Personalized Medicine - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 40

is no difference between the groups.) A p -value of 0.05 suggests that the observed
results could be attributed to chance only 5 % of the time [ 15 ]. The value of 0.05 is
an arbitrary, but commonly accepted, threshold below which a test is said to be sta-
tistically signifi cant . Typically, the more samples one uses to make a determination,
of signifi cance the more statistically signifi cant a fi nding will be. Clinical signifi -
cance, on the other hand, refl ects whether that difference has any impact on clinical
care. As an example, one might hypothesize that a given genotype confers increased
risk for heart attack. If, hypothetically, one were to test that gene in one million cases
and one million controls, it may be shown that individuals with the genotype in
question were 1.1 times as likely as those without it to have a heart attack. With such
a large sample size, the p -value for this fi nding could be <0.00001, a statistically
signifi cant result. However, from a health care perspective, it changes nothing. This
hypothetical test thus lacks clinical signifi cance.
Analytical validity refers to the accuracy and reliability of a test itself, e.g. how
well a given test predicts the presence or absence of a particular genetic change.
Criteria for analytic validity include not only sensitivity and specifi city of the test
itself, but also factors such as reproducibility, quality control, and limits of quantita-
tion [ 16 ]. Clinical validity, on the other hand, is a measure of the degree to which the
test result refl ects the presence, absence, or risk of disease. Clinical validity relies on
analytical validity, but also incorporates disease penetrance and prevalence and the
concepts of positive and negative predictive value, i.e. if the test indicates a person has
the disease, how likely is it that the person actually has the disease, and vice versa.
Clinical utility is related to the concept of clinical signifi cance. It is a measure of
how useful that information actually is in informing clinical care. A biomarker test
may be 100 % accurate, and perfectly indicative of a given disease, but if no treat-
ment for that condition exists, the test lacks clinical utility. Alternatively, a test may
be suggestive for specifi c treatment course, but cost of treatment or severity of side
effects may outweigh that recommendation in light of uncertainty. In addition, clini-
cal utility must be evaluated in the larger clinical context. An advanced biomarker
is only useful if it provides additional information beyond what could be gleaned
through standard, more easily obtained, observations. For example, in order to have
clinical utility, a molecular biomarker for risk of heart attack would need to be more
accurate than a standard clinical model that takes into account BMI, smoking status,
family history, etc.

3.3.3 Stratifi cation

Biomarkers can be used to stratify patients along a number of different axes. This
stratifi cation may or may not be actionable. For example, some people respond well
to certain drugs while others do not respond at all, or worse, have an adverse event.
Knowing which of these groups a patient belongs to can help inform pharmaceutical
intervention. Biomarkers may also help to group people by diagnosis where macro-
scopic observations cannot differentiate. For example, it can be diffi cult to know
whether fl u-like symptoms are caused by a virus or a bacterial infection, but if