Expert Review of Molecular Diagnostics

ISSN: 1473-7159 (Print) 1744-8352 (Online) Journal homepage: http://www.tandfonline.com/loi/iero20

Predictive and prognostic cancer biomarkers revisited Kenneth PH Pritzker To cite this article: Kenneth PH Pritzker (2015) Predictive and prognostic cancer biomarkers revisited, Expert Review of Molecular Diagnostics, 15:8, 971-974 To link to this article: http://dx.doi.org/10.1586/14737159.2015.1063421

Published online: 01 Jul 2015.

Submit your article to this journal

Article views: 131

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=iero20 Download by: [186.5.206.131]

Date: 16 September 2015, At: 11:42

Editorial

Predictive and prognostic cancer biomarkers revisited Expert Rev. Mol. Diagn. 15(8), 971–974 (2015)

Downloaded by [186.5.206.131] at 11:42 16 September 2015

Kenneth PH Pritzker Departments of Laboratory Medicine and Pathobiology, Surgery University of Toronto, Toronto, Ontario, Canada and Banting Institute, 100 College Street, Suite 252, M5G 1L5, Toronto, Ontario, Canada Tel.: +1 647 221 6909 Fax: +1 416 486 9460 [email protected]

While both prognostic and predictive cancer biomarkers predict clinical outcome, the term ‘predictive biomarker’ is reserved for the association of a specific therapy with a specific clinical outcome. The advent of genomic signatures and next generation sequencing as candidate predictive biomarkers has led to lengthy and expensive processes for biomarker qualification. The urgency to bring novel predictive cancer biomarkers to practice faster and cheaper requires strategies to lower the bar to biomarker implementation. Three strategies are suggested: identify biomarkers closely coupled to biologic mechanism associated with the clinical endpoint and scalable from cells to humans; identify biomarkers that can be reliably detected and quantified; and assess biomarkers by capacity to reduce toxicity as well as to increase therapy efficacy. Biomarker selection directly and closely related to production of end points by biologic mechanism demonstrated by a ladder of evidence should require less burden of proof clinically than biomarkers that are merely associative.

A casual Pubmed search, April 2015, revealed a very deep and often dense literature, 43,000 references to predictive biomarkers since 1968, and 4500 references to predictive cancer biomarkers since 1989. Writing in 2006, Clark et al. formalized the distinction between prognostic and predictive cancer biomarkers for therapeutic clinical trial purposes. “The terms ‘ prognostic’ and ‘ predictive’ have been used in numerous publications to describe relationships between biomarkers and clinical outcomes; however, these terms are seldom defined and are often used interchangeably. In this assessment, we will use the definitions proposed by Clark et al. 1994, 2001 and Hayes 1998. A prognostic factor is a measurement that is associated with clinical outcome in the absence of therapy or with the application of a standard therapy that all patients are likely to receive. It can be thought of as a measure of the natural history of the disease. A control group from a randomized clinical trial is an ideal setting for evaluating the prognostic significance of a biomarker. A predictive factor is a measurement that is associated with response or

lack of response to a particular therapy. Response can be defined using any of the clinical endpoints commonly used in clinical trials” [1]. Over the past decade, these definitions have evolved to become convention not only for clinical trials but also for oncology practice. However, although urgently needed, identifying new, clinically useful predictive cancer biomarkers remains challenging [2,3]. Therefore, review of these biomarker concepts is warranted. Both prognostic and predictive biomarkers are used to predict clinical outcome. ‘Prognosis’ as a term has been defined for >2500 years as ‘prediction of clinical outcome’ [4] irrespective of whether the markers were observed by visual inspection of entrails or more recently, by histopathology or genomic signatures. The term, ‘predictive biomarker’ continues to be reserved for a biomarker, for example, Her2 in breast cancer, that is used to associate a specific therapy, for example, trastuzumab, with a specific clinical response or lack of response [5]. Predictive biomarkers have evolved from single analytes to include

KEYWORDS: biomarker qualification . cancer . genomic signature . ladder of evidence . predictive biomarker

informahealthcare.com

.

prognostic biomarker

10.1586/14737159.2015.1063421

Ó 2015 Informa UK Ltd

ISSN 1473-7159

971

Downloaded by [186.5.206.131] at 11:42 16 September 2015

Editorial

Pritzker

multivariate genomic signatures [6] and now, variants of whole genome sequencing [7]. Recently, molecular pathological epidemiology, a directed and biological mechanism based approach to predictive biomarker discovery integrating large-scale molecular biology data has proven to be useful for identifying PIK3CA mutation in colorectal cancer susceptible to aspirin therapy [8]. When the variables within the biomarker exceed the number of cases, the possibility of a falsely positive association of biomarker and outcome becomes very real [9]. To avoid errors commonly grouped as ‘overfitting’, biomarker clinical utility is being evaluated within increasingly elaborate and rigid clinical trial frameworks [10,11]. As Yu et al. have noted [11], the rationale for this framework reflects the industrial application of complex genomic biomarkers which can be associated with clinical outcomes only by elaborate validation and clinical utility technologies. These biomarker qualification strategies are considered necessary because the link between the candidate biomarker signature and the biology specific to the outcome is often tenuous at best. In plain terms, biomarker signature components such as cell proliferation, endocrine sensitivity, or even mutations, components considered as cancer hallmarks [12], may have tenuous associations with clinical outcome because they occur not only with cancer or drug effects in cancer but also with other biologic circumstances. As well, the biomarker signature containing these hallmarks may not assess cell adaptation to carcinogenic factors in the same detail as the pro-neoplastic components. To increase association probability, multiple components are used but then the analytical and biologic uncertainty around each component must be accounted for using statistics and large comparative data sets. While there is broad agreement on the necessity for extensive formal pathways for biomarker qualification, current approaches have two major limiting obstacles: time and money. To test a novel candidate cancer biomarker in a well-controlled prospective trial where the clinical outcome might be observed years later requires extensive number of patients, often from multiple centers, with associated costs of tracking, collating and analyzing clinical data over prolonged time. Even if the biomarker testing is ancillary to a therapeutic trial, the resources and costs can be very large. Because of the difficulty in assembling resources and the costs involved, massive complex trials using novel biomarkers for practical purposes may not be reproducible or replicable in the real world in finite time [13]. Hence the science, however well conceived, is far from optimal [14]. Biomarkers which initially achieve clinical acceptance by following peer accepted development criteria can fail their intended use because further studies show lack of specificity or because the initial claims are not substantiated or worse, early success is not followed by rigorous further clinical evidence or worst of all on the basis of very inadequate evidence, the biomarker is promoted vigorously by various commercial means including direct consumer testing [15]. To return to clinical utility of predictive biomarkers, it is feasible to utilize genomic markers to predict that targeted drugs will not be efficacious as has been shown for K-RAS mutations and anti-EGFR therapy such as cetuximab and panitumumab in colon cancer [16]. Less often genomic biomarkers 972

can indicate that drugs that might work as has been demonstrated for anaplastic lymphoma kinase positivity and crizotinib therapy in non-small cell lung cancers [17]. In most such cases, when the biomarker is effective, there is a very direct and close link between the biomarker and the mechanism of drug response. In these examples, two principles emerge at the forefront: reduced toxicity by reducing ineffective therapy and tight linkage of biomarker to biologic mechanisms specific for the drug action. These are clues to a renewed perspective for predictive biomarker qualification. The fundamental clinical properties of an effective predictive cancer biomarker are that the association of biomarker results with clinical outcome can be trusted scientifically and that adverse results (false negative or false positive) are very unlikely to cause harm. The potential for the biomarker causing harm by guiding a particular therapy needs to be weighed against harm related to disease natural history, as well as harm (side effects, lack of efficacy) by therapy applied without using the biomarker. One laudable approach towards increasing confidence in biomarker validation and qualification is freeing the data from the tyranny of blindly using statistical p values by more careful consideration of the statistical characteristics of the compared populations [18–21]. This includes comparing the characteristics of the novel biomarker to those of optimized modeling of all other predictors in practice [20] as well as comparing the costs and comparative effectiveness of existing biomarkers in optimized configuration. Furthermore, for biomarkers that support clinical decision making, decision analytic methods such as net benefit should be used and take priority over general judgements such as area under the curve analysis [21]. To examine other possible approaches, let’s look back at the work of Hayes et al. on assessing clinical utility of cancer biomarkers [22,23]. These papers proposed assessing biomarker strength based not only on studies that associated biomarker with clinical outcome but also on the capacity to reliably analyze the biomarker and to associate the marker with the tumor. In this paradigm, a biomarker from tumor tissue would be intrinsically more reliable for prediction of therapy response than a cell or plasma biomarker from blood; a biomarker that can be measured directly, for example, direct spectroscopic measurement, would be intrinsically more reliable than dependence on indirect measurement such as antibody binding. The quality of predictive factors was based on devised Levels of Evidence I–V framework, which surprisingly in retrospect, was based on clinical epidemiologic criteria for the Canadian Periodic Health Examination, 1979 [24]. Levels of Evidence I–V are now widely adopted for cancer biomarkers [25] but this framework ignores completely what can be often assessed today, 35 years later, the tightness of linkage between the biologic mechanism and the clinical outcome [8,26]. Quite simply, a biomarker that measures the same effect in the same way, for example, marker for impending tumor cell death in cell cultures, experimental tumors, and human cancers by a recognized biologic mechanism provides a Expert Rev. Mol. Diagn. 15(8), (2015)

Downloaded by [186.5.206.131] at 11:42 16 September 2015

Predictive and prognostic cancer biomarkers revisited

biologic ladder of evidence with high intuitive trust because the biomarker has been demonstrated to be effective at each biologic level including clinical outcome. Logically, this approach appears stronger at the same level of statistical significance than biomarkers such as genomic signatures which require large parallel population data sets. The linkages between association and causal inference (biologic mechanism) fostered by Bradford Hill 50 years ago [27] are now formally in development at the intersection of philosophy and mathematics [28]. These interdisciplinary studies can only strengthen ways of assessing the intrinsic quality of cancer biomarkers. Recognizing the urgency of finding clinically useful cancer biomarkers and the recent partial failure of pathological complete response [29] an established predictive biomarker for breast cancer chemotherapy, what are the cheaper, faster, better ways that cancer biomarkers, particularly predictive biomarkers, can be brought to clinical practice? Three steps are suggested. First, the biomarker should have a clear rationale ,that is, be tightly coupled to a specific mechanism of drug action that incapacitates or kills cells. The same biomarker should be demonstrable and quantifiable at cell, experimental and human biologic levels. Second, the biomarker should be easily detected and quantified [22]. This is a quality that becomes more difficult with multivariate biomarkers. Third, and perhaps most controversial, is a cultural change for biomarker qualification. Presently, cancer biomarkers not only need to demonstrate analytical validity, clinical validity, and clinical utility but also comparative effectiveness beyond existing practice, a very high bar. If the biomarker qualification bar can be reduced to “prospect for harm is low and can be mitigated; prospect for clinical good, for example, reducing toxicity, References 1.

Clark GM, Zborowski DM, Culbertson JL, et al. Clinical utility of epidermal growth factor receptor expression for selecting patients with advanced non-small cell lung cancer for treatment with erlotinib. J Thorac Oncol 2006;1(8):837-46

2.

Pritzker KP. Cancer biomarkers: easier said than done. Clin Chem 2002;48(8):1147-50

3.

Pritzker LB, Pritzker KPH. Molecular signatures as cancer biomarkers: methodologic issues in discovery, validation, qualification, and standardization. In: Dakubo G, editors. Field cancerization: basic science and clinical applications. Nova Publishers; Hauppauge, New York, USA: 2011. p. 537-57

4.

Prognosis. Available from: http://oed.com/ view/Entry/152202

5.

Pritzker KP, Pritzker LB. Bioinformatics advances for clinical biomarker development. Expert opin Med Diagn 2012;6(1):39-48

informahealthcare.com

is high”, then biomarker clinical trial designs and implementation will become simpler and faster. A predictive biomarker that predicts outcome for a drug should be measurable during therapy so that if the drug is ineffective, adaptive therapeutic strategies, such as response guided therapy, can be considered. Clinical trials must fit within practice guidelines, usually current guidelines. However, some earlier guidelines employed drug regimens that were less toxic and less effective in populations but nevertheless could be effective in individuals. Therefore, novel predictive biomarkers might be brought to practice without harm by starting with a less toxic drug regimen. Then, if early in therapy, the biomarker shows that the drug is not working, adaptation to alternate therapy can be made. The need for better predictive cancer markers is urgent; rethinking the pathways to biomarker qualification can shorten the time and lower the cost to predictive cancer biomarker clinical adoption. With success, this will enable more effective individual patient centered cancer therapies, a key goal of personalized medicine. Acknowledgements

The author thanks Drs. Amadeo Parissenti, Laura Pritzker, Sanaa Noubir, Paul Walfish, Ranju Ralhan and Mr. John Connolly for their helpful discussions. Financial & competing interests disclosure

The author is CEO of Rna Diagnostics Inc. and of Proteocyte Diagnostics Inc. (both in Ontario, Canada). The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

6.

Marrone M, Stewart A, Dotson WD. Clinical utility of gene-expression profiling in women with early breast cancer: an overview of systematic reviews. Genet Med 2014. [Epub ahead of print]

7.

Tripathy D, Harnden K, Blackwell K, Robson M. Next generation sequencing and tumor mutation profiling: are we ready for routine use in the oncology clinic? BMC Med 2014;12:140

8.

Editorial

Ogino S, Lochhead P, Giovannucci E, et al. Discovery of colorectal cancer PIK3CA mutation as potential predictive biomarker: power and promise of molecular pathological epidemiology. Oncogene 2014; 33(23):2949-55

9.

Simon R. Clinical trials for predictive medicine: new challenges and paradigms. Clin Trials 2010;7(5):516-24

10.

Altman DG, McShane LM, Sauerbrei W, Taube SE. Reporting recommendations for tumor marker prognostic studies (REMARK): explanation and elaboration. BMC Med 2012;10:51

11.

Yu PP, Hoffman MA, Hayes DF. Biomarkers and oncology: the path forward to a learning health system. Arch Pathol Lab Med 2015;139(4):451-6

12.

Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; 144(5):646-74

13.

Leek JT, Peng RD. Opinion: Reproducible research can still be wrong: adopting a prevention approach. Proc Nadl Acad Sci USA 2015;112(6):1645-6

14.

Roth KA, Cox AE. Science isn’t science if it isn’t reproducible. Am J Pathol 2015; 185(1):2-3

15.

Ioannidis JP. Biomarker failures. Clin Chem 2013;59(1):202-4

16.

Perincheri S, Hui P. KRAS mutation testing in clinical practice. Exp Rev Mol Diagn 2015;15(3):375-84

17.

Isozaki H, Takigawa N, Kiura K. Mechanisms of Acquired Resistance to ALK Inhibitors and the Rationale for Treating ALK-positive Lung Cancer. Cancers 2015; 7(2):763-83

973

Editorial

Pritzker

18.

Velickovic VM. What Everyone Should Know about Statistical Correlation. Am Sci 2015;103(1):26-9

19.

Leek JT, Peng RD. Statistics: P values are just the tip of the iceberg. Nature 2015; 520(7549):612

20.

Kattan MW. Judging new markers by their ability to improve predictive accuracy. J Natl Cancer Inst 2003;95(9):634-5

21.

Hayes DF, Bast RC, Desch CE, et al. Tumor marker utility grading system:

23.

Hayes DF, Trock B, Harris AL. Assessing the clinical impact of prognostic factors: when is "statistically significant" clinically useful? Breast Cancer Res Treat 1998; 52(1-3):305-19

24.

The periodic health examination. Canadian Task Force on the Periodic Health Examination. CMAJ 1979;121(9):1193-254

25.

Hayes DF, Allen J, Compton C, et al. Breaking a vicious cycle. Sci Transl Med 2013;5(196):196cm6

26.

Floyd E, McShane TM. Development and use of biomarkers in oncology drug

development. Toxicol Pathol 2004; 32(Suppl 1):106-15 27.

Hill AB. The Environment and Disease: Association or Causation? Proc R Soc Med 1965;58:295-300

28.

Pearl J. Causal inference in statistics: An overview. Statist Surv 2009;3:96-146

29.

Cortazar P, Zhang L, Untch M, et al. Pathological complete response and long-term clinical benefit in breast cancer: the CTNeoBC pooled analysis. Lancet 2014;384(9938):164-72

Downloaded by [186.5.206.131] at 11:42 16 September 2015

22.

Steyerberg EW, Pencina MJ, Lingsma HF, et al. Assessing the incremental value of diagnostic and prognostic markers: a review and illustration. Eur J Clin Invest 2012; 42(2):216-28

a framework to evaluate clinical utility of tumor markers. J Natl Cancer Inst 1996; 88(20):1456-66

974

Expert Rev. Mol. Diagn. 15(8), (2015)

Predictive and prognostic cancer biomarkers revisited.

While both prognostic and predictive cancer biomarkers predict clinical outcome, the term 'predictive biomarker' is reserved for the association of a ...
532KB Sizes 1 Downloads 5 Views