Learning Center
Plans & pricing Sign in
Sign Out

Biomarkers and gene tests distinct disciplines now merging


									Biomarkers and Gene Tests - A Personal Insight by Dr Paul Debenham

Executive Summary

There is a bewildering array of biological markers that have been, and continue
to be, identified and associated with disease prediction and prognosis. Genetic
tests for gene mutations in well characterised biological processes are
increasingly obvious candidates to add into the panoply of tests available for this

The computation of well-being prediction based on biomarker data is becoming a
mature science, at the same time the increasingly manageable access to gene
analysis enables the partnering of all such information in the future. Thus genetic
information, even from unknown genes amongst micro-array data, may be
assimilated into calculations as just biomarker data points.

Historically genetic data has been synonymous with monogenetic disorders, the
low prevalence of which in the population at large has meant that such tests have
not had a significant value to the insurance industry. However an increasing
association with common disorders may change this stance. Further the
potentially extensive use of gene markers tests in the drive to make better, safer
drugs could bring many such tests to become commonplace and central to health
care, and thus of interest to the insurance industry.

As gene tests become just biomarker data points the insurance distinction
between predictive and diagnostic classification may need to be re-examined.

1. Biomarkers

Biomarkers have been defined in many different ways, for instance:
   (i)   A characteristic that can be measured and evaluated as an indicator of
         normal biological processes, pathological processes or pharmacologic
         responses to therapeutic interventions (NIH Biomarkers Definitions
         Working Group, 1998)
   (ii)  Any substance, structure or process that can be measured in the body
         or its products and can influence or predict the incidence of outcome of
         disease. (WHO International Programme on Chemical Safety)

In the context of a drug development application the definition can have a slight
change in emphasis:
    (iii) Any characteristic that can be objectively measured and evaluated as
          an indicator of biological processes (either normal or pathogenic), or to
          evaluate pharmacological response to therapeutic intervention.
          (Boguslavsky 2004)

Biomarkers are considered less relevant reference points compared to clinical
end points that reflect how a patient feels, functions or survives. However a
biomarker may become a suitable endpoint if it has gained the status of a
surrogate marker.

A surrogate marker is variously defined, but in essence is a marker that whilst not
necessarily directly linked to the event yet is considered to reflect the outcome of
importance, e.g. blood pressure is not directly linked to a heart attack event but it
may be used as a surrogate marker for this because it is a risk factor for
cardiovascular disease. A surrogate marker is expected to predict clinical benefit
(or harm or lack of benefit and harm) based on epidemiological, therapeutic,
pathophysiological or other scientific evidence. The NIH working group points out
that probably only a few biomarkers are likely to achieve a consensus surrogate
end point status. Perhaps the distinction between a biomarker and its elevated
status as a surrogate end point is the extent to which the observer can place the
emphasis on its role to indicate, or to predict, the biological end point under

A subtlety hidden within the definitions must be the empowerment achieved by
statistical analysis behind any specific association identified between the
biomarker and the clinical end point. The reality of any study is that there are
many more potential analytes than may be considered relevant for the biological
end point than can be effectively validated and measured. Thus the balance
between physiological endpoints employed such as heart rate, blood pressure,
bone density etc., and laboratory analysis end points such as blood glucose,
cholesterol, calcitonin etc. not only reflect the art of the possible but the practical
variability and accuracy associated with those measurements.

The potential multiplicity of biomarkers is bewildering as it includes enzymes,
structural proteins, protein breakdown products, peptides, amino-acids, vitamins
and derivatives, enzyme and co-factor conjugates, antibodies, lipids, lipoproteins,
hormones, coagulation factors, sugars, endogenous compounds and their
metabolites, nucleotides and nucleic acid intermediates, and micro-organisms
etc. Of course genetics adds a new layer of complexity with 30,000 genes and
their variations, transcription splice variants etc. Nonetheless genomics and
transcriptomics are on the new agendas for biomarkers.

Historically the choice of biomarkers to study to predict an aspect of health has
been influenced by the current scientific knowledge in the field. This process is
therefore biased and hypothesis-bound and potentially misleading. As an
example there has been recent published debate about the utility of screening for
Prostate Specific Antigen (PSA) as a surrogate for prostate cancer. This
association has held sway since the late 1980s but is now questioned as whilst
there is a correlation between PSA levels and the presence of cancer it appears
more informative when monitored through serial screening than one-off analysis.
PSA is a surrogate marker for benign prostatic hypertrophy and inevitably has

caused unnecessary biopsies and is now recognised as an imperfect cancer
marker. It can be argued that a non-directed approach, capturing new datasets
may be a valuable methodology for complex disease. However this rather hands
over well-being prediction to the power of an algorithm that sits in some
interested party's computer somewhere.

2. The more the merrier

For any study of well-being, forecast of future health, or disease state there are
probably minimum health biomarkers that we all provide to our doctors in general
practice, let alone for specific study. This may include age, height, weight,
gender, blood pressure, smoking, allergenic status, current medication and
exercise. But of course family history, and thus genetics is captured indirectly as

Within studies related to disease sensitivity or progression the drive is to increase
the evidenced-based statistical models to better predict the onset and
progression of chronic and preventable disease every bit as much as for cancers.
Historically a single biomarker would be studied to generate a range of relative
risk factors for disease and outcome. Increasingly a number of these risk
estimates from different relevant biomarkers can be combined to generate a
multi-variable prediction. Thus for coronary heart disease (CHD) the analysis of,
for example, exercise, fibrinogen, albumin, lipoprotein A, and homocysteine from
various different studies would be combined to give a variable concentration
range for each biomarker and its associated risk.

These ‘composite’ biomarkers are claimed to predict a disease state, or a drug's
effect on disease, through the effects on a panel of disease-related markers.
Thus the strategy proposed to evaluate the efficacy of, say, a statin in preventing
CHD would be to argue that a composite biomarker analysis is more accurate
than any single biomarker. Composite biomarkers are possible with sufficient
input data and acceptable assumptions. Critical assumptions may include that all
the biomarkers are essential to the causal pathways to disease, and that the
relative weightings of the component biomarkers accurately reflect their biological

The use of composite biomarker panels coupled to statistical approaches are
claimed to combine as a new tool that can improve the discrimination power,
precision and efficiency of underwriting practice. By detecting subtle changes in
the physiology that precede disease the approach is argued to predict the onset
of chronic diseases by up to 10 years in advance. Models exist for CHD, stroke,
type 2 diabetes, colon cancer, lung cancer, breast cancer, prostrate cancer,
osteoporosis and chronic obstructive pulmonary disease. Underwriting remains a
mixture of subjective analysis and numerical rating systems but one can imagine

that the new composite biomarker approach is attractive if it can offer more
consistency and accuracy in risk classification.

It is unclear whether there is any meaningful distinction in the nature or type of
biomarker that might be used as a predictor of disease and those that may be
used as markers for drug efficacy. Simplistically if, for example, the concentration
of a particular biomarker in the blood is tightly correlated with risk of heart
disease, then the lowering of that same biomarker’s concentration might be a
suitable target point to assess the efficacy of a new drug for the treatment of that
disease. The insurance industry could equally use that biomarker concentration
as a predictor of patient risk, or as an indicator of a suitable medicinal regime.
That being said, there are biomarkers that are of interest to the insurance
industry that are not directly associated with conventional disease or its
treatment. Thus the measurement of nicotine or cotinin from oral saliva is used
as a marker for smoking cessation, or lack thereof. Other tests may look at IgG
antibodies to HIV in saliva, or for traces of drugs of abuse.

The long history of biomarkers for disease prediction has evolved to a large
extent completely independently of that of genetic analysis. However the
blueprint behind the family history must become increasingly relevant as we
become competent at teasing out the detail. With the completion of the human
genome sequencing in recent times, and the
co-evolved technology that can now capture this diversity in tangible approaches,
it is to be expected that genetics will start to contribute more significantly to the
understanding of disease prediction.

3. Current gene markers and insurance

Amongst those who die before their 65th year inherited disorders are the fifth
most frequent cause of death. Most of such deaths result from inherited heart
disorders followed by anomalies of the central nervous system as well as
urogenital and gastrointestinal anomalies.

Terminology can be confusing with respect to changes in DNA sequence. Any
change in the DNA sequence from an accepted reference norm is termed a
polymorphism, although a subset called single nucleotide polymorphisms (or
SNPs) require any such change to be commonly observed (greater than 1%).
Only polymorphisms causing a disruption to a gene’s function, or to the protein it
codes for, are considered as mutations. The consideration of any newly
discovered DNA sequence polymorphism in a gene with respect to whether it is a
mutation or not requires either, or a combination of, molecular modelling, clinical
data or disease association in family studies. Thus as biomarkers for disease the
emphasis has been traditionally placed on mutations with an established

At least 6500 genes are known in which mutations lead to monogenetic
disorders. These range from the disorders affecting lifestyle such as colour
blindness, or short-sightedness, scaly skin, albinism and deafness through to the
life threatening syndromes predisposing or causing heart disease, muscle
atrophy and cancers. Currently of the order of 1000 genes have molecular
genetic analyses possible on these diseases. However only a small number of
genes have common mutations associated with predictable disease progression.

Cystic Fibrosis (CF) is often cited as an example with one mutation in the CFTR
gene, the delta 508 mutation, being the basis of CF in ~75% or more of cases in
white Caucasians. Commercial tests available for CF screening have been
designed to detect this mutation and around 30 to 70 less common mutations.
However approximately 1390 different mutations have been recorded in the
CFTR gene to date. Thus in theory the application of one of the commercial test
regimes will miss illness causing mutations. Full gene sequence analysis is
currently still considered time consuming and expensive. Very few genes are
sufficiently well characterised to the extent of cataloguing all observed mutations
with clinical prognosis or drug-responsiveness.

This lack of current knowledge about mutation and disease prognosis means that
that it remains difficult to predict the value of a genetic screen or test in most
cases. Within the current knowledge base one sees Huntington’s Disease
associated with the specific expansion of a triplet DNA base sequence in the
gene. Conversely, whilst BRCA 1 and BRCA 2 gene mutations are associated
with predisposition to breast and ovarian cancers, without family association to
the disease one would need to resort to molecular modelling to predict the
consequence of any new mutation identified. Within a family there may be
sufficient family history to provide a one-off association between a specific
genetic change and the disorder but these familial cases represent only a sub
group of the total potentially affected population.

 Statistical analysis from population data however may provide sufficient insight
that the technical validity of any test may not need to be 100% for it to be worth
considering from an insurance perspective. An often-cited example is the
mutation in the HLA DP B1 gene associated with occupational immune-mediated
chronic beryllium disease (CBD). The specific position 69 mutation has high
sensitivity (~94%) but a lower specificity (~70%) with respect to CBD amongst
individuals occupationally exposed to respirable beryllium. Such a test could be
used as a required job placement screen, or to reduce the frequency of
lymphocyte proliferation tests, or as a voluntary job placement screen.
Interestingly despite the poor specificity of the test one probabilistic analysis
considered that such a test has significant utility in avoiding cases of CBD on a
value-for-information base. This analysis was in terms of risk management and
cost reduction programmes and not in the context of the ethical considerations
and instigation of appropriate health and safety precautions.

Many commentators on the current use of genetics and health/employment
insurance quote the Burlington Northern- Santa Fe Railroad company that tested
employees for a deletion in the peripheral myelin protein 22 gene associated with
carpal tunnel syndrome. The apparent intention was to try and identify a
predisposition and avoid compensation for workers who develop this syndrome.
This dispute was brought before the USA’s Equal Employment Opportunity
Commission who found in the workers favour.

In Germany the Kaufmannische Krankenkasse insurance company offered all its
~2 m clients haemochromatosis mutations test so that they could receive early
treatment for the disease. 67 customers of the 4000 that took up the offer were
told they were at high risk of the disease. In this case the results were only given
to the clients and the results were not provided to the company.

More generally, there is limited information about genetic tests offered by
insurance companies to their clients, or companies to their employees. An
Australia Institute of Actuaries report surveying two years (2000 -2002) of genetic
testing in the country reported a total of 235 applications with the majority being
for haemochromatosis (170), then Huntington’s (22), breast cancer (10), CF(8),
Factor V Leiden (5) and then a few tests for each of myotonic dystrophy, Familial
AdenomatousPolyposis, colorectal cancer, polycystic kidney disease, and single
tests for Marfan Syndrome, Non Polyposis Colorectal Cancer, Muliple Endocrine
Neoplasia, Charcot Marie Tooth Disease, Prothrombin gene mutation,
Epidermolysis Bullosa, Tay Sachs Disease, Spinocerebellar Ataxia and Tuberous
Sclerosis Complex.

Given the near-religious zeal of anti-genetic testing campaigners to scrutinise the
life insurance sector it is unlikely that extensive genetic biomarker testing is being
undertaken and hidden from the public eye. Thus, despite the significant number
of potential genetic tests that could be employed at this time, it would seem a
deliberate strategy to not utilise, or cautiously use genetic biomarkers so far. A
sense of realism prevails that there are only a small number of highly predictive
genetic tests that might be of utility. Further it implies that either the power of
probabilistic calculations cannot sufficiently rectify any shortfall in predictive
accuracy for many of the tests, or that the financial or ethical framework for life
insurance, or employment insurance has not justified such an approach to date.
Importantly most opportunities for genetic testing in the population - mostly
associated with family planning, or limited by acceptance criteria for cancers –
are still only rarely taken. Thus the anti-selection potential for the insurance
sector, in which the individual takes insurance after an undisclosed test, is
currently low.

However one can expect things to change. A growing discussion with respect to
the value of newborn screening programmes heightens the potential for these
tests to figure in future life history disclosures. In addition, and perhaps of more
immediate impact is the growing move to involve genetic analysis in drug

development programmes. The expected consequence is that we are
approaching a future in which the drug prescription will be associated with a
diagnostic test to establish efficacy and/or avoidance of severe adverse reaction.
The weight of exploratory genetics being undertaken by the pharmaceutical and
academic communities associated with drug discovery should create drug-gene
mutation linkages that will transform the public application of tests. The
occurrence of such tests, their prevalence and their impact on well-being could
radically change the insurance perspective on genetic tests.

4. Drug discovery gene tests

In March 2005 the US Food and Drug Administration published its long awaited
Guidance for Industry on Pharmacogenetic Data Submissions. This Guidance
clearly signals the FDA's interest to take account of pharmacogenomic data but
starts from the premise that this science is relatively new and that aspects of its
methodology may mean that for certain cases the information is not yet suitable
for regulatory decision making. However, it declared that some pharmacogenetic
tests, primarily those associated with drug metabolism, have well accepted
mechanistic and clinical significance suitable for integration into drug
development decision making. These tests are defined in the Guidance as known
valid biomarkers and additionally it introduces a less well-defined category of
probable valid biomarkers, and then reflects on the broader field of exploratory
pharmacogenomics. The probable valid biomarkers are referred to as where
there is an available body of evidence to support clinical relevance but this has
not as yet been widely accepted or validated in the scientific community. Often
presentations in this field will show lists of twenty or so CYP P450 and 'phase II'
liver metabolism genes, and these may well constitute the 'probable' range along
with other well cited receptor or drug uptake gene systems.

When it comes to the more exploratory pharmacogenomics analyses this may
often relate to data generated from micro-arrays scanning large number of gene
markers for change in expression co-ordinating with disease progression, drug
response etc. There remains some debate about the accuracy, reproducibility
and comparability between different array systems. Thus the validity of such
exploratory data may well need to have defined methodological and control
specification, let alone identification of the pertinent gene loci that provide the
diagnostic information.

Thus, whereas the life insurance approach to genetic tests has been defined by
utility to predict a clinical end point, the drug discovery utility of a gene test (or an
expression signature of a multiplex array of gene loci) is wider including the
prediction of a synthetic compound’s ability to influence a transition in a clinical
process, or the avoidance of adverse reactions to a new chemical entity. These
are not conventional biomarker endpoints and as such open up the potential list
of relevant genes for scrutiny in an unlimited and haphazard way. The ability of a

genetic test to predict the efficacy of a drug will lead to a requirement for the
typing to be undertaken to ensure the best patient prognosis. Equally the ability
of a genetic test to predict the avoidance of an adverse reaction for a drug may
soon be required so that doctors are seen to not be liable for inappropriate care
in their prescription management. It is claimed that approximately 10,000 deaths
a year in the UK are related to adverse drug reactions and even that it is the
fourth biggest killer from a meta-study of deaths in US hospitals. One can
imagine therefore that these tests could actually constitute tests that the life
insurance industry would encourage their clients to consider.

Inevitably some of these pharmacogenetic and pharmacogenomic tests will
involve gene mutations that have links to apparently unrelated disease or
biological trait. It may be that certain mutations might simultaneously predispose
the patient to cancer from smoking, or behavioral alterations or some late onset
ill-heath, etc. The future prevalence of these tests and their possible link to
disease might upset the anti-selection balance that has kept the insurance
industry relatively quiescent to date.

5. Gene information with biomarkers

The drive to ever-more accurately define clinical end points must inevitably take
advantage of the combination of gene tests and biomarkers. Thus, whereas P53
protein has been a marker for cancer for over ten years and variants are
associated with about 50% of cancers, or the KIT tyrosine kinase is associated
with a number of cell malignancies etc, the specific screening of genes
associated with the cancer in question should identify specific genetic changes
that can be used to predict the specific nature of the disease. Used in
combination with databases, developed over time to link genetic changes to
clinical prognosis or cancer markers etc, the genetic biomarkers will become
invaluable for prediction of disease and thus essential to be undertaken as part of
the study.

Another example may develop from research into the performance of statins to
lower cholesterol and other lipid chemistries which is identifying particular gene
variants that affect statin efficacy. Thus the measurement of the conventional
biomarker, such as cholesterol, with respect to cardiovascular disease could in
the future be coupled with gene information in order to best tailor the dosage and
type of treatment to be utilised.

In essence one can expect that the systematic combination of biomarkers and
gene tests, accompanied by good databasing of illness progression and clinical
outcome, will dissect the broad clinical definition of any disease into many sub-
sets, each potentially with a specific strategy of treatment.

Through the progression and acceptance of genomic micro-array analyses one
can foresee that a key marker for a disease state or prediction may be a gene

expression change in an essentially unknown gene, observed perhaps as a
change in intensity of a fluorescent spot in a gene micro-array. Whilst the
understanding of what this gene exactly does may take years, its use as a pivotal
maker may advance far more rapidly than the understanding of its relevance, let
alone its association with any other human trait. Thus the past emphasis on
understanding the gene and its function before placing weight on its use may be
superseded by the unknown gene’s use as a reliable marker.

Importantly the application of genetic studies to complement those traditionally
predicting disease through biomarkers alone will provide individual patient
perspectives rather than just generalised disease stratification. However such
analyses may also track genes and the disease into the family group, whereas
previously the use of the biomarker alone and the family history record may be
insufficient to provide such information. Thus whilst a blood pressure test or trend
does not raise ethical issues for the patient’s relatives, the investigation of his/her
Factor V gene for mutations associated with thrombosis certainly does. These
are ethical issues that have dogged the use of genetics, but would now need to
be considered in many studies of well-being if the use of genetic tests became
commonplace alongside conventional biomarkers.

6. The future

There are two parallel tracks of analytical development in play at this time which
may influence the insurance agenda in the mid term future, say within the next
five years. Firstly, whether driven by the FDA expectation, or public pressure,
there will be an increasing use of specific pharmacogenetic tests for the public at
large linked to disease diagnosis and efficacy of treatment. These will be
increasingly associated with common disorders such as heart disease. These
tests will not just be about well characterised mutations in drug metabolising
gene activity, but will also involve gene mutations regularly associated with
disease progression. In the context of pharmacogenetics it is unclear whether
such tests will be considered predictive or diagnostic and that may be a definition
that is pivotal with respect to insurance access to such data.

In the UK a moratorium exists on predictive genetic tests until November 2011.
However other genetic tests used to confirm diagnosis are considered equivalent
to other clinical technologies. In this context Insurers are permitted to seek, with
customers consent, access to certain diagnostic genetic test results along with
other clinical and family data.

Secondly the capabilities of the biomarker algorithms will continue to improve at
predicting disease outcome from an increasing number of biomarkers in
combination. In this context unidentified gene markers in arrays associated with
illness or disease progression may become key biomarkers. These exploratory
gene markers, or gene expression markers, may even become linked to a
particular medication monitoring requirement. If the status of these tests is

categorised as a biomarker is it therefore considered as just clinical data for
insurance purposes?

On a longer time scale there will develop a detailed association between the
multitude of specific mutations in any particular gene, their clinical consequences
and their responsiveness to particular drug therapies. Within any gene it is
probable that certain mutations will have high predictive value irrespective of
other genetic or environmental factors, whilst other mutations are only weakly
predictive of disease or prognosis. Thus the eventual complexity of biomarker
and genetic information could leave the assessment of risk in the hands of the
truly empowered and motivated, that is the insurance industry.

The insurance industry to date has not acted generally in a rash manner and
presently indicates it is too early to determine the implications of future genetic
capabilities. The perhaps inevitable nature of the developments described in this
paper would imply that no-one can remain complacent.

Paul Debenham
August 05

Post script
This paper is written as a personal, rather than a professional, insight into the
complex interplay of biomarkers, illness and insurance.


To top