Evaluation of serum-based cancer biomarkers: A brief review from a clinical and computational viewpoint
Introduction
Biomarkers are believed to increase the accuracy of diagnosis to precisely characterize the disease in a diagnostic or prognostic level. Biomarkers predict the response of the patient, helping to guide a more tailored treatment for the patient. Serum biomarkers are more appealing due to their simplicity of obtaining the blood samples. There are several serum cancer biomarkers, which are routinely used in clinical oncology, e.g. prostate specific antigen (PSA) for prostate cancer and cancer antigen (CA)-125 for ovarian cancer. However, their applications have significant limitations, because of low specificity, i.e. small probability of samples with no biomarkers in all non-diseased samples. In fact, the issue of specificity has become much more acute, since more than 30% or higher circulatory PSA level patients have to go for extensive testing and treatment, indicating its lack of specificity of prostate tumor detection [1]. In summary, lack of specific serum biomarkers has impeded the change in morbidity and mortality in cancer patients.
The traditional “a priori” approach for biomarker development needs a well-established biological procedure, being subjected to two-step clinical validation: (1) simple test with a high level of quality control, and (2) planned statistical prospective evaluations within the validation pilot studies to prove an established clinical impact [2]. Contrarily, more recently “a posteriori” approaches evaluate the clinical rationale of a “biological indicator” through a systematic discovery of various screening tools (e.g. microarray, bioinformatics, High-throughput DNA sequencing). These biological instruments are “black boxes”, meaning that a clinical usage can be discovered through research pilot studies. Computational approaches give possible candidates for detecting certain diseases, by “sensitivity” and “specificity”, within a patient population, but the proper quantification of a single biomarker in serum is limited to the evaluation technicalities. Therefore this review will focus on the latter approach since the recent technologies provide a plethora of potential candidates which are in proper need of evaluation.
Over the past twenty years, biomarkers have shown significant promise in the mechanism of how it will transform a patient's treatment. Therefore, biomarker research has been aimed towards the development of personalized targeted therapy. Despite the recent technological advancements, there are still relatively few biomarkers that are in routine clinical use today [3]. With a growing number of complex genomic tests for biomarker signatures becoming commercially available, the promise of personalized medicine is fast becoming a reality. Much attention has to be placed on the reason why the promising biomarkers and the biomarkers signatures entering the clinic is a long road ahead [4].
Section snippets
Biomarker discovery validation: Three pitfalls
In this section, we first briefly show the most widely accepted guideline for evaluation and validation of biomarkers (Diagram 1): “Early Detection Research Network (EDRN)” developed by National Cancer Institute [5]. We then explain typical pitfalls and their examples of clinical biomarker evaluation failures, mainly caused by poor experimental design and inappropriate choice of the diagnostic assay.
Phase I of EDRN is the discovery of biomarkers through knowledge—based gene selection, gene
Computational measures for evaluating biomarkers
Diagram 2 is a 2 × 2 contingency table, showing cross-correlate disease status with biomarker presence. This table is obtained, under a pre-determined threshold for the biomarker test, to examine the usefulness of the candidate biomarker in the diagnostic, prognostic, predictive appraisal of the disease. This table has four cells with samples labeled by True Positive (TP), False Negative (FN), False Positive (FP) and True Negative (TN), which are mainly used to compute the measures introduced
Current biomarkers
Table 1 provides a comprehensive comparison of current potential and approved biomarkers, including 26 biomolecules, 2 metabolic biomarkers and 4 cell biomarkers. A unique feature of this table is positive and negative predictive values and DOR, which are extracted from corresponding references and attached to all biomarkers. Furthermore, one marker can be used for evaluating more than one cancer type. For example, heat shock proteins (HSPs) are used for five types: gastric, prostate,
Discussion
As explained in Section 3.8, OVA1 (approved in 2009) for ovarian carcinoma with five biomarkers including CA-125 had to show weaker performance in terms of specificity and positive predictive value than CA-125 alone (approved in 1997). This implies the need of improved statistical quantification methods to decrease the lack of specificity in the current instrumentation methods for low abundance proteins. This point brought a mentality shift focusing on a more robust statistical predictive tool,
Conflicts of interest
Disclosure of potential conflicts of interest.
The authors declare that they have no potential conflict of interest disclosed.
Funding sources
This work was partially supported by JSPS KAKENHI #2430054 and #26-381. S.Y. is supported by Grant-in-Aid for JSPS Fellows.
Funding disclosures
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Reviewers
Professor Christoph Thomssen: Martin-Luther-University Halle-Wittenberg, Gynaecology, University Hospital Halle, Ernst-Grube-Strasse 40, D-06097- Halle an der Saale, Sachsen-Anhalt, Germany.
Dr Dan Ruderman: Department of Medicine, USC Keck School of Medicine, 2250 Alcazar St., IGM (CSC) 243, Los Angeles, CA 90033, USA.
Acknowledgments
We acknowledge stimulating discussions with Dr. Ajit Bharti (Boston University, USA) concerning the importance of evaluation techniques of biomarkers and how it plagues the potential candidates in the pipeline. We would also like to thank Dr. Edda Klipp (Humboldt Universitat zu Berlin, Germany) for her thoughtful review and comments on many points in this manuscript.
We are grateful to the Editor as well as the two reviewers for their very helpful comments.
Sohiya Yotsukura is a PhD candidate of the School of Pharmaceutical Sciences at Kyoto University. Her research interests include computational biology and machine learning techniques to analyze the functional impact of SNPs and tumorigenic somatic mutations on biological systems.
References (107)
- et al.
Guidelines on prostate cancer
Eur Urol
(2008) - et al.
Early detection of prostate cancer in 2007. Part 1: PSA and PSA kinetics
Eur Urol
(2008) - et al.
EPCA-2: a highly specific serum marker for prostate cancer
J Urol
(2007) - et al.
Cystic fibrosis transmembrane conductance regulator gene mutation and lung cancer risk
Lung Cancer
(2010) - et al.
The diagnostic odds ratio: a single indicator of test performance
J Clin Epidemiol
(2003) Optimal choice of a cut point for a quantitative diagnostic test performed for research purposes
J Clin Epidemiol
(2003)- et al.
Strategies combining total and percent free prostate specific antigen for detecting prostate cancer: a prospective evaluation
J Urol
(2002) Screening for hepatocellular carcinoma: survival benefit and cost-effectiveness
Ann Oncol
(2003)- et al.
Induction of ovulation and ovarian cancer: a critical review of the literature
Fertil Steril
(2006) - et al.
Expressiom of heat shock proteins in osteosarcoma and its relationship to prognosis
Pathol Res Pract
(2000)
High prevalence of wild-type transthyretin deposition in patients with idiopathic carpal tunnel syndrome: a common cause of carpal tunnel syndrome in the elderly
Hum Pathol
Decreased levels of alpha-1-acid glycoprotein are related to the mortality of septic patients in the emergency department
Clinics (Sao Paulo)
Cyclophosphamide, fludarabine, alemtuzumab, and rituximab as salvage therapy for heavily pretreated patients with chronic lymphocytic leukemia
Blood
The predictive potential of the sweat chloride test in cystic fibrosis patients with the G551D mutation
J Cyst Fibros
Prognostic factors and treatment decisions in axillary-node-negative breast cancer
N Engl J Med
Translation of proteomic biomarkers into FDA approved cancer diagnostics: issues and challenges
Clin Proteomics
The failure of protein cancer biomarkers to reach the clinic: why, and what can be done to address the problem?
BMC Med
Phases of biomarker development for early detection of cancer
J Natl Cancer Inst
Biomarkers of diseases in medicine
Curr Trends Sci
Pitfalls and limitations in translation from biomarker discovery to clinical utility in predictive and personalised medicine
EPMA J
Why your new cancer biomarker may never work: recurrent patterns and remarkable diversity in biomarker failures
Cancer Res
Are biomarkers for bladder cancer beneficial?
J Urol
Lysophosphatidic acid as a potential biomarker for ovarian and other gynecologic cancers
JAMA
Cancer biomarkers: can we turn recent failures into success?
J Natl Cancer Inst
Plasma lysophosphatidic acid concentration and ovarian cancer
JAMA
Pitfalls and limitations in translation from biomarker discovery to clinical utility in predictive and personalised medicine
EPMA J
A review on meta-analysis of biomarkers: promises and pitfalls
Clin Chem
A novel knowledge representation framework for the statistical validation of quantitiative imaging biomarkers
J Digit Imaging
Early prostate cancer antigen-2 (EPCA-2): a controversial prostate cancer biomarker?
Clin Chem
A limitation of the diagnostic-odds ratio in determining an optimal cut-off value for a continuous diagnostic test
Stat Methods Med Res
Moving Beyond sensitivity and specificity: using likelihood ratios to help interpret diagnostic tests
Aust Prescr
Analysis of biomarker data: logs, odds ratios and ROC curves
Curr Opin HIV AIDS
ROC analysis: applications to the classification of biological sequences and 3D structures
Brief Bioinform
A roadmap for biomarker qualification
Nat Biotech
Application of the time-dependent ROC curves for prognostic accuracy with multiple biomarkers
Biometrics
Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond
Stat Med
The meaning and use of the area under a receiver operating characteristic (ROC) curve
J Radiol
Making sense of odds and odds ratios
Obstet Gynecol
Reliability and validity of the Thai version of the PHQ-9
BMC Psychiatry
Biomarker adaptive threshold design: a procedure for evaluating treatment with possible biomarker-defined subset effect
J Natl Cancer Inst
A readers’ guide to the interpretation of diagnostic test properties: clinical example of sepsis
J Intensive Care Med
Integrative the predictiveness of a marker with its performance as a classifier
Am J Epidemiol
Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker
Am J Epidemiol
OVA1 test for preoperative assessment of ovarian cancer
Community Oncol
Effectiveness of a multivariate index assay in the preoperative assessment of ovarian tumors
Obstet Gynecol
Preoperative diagnostic performance of ROMA (Risk of Ovarian Malignancy Algorithm) in relation to etiopathogenesis of epithelial ovarian tumors
J Mol Biomark Diagn
The ROMA (Risk of Ovarian Malignancy Algorithm) for estimating the risk of epithelial ovarian cancer in women presenting with pelvic mass: is it really useful
Clin Chem Lab Med
Association of mitochondrial DNA copy number in peripheral blood leukocytes with risk of esophageal adenocarcinoma
Carcinogenesis
Identifying cancer biomarkers by network-constrained support vector machines
BMC Syst Biol
The new data on prostate cancer screening: what should we do now?
Cleve Clin J Med
Cited by (0)
Sohiya Yotsukura is a PhD candidate of the School of Pharmaceutical Sciences at Kyoto University. Her research interests include computational biology and machine learning techniques to analyze the functional impact of SNPs and tumorigenic somatic mutations on biological systems.
Hiroshi Mamitsuka is a professor at the Institute for Chemical Research in Kyoto University, jointly appointed as a professor at the School of Pharmaceutical Sciences of the same university. His research interests include machine learning, data mining and their applications in bioinformatics and chemoinformatics.