Personalized Medicine - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 39

a b

Fast caffeine
metabolizer

IFN responder

Sensitive to
warfarin

At risk for
diabetes

Adverse reaction
to Tamoxifen

Fig. 3.4 Personalized medicine: ( a ) A group of individuals may appear homogeneous upon mac-
roscopic observation through traditional methods. ( b ) Novel molecular assays enable discernment
of underlying physiological differences. This stratifi cation can inform decisions regarding life-
style, disease prevention, and therapeutic interventions (Reprinted from [ 13 ], used with permission
from Springer)

3.3.2.1 Predictive vs. Mechanistic Biomarkers

Different players have different motivations for probing underlying biological state
through the use of biomarkers, and different types of biomarkers may be used for
different purposes. Health care providers are primarily interested in predictive bio-
markers. The biomarker itself may be causal for a disease or phenotype, or simply
correlated with the phenotype because both are physiologically downstream of the
actual cause. As long as the marker can be used as a reliable indicator for, e.g. diag-
nosis, prognosis, or therapeutic response, it can be useful from a clinical perspec-
tive. Mechanistic biomarkers, in contrast, are useful to researchers to devise methods
for intervention and prevention. Understanding underlying disease mechanism can
help identify causal pathways, which may in turn suggest new directions for
hypothesis-driven research or new leads for drug targeting.

3.3.2.2 From Statistical Signifi cance to Clinical Utility

In any discussion of biomarkers, it is critical to recognize the differences between
statistical and clinical signifi cance, analytic and clinical validity, and clinical utility.
Statistical signifi cance is a measure of confi dence that a fi nding from a statistical test
is actually true. Some common statistical tests performed in the biomedical context
include the t-test, Chi-squared test, or analysis of variance (ANOVA). In each case,
the test essentially asks whether different sets of values likely came from the same
distribution, or different distributions. Statistical signifi cance is often expressed in
the form of a p -value, which represents the probability of being wrong if one were
to conclude that the distributions are indeed different. (Drawing this conclusion is
commonly described as “rejecting the null hypothesis”, i.e. the hypothesis that there