When the stimulus bill passed last year–allocating $20 billion to help doctors and hospitals adopt electronic medical records (EMRs)–many scientists were excited about the possibilities for medical research. EMRs provide vast amounts of medical information that can be combed automatically and used to ask questions that are too expensive or perhaps unethical to study in traditional clinical trials, such as whether newer, more expensive treatments are more effective than older ones.
“There is a lot of federal funding right now supporting the development of the infrastructure to do that kind of work, as well as to look at comparative effectiveness research using databases,” says Richard Tannen, a physician at the University of Pennsylvania, in Philadelphia. “But it’s a complex and difficult problem, in some ways more difficult than people appreciate.”
While the idea of using electronic medical records for research has been around for more than a decade, it’s only recently started to take off. Scientists and physicians are now scouring the growing number of electronic medical records and genomic databases to figure out how to use this vast medical resource to answer a number of questions in medicine, such as why patients can respond so variably to treatment, and how genetics or other factors might contribute to this.
It has been necessary to invent new analysis methods to glean useful data from often disparate databases, and to make sure that the results produced aren’t biased. Studies based on data from EMRs are subject to the same concerns as observational studies, in which scientists look for links between an individual’s natural behavior and their health. It was observational study that suggested that hormone replacement in postmenopausal women reduced risk of heart attack, while subsequent clinical trials found that the treatment increased risk of heart disease and stroke.
Dan Roden, a clinical pharmacologist at Vanderbilt University, in Nashville, TN, is beginning to address some of those challenges in a pilot project linking EMRs to genomics databases. While he ultimately wants to use EMRs to better understand why different patients can react so differently to the same drug, the project is starting with the most basic questions. “We wanted to ask what genetic information would you want to access to take care of someone, what are the informatics challenges, and what are the ethical challenges in storing people’s information?” says Roden.
His team began by building a DNA database in 2007, extracting DNA from clinical samples collected for other research projects. (Thanks to the way the Vanderbilt medical system is organized, researchers can use such samples for multiple purposes and link that information to the patient’s medical record, while the patient’s identity remains hidden.) The team analyzed DNA from 10,000 people, searching for 21 specific single-letter variations that had been previously linked to different diseases. Using a technique called natural language processing–a sophisticated way of analyzing information–researchers developed a method to reliably identify patients with specific diseases solely from their medical records. The task is more challenging than one might expect; for example, someone may see a rheumatologist for evaluation without actually having rheumatoid arthritis.
By searching for genetic variations that are more common in people with specific diseases, the team confirmed a number of previously identified gene-disease links. The findings, published last week in the American Journal of Human Genetics, show that this type of research can yield useful results.
The team has now expanded the database to 81,000 samples and plans to use it to ask more complex questions. Roden will to try to find genetic predictors of drug response–specific variations that predict whether a patient is unlikely to respond to a specific drug, or more likely to suffer a dangerous or debilitating side effect. “The outcome will be a set of genetic variants that we think will be important to incorporate into medical record,” says Roden. “We want to be able to say, ‘Here’s a person who won’t respond to beta blocker, so they should get a diuretic.’ “
According to Penn’s Tannen, it will likely take years to build up the databases needed to conduct broader clinical research. He estimates that a database of about 50 million people is necessary to ask the types of questions he is most interested in, such as whether patients older than 75 react the same way to a particular therapy as do those who are in their 40s. “That’s the potential great power of database studies,” he says.