Design and evaluation of a pharmacogenomics information ... · types uncertain of their ability to...
Transcript of Design and evaluation of a pharmacogenomics information ... · types uncertain of their ability to...
Research and Applications
Design and evaluation of a pharmacogenomics
information resource for pharmacists
Katrina M Romagnoli,1 Richard D Boyce,1 Philip E Empey,2 Yifan Ning,1 Solomon
Adams,2 and Harry Hochheiser1,3
1Department of Biomedical Informatics, School of Medicine, University of Pittsburgh, Pittsburgh, PA, USA, 2School of Pharmacy,
University of Pittsburgh and 3Intelligent Systems Program, University of Pittsburgh
Corresponding Author: Katrina Romagnoli, 5607 Baum Blvd, 417D, Pittsburgh, PA 15206, USA. E-mail: [email protected].
Received 4 August 2016; Revised 20 September 2016; Accepted 11 January 2017
ABSTRACT
Objective: To develop and evaluate a pharmacogenomics information resource for pharmacists.
Materials and Methods: We built a pharmacogenomics information resource presenting Food and Drug Admin-
istration (FDA) drug product labelling information, refined it based on feedback from pharmacists, and con-
ducted a comparative usability evaluation, measuring task completion time, task correctness and perceived
usability. Tasks involved hypothetical clinical situations requiring interpretation of pharmacogenomics informa-
tion to determine optimal prescribing for specific patients.
Results: Pharmacists were better able to perform certain tasks using the redesigned resource relative to the Phar-
macogenomic Knowledgebase (PharmGKB) and the FDA Table of Pharmacogenomic Biomarkers in Drug Labeling.
On average, participants completed tasks in 107.5 s using our resource, compared to 188.9 s using PharmGKB and
240.2 s using the FDA table. Using the System Usability Scale, participants rated our resource 79.62 on average,
compared to 53.27 for PharmGKB and 50.77 for the FDA table. Participants found the correct answers for 100% of
tasks using our resource, compared to 76.9% using PharmGKB and 69.2% using the FDA table.
Discussion: We present structured, clinically relevant pharmacogenomic FDA drug product label information
with visualizations to help explain the relationships between gene variants, drugs, and phenotypes. The results
from our evaluation suggest that user-centered interfaces for pharmacogenomics information can increase
ease of access and comprehension.
Conclusion: A clinician-focused pharmacogenomics information resource can answer pharmacogenomics-
related medication questions faster, more correctly, and more easily than widely used alternatives, as perceived
by pharmacists.
Key words: pharmacogenomics, structured product labeling, pharmacist, evaluation, usability, clinical decision making
BACKGROUND AND SIGNIFICANCE
As pharmacogenomics-guided drug decision-making becomes more
common, pharmacists seek trustworthy, clinically actionable pharma-
cogenomics information to guide their decisions.1–4 Using information
about the interaction between genotypes and medications to inform
medication and dosage choices can optimize treatment and prevent ad-
verse medication events,5 yet that knowledge is not regularly applied.6,7
Insufficient pharmacogenomics training, complicated terminology, and
a rapidly moving field complicate the task, making clinicians of all
types uncertain of their ability to make informed decisions about medi-
cations with pharmacogenomics implications.7–9
Consequences of not knowing a patient’s genotype and resulting
phenotype can be severe, even fatal, as in the death of a breastfed
newborn whose mother was an ultra-rapid metabolizer of codeine,
which caused her breast milk to contain toxic levels of codeine’s
VC The Author 2017. Published by Oxford University Press on behalf of the American Medical Informatics Association. All rights reserved.
For Permissions, please email: [email protected]
1
Journal of the American Medical Informatics Association, 0(0), 2017, 1–10
doi: 10.1093/jamia/ocx007
Research and Applications
active metabolite, morphine.10 The Food and Drug Administration
(FDA) now requires the boxed warning section of clopidogrel’s
product label to warn that poor metabolizers of clopidogrel may not
be effectively treated by the drug and may have a higher risk of heart
attack, stroke, and death than normal metabolizers,11 implying that
genetic testing may be appropriate prior to prescription. With the
help of genetic tests, clinicians can ideally prescribe the right dose of
the right drug at the right time to the right patient. In some cases,
such as with warfarin1 and clopidogrel,3 genotyping already informs
drug and dose selection in select hospitals.
However, medication-related genetic testing is still in its nascent
phase,9,12 and there is no consensus yet on the usefulness of pharmaco-
genomics vs other methods to determine drug choice or dosage,13–16
further complicating how to use genetic information. In 2011, the
Clinical Pharmacogenetic Implementation Consortium (CPIC) began
developing clinical guidelines to support biomarker-based clinical
decision-making in conjunction with the Pharmacogenomics Knowl-
edgebase and the National Institute of Health’s Pharmacogenomics
Research Network (www.pgrn.org).17 As pharmacogenomics be-
comes more integrated into standard practice, clinicians have an ever
greater need for clinically relevant, effectively presented, searchable
information that supports accurate interpretation and adoption of
treatment recommendations.18
Multiple efforts exist to support the integration of patient-
specific, clinically actionable pharmacogenomics information into
practice. Eleven sites in the Electronic Medical Records and
Genomics (eMERGE) Network are implementing pharmacogenom-
ics clinical decision support. A recent article found that while these
sites have experienced delays and barriers to implementation of
pharmacogenomics clinical decision support, these are generally sur-
mountable.12 The eMERGE group is developing an infobutton that
will write genomic medicine information to share genetic testing re-
sults with clinicians within the eMERGE network.19 Trustworthy,
clear pharmacogenomic information presented in a clinically useful
way and combined with patient data and local or CPIC practice
guidelines has the potential to move pharmacogenomics into clinical
practice.
As drug information experts poised to support the pharmacoge-
nomics information needs of other clinicians, pharmacists play a key
role in pharmacogenomics-informed drug decisions.20–24 Meeting
pharmacists’ information needs is critical to effectively incorporate
pharmacogenomics information into decision-making.22 Multiple re-
sources are available, including the FDA Table of Pharmacogenomic
Biomarkers in Drug Labeling25 and FDA labeling information in the
Pharmacogenomics Knowledge Base (PharmGKB),26 but none of these
resources are currently able to provide clinically relevant information
at the point of care. The FDA table lists the drugs that are mandated by
the FDA to contain relevant pharmacogenomics information, but it re-
quires a minimum of 4 nonobvious clicks to access a product label.
The user still has to scan through the entire product label to locate the
pharmacogenomic information, which is dispersed throughout the la-
bel in multiple sections. PharmGKB aims to provide comprehensive
pharmacogenomics information, with an emphasis on the CPIC guide-
lines,27 but the utility of practicing pharmacists using it to make pre-
scribing decisions has not been evaluated.
In previous work, we developed a semantic model of pharmaco-
genomics information in product labels and designed a preliminary
prototype interface to explore that information.28 We extracted
pharmacogenomics statements from drug product labels and pre-
sented that information in a prototype user interface, as a first step
toward the goal of providing clinically actionable pharmacogenom-
ics information at the point of care. We also conducted a series of in-
terviews with pharmacists to assess their pharmacogenomics
information needs and resource requirements.29
In this study, we redesigned our prototype pharmacogenomics
information resource based on the results of that assessment and
conducted a user study comparing it to the original prototype, the
FDA Table of Pharmacogenomic Biomarkers in Drug Labeling,25
and the FDA labeling information in the Pharmacogenomics Knowl-
edge Base.30 Here, we present the redesigned prototype interface
and the results of our evaluation.
MATERIALS AND METHODS
Version 1.0 designIn the initial version of the prototype, the top table provides infor-
mation about the gene variants, the phenotype, and whether the im-
pact is on the drug’s efficacy or toxicity. A pathway diagram
illustrates how the drug interacts with the gene product to influence
the drug-related phenotype (Figure 1). Below are tabs containing
pharmacogenomics information from the annotated structured
product labels (SPLs), and recommendations regarding how to
change the patient’s treatment (Figure 2).
Initial usability studyAs part of a previously published study,29 we asked 14 pharmacists
working in a variety of settings to use our initial prototype to answer
pharmacogenomics-related medication questions while we observed
their behavior. We recruited pharmacists through professional con-
nections of the authors, using a convenience sampling strategy de-
signed to represent the diversity of needs and perspectives associated
with different care contexts, including tertiary-care centers, commu-
nity practices, and private nursing home consulting pharmacies. We
used the results of this preliminary usability study to identify prob-
lems with the pharmacogenomics information resource interface.
These problems, along with the results of our inquiry into pharma-
cists’ perceived information needs, informed a redesign of the proto-
type. The study was approved by the University of Pittsburgh
Institutional Review Board.
Pharmacogenomics resource usability evaluationWe conducted laboratory-based user studies with 16 pharmacists,
comparing both versions of our prototype to potential alternative
sources of pharmacogenomics information contained in FDA drug
product labelling. The FDA table was selected as the authoritative
source issued by a key regulatory agency. PharmGKB was selected
based on its history as the most comprehensive and well-established
pharmacogenomics information resource providing pharmacoge-
nomics information from product labels.
Study participants completed 8 pharmacogenomics information
tasks that could be answered in all 4 information resources. These
tasks asked participants to consider drug and dosage choices appro-
priate for individuals with various genotypes, along with questions
regarding the applicability of testing. The tasks were designed to
cover a range of pharmacogenomics concepts that have been identi-
fied as important competencies for pharmacists by the American So-
ciety of Health System Pharmacists and the Pharmacogenomics
Education Program (PharmGenEd) at the University of California,
San Diego.31 The tasks and correct answers were written based on
information provided solely within the FDA product label for
each relevant drug, as that is the source of pharmacogenomics
2 Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0
information provided by all assessed resources. Four of the tasks in-
volved only 1 drug question, and 4 were multidrug questions. The
tasks are listed in Table 1.
The order in which tasks and resources were presented was var-
ied among participants using a pseudo-randomization technique, to
avoid order effects. The order of tasks was randomly assigned to the
first participant and was assigned to each subsequent participant
based on that participant, to ensure that each task was performed
on each resource an equal number of times. The order of resources
was assigned randomly to each participant.
The FDA table and PharmGKB resources were downloaded to
the study laptop to guarantee that task completion times would not
be dependent on Internet connectivity and network latency. We
used HTTrack Website Copier32 to download the FDA table (down-
loaded June 2, 2015) and the FDA Drug Labeling portion of
PharmGKB.org (downloaded June 5, 2015).
The pharmacists were instructed to use only the assigned re-
source for each task and not seek the information on any other
sites. We recorded their activity on the screen and their voices
simultaneously using SnagIt, a Chrome web browser extension
for screen capture.33 Participants were instructed to highlight the
text of the answer to each task on the screen when they found it.
Each task had a maximum completion time of 10 min. Tasks that
were not completed in this time were marked as incorrect. The
task completion time for incorrect and incomplete tasks was mea-
sured as the time it took to find the incorrect answer, give up, or
reach 10 min. After all 8 tasks were finished, participants com-
pleted the System Usability Score (SUS) questionnaire34 for each
resource, to assess perceived usability. Measurements included
the time to complete each task from start to finish, the correctness
of each answer as determined by the investigators, and the SUS
score for each resource.
Evaluations took place during July and August 2015 and were
conducted at the participants’ workplaces. The study was approved
by the University of Pittsburgh Institutional Review Board.
Data analysisData were analyzed using MiniTab 1735 and R version 3.2.3.36
Descriptive statistics were used to summarize participant demographics.
We performed a regression analysis to examine task completion
time vs task number to assess order effect. For task completion time,
we assessed the distribution of the data, which was lognormal. Due
to the lognormal distribution, we used the Kruskal-Wallis test, a
nonparametric test for assessing the variance of data, and the Dunn
test for nonparametric pairwise correction to analyze the signifi-
cance of task completion time, comparing the 4 resources across all
8 tasks. For the SUS scores, we assessed the distribution of the data.
Due to the normal distribution, we performed a one-way analysis of
variance (ANOVA) and Games-Howell test for post hoc pairwise
comparison to analyze the significance of SUS scores of all 4 re-
sources. Games-Howell test was used due to unequal variances
among the 4 resources. Finally, for the correctness of answers for
each resource, we performed a Chi-square test with a post hoc
Figure 1. Drug page of version 1.0
Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0 3
pairwise comparison Bonferroni correction. MiniTab was used for
all analyses except the Kruskal-Wallis and Dunn tests, for which R
was used. P values were considered significant at .05. We considered
3 types of incorrect answers: the user gave an answer that was
wrong, the user failed to find an answer before 10 min, and the
user gave up before finding an answer but before the 10-min mark.
All of these are considered incorrect, as they are a failure to find
the correct answer.
RESULTS
Usability issuesWe organized usability issues identified in our initial usability study
into 3 categories: layout, information display, and information. The
layout was problematic, because the accordion design of the section
containing the information from product labels seemed almost invis-
ible. Participants also often failed to notice the Drug Info Links ta-
ble. As a result, many of the pharmacists were unable to access
product label information, and they requested links to other sources
despite those links being available. While they liked the illustrations
of interactions between drugs and genes, they were often confused
by the table containing information about efficacy and toxicity, or
how that influenced the phenotype. Pharmacists were also confused
by the notation of the gene variants (“star” notation) and they
wanted clear recommendations. They were concerned about the
source of the information, as it was not clear to them that all the
statements originated from FDA-approved product labels. Finally,
pharmacists also wanted more general background information
about pharmacogenomics, specifically about the drugs and genes, to
understand the information presented to them.
Figure 2. Expanded FDA label annotations section of PGx@Pitt version 1.0
4 Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0
Version 2.0 redesignThe problems with version 1.0 of the prototype identified in the ini-
tial usability study and the changes made in response are summa-
rized in Table 2.
The most prominent change to the drug pages (Figure 3) was
combining the pathway portion of the page with the table to form a
visual map, leading users from interactions between drugs and vari-
ants to information about phenotypes, and finally to FDA recom-
mendations. This change was made in response to pharmacists’
difficulty in interpreting the table with information about genotype
and phenotype. Specifically, our participants requested assistance
with interpreting the relationship between genotype, phenotype, and
medication recommendations. The visual map was designed to pro-
vide a clear and predictable structure of this information for all
drugs, including drugs like warfarin that have multiple relevant bio-
markers (Figure 4). The tabs were reordered, placing boxed warn-
ings and recommendations first. The remaining information was
provided in the second tab, organized by product label section. This
layout was intended to draw attention to the most relevant informa-
tion first, while also providing access to the label information in the
second tab. The box including links to other resources, including to
the original product label used for the annotation, was modified to
be visible at all times.
Usability evaluationSixteen pharmacists representing a variety of clinical practice envi-
ronments, with experience ranging from 2 months to >20 years,
participated in the usability evaluation. Most of them were clinical
pharmacists (9). Of those, a majority (5) were nursing home consul-
tant pharmacists (Table 3). All pharmacists who agreed to partici-
pate completed the evaluation.
Task completion time
To verify that the order of presentation was not a factor in task
completion, we confirmed that task completion time and order
were not correlated (r2¼0.02, P¼ .08). Task completion times
followed a lognormal distribution, requiring the use of Kruskal-
Wallis instead of ANOVA. Kruskal-Wallis results indicated signif-
icant differences between the resources (Figure 5) (Kruskal-Wallis,
P¼ .001). Search tasks were completed more quickly with the ver-
sion 2.0 resource (mean 107.5 s) as compared to both the FDA ta-
ble (mean 240.2 s; Dunn, P¼ .003) and PharmGKB (mean 188.9 s;
Dunn, P¼ .0087). Search tasks using the original PGx 1.0 proto-
type (mean 143.3 s) were completed significantly faster than the
FDA table (Dunn, P¼ .047). All other pairs were not statistically
significantly different.
Task correctness
A Chi-square test with a Bonferroni correction showed a difference
among the resources in rate of task correctness (N¼128,
v2¼13.393 [df¼3], P¼ .004). Version 2.0 performed statistically
significantly better than all other resources (P¼ .006) and was the
only resource in which 100% of answers were found correctly
(Table 4). Of the incorrect answers (22), 2 involved timeouts (maxi-
mum 10 min), 6 involved participants giving up, and in the remain-
ing 14, they found answers they incorrectly believed were correct.
System usability scale scores
The SUS score data were normally distributed, allowing for use of
ANOVA. Significant differences in SUS scores exist between the var-
ious resources (Figure 6) (ANOVA, P< .0005). As variances were
not equal, Games-Howell was used for post hoc analysis.37 SUS
scores were higher for version 2.0 (mean SUS score 79.62) than ei-
ther the FDA table (mean SUS score 50.77; Games-Howell,
Table 1. Pharmacogenomics Evaluation Tasks
Task Answer
1. A new patient with a prescription for warfarin is genotyped for all major pharmaco-
genes. Results indicate that he is CYP2C9 *2/*3, VKORC1 G-1639A AA. Using this re-
source, what starting dose is indicated for this patient, based solely on the genotype?
0.5–2 mg, depending on other factors like weight, age,
etc.
2. A physician has a patient with rheumatoid arthritis. She has had genotyping performed
for all major pharmacogenes, and results indicate she is heterozygous *3A (1 copy of
*3A, 1 copy of *1) for TPMT. The doctor wants to know if she can prescribe azathio-
prine for this patient.
Consider lower dose
3. Why should a patient with a CYP2C19 genotype other than *1 limit his exposure to cita-
lopram to 20 mg/day?
Poor metabolizers are at risk for QT prolongation
4. A postpartum breastfeeding mother tried to pick up her medications after being released
from the hospital. The community pharmacist refused to fill the prescription for codeine
because of a possible genetic issue. Why is the community pharmacist refusing to fill a co-
deine prescription for this patient?
Codeine should not be given to breastfeeding mothers
due to the risk of neonate morphine overdose if the
mother is an unknown ultra-rapid metabolizer
5. A new patient is genotyped for all major pharmacogenes. Results indicate he is a
CYP2D6 poor metabolizer. The patient is taking lisinopril, citalopram, metformin, war-
farin, and zolpidem. Which of those drugs, if any, are affected by CYP2D6 genes?
None
6. If a patient is positive for HLA-B*5701, which of these drugs can she take: carbamaze-
pine or abacavir? Both? Neither?
She can take carbamazepine, but not abacavir
7. A cardiologist wants to know if genetic testing is available by the FDA prior to initiating
cardiology drugs he frequently prescribes, such as clopidogrel, warfarin, atorvastatin,
and prasugrel
Clopidogrel and warfarin
8. A patient who is predicted to be a CYP2D6 extensive (*1/*1) metabolizer based on geno-
type alone is taking fluoxetine. If her physician is considering adding aripiprazole, what
is the recommendation regarding dosing?
50% dose reduction
Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0 5
P¼ .001) or PharmGKB (mean SUS score 53.27; Games-Howell,
P¼ .013), and higher for version 1.0 (mean SUS score 75) than the
FDA table (Games-Howell, P¼ .01). All other pairs were not statis-
tically significantly different. SUS scores above 68 (on a scale of 0–
100) are considered above average, while scores below 68 are con-
sidered below average.
Pharmacists’ reactions to pharmacogenomics information resources
Participants in both the qualitative inquiries and the evaluations in-
dicated that while the information provided by our pharmacoge-
nomics resource was important, a stand-alone web interface would
not be adequate for effective use of pharmacogenomics information
in the clinic. Ideally, they would prefer a resource that is integrated
into the electronic health record (EHR) and connects actual patient
genotype data with FDA-approved drug label information. They
also felt that the information provided by the FDA did not always
give them satisfactory recommendations on how to proceed with
specific patients. Recommending that a dose be lowered is not as
helpful as recommending a specific dose; recommending that an al-
ternative drug be given is not as helpful as recommending a list of al-
ternatives. Furthermore, for many drugs, other nongenomic factors
have a significant impact on the optimal dose for a given patient,
such as age, weight, renal function, and other clinical factors. As a
result, pharmacists felt that the information and recommendations
provided by the FDA product labels were insufficient without
patient-specific clinical data and recommendations based on that
data in addition to genotype data.
DISCUSSION
Clinicians of all types struggle with pharmacogenomics informa-
tion.7,9,21,38,39 To make a decision regarding the applicability of
drugs for patients, clinicians must incorporate information about
genes, how they interact with drugs, the predicted phenotypes of
given gene variants, and the phenotypes associated with those vari-
ants with respect to affected drugs.39 Interpreting this information
presents multiple challenges, including unfamiliar and potentially
confusing notation for describing gene variants and phenotypes.
Studies have shown that clinicians feel poorly informed about phar-
macogenomics testing, and that they have a difficult time interpret-
ing test results7,9 and the complex interactions between drugs and
genes.39 Efforts at meeting clinicians’ pharmacogenomics informa-
tion needs are under way, such as the work of Devine et al.40 to de-
velop pharmacogenomics alerts within a computerized provider
order entry system. Devine, et al. found in their recent usability
study with clinicians found that clinicians seek trustworthy, rele-
vant, dosing-related recommendations.40 Another recent study
found that clinicians welcome pharmacogenomics information for
the clopidogrel–CYP2C19 drug-gene interaction, and that they
would like supplementary information about the interaction due to
insufficient knowledge about pharmacogenomics.41 Pharmacists, as
trained drug information experts, are naturally poised to support
the incorporation of pharmacogenomics into clinical practice, but
they also struggle with pharmacogenomics information.22,29,39,42
Many groups, including the Pharmacogenomics Research Network,
CPIC, and eMERGE, as discussed previously, are implementing clin-
ical pharmacogenomics testing, producing evidence-based guide-
lines, and incorporating genetic testing results into clinical decision-
making. All of these efforts have hurdles to overcome related to howTab
le2.U
sab
ilit
yO
bse
rva
tio
ns
inV
ers
ion
1.0
an
dD
esi
gn
Ch
an
ge
sfo
rV
ers
ion
2.0
Layout
Info
rmati
on
Dis
pla
yIn
form
ati
on
Obse
rvati
on
Use
rsdid
not
noti
ce
acc
ord
ion
sect
ions
Use
rsappea
red
not
to
noti
ceD
rug
Info
Lin
ks
table
and
links
Effi
cacy
and
toxic
ity
not
wel
lunder
-
stood,phen
oty
pe
expla
nati
ons
wer
e
confu
sing
“Sta
r”nota
tion
uncl
ear
Part
icip
ants
wante
d
clea
rre
com
men
da-
tions
Part
icip
ants
wante
d
toknow
info
rma-
tion
sourc
e
Part
icip
ants
reques
ted
back
gro
und
info
to
be
able
tounder
-
stand
pharm
aco
-
gen
om
ics
Change
Change
tab
des
ign
of
SPL
sect
ions
Change
“For
more
info
”box
Com
bin
eto
xic
ity
and
effica
cyin
form
ati
on
wit
hpath
way
dia
gra
min
new
ta-
ble
dia
gra
m
Pro
vid
eso
urc
eof
in-
form
ati
on
more
clea
rly
Pro
vid
eback
gro
und
info
rmati
on
Spec
ific
Changes
Reo
rganiz
eta
bs:
boxed
warn
ing,
Pharm
aco
dynam
ic/
pharm
aco
kin
etic
(PD
/PK
),and
rec-
om
men
dati
ons
in
Sum
mary
,re
stin
seco
nd
tab
Move
tori
ght
of
scre
en,per
ma-
nen
tly
open
Rep
lace
effica
cyand
toxic
ity
info
rma-
tion
wit
hphen
o-
type
info
rmati
on
Have
“vari
ant”
col-
um
nin
new
dia
-
gra
m
Pro
vid
ere
com
men
da-
tion,if
appro
pri
ate
,
wit
ha
vis
ualaid
Make
links
box
easi
er
tofind,w
ith
clea
r
link
toso
urc
epro
-
duce
label
at
Dai-
lyM
ed
Add
links
tooth
erre
-
sourc
es,“pri
mer
”
page
6 Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0
to structure and present genetic data and evidence in a way that is
clinically actionable.12
One solution to managing this overwhelming amount of informa-
tion is to present structured, concise, relevant information, particu-
larly with visual depictions of the relationships between gene
variants, drugs, and phenotypes. Our evaluation suggests that these
techniques can help pharmacists to access, understand, and incorpo-
rate pharmacogenomics information into their decision-making pro-
cess, as compared to other pharmacogenomic information resources.
Version 2.0 supported pharmacists in answering pharmacogenomics-
related drug questions more successfully than the FDA table on all 3
measures (task completion time, correctness, and perceived usability)
and more successfully than PharmGKB on 2 out of 3. Participants
saved 81.4 s, on average, when completing tasks using version 2.0
compared to PharmGKB, a 75.7% decrease, and 132.7 s compared to
the FDA table, a 123.4% decrease. Version 2.0 was the only resource
on which users answered every question correctly.
Some of the differences in performance may be due to a differ-
ence in focus among the resources. For instance, the FDA infor-
mation provided by these resources is not the only clinically
important pharmacogenomics information available: PharmGKB is
built around the clinical guidelines developed by the CPIC.17
Incorporating CPIC guidelines into our resource and integrating
them alongside the drug label information is a potential future im-
provement. Also, the FDA table is not intended to be a clinical in-
formation resource providing practice recommendations or
supporting clinical decision-making. In contrast, our resource is a
novel implementation of pharmacogenomics information, struc-
tured by a clinically informed, detailed semantic model that allows
us to present the information in a highly usable, searchable way,
while also providing the opportunity to connect the information
to other sources of information, including guidelines, evidence,
and patient-specific data.
The goal of this project was to develop an effective method of im-
parting pharmacogenomics information to clinicians, specifically
pharmacists, by providing clinically actionable pharmacogenomics
drug information connected to relevant FDA-mandated product label
sources. We aimed to access and present that information faster,
Figure 3. Version 2.0, drug page
Figure 4. Generic model presentation of pharmacogenomics information
Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0 7
more completely, with a more satisfactory user experience than avail-
able alternatives. However, a key piece of information missing from
our resource, as well as all other currently available resources, is accu-
rate patient data, both clinical and genotype. The semantic model on
which this resource is built is capable of connecting patient data from
EHRs to evidence from sources such as product labels, guidelines,
and institutional policies. We have achieved the first necessary steps
in building the resource, designing a highly usable way of presenting
pharmacogenomics information, and evaluating it against alternative
resources, with positive results in performance and usability. Our
next step is to connect it to patient data so pharmacists have access to
patient-specific, clinically relevant, actionable information at the
point of care. Furthermore, pharmacists indicated throughout this
project that a stand-alone system would not fully support their needs;
they want the information integrated into their existing systems, in-
cluding computerized provider order entry systems and EHRs.
Understanding how best to achieve this requires careful study of cur-
rent workflows of pharmacists and other clinicians as they engage
with medication-related pharmacogenomics information.
LIMITATIONS
Limitations include a small sample size and number of tasks, which
limit the generalizability of the conclusions. We cannot claim that
any one of the interface changes definitely improved usability or per-
formance. Clinical implementation of genetic testing and genotype-
guided clinical decision support is necessary to determine improved
clinical outcomes. Some pharmacogenomics information from prod-
uct labels might be of limited value due to the lack of actionable rec-
ommendations and connection to patient-specific genetic data.
Similarly, other consensus PGx guidelines (CPIC) exist and are com-
monly used to drive decision support. However, we chose to focus
on FDA product label information as a starting point as it is the US
regulatory standard for drug information. The information was pro-
vided by all assessed resources, and the tasks and correct answers
were written based solely on that information. Participants used
only the FDA labeling portion of the PharmGKB.org website rather
than the complete site to ensure that they were using the same type
of information to complete tasks across each resource. We also did
Table 3. Demographics
Characteristics N¼ 16
Sex (N)
Female 12
Male 4
Ethnicity (N)
Caucasian 13
Asian 1
Hispanic 1
Other 1
Job Title (N)
Pharmacy student 1
Pharmacy resident 2
Pharmacy faculty 2
Clinical pharmacist, nursing home consultant 5
Clinical pharmacist, unit based 3
Clinical pharmacist, director of patient care 1
Pharmacy director 1
Pharmacist, PhD candidate 1
Time as Pharmacist (years)
Mean 7.5
Median 3
SD 8.9
Figure 5. Box plot of time (s) by resource: boxes and whiskers represent median interquartile ranges, asterisks represent outliers, solid circles represent mean
task completion times
Table 4. Task correctness (%) data
Resource Correct Incorrect Total Percent Correct
PGx 1.0 27 5 32 84.6
PGx 2.0 32 0 32 100
PharmGKB 26 6 32 76.9
FDA 21 11 32 69.2
8 Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0
not compare tertiary resources that provide pharmacogenomics in-
formation because we were focusing on sources that provide prod-
uct label statements directly. Finally, further development and
implementation of the system as an integrated resource will require
long-term effort and institutional support. Much of the expense lies
in the translation of FDA pharmacogenomics information from free
text into a structured semantic model.28 Community adoption of a
common model for representing key pharmacogenomics details
would reduce this expense.
Future work will include exploring how this resource can fit
within pharmacists’ workflow; expanding the information and sour-
ces of information to include CPIC guidelines, local practice guide-
lines, and tertiary information sources; and, most importantly,
connecting it to patient genetic and clinical data to provide patient-
specific recommendations.
CONCLUSIONS
Our pharmacist-focused resource supports pharmacogenomics in-
formation tasks more successfully than alternative resources. In the
process of designing this resource, we identified a model for display-
ing pharmacogenomics information that pharmacists find under-
standable and useful while seeking this information to make drug
decisions. The model for presenting pharmacogenomics information
and the semantic model on which the resource is built are necessary
for effective implementation of patient-specific pharmacogenomics
decision support connected to EHRs. Researchers and designers of
alerts and practice guidelines related to pharmacogenomics may find
this model to be a strong guideline for presenting pharmacogenom-
ics information to clinicians in a concise and accessible way.
CONTRIBUTORS
KMR designed and implemented the work; acquired, analyzed, and
interpreted the data; and wrote the manuscript. RDB provided sub-
stantial contributions to the design of the work, interpretation of the
data, and drafting of the manuscript, and assessed the manuscript
critically for intellectual content. PEE provided substantial contribu-
tions to the design of the work and interpretation of the data, and
assessed the manuscript critically for intellectual content. YN pro-
vided substantial contributions to the design and implementation of
the work and assessed the manuscript. SA provided substantial con-
tributions to the design of the work and assessed the manuscript crit-
ically for intellectual content. HH provided substantial
contributions to the design of the work, interpretation of the data,
and drafting of the manuscript, and assessed the manuscript criti-
cally for intellectual content.
FUNDING
This research was supported by the National Library of Medicine
(R01LM011838 and T15 LM007059), the National Institute of Aging (NIA
K01 AG044433), and the Agency for Healthcare Research and Quality
(K12HS019461).
COMPETING INTEREST
The authors have no competing interests to declare.
ACKNOWLEDGMENTS
The authors would like to thank the pharmacists who participated in this
work for their valuable contributions to the annotation and evaluation of
pharmacogenomics information.
REFERENCES
1. Kangelaris KN, Bent S, Nussbaum RL, et al. Genetic testing before antico-
agulation? A systematic review of pharmacogenetic dosing of warfarin.
J Gen Intern Med. 2009;24:656–64.
Figure 6. Box plot of System Usability Scales by resource: boxes and whiskers represent median and interquartile ranges, asterisks represent outliers, solid cir-
cles represent mean SUS scores
Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0 9
2. Caudle KE, Klein TE, Hoffman JM, et al. Incorporation of pharmacoge-
nomics into routine clinical practice: the Clinical Pharmacogenetics Imple-
mentation Consortium (CPIC) guideline development process. Curr Drug
Meta.b 2014;15:209–17.
3. Pulley JM, Denny JC, Peterson JF, et al. Operational implementation of
prospective genotyping for personalized medicine: the design of the Van-
derbilt PREDICT project. Clin Pharmacol Ther. 2012;92:87–95.
4. Johnson JA, Elsey AR, Clare-Salzler MJ, et al. Institutional profile: University
of Florida and Shands Hospital Personalized Medicine Program: clinical im-
plementation of pharmacogenetics. Pharmacogenomics. 2013;14:723–26.
5. Empey PE. Genetic predisposition to adverse drug reactions in the inten-
sive care unit. Crit Care Med. 2010;38:S106–16.
6. Sheffield LJ, Phillimore HE. Clinical use of pharmacogenomic tests in
2009. Clin Biochem Rev. 2009;30:55–65. http://www.pubmedcentral.nih.
gov/articlerender.fcgi?artid¼2702214&tool¼pmcentrez&rendertype¼ab
stract. Accessed April 5, 2013.
7. Crews KR, Hicks JK, Pui C-H, et al. Pharmacogenomics and individual-
ized medicine: translating science into practice. Clin Pharmacol Ther.
2012;92:467–75.
8. Johansen Taber KA, Dickinson BD. Pharmacogenomic knowledge gaps
and educational resource needs among physicians in selected specialties.
Pharmgenomics Pers Med. 2014;7:145–62.
9. Stanek EJ, Sanders CL, Taber KAJ, et al. Adoption of pharmacogenomic
testing by US physicians: results of a nationwide survey. Clin Pharmacol
Ther. 2012;91:450–58.
10. Madadi P, Ross CJD, Hayden MR, et al. Pharmacogenetics of neonatal
opioid toxicity following maternal use of codeine during breastfeeding: a
case-control study. Clin Pharmacol Ther. 2009;85:31–35.
11. George J, Doney A, Palmer CN, et al. Pharmacogenetics testing: implica-
tions for cardiovascular therapeutics with clopidogrel and warfarin. Car-
diovasc Ther. 2010;28:135–38.
12. Herr T, Denny J, Hakonarson H, et al. Practical considerations in genomic
decision support: The eMERGE experience. J Pathol Inform. 2015;6:50.
13. Nutescu EA, Drozda K, Bress AP, et al. Feasibility of implementing a com-
prehensive warfarin pharmacogenetics service. Pharmacotherapy.
2013;33:1156–64.
14. Klein TE, Altman RB, Eriksson N, et al. Estimation of the warfarin dose
with clinical and pharmacogenetic data. N Engl J Med. 2009;360:753–64.
15. Kimmel SE, French B, Kasner SE, et al. A pharmacogenetic versus a clini-
cal algorithm for warfarin dosing. N Engl J Med. 2013;369:2283–93.
16. Drozda K, Wong S, Patel SR, et al. Poor warfarin dose prediction with
pharmacogenetic algorithms that exclude genotypes important for African
Americans. Pharmacogenet Genomic.s 2015;25:73–81.
17. Relling MV, Klein TE. CPIC: Clinical Pharmacogenetics Implementation
Consortium of the Pharmacogenomics Research Network. Clin Pharma-
col Ther. 2011;89:464–67.
18. de Graaff LCG, van Schaik RHN, van Gelder T. A clinical approach to
pharmacogenetics. Neth J Med. 2013;71:145–52. http://www.ncbi.nlm.
nih.gov/pubmed/23712814. Accessed November 1, 2013.
19. Overby CL, Rasmussen LV, Hartzler A, et al. A Template for Authoring
and Adapting Genomic Medicine Content in the eMERGE Infobutton
Project. AMIA Annu Symp Proc. 2014;2014:944–53. http://www.pub
medcentral.nih.gov/articlerender.fcgi?
artid¼4419923&tool¼pmcentrez&rendertype¼abstract. Accessed Octo-
ber 15, 2015.
20. El-Ibiary SY, Cheng C, Alldredge B. Potential roles for pharmacists in
pharmacogenetics. J Am Pharm Assoc. 2003;48:e21–29, e30–32.
21. Tuteja S, Haynes K, Zayac C, et al. Community pharmacists’ attitudes to-
wards clinical utility and ethical implications of pharmacogenetic testing.
Per Med. 2013;10:793–800.
22. Dias MM, Ward HM, Sorich MJ, et al. Exploration of the perceptions,
barriers and drivers of pharmacogenomics practice among hospital phar-
macists in Adelaide, South Australia. Pharmacogenomics J. Published
Online First: September 10, 2013, doi:10.1038/tpj.2013.31
23. Reiss SM. Integrating pharmacogenomics into pharmacy practice via med-
ication therapy management. J Am Pharm Assoc. 2003;51:e64–74.
24. American Society of Health System Pharmacists. ASHP Guidelines on the
Pharmacist’s Role in Providing Drug Information. In: ASHP Best Practices
2014–2015. 2014. 355–59. http://www.ashp.org/DocLibrary/BestPracti
ces/SpecificGdlMedInfo.aspx
25. FDA. Genomics: Table of Pharmacogenomic Biomarkers in Drug Labels.
http://www.fda.gov/drugs/scienceresearch/researchareas/pharmacoge
netics/ucm083378.htm. Accessed March 31, 2015.
26. Hewett M, Oliver DE, Rubin DL, et al. PharmGKB: the Pharmacogenetics
Knowledge Base. Nucleic Acids Res. 2002;30:163–65.
27. Altman RB. PharmGKB: a logical home for knowledge relating genotype
to drug response phenotype. Nat Genet. 2007;39:426.
28. Boyce RD, Freimuth RR, Romagnoli KM, et al. Toward semantic modeling
of pharmacogenomic knowledge for clinical and translational decision sup-
port. AMIA Summits Transl Sci Proc. 2013;28–32. http://www.pubmedcen
tral.nih.gov/articlerender.fcgi?artid¼3814496&tool¼pmcentrez&renderty
pe¼abstract. Accessed September 3, 2014.
29. Romagnoli KM, Boyce RD, Empey PE, et al. Bringing clinical pharmaco-
genomics information to pharmacists: a qualitative study of information
needs and resource requirements. Int J Med Inform. 2016;86:54–61.
30. Drug Labels. https://www.pharmgkb.org/view/drug-labels.do. Accessed
July 6, 2016.
31. Kuo GM, Ma JD, Lee KC, et al. Institutional Profile: University of Califor-
nia San Diego Pharmacogenomics Education Program (PharmGenEdTM):
bridging the gap between science and practice. Pharmacogenomics.
2011;12:149–53.
32. Roche X. HTTrack Website Copier. 2015. https://www.httrack.com/.
Accessed June 2, 2015.
33. TechSmith. Snagit. 2015. https://www.techsmith.com/snagit.html.
Accessed June 2, 2015.
34. Brooke J. SUS: A ‘quick and dirty’ usability scale. In: Jordan PW, Thomas
B, Weerdmeester BA et al., eds. Usability Evaluation in Industry. Bristol,
PA: Taylor & Francis; 1996:189–94.
35. MiniTab. MiniTab 17. State College, PA: Minitab, Inc.;2015.
36. Team RC. R: A Language and Environment for Statistical Computing.
2015. http://www.r-project.org/. Accessed August 1, 2015.
37. Games PA, Howell JF. Pairwise multiple comparison procedures with unequal
N’s and/or variances: a Monte Carlo study. J Educ Behav Stat. 1976;1:113–25.
38. Overby CL, Erwin AL, Abul-Husn NS, et al. Physician attitudes toward
adopting genome-guided prescribing through clinical decision support.
J Pers Med. 2014;4:35–49.
39. McCullough KB, Formea CM, Berg KD, et al. Assessment of the pharmacoge-
nomics educational needs of pharmacists. Am J Pharm Educ. 2011;75:51.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid¼3109805&tool¼pmcentrez&rendertype¼abstract. Accessed November 19, 2012.
40. Devine EB, Lee C-J, Overby CL, et al. Usability evaluation of pharmacoge-
nomics clinical decision support aids and clinical knowledge resources in a
computerized provider order entry system: a mixed methods approach.
Int J Med Inform. 2014;83:473–83.
41. Nishimura AA, Shirts BH, Salama J, et al. Physician perspectives of
CYP2C19 and clopidogrel drug-gene interaction active clinical decision
support alerts. Int J Med Inform. 2016;86:117–25.
42. de Denus S, Letarte N, Hurlimann T, et al. An evaluation of pharmacists’ ex-
pectations towards pharmacogenomics. Pharmacogenomics. 2013;14:165–75.
10 Journal of the American Medical Informatics Association, 2017, Vol. 0, No. 0