Because the use of personalized medicine relies on the availability of lots of data, the major IT headache would seem at first glance to be the same storage woes that sequencing centers are grappling with. But this is not solely the case. Housing personalized genomics data is only one of the challenges for personalized medicine-centric IT. The integration of genomic information into health care IT infrastructures is also stymied by the lack of seamless clinical interpretation, health care database interoperability and connectivity, and the need for an overhaul of electronic personal health records.
"What I'm seeing, more generally, is that biomarkers of a certain kind may be incorporated into certain practices, but at the moment, if you go to the commercial personal health records, like the Microsoft Health Vault or Google Health, there isn't a specific field that says: 'Type in your genome here,'" says John Halamka, chief information officer at Beth Israel Deaconess Medical Center and dean for technology at Harvard Medical School. "So as a doctor, do you really want the 30 million base pairs? No, you want access to the four or five probabilities of diseases that would impact a patient's treatment options and wellness considerations. We're really on the cusp of figuring out the IT pipeline — of going from the sequence to the interpretation into clinical decision-making. For the moment, there are just a couple of departments doing it, but within the next year or two, we'll start to see the application of those biomarkers being part of a patient's chart."
However, just because clinicians may not want access to a patient's entire genome during one particular visit doesn't mean they won't want to explore other parts of that patient's genome at a later date. "It's unlikely that many clinicians will want to sift through two to five million variants in the course of a patient visit, but that doesn't mean that you don't need all the variants associated with that patient," says Sandy Aronson, executive director of IT at the Harvard Partners HealthCare Center for Personalized Genetic Medicine. "The clinician may want to look at a particular patient's genome from three years ago for hearing loss, and then, three years later, order a cardiac interpretation for that patient's genome." The physical genome shouldn't have to be re-run, only the interpretation of the variants, he adds.
Another role IT plays in personalized genomics, Aronson says, is to support the mission of geneticists and genetic counselors not just with genomics data, but with genomics data that is delivered in a timely fashion. To this end, Aronson — together with HP — developed the Gateway for Integrated Genomics-Proteomics Applications and Data, or GIGPAD, an IT infrastructure management tool that supports both research and clinical processes on the same platform. He and his colleagues have also implemented a collection of IT tools, called the GeneInsight Suite, which facilitate rapid clinical interpretation of data and genetic test results management for clinicians.
"We regularly have to quickly generate a report that explains to the clinician what's known about the implications, and you also need to make sure that it's generated as quickly as possible because there's a shortage of geneticists and make sure that you optimize their time, so that they can report out on as many cases as possible," Aronson says. "That requires a very sophisticated IT infrastructure. You need a knowledge base, templating tools, a reporting engine, and it all needs to be genetic- and genomic-specific because clinicians need to manage the patients' profiles, they need to know what tests have been performed on what patients, what variants were found, and most importantly, they need to be alerted when something new is learned about a variant that is previously identified in one of their patients."
Some health care IT managers, like Halamka, say that incorporating genomic data into a clinician's IT infrastructure — where this information will persist and would most likely not be deleted during the course of a patient's life — should not really be a cause for concern at all. "What's the size of my genome? 750 megabytes, which is not significant, so I don't see the storage of the data as the real issue with the cost of a gigabyte at about 50 cents," Halamka says. "The cost to keep the data for a patient for their lifetime is very cheap. Many folks are using cloud-based solutions. You can assume that [personalized medicine IT] is going to be a software-as-a-service model, and the kind of data volumes we're talking about, where it costs 50 cents a patient to store, the issue is more about turning the data into knowledge, not storage."
At Beth Israel Deaconess Medical Center's electronic records hosting center, Halamka and his colleagues have chosen to use a private cloud with eClinical Works EMR that's available to users as a software-as-a-service model. Despite these efforts, there are still concerns about security and the ability to build HIPAA-compliant clouds — be they private or public — where EHRs can be stored indefinitely. For many, cloud computing seems to be the go-to IT solution, as it is inherently scalable and can, in theory, provide clinicians and researchers with an entire IT infrastructure on the fly, complete with storage and analytics. Allowing researchers and clinicians the freedom to access patient data and analysis results from various locations, while at the same time maintaining security and anonymity of patient data, is a key motivator for Halamka and many other IT health managers looking toward internal clouds versus those that are publicly hosted, such as Amazon's EC2 service.
"Why we keep talking about this cloud computing stuff is that users want storage space and lots of computing power to do this preprocessing and then lots of analysis, [but] at the same time I need to be very sure that this data is properly protected," says Krishna Sankhavaram, director of research information systems and technology development at the University of Texas MD Anderson Cancer Center. "Sometimes it could be de-identified patient data or research data that's unpublished. So we have some very well defined IT governance and oversight at MD Anderson, and very formal security policies that we all stick with."
At MD Anderson, researchers and clinicians have access to a private cloud to help translate genomic, transcriptional, and proteomic patient data into personalized treatment programs. The center — which in the near future expects to handle roughly 10 to 15 petabytes of 'omics data from patients per year — currently boasts an 8,064-core blade cluster for next-generation sequence analysis, dosing calculations for radiation therapy, and outcomes research, as well as a 192-core server cluster for initial next-gen sequencing pre-processing and molecular modeling. However, solutions to the IT challenges of personalized medicine will not be found in IT alone. "You cannot just throw IT technology at this and say, 'I know how to connect these databases and run some code or store everything in some semantic store' — it doesn't work that way. People should look at what needs to be done and then have IT come in and streamline it," Sankhavaram says. "Personalized medicine is not easy or trivial at all — it's very complicated. For example, the vocabularies are a huge challenge because everyone has their own way of doing things — even in a hospital like ours — and they don't match."
One approach that could be a boon for personalized medicine IT and the data connectivity hurdles described by Sankhavaram — where ontologies and medical definitions are not standardized — is semantic Web technology. System-to-system interconnection of information — which is especially crucial in personalized medicine, where the complexity of analytic results must be made tractable in the clinic — is dependent upon syntactic and semantic interoperability. "We actually believe that one of the critical variables for the health system to address is: How do we generate information that can be reused by others? And that's why we're worried about semantics, a standards-based representation of information," says Ken Buetow, laboratory chief at the National Cancer Institute. "I think that [semantic technology] has come an incredible distance in the last few years and has the potential to be leveraged in really powerful ways. That said, there are still barriers to its implementation, partially having to do with the large-scale legacy frameworks that exist both in health biomedicine and research as well as the pretty heavy lifting that's necessary to represent information in a semantically rigorous way."
An example of how research and clinical data can come together is the Cancer Biomedical Informatics Grid, which Buetow initiated and currently oversees. By uniting the cancer research community within a traditional grid infrastructure, caBIG demonstrates one scenario in which biomedical data can be securely shared by both organizations and individuals with the type of connectivity that personalized medicine will require. "A lot of the technical IT challenges for personalized medicine are going to go away, and then start to move more towards the next layer of challenges," Buetow says. "These include: How does one connect your genome type to health information to make ... it interpretable to physicians and patients, or even healthy individuals, so that they can make health decisions?"
Part of the problem
As IT teams at research institutes and university hospitals begin to acquaint themselves with exactly what it means to connect the world of genomics to the bedside, it is becoming increasingly clear that storage is only part of the personal genomics quagmire. Trailblazing initiatives aimed at delivering on the promise of personalized medicine, such as the recently established Indiana University Personalized Medicine Institute, are finding that out first hand. "These challenges include the storage of data with integrity, and retrieval with integrity that can be readily validated," says Indiana University Professor David Flockhart, who directs the new institute. "Statistical analysis of high-volume data, including attention to missing fields, multiple sample comparisons, modeling outcomes, and weighting phenotypic data appropriately [is also a challenge] as some phenotypic data are of more integrity or depth and should receive greater weighting than less robust phenotypes."
As far as security goes, the IU Personalized Medicine Institute IT staff will be looking toward already existing legal guidelines. Flockhart says that personalized medicine does not have increased database security requirements. "Treat [personal genomics data] like personal medical data, using all the normal HIPAA guidelines if patients are identified, and this is a minority of the time at the moment," Flockhart says. "Research can be significantly impeded when anonymized sample data are treated as if they represent a threat to the public health and, in general, they do not. That data kept and stored and used are basically personal medical information that is protected by the laws and rules about the protection of medical records, so we do not need to re-invent this [from an IT perspective]."
To some extent, worrying about expanding or modifying IT infrastructure to accommodate personal genomics data is putting the cart before the horse. As Halamka points out, a significant number of primary care physicians in the US are using electronic medical health records, but that number is low when compared to physicians in the Netherlands, for example. "The US has traditionally had fewer electronic health records than other countries. … The average health records system cost about $40,000 to implement. The average family practitioner makes about $100,000 a year, so the economics of buying the application and its implementation have just not aligned," he says.
That physicians need incentives to incorporate IT into their practice points to the need for a massive IT policy sea change in the US health care system before personalized genomics data can be accommodated and exploited in the clinic in any widespread manner. Programs such as the US Department of Health and Human Services' Medicare and Medicaid Electronic Health Records Incentive Program — an effort aimed at encouraging physicians and health professionals to demonstrate meaningful use of electronic health record technology for reimbursement — are examples of those incentives.
Despite the appearance of the current state of US health care as needing to learn to take baby steps into the world of IT, Halamka says that the future is bright for personalized medicine. "We're on the road to exchanging data between electronic health records and personal health records, and I don't think we'll simply have to rip and replace what we have today, we just have to be wise and go forward to make sure that we're sharing data and sharing knowledge."