IJC International Journal of Cancer

Transcription

1 IJC International Journal of Cancer Detection of colorectal neoplasia: Combination of eight blood-based, cancer-associated protein biomarkers Michael Wilhelmsen 1, Ib J. Christensen 1, Louise Rasmussen 1, Lars N. Jørgensen 2, Mogens R. Madsen 3, Jesper Vilandt 4, Thore Hillig 5, Michael Klærke 6, Knud T. Nielsen 7, Søren Laurberg 8, Nils Br unner 9, Susan Gawel 10, Xiaoqing Yang 10, Gerard Davis 10, Annemieke Heijboer 11, Frans Martens 11 and Hans J. Nielsen 1 1 Department of Surgical Gastroenterology, Hvidovre Hospital, University of Copenhagen, Hvidovre, Denmark 2 Department of Surgical Gastroenterology, Bispebjerg Hospital, University of Copenhagen, København, Denmark 3 Department of Surgical Gastroenterology, Herning Hospital, University of Copenhagen, Herning, Denmark 4 Department of Surgical Gastroenterology, Hillerød Hospital, University of Copenhagen, Hillerød, Denmark 5 Department of Clinical Biochemistry, Hillerød Hospital, Denmark 6 Department of Surgical Gastroenterology, Horsens Hospital, University of Copenhagen, Horsens, Denmark 7 Department of Surgical Gastroenterology, Randers Hospital, University of Copenhagen, Randers, Denmark 8 Department of Surgical Gastroenterology, Aarhus Hospital, University of Copenhagen, Aarhus, Denmark 9 Department of Veterinary Disease Biology, Faculty of Health and Medical Sciences, University of Copenhagen, Denmark 10 Cancer Core R&D, Abbott Diagnostics Division, Abbott Park, Chicago, IL, USA 11 Department of Clinical Biochemistry, VU medical center, Amsterdam, The Netherlands Serological biomarkers may be an option for early detection of colorectal cancer (CRC). The present study assessed eight cancer-associated protein biomarkers in plasma from subjects undergoing first time ever colonoscopy due to symptoms attributable to colorectal neoplasia. Plasma AFP, CA19-9, CEA, hs-crp, CyFra21-1, Ferritin, Galectin-3 and TIMP-1 were determined in EDTA-plasma using the Abbott ARCHITECTVR automated immunoassay platform. Primary endpoints were detection of (i) CRC and high-risk adenoma and (ii) CRC. Logistic regression was performed. Final reduced models were constructed selecting the four biomarkers with the highest likelihood scores. Subjects (N 5 4,698) were consecutively included during Colonoscopy detected 512 CRC patients, 319 colonic cancer and 193 rectal cancer. Extra colonic malignancies were detected in 177 patients, 689 had adenomas of which 399 were high-risk, 1,342 had nonneoplastic bowell disease and 1,978 subjects had clean colorectum. Univariable analysis demonstrated that all biomarkers were statistically significant. Multivariate logistic regression demonstrated that the blood-based biomarkers in combination significantly predicted the endpoints. The reduced model resulted in the selection of CEA, hs-crp, CyFra21-1 and Ferritin for the two endpoints; AUCs were 0.76 and 0.84, respectively. The postive predictive value at 90% sensitivity was 25% for endpoint 1 and the negative predictive value Key words: colorectal cancer, blood-based biomarkers, detection Additional Supporting Information may be found in the online version of this article. Grant sponsor: The Andersen Isted Fund; Grant sponsor: The Augustinus Foundation; Grant sponsor: The Beckett Fund; Grant sponsor: The Inger Bonnen Fund; Grant sponsor: The Hans & Nora Buchard Fund; Grant sponsor: CEO Jens Bærentsen (private donation); Grant sponsor: The Walter Christensen Family Fund; Grant sponsor: The P.M. Christiansen Family Fund; Grant sponsor: The Aase & Ejnar Danielsen Fund; Grant sponsor: The Erichsen Family Fund; Grant sponsor: The Knud & Edith Eriksen Fund; Grant sponsor: The Svend Espersen Fund; Grant sponsor: The Elna and Jørgen Fagerholt Fund; Grant sponsor: The Sofus Carl Emil Friis Fund; Grant sponsor: The Torben & Alice Frimodt Fund; Grant sponsor: The Eva & Henry Frænkel Fund; Grant sponsor: The Gangsted Fund; Grant sponsor: The Thora & Viggo Grove Fund; Grant sponsor: The H Foundation; Grant sponsor: The Erna Hamilton Fund; Grant sponsor: The Sven & Ina Hansen Fund; Grant sponsor: The Søren & Helene Hempel Fund; Grant sponsor: The Henrik Henriksen Fund; Grant sponsor: The Jørgen Holm Family Fund; Grant sponsor: Foundation Jochum; Grant sponsor: The KID Fund; Grant sponsor: The Kornerup Fund; Grant sponsor: The Linex Fund; Grant sponsor: The Dagmar Marshall Fund; Grant sponsor: The Midtjyske Bladfund ; Grant sponsor: The Axel Muusfeldt Fund; Grant sponsor: The Børge Nielsen Family Fund; Grant sponsor: The Michael Hermann Nielsen Fund; Grant sponsor: The Arvid Nilsson Fund; Grant sponsor: The Obel Family Fund; Grant sponsor: The Krista & Viggo Petersen Fund; Grant sponsor: The Willy & Ingeborg Reinhard Fund; Grant sponsor: The Kathrine & Vigo Skovgaard Fund; Grant sponsor: The Toyota Fund; Grant sponsor: The Vissing Fund; Grant sponsor: The Wedell-Wedellsborg Fund; Grant sponsor: Hvidovre University Hospital (The Capital Region of Denmark) DOI: /ijc History: Received 15 July 2016; Accepted 8 Nov 2016; Online 8 Dec 2016 Correspondence to: Michael Wilhelmsen, Department of Surgical Gastroenterology, Hvidovre Hospital, University of Copenhagen, 2650 Hvidovre, Denmark, Wilhelmsen@outlook.Dk; Tel:

2 Wilhelmsen et al was 93%. For endpoint 2, the postive predictive value was 18% and the negative predictive value was 97%. Combinations of serological protein biomarkers provided a significant identification of subjects with high risk of the presence of colorectal neoplasia. The present set of biomarkers could become important adjunct in early detection of CRC. What s new? How can colorectal cancer be detected earlier? Colonoscopies for everyone might be effective, but it is far too expensive. Instead, these authors propose a set of eight protein biomarkers that could signal the presence of cancer. They tested the blood-based biomarker screening in patients who were already seeking a colonoscopy after experiencing bowel-related symptoms. Of the nearly 4,700 subjects tested, colonoscopy uncovered 512 colorectal cancers and 399 high-risk adenomas. The panel of biomarkers, in combination, successfully predicted the cancers. Next, the markers will need to be tested on asymptomatic subjects, to determine their usefulness as a general screening tool. Introduction Globally, colorectal cancer (CRC) is the third most frequent cancer disease with >1.4 million new cases and >690,000 deaths annually. 1 Although 80% of the patients will undergo intended curative resection, half of these will be diagnosed with recurrent or metastatic disease within the ensuing 5 years. 2 Patients, who undergo intended curative resection have a 5-year stage-dependent survival rates ranging from 35 to 90%, whereas long-term survival is rarely seen in patients with stage IV disease. Obviosly, the survival rates would be significantly improved if more patients could be identified with early stage disease. 3 Hitherto, it has been possible to increase the number of patients with early stages by various screening methods. Direct colonoscopy appears to be the optimal method, but due to limited cost/benefit and compliance issues the method cannot be recommended for screening of entire populations and must be restricted to at risk subjects. 4,5 Such subjects may be identified by using various stool-based screening methods, of which the guaiac fecal occult blood test (gfobt) has been used for more than a decade in Europe and USA. The test have specificities of 85 90%, while the sensitivity range between 40 and 90%. 4 However, these high ranges have been contradicted by new data from a major observational study showing sensitivites for nonadvanced adenoma, advanced adenoma and CRC of 5.3, 8.6 and 24.2%, respectively. 6 In addition, compliance rates using gfobt are often limited to <50% and thereby the clinical sensitivity (test sensitivity X compliance) is restricted to 12 33% for CRC lesions. 4 Interpretation shows that the majority of not yet diagnosed CRCs and in particular adenomas will be missed by gfobt screening. The fecal immunological test (FIT) may howeve, be advantageous compared to gfobt with increased specificity, sensitivity and compliance rates 7 leading to an increased impact on CRC incidence and mortality. 8,9 In spite of the present advantageous FIT results, screening for CRC is far from optimal. It is, therefore, neccessary to identify and validate new screening methods, which may improve compliance 5 and thereby the clinical sensitivity and certainly also optimize the cost-effectiveness. One such possible option is to develop test concepts based on analysis of cancerassociated proteomics, genomics, epigenomics, transcriptomics and/or metabolomics, which may detect at risk individuals, who must be offered subsequent colonoscopy. 4,10 13 Recent results have shown single serological biomarkers and combinations of a variety of such biomarkers may have the potential to be included in future research towards development of a clinically useful blood-based screening concept. Thus, various proteins, cell-free DNA, mirna, 22 nucleosomes 12 and metabolomes 10 are among the biomarkers currently in focus. Therefore, not only detailed biomarker development is needed but also the statistically objectives must be improved to calculate the enormous amount of informations achieved by blood-based screening; this includes data fusion 10,23 and bioinformatics. 24 Previously, we have focused on the efficacy of the two plasma proteins tissueinhibitor of metalloproteinases-1 (TIMP-1) and CEA in early detection of CRC. 15,25 The results have been rather uninterpretable due to the retrospective 25 versus the prospective 15 nature of the two studies. Therefore, we established an additional prospective, well-powered clinical study with focus on subjects referred to colonoscopy due to symptoms that may be attributable to CRC or other large bowel neoplasms. These symptoms included pain, changed bowel movement habits, visible blood on the stool, weight loss, anemia anal bleeding. Some subjects had more than one symptom. None of the subjects underwent preexamination FOBT or FIT. The Tumor Marker Utility Grading System guidelines 26 suggest that the validation of retrospectively achieved results must be prospectively validated in order for biomarkers to reach clinical acceptance and subsequent implementation. 27 The present prospective, population-based study included subjects scheduled for colonoscopy due to symptoms of large bowel neoplasia. Thus, the subjects should be considered as representing a high-risk study population and not a general screening population. The study aims to evalutate a variety of protein biomarkers in detection of bowel neoplasia. The primary endpoints were as follows: 1. CRC (all stages) plus high-risk adenomas versus all others, but excluding extracolonic cancer

3 1438 Detection of colorectal neoplasia 2. CRC (all stages) versus all others, but excluding extracolonic cancer The secondary endpoints were as follows: 1. CRC (all stages) plus extracolonic cancer versus all others. 2. Extracolonic cancer versus all others but excluding CRCs plus high-risk adenomas. Materials and Methods Study population Subjects scheduled for first time ever colonoscopy due to symptoms attributable of colorectal neoplasia were included from May 1, 2010, to November 30, 2012, at one of the seven collaborating hospitals located in various parts of Denmark. The following three centers were university hospitals: Aarhus, Hvidovre and Bispebjerg; while the following four centers were regional teaching hospitals: Herning, Hillerød, Horsens and Randers. At each center, dedicated research nurses were employed to carry out the study logistics including recruitment, recording of data, collection and handling of blood samples. Inclusion criteria were age 18 years, symptoms of colorectal neoplasia and signed informed consent. Exclusion criteria were previous malignant disease, previous large bowel adenoma, previous large bowel endoscopy, member of a hereditary nonpolyposis colorectal cancer (HNPCC) or familiar adenomatosis polyposis (FAP) family, surgical intervention within 3 months before inclusion and/or unable to understand the Danish study information. Demographic data including data on previous and/or concurrent diseases (comorbidity) were recorded before colonoscopy. All subjects unable to undergo complete colonoscopy and subjects with complete colonoscopy, but without bowel pathology and persisting symptoms were offered additional examination using combinations of X-ray with barium enema, ultrasonography (US), computer-assisted tomographhy (CAT) and/or magnetic resonance imaging (MRI). Subjects with findings that could not be completed during the primary colonoscopy were referred to subsequent intervention, including operation. Excised or resected tissues were histologically examined, and the final diagnoses were recorded in the database; all pathological findings were recorded using the International Classification of Disease-10 coding system. High-risk adenomas were classified as follows: size 1 cm, or number 3, or high-grade neoplasia; or 20% villous histology. After final inclusion the database recordings were rigorously audited by on-site visits and by using the unique Danish database systems including disease diagnosis and pathology findings. Patients were tracked using the Danish computerized, unique central personal registration (CPR) number. A flow chart showing the study population is seen in Figure 1. Sample collection, handling and storage According to a validated Standard Operating Procedure (SOP), 29 all blood samples for serum, EDTA plasma, citrate plasma and buffy-coats were collected with light tournique from an antecubital vein just before the colonoscopy using endotoxin-, DNA se- and RNA se-free collection and handling equipment, collection tubes and storage vials from Becton-Dickinson (Franklin Lakes, NJ) and Almeco A/S (Esbjerg, Denmark), respectively. The blood samples were centrifuged at 3,000g, 10 min, at 218C, and serum and plasma were immidiately separated from the red cell and buffycoat layer. Contamination with white cells and platelets were Figure 1. Chart showing the flow of accrued individuals in the study.

4 Wilhelmsen et al Table 1. Overview of various biomarkers in this study Marker Rationale References CEA CEA is a well characterized marker related to 12,15,34 CRC and has been clinically approved for monitoring of GI cancers including CRC and its performance to detect some CRC has been characterized CyFra 21-1 Several reports support that Cyfra 21-1 and 12,14,34 other cytokeratin fragments are elevated in many CRC patients CA19-9 Indications are that elevated levels may be 12,34 informative for some CRC cancers AFP Analyses of unpublished results have suggested that AFP may have prognostic utility in CRC Danish Serum Institute via H. Nielsen TIMP-1 Galectin-3 hs-crp Ferritin Considerable research supports that TIMP-1 is elevated in plasma of many CRC patients Research phase clinical studies support that Galectin-3 ligand is elevated in colorectal cancers and that its binding partner galectin-3 is also increased in cancers Inflammation is an early hallmark in cancer development, and indication is that CRP measurement may be helpful in assessment for patient risk for having screening relevant findings Ferritin, like CRP, is considered an acute phase reactant, with levels increasing in various clinical conditions including infections and cancers reduced by leaving 0.5 cm untouched serum or plasma above the buffy-coat, which subsequently was transferred to separate freezing tubes. All separated samples were marked with unique barcodes for storage identification, which was performed using the FreezerWorksVR (Seattle, WA) pc-based tracking system. All separated samples were frozen at 2808C under 24/7 electronic surveillance; the entire procedure was finalized within 2 hr. When all samples were collected EDTA plasma levels of AFP, CA19-9, CEA, hs-crp, CyFra21-1, Ferritin, Galectin-3 and TIMP-1 were determined by using the Abbott ARCHI- TECT R i2000 automated immunoassay platform utilizing a two-step dual monoclonal immunoassay (Table 1). The determinations were performed at the Abbott Center of Excellence at VU Medical Center (VUMC) (Amsterdam, the Netherlands). Abbott in-house research prototype and on-market reagents were used for the determinations. All achieved analysis results were transferred to the study statisticians and included in the database of demografic and findings data. Sample size calculation and ethics The expected incidence of CRC in the included population was 5 6%. Therefore, in order for the area under the receiver operating characteristic (ROC) curve as a measurement of discrimination be estimated with a half-width of 5% points with 80% coverage probability, a sample size of 4,500 subjects was needed. The Ethics Committee of the Capital Region of Denmark (H ) and the Danish Data Protection Agency ( ) approved the study, which was performed according to the Declaration of Helsinki II. The REMARK guidelines for running and reporting the study were followed whenever applicable. 28 Statistics All endpoints considered are binary. Logistic regression analyses with the endpoints as the dependent variables and the biomarkers as explanatory variables have been done. Biomarkers were log transformed (base 2) for analysis. Age and gender have also been included as explanatory covariates. Supplementary analyses have been performed including comorbidity and life style variables as explanatory variables. The logistical regression analyses have been validated using residuals and goodness of fit tests. Initial analyses were done for each explanatory variable as a univariable analysis, and then multivariable analysis was performed including all explanatory variables as well as age and gender. The results are presented by the odds ratios (OR) with 95% confidence intervals (CI) and receiver operating characteristic curves with area under curve (AUC) as a measure of discrimination. An effect for clustering by center was included in the multivariable analyses. Then a model reduction was done for the primary endpoints by selecting the best four biomarkers as 4,34 14, ,48

5 1440 Detection of colorectal neoplasia Table 2. Overview of univariable analysis Non-CRCs versus all others, but excluding CRCs plus high-risk adenomas CRC (all stages) plus non-crcs versus all others CRC (all stages) versus all others, but excluding non-crcs CRC (all stages) plus high grade adenomas versus all other findings, but excluding non-crcs OR CI p values AUC OR CI p values AUC OR CI p values AUC OR CI p values AUC AFP < CA < < < < CEA < < < < CyFra < < < < Ferritin < < < Galectin < < < < hs-crp < < < < TIMP < < < < Abbreviations: AFP, alpha foeto protein; CA 19-9, cancer antigen 19-9; CEA, carcino embryogenic antigen; CyFra21-1, cytokeratin fragment 21-1; hs-crp, high sensitivity C-reactive protein; TIMP-1, tissue inhibitor of metalloproteinases-1. well as age and gender identifying the model with the highest likelihood score. 29 This model was then used for the secondary endpoints. In addition, the estimated specificities as well as positive and negative predictive values are shown for 60, 70, 80 and 90% sensitivities. In addition the estimated sensitivities at 90% specificity are presented. The impact of comorbidity has been evaluated by testing for interaction between the specified comorbidities and the biomarkers. The final validation of these models was done using 10-fold cross-validation in order to compare the estimated regression coefficients to these. Model validation using cross validation methods confirmed the models. p-values <5% were considered significant. Database management and calculations have been done using SAS (v9.3, SAS Institute, Cary, NC) and R (R Development Core Team, Vienna, Austria; Results A total of 4,698 subjects were included in the present study and the distribution of the results of the colonoscopy are shown in Figure 1. In particular, 319 subjects were diagnosed with colon cancer (CC), 193 with rectal cancer (RC), 177 with primary extracolonic cancer and 689 with adenomas, of whom 399 were categorized as high-risk adenomas. The 177 extracolonic cancers were as follows: 33 lung cancers, 22 pancreatic cancer, 17 B-cell lymphoma, 16 ovarian cancer, 16 unkonwn primary cancer, 10 prostate cancer, 9 gastric cancer, 8 gallbladder cancer, 7 renal cancer, 6 anal cancer, 5 malignant myeloma, 4 breast cancer, 3 small intestine cancer, 3 mesothelioma, 3 esophageal cancer, 3 larynx cancer, 3 chronic lymphatic leukemia, 2 neuoendocrine cancer, 2 malignant melanoma, 1 uterine cancer, 1 testicular cancer, 1 bladder cancer and 1 adrenal gland cancer. Univariable analysis of the 8 blood-based protein biomarkers with focus on endpoint 1 (CRC 1 high-risk adenoma) demonstrated that all were statistically significant. The results are presented in Table 2 by the OR s for a twofold difference in the levels and by the AUCs as well. High levels of the biomarkers were associated to the endpoint with exception of Ferritin and only moderate discrimination was observed with the two best AUCs of 0.65 (CyFra21-1 and CEA). Similar results were observed for the other three endpoints, with the exception of Ferritin, where the OR is >1 for the fourth endpoint. ROC curves for the primary endpoints for each biomarker are shown in Figure 2a d. The levels of each biomarker are shown in Figure S1. Multivariable analysis of the biomarkers and including age and gender showed that all blood-based biomarkers demonstrated independent significance except TIMP-1 for the first endpoint (CRC 1 high-risk adenoma); the model had an AUC of Similar analysis for the second endpoint (CRC) demonstrated that all biomarkers had independent significance; however, the OR for TIMP-1 and galectin-3 were reversed, i.e., <1; the model had an AUC of

6 Wilhelmsen et al Figure 2. a-d: The ROC curves for each of the 4 endpoints are shown for each marker as well as the multivariable model. The AUC s are shown for each model in the same color as the ROC curve. The ROC curve for the mltivariable model is shown in solid black For the third endpoint (CRC 1 extracolonic cancer) TIMP-1 was not significant, whereas all others had independent significance (Table 3); the model had an AUC of Ferritin had no independent significance in the analysis of the fourth endpoint (extracolonic cancer) or had TIMP-1, while all others had independent significance; the model had an AUC of The ROC curves for the multivariable analyses are shown in Figure 2a d, and p-values for multivariable analyses are shown in Table 3. The blood-based biomarkers are associated with rank correlations between TIMP-1 and hs-crp, galectin-3 and CyFra21-1 >0.40 (data not shown), which suggests a need for caution in the interpretation of the full model. The results of a reduction of the model for each primary endpoint based on 4 biomarkers as well as age and gender choosing the models with the highest likelihood score identified the same four biomarkers for each of these endpoints. These were CEA, hs-crp, CyFra21-1 and Ferritin. This model was then applied to the secondary endpoints. The AUCs for each endpoint were only slightly reduced when compared to the full model endpoint 1: AUC ; endpoint 2: AUC ; endpoint 3: AUC and endpoint 4: AUC A subgroup analysis restricting endpoint 2 to stage I or II CRC (stage I: n and stage II: n 5 163) and to stage III or IV CRC (stage III: n and stage IV: n 5 108; 1 not available) resulted in AUC (p < ) and AUC (p < ), respectively. The specificities for a number of fixed sensitivities were estimated, based on the linear predictor obtained by multivariable logistic regression (Table 4): for the first endpoint, with 90% sensitivity the specificity was 33% with a positive predictive value (PPV) of 25% and negative predictive value (NPV) of 93%. The same analysis for the second endpoint showed a specificity of 48% at 90% sensitivity with a PPV equal to 18% and NPV equal to 97%. The third endpoint had a specificity of 40% at a sensitivity of 90% and PPV equal to 21% and NPV equal to 96%. Table 4 also includes

7 1442 Detection of colorectal neoplasia Table 3. Multivariable analysis of each endpoint including all eight biomarkers as well as age and gender Endpoint Parameter Model AUC OR Lower 95% CI Upper 95% CI p values 1: CRC and high-risk adenoma AFP Age pr 10 years < CA < CEA < CyFra < Ferritin < Galectin < Gender F versus M < hs-crp < TIMP AUC : CRC AFP Age pr 10 years < CA CEA < CyFra < Ferritin < Galectin < Gender F versus M < hs-crp < TIMP AUC : CRC and non-crc AFP Age pr 10 years <0.01 CA <0.01 CEA <0.01 CyFra <0.01 Ferritin <0.01 Galectin <0.01 Gender F versus M <0.01 hs-crp <0.01 TIMP AUC : Non-CRC AFP Age pr 10 years CA CEA CyFra <0.01 Ferritin Galectin <0.01 Gender F versus M hs-crp <0.01 TIMP AUC 0.82

8 Wilhelmsen et al Table 4. Reduced multivariable model including 4 biomarkers, age and gender Endpoint 1: CRC and high-risk adenoma Parameter Age pr 10 years Model AUC OR Lower 95% CI Upper 95% CI p values < CEA < CyFra < Ferritin < Gender F < versus M hs-crp < AUC % 70% 80% 90% 2: CRC Age pr < years CEA < CyFra < Ferritin < Gender F versus < M hs-crp < AUC % 70% 80% 90% 3: CRC and non-crc Age pr < years CEA < CyFra < Ferritin < Gender F versus < M hs-crp < AUC % 70% Specificity (%) Positive predicted value (%) Negative predicted value (%)

9 1444 Detection of colorectal neoplasia Table 4. Reduced multivariable model including 4 biomarkers, age and gender (Continued) Lower 95% CI Upper 95% CI Endpoint Parameter Model AUC OR p values 80% 90% 4: Non-CRC Age pr years CEA CyFra < Ferritin Gender F versus M hs-crp < AUC % 70% 80% 90% Specificity (%) Positive predicted value (%) Negative predicted value (%) these results for 60, 70 and 80% sensitivity. At 90% specificity, the sensitivities for the four endpoints were 41, 59, 56 and 57%. Univariable analysis of each comorbidity for each of the endpoints demonstrated that only lung disease was associated to endpoints 2 and 3 (p-values <4.5%). It has been hypothesized that comorbidity is a moderator variable affecting the predictive value of the biomarkers. Tests for interaction between the markers and comorbidities showed that only cardiovascular disease with AFP, CyFra21-1, Ferritin, Galectin-3, hs-crp and TIMP-1 were significantly associated. The results of analyzing each biomarker include comorbidities with interaction for endpoint 2 demonstrated that significant results were only found with cardiovascular disease(table S1). Similar results were obtained for all endpoints(only endpoint 2 is shown). Discussion First time ever colonoscopy of 4,698 subjects included in the present study identified 512 patients with CRC, 689 with adenomas and 177 with extracolonic cancer, reflecting that this is a high risk population. The univariable analyses showed limited, but significant associations between the respective serological protein biomarkers and study outcomes of all four endpoints. In the multivariable analysis however, all biomarkers except TIMP-1 showed statistically significant results. Combination of all eight biomarkers improved the results significantly with AUCs ranging from 0.76 to 0.84 in all four endpoints. The reduced model including combination of the four biomarkers (CEA, CyFra21-1, Ferritin and hs- CRP) with the highest individual detection rates plus age and gender showed almost similar AUCs as the complete model. No adjustment for multiple testing was performed; however, all primary analyses remain highly significant when applying the Bonferroni correction. Although the protocol included a number of predefined comorbidities that potentially may have influence on the biomarker levels, 15 only lung disease was associated to any endpoint and the influence of comorbidiy on the predictive effect of the markers was limited to cardiovascular disease. Therefore, comorbidity appears to be of limited importance for the detection rates of CRC, adenomas or extracolonic cancer. In addition, the inclusion of comorbidity would complicate the use of the algorithm. The results of the present prospective, population-based study support recent results that combinations of serological biomarkers are valuable in early detection of CRCs and adenomas. 10,15,18,30 34 Considerations on the clinical value of using the present serological biomarker profile in screening must include that the study population was at-risk due to the symptoms that lead to the colonoscopy. The number of subjects identified with CRC, adenomas and extracolonic cancers will be lower in a screening population. 18,35,36 In addition, it must be evaluated whether blood-based testing is to replace direct colonoscopy or to screen and then lead to subsequent colonoscopy. In the event of replacing direct colonoscopy the test requires high sensitivity, which the present study may

10 Wilhelmsen et al demonstrate. The results appear to be similar or in some instances improved compared to results of screening with gfobt, 4,6,9 while results of FIT screening may still be advantageous. 9,37,38 However, both the PPVs and in particular the NPVs appear to be similar to the FIT screening results. 39 Recent results have also indicated that serological biomarkers may detect bowel neoplasia with an efficacy similar to both gfobt and FIT screening. 18,34 These two studies have however not followed the TMUGS 26 and REMARK guidelines 28 in detail. Therefore, additional prospective studies are needed to validate the achieved results. One major limitation of stool-based screening is the compliance rates. This may be improved with blood based testing; 4,5 the compliance rate achieved in the present study was >90%. It should be noted that the subjects were referred to diagnostic colonoscopy due to symptoms of colorectal neoplasia, and presumably due to the direct contact by the subjects and the research persons, the compliance was unexpectedly high. The eight protein biomarkers were chosen due to their association with GI malignancy. Additional blood-based biomarkers such as genomics, 19 21,40,41 epigenomics 12,22 and metabolomics 10 are currently under intense evaluation in various clinical studies. Methylated genes such as SEPT9 42 may be valuable in early detection of colorectal neoplasia. 36 Hitherto, methylated SEPT9 has been shown to be specific for CRC neoplasia and a m SEPT9 test is on market and used for screening; however, recent results indicate that the SEPT9 test may also to a certain extend identify subjects with nonmalignant diseases. 43 Therefore, future research must focus on development of biomarker panels, which are both sensitive and have high specificity values, and thereby reduce unnecessary colonoscopies. Such panels might include proteomics, genomics, epigenomics, transcriptomics and metabolomics in various combinations. Results of a previous study showed 15 that symptoms associated with CRC may also be attributable to extracolonic malignancy, and in the present study, 177 subjects were diagnosed with such cancer diseases. The study showed that those patients could be detected by the included protein biomarkers. As all of the included biomarkers to some degree were nonspecific, it would not have been possible to avoid colonoscopy of those patients. Testing with biomarkers could however, lead to subsequent clinically relevant examinations or surveillance as such patients with increased biomarker levels and subsequent non-neoplastic colorectum may have an extracolonic cancer or may be at high risk of developing a REFERENCES subsequent primary malignancy either intracolorectal or extracolorectal. 34,44,45 It must be accepted that such detections will not be achieved by stool-based screening tests and that blood-based testing may be easily performed. Developments of blood-based biomarker screening tests must focus on detection of early-stage disease and advanced/ high-risk adenomas. Thereby, the survival rates after diagnosis of CRC will be improved substantially. Detection of subjects with high-risk adenomas must be offered frequent follow-up colonoscopies, which eventually will decrease the incidence of CRC. Interpretation of specific results of endpoint 1 and endpoint 2 indicates that the present blood-based protein biomarkers could detect high-risk adenomas, early-stage CRC and late-stage CRC with similar accuracy even when taking the tendency to stage dependent increasing biomarker levels into account as shown with subgroup analysis of stage I or II CRC versus stage III or IV CRC. However, the results must be considered as one of the first steps towards the development of blood-based CRC screening tests that may form the basis for additions of other biomarkers of proteomic, genetic, epigenetic, transcriptomic and metabolomic origin. Such combinations may lead to improved screening tests that may provide significant outcome of screening for CRC. Therefore, the present results should be validated in a prospective clinical trial. In conclusion, the results of the present prospective, population-based, clinical study indicate that a combination of eight or four plasma-based protein biomarkers can detect high-risk adenomas, CRC and extracolonic cancer with a high sensitivity and a high NPV. Results of additional combinations of various genetic, epigenetic, transcriptomic and metabolomic biomarkers currently analyzed in samples from the study are awaited with interest. Moreover, results from an additional current, prospective screening-associated clinical study of 8,000 FIT-positive and 36,000 FIT-negative subjects are currently being accrued. Acknowledgements The research nurses, secretaries and technicians at the participating hospital departments and laboratories are thanked for their skillful work. Abbott Laboratories Inc. (Abbott Park, IL, USA) sponsored the protein analyses at their Center of Excellence, VUMC, Amsterdam, the Netherlands. Conflict of Interest Susan Gawel, Xiaoqing Yang and Gerard Davis are the employees of Abbott Laboratories Inc. All others have no disclosures. 1. Ferlay J, Soerjomataram II, Dikshit R, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN Int J Cancer 2014;136:E Wilhelmsen M, Kring T, Jorgensen LN, et al. Determinants of recurrence after intended curative resection for colorectal cancer. Scand J Gastroenterol 2014;49: Lindebjerg J, Osler M, Bisgaard C. Colorectal cancers detected through screening are associated with lower stages and improved survival. Dan Med J 2014;61:A Nielsen HJ, Jakobsen KV, Christensen IJ, et al. Screening for colorectal cancer: possible improvements by risk assessment evaluation? Scand J Gastroenterol 2011;46: Adler A, Geiger S, Keil A, et al. Improving compliance to colorectal cancer screening using blood and stool based tests in patients refusing screening colonoscopy in Germany. BMC Gastroenterol 2014;14: Brenner H, Hoffmeister M, Birkner B, et al. Diagnostic performance of guaiac-based fecal occult blood test in routine screening: state-wide

11 1446 Detection of colorectal neoplasia analysis from Bavaria, Germany. Am J Gastroenterol 2013;109: Digby J, Mcdonald PJ, Strachan JA, et al. Use of a faecal immunochemical test narrows current gaps in uptake for sex, age and deprivation in a bowel cancer screening programme. J Med Screen 2013;20: Zorzi M, Fedeli U, Schievano E, et al. Impact on colorectal cancer mortality of screening programmes based on the faecal immunochemical test. Gut 2015;64: Young GP, Symonds EL, Allison JE, et al. Advances in fecal occult blood tests: the FIT revolution. Dig Dis Sci 2015;60: Bro R, Nielsen HJ, Savorani F, et al. Data fusion in metabolomic cancer diagnostics. Metabolomics 2013;9: Jimenez CR, Knol JC, Meijer GA, et al. Proteomics of colorectal cancer: overview of discovery studies and identification of commonly identified cancer-associated proteins and candidate CRC serum markers. J Proteomics 2010;10: Holdenrieder S, Dharuman Y, Standop J, et al. Novel serum nucleosomics biomarkers for the detection of colorectal cancer. Anticancer Res 2014;34: Ganepola GA, Nizin J, Rutledge JR, et al. Use of blood-based biomarkers for early diagnosis and surveillance of colorectal cancer. World J Gastrointest Oncol 2014;6: Bresalier RS, Byrd JC, Tessler D, et al. A circulating ligand for galectin-3 is a haptoglobin-related glycoprotein elevated in individuals with colon cancer. Gastroenterology 2004;127: Nielsen HJ, Br unner N, Jorgensen LN, et al. Plasma TIMP-1 and CEA in detection of primary colorectal cancer: a prospective, population based study of 4509 high-risk individuals. Scand J Gastroenterol 2011;46: Ladd JJ, Busald T, Johnson MM, et al. Increased plasma levels of the APC-interacting protein MAPRE1, LRG1, and IGFBP2 preceding a diagnosis of colorectal cancer in women. Cancer Prev Res 2012;5: Chen H, Werner S, Tao S, et al. Blood autoantibodies against tumor-associated antigens as biomarkers in early detection of colorectal cancer. Cancer Lett 2014;346: Werner S, Krause F, Rolny V, et al. Evaluation of a 5-marker blood test for colorectal cancer early detection in a colorectal cancer screening setting. Clin Cancer Res 2016;22: Haber DA, Velculescu VE. Blood-based analyses of cancer: circulating tumor cells and circulating tumor DNA. Cancer Discov 2014;4: Heichman KA. Blood-based testing for colorectal cancer screening. Mol Diagn Ther 2014;18: Reinert T, Schøler LV, Thomsen R, et al. Analysis of circulating tumour DNA to monitor disease burden following colorectal cancer surgery. Gut 2016;65: Luo X, Stock C, Burwinkel B, et al. Identification and evaluation of plasma micrornas for early detection of colorectal cancer. PLoS One 2013;8: e Acar E, Lawaetz AJ, Rasmussen MA, et al. Structure-revealing data fusion model with applications in metabolomics. Conf Proc IEEE Eng Med Biol Soc 2013;2013: Zhu J, Shi Z, Wang J, Zhang B. Empowering biologists with multi-omics data: colorectal cancer as a paradigm. Bioinformatics 2015;31: Holten-Andersen MN, Christensen IJ, Nielsen HJ, et al. Total levels of tissue inhibitor of metalloproteinases 1 in plasma yield high diagnostic sensitivity and specificity in patients with colon cancer. Clin Cancer Res 2002;8: Hayes DF, Bast RC, Desh CE, et al. Tumor marker utility grading system: a framework to evaluate clinical utility of tumor markers. J Natl Cancer Inst 1996;88: Schrohl A-S, Wurtz S, Kohn E, BanksÔ RE, et al. Banking of biological fluids for studies of diseaseassociated protein biomarkers. Mol Cell Proteomics 2008;7: Altman DG, McShane LM, Sauerbrei W, et al. Reporting recommendations for tumor marker prognostic studies (REMARK): explanation and elaboration. PLoS Med 2012;9:e Furnival GM, Wilson RW. Regressions by leaps and bounds. Technometrics 1974;16: Ransohoff DF, Pignone M. Colon cancer screening models: lessons and challenges. Clin Gastroenterol Hepatol 2013;11: Vatandoost N, Ghanbari J, Mojaver M, et al. Early detection of colorectal cancer: from conventional methods to novel biomarkers. J Cancer Res Clin Oncol 2016;142: Yamada A, Horimatsu T, Okugawa Y, et al. Serum mir-21, mir-29a, and mir-125b are promising biomarkers for the early detection of colorectal neoplasia. Clin Cancer Res 2015;21: Christensen IJ, Br unner N, Dowell B, et al. Plasma TIMP-1 and CEA as markers for detection of primary colorectal cancer: a prospective validation study including symptomatic and nonsymptomatic individuals. Anticancer Res 2015;35: Wild N, Andres H, Rollinger W. Colorectal cancer: a combination of serum markers for the early detection of updated version. Clin Cancer Res 2010;16: Mandel JS, Church TR, Bond JH, et al. The effect of fecal occult-blood screening on the incidence of colorectal cancer. N Engl J Med 2000;343: Church TR, Wandell M, Lofton-Day C, et al. Prospective evaluation of methylated SEPT9 in plasma for detection of asymptomatic colorectal cancer. Gut 2014;63: Fraser CG, Digby J, McDonald PJ, et al. Experience with a two-tier reflex gfobt/fit strategy in a national bowel screening programme. JMed Screen 2012;19: Tinmouth J, Lansdorp-Vogelaar I, Allison JE. Faecal immunochemical tests versus guaiac faecal occult blood tests: what clinicians and colorectal cancer screening programme organisers need to know. Gut 2015;64: Shah R, Jones E, Vidart V, et al. Biomarkers for early detection of colorectal cancer and polyps: systematic review. Cancer Epidemiol Biomarkers Prev 2014;23: Spindler K-LG, Appelt AL, Pallisgaard N, Andersen RF, Brandslund I, Jakobsen A. Cell-free DNA in healthy individuals, noncancerous disease and strong prognostic value in colorectal cancer. Int J Cancer 2014;12: Pedersen SK, Symonds EL, Baker RT, et al. Evaluation of an assay for methylated BCAT1 and IKZF1 in plasma for detection of colorectal neoplasia. BMC Cancer 2015;15: Song L, Li Y. SEPT9: a specific circulating biomarker for colorectal cancer. Adv Clin Chem 2015;72: Ørntoft M-BW, Nielsen HJ, Ørntoft TF, et al. Performance of the colorectal cancer screening marker Sept9 is influenced by age, diabetes and arthritis: a nested case-control study. BMC Cancer 2015;15: Kring TS, Piper TB, Jørgensen LN, et al. Bloodbased biomarkers at large bowel endoscopy and prediction of future malignancies. Biomark Cancer 2015;7: Hvolris MH, Piper TB, Hammer E, et al. Increased serological cancer-associated biomarker levels at large bowel endoscopy and risk of subsequent primary malignancy in patients diagnosed with diverticula. Scand J Gastroenterol 2016;51: Greco C, Vona R, Cosimelli M, et al. Cell surface overexpression of galectin-3 and the presence of its ligand 90k in the blood plasma as determinants in colon neoplastic lesions. Glycobiology 2004;14: Mazhar D, Ngan S. C-reactive protein and colorectal cancer. QJM 2006;99: Nelson RL. Elevation of serum ferritin is superior to fecal occult blood testing as a screening test for colonic adenoma... and not only because patients do not have to handle their own stool. Dis Colon Rectum 1996;39: Guo Y-Z, Pan L, Du C-J, et al. Association between C-reactive protein and risk of cancer: a meta-analysis of prospective cohort studies. Asian Pac J Cancer Prev 2013;14: Crockett SD, Mott LA, Barry EL, et al. C-reactive protein and risk of colorectal adenomas or serrated polyps: a prospective study. Cancer Prev Res 2014;7: