Important Points Highlighted by Individual Speakers
• Pharmacogenetic results can be important for patient care, but data need to be carefully integrated into patient records and care processes.
• If patients are empowered to make their privacy preferences available to caregivers and researchers, the delivery of care and the use of patient data for research could both be enhanced.
Genomic data are of no value unless they can be communicated in an effective way to people who can act on that information. Understanding what the message needs to be is only part of the challenge, said Greg Feero of the National Human Genome Research Institute, who moderated the workshop session on communicating genomic data. The actual delivery of that information to both the health care professional and the consumer is essential to improving outcomes. Will such information be delivered at the point of care or in some other setting? What infrastructure is needed to deliver genomic information? Who has the responsibility for delivering the information and ensuring that it is understood?
St. Jude Children’s Research Hospital, in association with the Pharmacogenomics of Anticancer Agents Research 4Kids program under the Pharmacogenomics Research Network (PGRN) at the NIH, has been using well-known genetic polymorphisms to adjust drug therapy in patients in real time. St. Jude has the advantage of being able to overcome (or ignore) many obstacles to preemptive genotyping, said Mary Relling of St. Jude. The hospital covers all patient care costs and provides all medications for 5,000 unique, high-risk patients per year, 80 percent of whom have cancer and 20 percent of whom have sickle cell disease, HIV infection, or other life-threatening diseases. St. Jude has a collaborative approach to patient care, in which pharmacists are integrated into the team that delivers care and are responsible for signing off on every consult associated with a pharmacogenetic test result. St. Jude also maintains comprehensive EMRs that fully integrate every aspect of outpatient and inpatient care, which Relling said was a large motivating factor for St. Jude to incorporate genetic testing into the patient record.
For pharmacogenomic testing, St. Jude is now using the Affymetrix DMET-plus array, which tests for more than 1,900 polymorphisms in 225 genes that are likely to be important for pharmacogenetics. Previously it had been screening for only two of the genes included on that panel and doing so at a higher cost. The medical staff would rather not have the medical record populated with genomic information of uncertain clinical utility, Relling said. As a result, the hospital obtains consent from patients to withhold results that are not clinically interpretable, though it also discusses with patients the possibility that these findings may in the future have implications for disease risk.
A new program known as PG4KDS has the goal of migrating a larger number of pharmacogenetic tests from the laboratory into routine patient care so that results are available for preemptive use. The primary program objective is to estimate the proportion of patients who have high-risk or actionable pharmacogenetic results entered in their EMR with decision support (automated information alerts generated to assist health care providers in making decisions about a patient’s care). The secondary objectives are to use systematic procedures for prioritizing and migrating pharmacogenomic test results to the EMR, to incorporate clinical decision-support tools linking test results to medication use, and to assess the attitudes and concerns of research participants and clinicians.
An educational video made with the hospital’s Family Advisory Council is available to provide information for families. Information is also available on the program’s website (www.stjude.org/pg4kds) which lists which
genes are tested for and which genes are reportable, along with links to more detailed information for clinicians.
After patients are enrolled, their DNA is genotyped, and the results that pass quality control thresholds are posted in the research database. Only the small portion that meets the clinical threshold is migrated to the EMR. A limited number of the clinical results that are based on highrisk genotypes and high-risk drugs have decision-support rules that send automatic alerts to clinicians. The study investigators, with the input of an oversight committee, decide how to update the information that is entered into the chart, including the addition of new clinically actionable variants and which results are designated to receive decision support, starting with widely accepted results for high-risk genotypes. “We won’t tackle the most controversial ones until later, or maybe never. We do a big disservice to implementing clinical genomics by trying to implement stuff too fast. We should concentrate on the home runs first,” Relling said.
When it is decided to put a new gene into the EMR, the results for that gene are inserted into the medical record for all past and future patients, with participants having the choice of getting a letter for each new gene that is posted. “So far every single patient has asked for that information,” Relling said, “and of course that could be converted to something electronic in the future.”
To facilitate access to information, the EMR at St. Jude Children’s Research Hospital has been customized with a pharmacogenetics tab, where genotypes are entered along with a detailed laboratory report of how the test was performed and clinically relevant gene-specific information explained. These are lifetime results, Relling said, and physicians should not have to search for them by date or order a test on a gene that has already been interrogated.
When the decision to prescribe a high-risk drug conflicts with the presence of a high-risk or high-priority genotype, a decision-support alert is sent to the clinician. For example, the EMR will automatically generate a warning if codeine is prescribed for the 10 percent of patients who are poor metabolizers based on their CYP2D6 profile. There are two types of warnings. The first is a post-pharmacogenetic test result: If the patient already has a high-risk genotype in the EMR, the clinician will get an alert. The second is a pre-genetic test warning: If thiopurine is ordered for a patient and the patient has not had thiopurine S-methyltransferase (TPMT) tested, an alert is issued. Alerts are also being linked to an explanation of why a high-risk drug–gene pair exists for use by clinicians who are interested in learning more.
Challenges in Implementation
Implementing this system has revealed several challenges that future genomic medicine initiatives will likely face, Relling said. First, she said, there is a lack of consensus or guidelines on which drug–gene diplotypes are most important, although she acknowledged that “it is better to have experts review the evidence and come up with some recommendations” even if they are not in agreement than to make each individual clinician have to synthesize his or her entire knowledge and medical practice experience. The PG4KDS program is addressing this issue through the PGRN. Specifically, a subgroup called the Clinical Pharmacogenetics Implementation Consortium has been formed to evaluate drug–gene pairs using standard grading systems, the peer-reviewed literature, and other information. Severity of disease, therapeutic alternatives, consequences of giving the wrong drug, and consequence of giving the right dose all have to be taken into account when making these decisions.
Another complication is that once a gene test is in the EMR, clinicians are obligated to use those results for all drugs affected by the test. Even rarely prescribed drugs need decision support. “It is going to take a lot of work to go through and make rules that we are all comfortable with,” Relling said.
Diplotypes are sometimes ambiguous because of the nature of the testing. For example, the DMET array often produces ambiguous calls, Relling said. “We have to look very carefully at the reasons for those ambiguous diplotype assignments. Is it the fact that you can’t phase haplotypes, so we can’t always distinguish between a heterozygous- and a homozygous-deficient patient? Or is it because of a simple no call of a probe for a rarely involved SNP? Someone has to decide whether those [ambiguous assignments are actually important] or not, and again that takes a high level of knowledge of the genes and the drugs.”
Using a software program called PHASE, a much higher percentage of diplotype assignments can be made non-ambiguous, but it is a judgment call as to whether to deliver that information to the clinician. Interpretation is complex and time consuming, and it changes over time, Relling observed. For example, even with the relatively simple TPMT diplotypes, about 8 percent of patients have an ambiguous diplotype. Some patients are homozygous deficient, and if they get a normal dose of thiopurine, there is a high probability they will die of toxicity, whereas other patients can tolerate doses for a much longer time period. “Basically, we have to write very specific [reports] to say what the caveat is in interpreting these kinds of results.”
Multiple testing of the same gene over the lifetime of the patient requires that someone check to see whether the results contain discrep-
ancies. One must also check other details such as whether the race of a patient in the EMR agrees with the patient’s self-declared race or whether the sex in the EMR agrees with the self-declared sex. “Flagging some of those genotypes and eventually manually approving them to move from the research laboratory into the EMR takes a lot of steps,” Relling said.
Another critical issue is who will pay for preemptive genotyping. Arraybased genotyping can be cheaper, easier, and more effective than testing one gene at a time even when no drug is being contemplated for treatment, but it may not be easy to get reimbursement for such tests. Finally, Relling said, until patients have universal lifetime EMRs, the fragmentation that affects all of health care is going to affect genomic medicine.
“Remember when using an ATM was a mysterious and improbable experience?” asked Robert Shelton of Private Access, Inc. “That’s about where health care data sharing is today.”
The first automated teller machines (ATMs) were introduced in 1967, but their use was limited until networks of ATMs enabled people to withdraw money from almost any machine. Over time ATMs also developed a compelling business justification because they saved money spent on human tellers and generated revenue from ATM fees. Today there are 2.2 million ATMs around the world, with a new one being added every 4 minutes.
Using a search engine also used to be a mysterious and improbable experience, Shelton observed. In 1990 a search engine called Archie offered access to a directory of directories. In July 2008 Google announced that it had indexed 1 trillion pages and was adding several billion new pages per day. As with ATMs, several intermediate technologies and changes made this massive growth possible, including natural language search of contents, analytics-assisted search that helped a user find the desired Web page despite the terms entered, and compelling business models.
“What if people could search for health information as easily as we do public documents?” Shelton asked. In that case physicians or patients could call up any information contained in an EMR as easily as they access information on the Web. But privacy needs to be protected. Thus, access provisions need to be interposed between a search query and a result to determine whether the searcher has the right to see the results of the search.
The tipping point for integrating privacy protections is occurring now, Shelton said. In December 2010, the President’s Council of Advisors on Science and Technology released a report that called for a “universal exchange language for health care information that enables health IT [information technology] data to be shared across institutions, along with network infrastructure that enables a patient’s data to be located and
accessed across institutional boundaries subject to strong, persistent, privacy preferences” (PCAST, 2010). Even before the release of the report, the company Shelton directs had built and tested such a system. The centerpiece of the system is that patients log onto a secure website and set their own privacy preferences, after which researchers can search for information. For example, Shelton said, if a researcher is looking for subjects for a clinical trial, patients “need to have their hand raised for interest in clinical trials already.”
Private Access does not actually hold patient data. Instead it holds the patient index and search capability, with the privacy directives acting as the switch for searches. For example, if a researcher logs in and searches for potentially relevant subjects through the search index, the system filters the results based on patients’ preferences. If a patient has indicated interest in being part of a trial, the researcher gets contact information and can make an offer.
The system has been tested in patient populations, and the lessons learned have been applied in successive generations of the system. First, patients need to be educated and empowered, not coerced. Consumers are more likely to engage in an environment of trust if they are referred by a trusted intermediary, a friend, a health care provider, or peers, Shelton said. Patients should control the use of their personal information for expressed purposes, they need guide-based assistance to help make informed choices, and the consent tools should be dynamic and granular. To make sure that patients have the information needed to set preferences, links should be available to more detailed educational sites.
On the searcher’s side, the system needs to be fast, easy to use, and powerful. It should use familiar Web-based search conventions such as bookmarks and automated alerts. And, Shelton said, researchers should be able to receive pre-authorized access to personal health data.
When using such a system, a patient who is undergoing the registration process at a hospital would also convey privacy preferences to the hospital, perhaps through a smartphone application. The patient would be asked if information could be deposited in a personal health record (PHR), whether or not the patient has established one previously. If the patient agrees, information from the hospital would be moved beyond the hospital’s firewall to a PHR where it could be searched through a search index.
Other data, such as pharmacy histories, laboratory results, or the results of genetic tests, can also be moved into the PHR. Data can also be moved into other repositories, such as EMRs or searchable databases in each institution visited.
Patients have heterogeneous preferences, Shelton said. Attitudes range from “It is okay for researchers to use my data without my consent at all” to “I am willing to give general consent in advance for the use of my data without being consented” to “Consent is not needed if my identity will never be revealed and the study is supervised by an IRB [institutional review board]” to “I want each study seeking to use my data to contact me in advance and get my specific consent each time” to “I would not want researchers to contact me or use my data under any circumstances.” Most patients fall in the middle of this distribution of preferences, but it used to be logistically impossible to satisfy their wishes. New technologies have now made it possible, Shelton said. Furthermore, he added, “It is empowering, and it creates the data liquidity that we all are looking for.”
In one example cited by Shelton, a researcher at the University of California, Los Angeles, looking for lupus patients found six within 25 miles of campus in under a minute. “Why? Because he was using a search engine to find them.”
Many resources are currently flowing to the basic science of genomic medicine, Shelton said, but resources also need to be expended on sociological research directed at how the science will be applied. This research in turn could inform the business models that will drive adoption.
The patient is the key, Shelton concluded. “We have to give them the tools, and we have to give them the chance.”
Increasing the use of genomic information by providers and consumers will be a slow process. Many people may be able to understand genetic data if they are provided with information, Relling said, but people who are old or poor or who can barely read will have much more difficulty. “There are a lot of people in this country who need us to be a little paternalistic,” she said. If highly trained physicians sometimes cannot understand the information linking genetic variation with clinical recommendations, creating such understanding among others may not be feasible in the short term. “We have to crawl before we walk,” she said. “We have to get this to work in health care institutions where we have highly trained people and get it understood and adopted, and then we can maybe start pushing it out more to consumers.”
Shelton added that he did not expect consumers to have the ability to understand the data that they are receiving. “What I am advocating is that a consumer knows that I don’t want Aunt Betty to see my data. ‘If you tell me that Aunt Betty can’t see it, then I am fine with anybody else who is a researcher seeing it.’ Consumers understand their privacy wishes. That is one thing they do understand and the law doesn’t.” Patients should be
empowered to sign up for a personal genome project if they wish, just as they should be empowered to forbid the use of any of their genetic data. “We don’t use sushi-grade data in the medical establishment. We use chum. It is deidentified and it is all ground up because we have to in order to come under the IRB requirements and in order to avoid a lot of the challenges of asking for permission. If we ask for permission, empower it, and make it easy, it will happen.”
Given the limited associations between genetic test results and most diseases, Shelton suggested that the business case for using health IT systems that incorporate genetic results rests on two motivations. The first is recruiting subjects for clinical trials. The second is the reduction of duplicate tests. In the future an advertising model may evolve as well, he added, as patients become willing to have their medical information used to make them aware of offers that are custom tailored to them.
More generally, he said, the most important piece of health care in the future will be data, which will make the IT department the profit center of health organizations. “The IT department in health care gets a very small fraction of what IT departments get in other industries,” he said. “The reason is because it is a cost center. It is not a profit center. Make it into a profit center, and budgets will go up.”