the diagnostic technology. In some cases, cost-effectiveness studies may not even
be necessary due to very small prevalence of a certain variation. This can be
illustrated by the example of Annemans et al. ( 2013 ) showing that if the prevalence
of HER2-positive tumors among breast cancer cases is known to be very low, then
the most cost-effective strategy will be not to use expensive treatment
(trastuzumab) since this yields the smallest net monetary benefit. If, on the other
hand, the prevalence of the characteristic is known to be high, then the most cost-
effective option will be to treat all patients with trastuzumab, even without testing
(hence no cost-effectiveness study of the companion diagnostic is necessary). If the
prevalence is between 10 % and 90 %, the “test-first” strategy may be the most cost-
effective, and it is reasonable to invest in determining the relative cost-effectiveness
of the companion diagnostic test. However, testing itself is neither without cost nor
without problems. The choice of the strategy used to identify HER2-positive
patients could significantly impact the predicted cost–effectiveness of trastuzumab
therapy if the rates of false-positive and false-negative diagnoses are high.^34 In fact,
retesting in the case of false positive or negative tests may represent additional costs
that may not even justify the incremental gain of accurate diagnoses. Hence, as
Ferrusi et al. ( 2009 ) argue, modeling the benefits and costs of testing procedures and
treatment options (whether combined or separately) should not ignore the influence
that inaccurate diagnosis has on the incremental cost-effectiveness of the test–
therapy combination, or the test itself, since this may lead to suboptimal conclu-
sions and decisions based on the results of economic evaluations.
Furthermore, to achieve its potential, personalized medicine will require large
amount of data and a vast network of infrastructure to manage, understand, and use
this data in a clinical setting—infrastructure such as biobanks that will contain
tissue specimens and link them to clinical outcomes. Many authors, such as Hewitt
( 2011 ) and Olson et al. ( 2014 ), point out that biobanking services must improve
rapidly to serve the needs of personalized medicine since personalized medicine
relies heavily on the technical requirements due to a large amount of evidence that
needs to be amalgamated as to produce meaningful conclusions from companion
diagnostics. This evidence includes the knowledge about the genotypic and pheno-
typic relationships (preferably causal relationships), clinical outcomes and bio-
marker information that need to be connected together and linked to public
databases and administrative datasets. Connecting complex sources of information
may allow researchers to better understand and assess the clinical importance of
genetic variation and testing for genetic variation in a vast array of diseases and
drug response. These processes are on the way. For example, as Hamburg and
Collins ( 2010 ) describe, the US National Institute for Health supports genome
analysis in participants in the Framingham Heart Study, obtaining biologic speci-
mens from babies enrolled in the National Children’s Study and performing
detailed genetic analysis of 20 types of tumors to improve our understanding of
their molecular basis. Harvard Medical School and Partners HealthCare System
(^34) Further details available in Ferrusi et al. ( 2009 ), pp. 193–215.
Economic Evaluations of Personalized Health Technologies: An Overview of... 123