2012 EDUCATIONAL BOOK - American Society of Clinical Oncology

Why Hasn’t Genomic Testing Changed the
Landscape in Clinical Oncology?
By Daniel F. Hayes, MD, Muin J. Khoury, MD, PhD, and David Ransohoff, MD
Overview: The “omics” revolution produced great optimism
that tumor biomarker tests based on high-order analysis of
multiple (sometimes thousands) of factors would result in truly
personalized oncologic care. Unfortunately, 10 years into the
revolution, the promise of omics-based research has not yet
been realized. The factors behind the slow progress in omicsbased
clinical care are many. First, over the last 15 years,
there has been a gradual recognition of the importance of
conducting tumor biomarker science with the kind of rigor that
has traditionally been used for therapeutic research. However,
this recognition has only recently been applied widely, and
therefore most tumor biomarkers have insufficiently high
levels of evidence to determine clinical utility. Second, omicsbased
research offers its own particular set of concerns,
TUMOR BIOMARKERS are used to determine a patient’s
current status and more importantly to predate
future events that might be modified by intervention. 1
Tumor biomarkers can be analyzed in cancer or healthy
tissue, in secretions, and circulating in blood. Tumor biomarkers
usually represent somatic changes that have
emerged during the process of carcinogenesis. However, it is
also reasonable to consider inherited germ-line differences
between individuals that predict higher risk of developing a
new malignancy or for estimating differential distribution,
metabolism, or response to a drug (“pharmacogenomics”). 2,3
Assays for tumor biomarkers can identify changes, or
individual differences, in nucleic acids (DNA, RNA), proteins,
lipids, whole cells, or tissue processes. Until recently,
an assay for a biomarker usually analyzed a single analyte,
or substance. Perhaps one of the best examples is the
development of assays for the estrogen receptor (ER) to
predict both prognosis and likelihood of responding to antiestrogen,
or “endocrine” therapies. 4,5 The biology of ER was
determined in the 1960s, and the first assay was a cumbersome
test to biochemically measure binding of radioactively
labeled estrogen to the receptor. Subsequent tests using
specific antibodies to perform either enzyme-linked immunosorbent
assays (ELISAs) and more recently immunohistochemistry
(IHC) replaced the original ligand-binding
assays, and are now almost uniformly used in clinical
medicine. However, recently, newer assays that measure
RNA expression have been introduced. Regardless, these
are all tests that assay for a single analyte, ER, and not for
several analytes that might be combined into a unified index
designed to make a clinical decision.
High-Dimensional Biomarkers
In addition to measuring a single analyte, assessment of
complex processes, such as counting vessels to determine
levels of angiogenesis, or development of a multifactoral
index, such as tumor grade (which combines estimates of
relative gland formation, nuclear appearance, and mitotic
rate), represent higher-order forms of a single test. In this
regard, during the last decade, advances in molecular biology,
technology, and bioinformatics have led to a new field
loosely described as “omics.” This field encompasses several
disciplines, generating high-dimensional data from global
e52
especially in regard to overfitting computational models and
false discovery rates. Researchers and clinicians need to
understand the importance of analytic validity, and the difference
between clinical/biologic validity and clinical utility. The
latter is required to introduce a tumor biomarker test of any
kind (single analyte or omics-based), and are ideally generated
by carefully planned and properly conducted “prospective
retrospective” or truly prospective clinical trials. Only
carefully planned studies, which take all three of these into
account and in which the investigators are aware and recognize
the enormous risk of unintended bias and overfitting
inherent in omics-based test development, will ultimately
result in translation of the exciting new technologies into
better care for patients with cancer.
sets of biologic molecules such as DNAs (“genomics”), RNAs
(“transcriptomics”), proteins (“proteomics”), and metabolites
(“metabolomics”). 6 Massive amounts of data are used to
produce a profile, or “signature,” generated by a computational
mathematical function model. This model may be
unsupervised, meaning that specimens are grouped computationally
by apparent similarities in the omics patterns,
without regard to preconceived biologic or clinical associations.
Alternatively, generation of the profiles can be “supervised”;
in this case, the signature is “pegged” to some sort of
prospectively defined biologic or clinical characteristic of
interest. Ultimately, at least in regard to clinical care, one or
more omics-based test that reputedly has clinical utility in
guiding patient care is generated.
In an online continuous horizon scanning review of the
literature from 2009 to the present, researchers from the
Centers for Disease Control and Prevention found more than
400 new genomic and other omics-based tests in transition
from bench to bedside, of which the vast majority are related
to cancer. 7 However, in 2012, few if any omics-based tests
have actually been widely adopted or embraced in the clinic.
Why not? There are several obstacles that block introduction
and use of an omics-based test. These relate to generation
and validation of tumor biomarker tests in general, but
in addition omics-based tests have special considerations
that have impeded progress in the field. 8 It is essential that
basic, translational, clinical, and computational scientists,
and importantly clinicians caring for patients with cancer,
understand these obstacles and work to overcome them so
that patients receive better, more personalized oncologic
care than they do now.
From the University of Michigan Comprehensive Cancer Center, Ann Arbor, MI; Epidemiology
and Genomics Research Program, Division of Cancer Control and Population
Sciences, National Cancer Institute, Bethesda, MD; Office of Public Health Genomics,
Centers for Disease Control and Prevention, Atlanta, GA; Departments of Medicine and
Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC.
Authors’ disclosures of potential conflicts of interest are found at the end of this article.
Address reprint requests to Daniel F. Hayes, MD, Breast Oncology Program, University
of Michigan Comprehensive Cancer Center, 6312 Cancer Center, 1500 E. Medical Center
Drive, Ann Arbor, MI 48109-0942; email: hayesdf@umich.edu.
© 2012 by American Society of Clinical Oncology.
1092-9118/10/1-10