, The Poul Martin Christiansen Fund, The Aase and Ejnar Danielsen Fund, The Danish Cancer Society, The Erichsen Family Fund, The Knud and Edith Eriksen Fund, The Svend Espersen Fund, The Elna and Jørgen Fagerholt Fund, The Sofus Friis Fund, The Torben and Alice Frimodt Fund, The Eva and Henry Frænkel Fund, The Gangsted Fund, The Thora and Viggo Grove Fund, The Erna Hamilton Fund, The Sven and Ina Hansen Fund, The Hede-Nielsen Family Fund, The Søren and Helene Hempel Fund, The Henrik Henriksen Fund, The Carl and Ellen Hertz Fund, The Elisa and Jørgen Holm Fund, The Humanitarian Foundation, The Ovita Juhl Fund, Foundation Jochum, The KID Fund, The Kornerup Fund, The Linex Fund, The Aage and Johanne Louis-Hansen Fund, The Mid Jutland Newspaper Fund, The Dagmar Marshall Fund, The Muusfeldt Fund, The Børge Nielsen Fund, The Inge and Jørgen Nielsen Fund, The Michael H. Nielsen Fund, The Arvid Nilsson Fund, The Obel Family Fund, The Orient Fund, The Krista and Viggo Petersen Fund, The Willy and Ingeborg Reinhard Fund, The Katrine and Vigo Skovgaard Fund, The Toyota Fund, The Tryg Fund, The Vissing Fund, The Else og Mogens Wedell-Wedellsborg Fund.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy or ethical restrictions.

Ethics approval and consent to participate

The study was performed in accordance with the Declaration of Helsinki, and approved by the Ethics Committee at the Capital Region of Denmark (H-4-2013-050) and the Danish Data Protection Agency (2007-58-0015/HVH-2013-022). All participants signed an informed consent form.

Consent to publish

Not applicable.

Competing interests

SG and GJD are employees of Abbott Laboratories Inc. The remaining authors declare no competing interests.