To describe the knowledge and attitudes of clinicians participating in a large pharmacogenomics implementation program.
Semi-structured interviews with 15 physicians and nurse practitioners were conducted.
Three categories of themes were identified: preparation and knowledge, pharmacogenomics usage in practice, and future management of genomic variants. Providers expressed an inability to keep up with the rapid pace of evidence generation and indicated strong support for clinical decision support to assist with genotype-tailored therapies. Concerns raised by clinicians included effectively communicating results, long-term responsibility for actionable results and hand-offs with providers outside the implementation program.
Clinicians identified their own knowledge deficits, workflow integration, and longitudinal responsibility as challenges to successful usage of pharmacogenomics in clinical practice.
The use of genomic variants to tailor medical therapy is becoming
increasingly relevant to routine clinical practice, as medications commonly used in
primary care and cardiology practice acquire new indications for pharmacogenomics
testing[
Translating research knowledge to clinical practice has historically
presented multiple challenges, requiring changes to process and organizational
culture[
The field of pharmacogenomics needs a better understanding of how clinicians
are responding to genomic data in routine care activities. As part of an evaluation
program for a large scale pharmacogenomics implementation, PREDICT (Pharmacogenomic
Resource for Enhanced Decisions in Care and Treatment)[
The study was conducted at Vanderbilt University Medical Center (VUMC),
which launched PREDICT in 2010[
We developed a purposive sampling plan for interviews along two axes: usage patterns and practice domain. The sampling plan solicited users from two types of practice, primary care and cardiology, selected because of the indications for the commonly used medications targeted by the program. Usage patterns were quantified by the number of orders for the PREDICT test. Low usage pattern was defined as < 10 orders summarized over the prior year, medium usage was defined as between 10 and 99 orders, and high usage patterns defined as > 100 orders for pharmacogenomics testing. We recruited subjects along the two sampling axes, contacting potential subjects directly by email or in person and requesting their interview participation. Interview subjects were compensated for their time. We continued with interviews within each subgroup until we reached a point of data saturation, where additional interviews did not yield significant additional knowledge.
Semi-structured interviews were selected to assess clinician attitudes
and knowledge based on prior experience evaluating health information technology
and program evaluations[
Interview questions were developed based on the research questions
motivating the study. We developed the interview instrument through discussion
and iterative refinement by the research team, including experts in qualitative
research approaches. Research questions were divided into categories: role and
computer use, meaning and use of pharmacogenomics, experiences with PREDICT,
pharmacogenomics nomenclature and open-ended feedback. Each question category
sought to elicit specific feedback regarding pharmacogneomic use in practice.
Subsequently, we pilot tested the instrument with two interview subjects. Pilot
testing led to minor changes to the instrument to clarify the phrasing and
content of the questions. Finally, we arranged interviews in locations
convenient to our interview subjects. Each semi-structured interview used the
same interview script (
Each interview was conducted by one or two researchers with experience in qualitative methods. All interviews were either audio or video recorded, with interview subjects allowed to choose between the two options. Interview subjects reviewed a written informed consent document prior to the interview beginning, and all interview subjects provided signed consent.
After interviews were completed, the audio or video files were transcribed. Transcribed files were then uploaded to Dedoose, a cloud-based data analysis package, specifically developed to support analysis of qualitative and mixed methods data. Users are able to upload files to Dedoose in a variety of formats, including text files containing, for example, transcribed content of audio- or video-recorded data. Once uploaded to Dedoose, the tool allows users to review file contents, tagging text elements and applying codes to them. Dedoose organizes the qualitative data throughout analysis and supplies aggregate views of codes and text excerpts coded by researchers.
Using Dedoose, we analyzed the data, applying a grounded theory approach
to data analysis. Grounded theory approaches allow theory to emerge from the
data, rather than applying existing theoretical frameworks to data analysis
[
Two researchers working independently reviewed each transcript in
Dedoose and applied codes to the data to identify elements of interest through
an open coding process. Working collaboratively, researchers then identified the
main themes, looking for recurrent patterns and key themes in the assigned codes
[
We recruited 15 clinicians from both internal medicine and cardiology and from each of the three usage categories. Nine cardiology and six primary care providers were interviewed representing four low usage, four medium usage, and seven high usage clinicians. More high usage clinicians were interviewed as they expressed a greater diversity of opinions regarding the testing and more interviews were required to achieve data saturation. The majority of interviewees (13) were attending physicians, as the majority of PREDICT users had this role. We also interviewed two nurse practitioners who actively prescribed medications targeted by PREDICT and interacted with clinical decision support.
Based on analysis of interview data, we identified three high-level theme categories in the data: preparation and knowledge, pharmacogenomics usage in practice, and future implementation challenges. Each category consisted of multiple themes incorporating related concepts.
None of the clinicians in our sample had specific coursework or other
training in pharmacogenomics prior to the program implementation, an expected
outcome given the early stages of translating pharmacogenomics to practice.
Despite the lack of formal training, clinicians developed knowledge and
understanding of pharmacogenomics concepts through various mechanisms. Specialty
care providers discussed developing initial pharmacogenomics knowledge prior to
the informatics intervention from research studies about the relationship
between clopidogrel and the gene encoding the metabolizing enzyme,
Interviewers asked each subject, “How do you define the term
‘pharmacogenomics’?”, eliciting a wide variety of
reactions and responses. Several interview subjects laughed at the question,
expressing uncertainty about the concept. For example, one respondent stated,
PREDICT implementation initially targeted specialty care providers in
cardiology. Cardiology clinicians cited outreach efforts by clinical and
informatics leaders to promote the upcoming pharmacogenomics tool implementation
as a key factor in their development of knowledge about and use of the testing.
Methods used for knowledge dissemination included Grand Rounds and smaller
practice group meetings. Clinicians typically discussed several key clinical and
research leaders who provided an initial introduction and ongoing dialog related
to pharmacogenomics, “Probably at the department level through larger
group presentations. I recall [program faculty leader] being an
influential voice to introduce this. I also worked closely with one
of the key members of the pharmacogenomics team…. So, that
has been a steady source of conversation over the last two
years.”
The types of initial exposure to pharmacogenomics discussed by primary care providers focused more on general communication channels, such as electronic medical center newsletters and journal articles.
Despite outreach efforts, questions remained about rapidly changing
pharmacogenomics knowledge. One cardiologist described the balance between
knowledge dissemination and the types of knowledge needed to apply information
in practice, “Well, I think we could be more informed of some
of the pharmacogenetic principles frankly in a more understandable
way than the state-of-the-art grand rounds lecture on one hand and
the patient communication piece on the other. There's
something in between that is clear, concise, pitched at the general
practitioner’s level that is missing in this whole
operation.”
Because of the rapid evolution and expansion of pharmacogenomics
knowledge, clinicians discussed the need for continuing education. Clinicians
discussed concerns about their knowledge becoming quickly obsolete. One
cardiologist summarized this concern by saying, “I wish I knew more because sometimes I think
we, I feel like we practice in a vacuum, especially on something so
super specialized as this. So, you know it, and you learn it, and
you know very well in six months what you know is not current. I
mean, there's no way that it is.”
In summary, we identified both positive aspects and gaps in the outreach efforts related to education and concerns by clinicians about continuous engagement to maintain their knowledge of research advancements.
A clear theme across interviews was that clinicians understood the
rationale for obtaining pharmacogenomics information, but integrating this
knowledge into healthcare practices raised complex questions and concerns.
One strong proponent of obtaining pharmacogenomics panel data summarized
this view, “I think more information is always better
about patients. So I believe that it's important to try
to obtain this genetic information, pharmacogenomic information
on my patients. That's step number one. Step number two
is what do you do with the information? We're still
learning.”
Standard laboratory reporting of genomic test results was sometimes
unclear to clinicians, leading them to seek answers from the interpretive
information present in other sections in the EHR. For example, “There is so much information that comes
back when [the laboratory test report] shows results that you
say, “Okay, I don't know what this means.
I'm going to go to [the EHR] where it's really
simple and it tells me it's a poor metabolizer or
intermediate metabolizer.” Those are words I can
understand as opposed to getting all the genetic
information.”
Some clinicians felt that even these distilled phenotype terms were
difficult to interpret, in part because the nomenclature that was familiar
to the clinical genetics research community was not transparent to
end-users. “I think poor metabolizer is a good word, a
good phrase… Indeterminate would suggest that we have no
idea what the mutation does. Whereas, intermediate… a
more suitable word might be partial
metabolizer.”
Due to the highly specialized content of pharmacogenomics tests,
some clinicians expressed concern about the clinical relevance and
information overload of reporting genomic information to the EHR. As one
clinician expressed,
Once tests results were reported, clinicians integrated
pharmacogenomics test data into their clinical decision-making processes to
varying degrees. Clinicians expressed strong desires for, as one clinician
described it,
Interview subjects provided suggestions for several different
approaches to CDS, focused on the idea that CDS needs to be clear and
concise but also provide the ability to seek out more information quickly
and easily if desired. One clinician laid out a rationale as follows,
"I'm a very quantitative person…
intermediate doesn't mean anything to me. So…
can you tell me poor metabolizer? Could you quantify that in
some way? 10, less than 10%? Some number that tells me
or even something that's just color coded and it says,
“Prescribe something else. Don't do these
drugs,”
Another physician suggested, “Most importantly, the information has to be
pushed to the ordering physicians so that the ordering physician
gets the information. With that push has to be very easy links
to… written advisor statements invented by the experts
that tell us what is recommended. And then, there should be
another link to the original data for people that want to know
exactly what the evidence is one way or the
other.”
Clinicians in our interview sample viewed pharmacogenomics data as
just another element to integrate into clinical decisions, much like routine
laboratory tests. At the same time, they pointed out that multiplexed
pharmacogenomic testing as assessed by PREDICT encompassed potential
pharmacogenomics interactions beyond ones that currently have clear
treatment guidelines in their field. Other subjects expanded on this
concept, to discuss how the current state of pharmacogenomics knowledge may
not give a full picture of the potential variation in drug response.
“That's part of the frustration at
this particular point with pharmacogenomics, in terms of having
a patient walk into the door and really not knowing that
information that you think may be vital to their care in terms
of trying to individualize their care at that particular point,
but as we go forward, as information starts to compile and
build, connects and make modifications in terms of
therapy.”
Clinicians reported that making medical decisions related to
pharmacogenomics data involved a complex effort to balance cost, risk, and
benefit. Alternative medications suggested by the literature and adopted by
the program for use within CDS could create higher out of pocket costs and
new safety concerns in addition to the promise of improved efficacy.
Clinicians discussed the challenges inherent in integrating program guidance
with the social situation of the patient and uncertainties with how much
genome tailored therapy would improve outcomes.
Initially, clinicians discussed pharmacogenomics testing in detail
with patients before ordering tests, but reduced the amount of explanation
over time. “So, at the beginning, we started all this,
I went through this detailed explanation of what we were
ordering, and what I found from patients is that the response
all along is, ‘Oh, please order it. It's stupid
not to order this particular test. I definitely want to know the
information.’ At this point, it's become a
shorter conversation in terms of, ‘I want to do this. I
think it's smart. This is why,’ and everybody
says, ‘Fantastic. Please do and can my daughter get it?
Can my uncle get it? Can my grandmother get
it?’”
Other clinicians felt that in-depth explanations of specific
pharmacogenomic testing details were unnecessary in initial decisions to
test. “I'm usually somebody that likes to
simplify things an awful lot for understanding for both my
patients and for me. So, you know, how can I make this as simple
as possible so that they get the big picture of why I'm
doing the test, but not overwhelm them with its
purpose.”
Clinicians discussed some of the language they used in explaining
pharmacogenomics testing to patients, “I try to explain that this is a piece of
the puzzle. That we can get lots and lots of people's
data and then we can be able to sort of make more, I
don't say responsible, but medically sound decisions
based on evidence and not guess work.”
Clinicians described a clear pattern that, as they became more
familiar with this type of testing, they began to view the test in a similar
light as other clinical tests in terms of explanation required before
testing. One substantial caveat to this explanation pattern is that during
the time interviews were conducted, the PREDICT test was institutionally
supported and offered free of charge to patients. Some clinicians expressed
reservations regarding whether patients would be receptive to genetic
testing once it was charged to their insurance plan and they were
responsible for co-pays and deductions. For example, one specialty care
provider stated,
When receiving test results, clinicians faced the challenge of
interpreting and communicating the information to patients and families.
Their level of familiarity with pharmacogenomics impacted this interaction.
“I think that you had a lot of clinicians
who were blindsided because all of a sudden, patients start
finding out they were intermediate metabolizers and this is
before anyone knew what to do with that. And so, I think, you
know, you had patients asking their doctors, ‘Well, I
got this, you know, this is what they said I am. You know, what
do I do?’ And the doctors would go, ‘Uh, I
don't know.’”
In some cases, a sense of lack of preparedness led to conversations
with patients being conducted in less detail than clinicians would normally
pursue. One clinician explained,
Secondly, specialty care providers felt underprepared to explain
drug-gene interactions that involved drugs they did not prescribe,
Providers expressed interest in a formal set of patient education
materials that anticipated questions and concerns.
Providers discussed how the persistent nature of pharmacogenomics
data presents new challenges related to long-term data ownership,
responsibility, and liability. For example, “Does that information [the full range of
PREDICT results] remain undiscovered if I don't actively
push it to the primary care physician or can it automatically
get to them so that they can use that information for the 48
other drugs that I'm not going to be
prescribing?”
Clinicians explained a gap between current policies and the range of
data in the informatics intervention, with several clinicians exploring the
need for formal clear policies to explain responsibility and ownership for
pharmacogenomics data. “I think it'd be nice if there were
some clarity about the responsibility for the ordering physician
in terms of notifying the other physicians involved in the
patient's care just so people know exactly
what's expected of them when they order the
test.”
While clinicians felt clear lines of responsibility and ownership
were necessary, they expressed concerns about the level of pharmacogenomics
knowledge among referring clinicians outside the academic medical center
environment. The need to educate busy community clinicians about the results
and recommended action was an area that some clinicians felt needed to be
explored in detail, “I think it's going to be important
to come up with good processes to educate referring physicians
as well as ordering physicians and specialists on how to handle
this information. Who do you need to notify? Who's
responsible for acting on the information? Who's
responsible for educating the patients on it as
well?”
Although many clinicians came to view pharmacogenomics testing as another routine laboratory test in their practices, there were clear concerns about challenges related to the persistence of pharmacogenomics data over time.
Regardless of how well the pharmacogenomics test ordering and results
were integrated into clinical practice, subjects discussed the need to continue
scientific exploration of outcomes related to treatment changes. One clinician
discussed the future of pharmacogenomics by saying, “So, [pharmacogenomics testing] has changed my
practice even though the outcomes data are not there yet. And I feel
comfortable about that because my change has been validated in a
cohort of patients outside of known genetic information. In
randomized trials. I also think it's important to get into
the mindset where we are, as, as clinicians, routinely thinking
about optimizing drug therapy for patients based on their
genetics.“
Closing the loop on the current approach to pharmacogenomics was
critical to multiple clinicians interviewed for this study. As one clinician
stated, “I’m not sure where it’s headed
as far as using it for science in terms of having a strong database
where we’re linking PREDICT data with clinical
outcomes.”
Continuing along the path to personalizing treatment decisions for patients based on genetic data requires demonstrating the value of this approach, particularly on improving patient outcomes.
This study provides insights into the barriers facing the dissemination of personalized medicine. First, clinicians acknowledged the complexity of genomic data; the unfamiliar representations and nomenclature used to describe results led to difficulties with interpreting, communicating, and applying the data to clinical care. Strong support was expressed for ongoing engagement with the implementation team to keep clinicians updated on the latest research results. Providers also strongly supported the use of thoughtfully designed and well-integrated CDS tools to facilitate genome-informed decisions. However, they identified gaps in the program related to long-term responsibility for genomic risks when patients leave the institution, and hand-offs to community providers.
Several prior qualitative and survey studies have identified
providers’ concerns about incorporating genomic information into their
practice. Interviews of hospital pharmacists working in Australia indicated their
knowledge, education, and time constraints were barriers to use of
pharmacogenomics.[
Pharmacogenomics testing was viewed by practitioners in our study as similar to other laboratory tests, particularly when explaining the need for such testing to patients. Practitioners identified, however, that pharmacogenomics results also had different attributes from routine laboratory results. Testing was to address specific treatment questions, but the PREDICT pharmacogenomics panel test covered a wide range of genetic variants. Pharmacogenomics testing creates persistent data whose meaning and interpretation will evolve over time. This persistent value requires assignment of long-term responsibility for interpretation and management. Ordering clinicians expressed concerns regarding hand-off of responsibility for managing drug-gene interactions for drugs not prescribed by the ordering clinician. Primary care physicians in the community had limited preparation to interpret and manage drug-gene interaction data, raising questions about how long-term responsibility for managing drug-gene data can best be transitioned from ordering practitioners to referring and general practitioners. Pharmacogenomics testing represents an important and emerging frontier in health data, requiring communication, coordination, and longitudinal follow up that is rarely handled effectively in the current fragmented structure of healthcare.
The study has several limitations. The themes were derived from a small sample of clinicians that may not be fully representative of all opinions within our institution or among other types of subspecialists who encounter genomic results. We have not compared attitudes between primary care physicians and specialists which would require additional data from a broader spectrum of clinicians. Indeed, we expect oncologists who have greater clinical experience applying molecular diagnostics in practice would be more comfortable with genomic results. Although much of the data gathered in this study has broad relevance, pharmacogenomics implementations vary widely and some details may be implementation specific. Interviews were conducted when the institution supported the cost of pharmacogenomics testing; however, clinicians in the study were already anticipating the evolution to testing reimbursed by insurance. Costs of testing and treatment alternatives may change rapidly with updates to program and insurance policies, and we anticipate further evolution of provider perspectives. Finally, all clinicians interviewed were affiliated with an academic medical center, leaving a significant area for future research: studying the perspectives of healthcare practitioners in the community.
A qualitative study of clinician views of pharmacogenomics defined gaps in the current implementation and suggestions for future improvement. In particular, pharmacogenomics implementations need to focus on education of both practitioners and patients. Continuous educational outreach may be required to assist with rapid pace of knowledge development. Clinical decision support and long-term responsibility for pharmacogenomics panel data are important areas to be addressed by new policies and new program features. With the emerging implementation of next generation sequencing of both somatic and germline variants, we anticipate attitudes will change as additional evidence is generated. Future investigations of clinicians’ views of genomic medicine should include a broad spectrum of specialists, including those who have already embraced targeted therapy and those who are poised to incorporate targeted therapy into their clinical practice.
A qualitative study of clinician views of pharmacogenomics highlighted ongoing interest in incorporating genomic information into routine clinical care and defined gaps in the current pharmacogenomics implementation and suggestions for future improvement. Study subjects reported the following themes:
Clinicians expressed support for the idea that pharmacogenomics is rapidly becoming part of standard practice
Clinicians found it challenging to keep pace with the rapid generation of new drug-gene interaction evidence without ongoing educational support
Clinicians expressed concerns about communicating to patients the rationale for applying pharmacogenomic results to prescriptions and the need to balance genomic information with other clinical, social, and financial factors
Clinicians expressed unease with taking long-term responsibility for genomic variation that was either not directly related to their care plan or outside their specialty.
Goal: understanding the interview subject’s role in healthcare delivery and interaction with health information technology.
Can you describe your current role? What types of patients do you primarily see? What clinical department(s) do you normally work in?
How do you record clinical provider notes?
What tools do you use for ordering tests and procedures?
Are there any other health information technology systems that you use?
How would you describe your use of computers in healthcare?
Goal: understanding how interview subjects conceptualize pharmacogenomics and the role of pharmacogenomics in healthcare.
How do you define the term “pharmacogenomics”?
How were you first introduced to pharmacogenomics?
Where have you learned the most about pharmacogenomic
testing?
What types of evidence or guidance do you feel is most persuasive in adjusting your clinical practice?
How has your understanding or interpretation of pharmacogenomics changed over time (in general or for a specific drug-gene interaction)?
Have you received any informal or formal training in pharmacogenomics? Can you tell us more about any training you’ve received?
What role does pharmacogenomic testing have in your healthcare practice currently? What role do you think it will have in the future? How prepared do you feel to order pharmacogenomic tests and apply the results?
Goal: gathering self-reported current usage of PREDICT, an example of how the subject currently uses PREDICT, and their anticipated future use.
How often do you think you order PREDICT tests right now? How often do you think you use the results of PREDICT tests?
Could you walk us through an example of a time that you used PREDICT to order a pharmacogenomic test? Why did you order the test?
What was the timeline for ordering the test?
Did you use the results of the test yourself, or did you pass the results onto another provider?
How did the patient respond to ordering the test?
Could you walk us through an example of a time you used the results from PREDICT tests in care?
Did the results change your care plan?
What would you have done without the results?
Was the patient aware that pharmacogenomic data was used in care planning?
Can you describe some of the reasons why you order PREDICT testing during clinical encounters?
The PREDICT testing is currently free for patients. What difference, if any, will it make to you when PREDICT testing is no longer free?
What circumstances would make you hesitate to order a PREDICT test or lead you to not act on recommended treatment changes?
Goal: understanding how phrasing of PREDICT prompts impacts provider understanding of those prompts.
What are your thoughts on the PREDICT guideline recommendation language that is currently displayed? For example, some terms that are used include “poor metabolizer” and “intermediate metabolizer.” What do you think of these terms? Are there other terms you would suggest for guidelines?
Do you think including other types of information such as quantitative estimates of risk (e.g., absolute risk, relative risk) would influence your response to PREDICT communications?
Based on your own experience with pharmacogenomic guideline
recommendations, could you rank these words from highest to lowest
degree of obligation?
Follow up question: could you discuss why you put these terms in this particular order?
Goal: wrapping up the interview, gathering any other open-ended comments subjects would like to share.
Do you have any suggestions on how to best integrate pharmacogenomic data into clinical workflow?
Is there any other feedback you’d like to give about your interaction with PREDICT?
Are there any questions about pharmacogenomic testing or about PREDICT that you think we should be asking that we haven’t asked?
May
Should Consider
Is Suggested
Should Be
Is Indicated
Is Recommended
Must
May Consider
Should
Should Be Considered
May Be
*