[ back ]
Patterns in the chaos
Where medicine meets math, some unexpected insights can emerge
The story of blockbuster drug-gone-wrong Vioxx is now infamous: The popular painkiller came on the market in 1999, was taken by millions of people, and was finally withdrawn in 2004 after it was found to dramatically increase the risk of heart attack and stroke. Could the link between the drug and these deadly side effects have been detected early, potentially saving thousands of lives? What would it have taken to make the connection?
Three elements are required: vast amounts of data, intricately designed computer software and a motley team of bioinformatics experts to make sense of it all. All these elements come together in the Children's Hospital Informatics Program (CHIP), a data laboratory at Children's Hospital Boston, with ties to Harvard and MIT, where researchers specializing in medicine, biology, statistical analysis and mathematics pull patterns from a sea of biological and clinical data. Instead of experimenting with microscopes, petri dishes and tissue samples, they use algorithms and mathematical calculations. Once they make a hypothesis, they can quickly create software systems to mine and manipulate data sets, turning the byproducts of the health care system into drivers of discovery.
The field these investigators work in—biomedical informatics research—is fast-growing, thanks to electronic access to mountains of information generated by the medical and research fields, from insurance claims to genome sequence databases to public health records. But keeping up with this constant stream of information can be an uphill battle—and envisioning how seemingly unconnected data sets might be integrated and examined isn't easy.
"What to do with all this data and how to look for patterns is still an emerging science," says CHIP investigator Kenneth Mandl, MD, MPH, who is also an attending emergency medicine physician at Children's. "But the idea is to take this great data the health system spits out every day and use it to monitor health and to see how our health system is doing." In the case of Vioxx, the data were ripe for the picking—it just took someone who could piece it all together.
Behind biosurveillance
A major interest of CHIP is biosurveillance, or monitoring for disease outbreaks—both naturally occurring and those caused by bioterrorism. In the late 1990s, Mandl and Ben Reis, PhD, created the Automated Epidemiological Geotemporal Integrated Surveillance system (AEGIS) to compile all the activity in Children's Emergency Department (ED) as it occurred, while maintaining each patient's anonymity. Still in use today, the software sifts through patient information and compares it with hundreds of thousands of previous ED visits, sounding an alarm if it detects unusual patterns of illness that could indicate an outbreak.AEGIS's success led Mandl to share the software with the government for use as a state-wide surveillance system. It now runs continuously in all EDs across Massachusetts, enabling public health officials to locate disease clusters, track how they spread and identify early signs of a possible biological attack. "What's exciting is that through surveillance, we can make public health recommendations based on real-time data to prevent and control outbreaks," says John Brownstein, PhD, a CHIP expert in infectious disease epidemiology. "On the flip side, we're collecting clinical data that can better help us understand all kinds of diseases."
Examining ED data from 2000 to 2004, for example, Brownstein and Mandl showed that 3- and 4-year-olds are the first to come into EDs with influenza-related illnesses, arriving as early as September (see graph below). This finding, indicating that preschoolers lead the annual flu epidemics, affected vaccination policies throughout the country: the Advisory Committee on Immunization Practices (ACIP) recently recommended that 2- to 5-year-olds get vaccinated each year, along with their parents and siblings.
In another flu project, examining how flu spreads between cities, Brownstein and Mandl combined data sets from the Centers for Disease Control and Prevention with airline flight data and found that the 2001-2002 flu season was delayed—coinciding with the halt in flights after the terrorist attacks of September 11. Usually, flu season peaks around February 17, but post-9/11, it was delayed until March 2 (see graph below). This analysis has powerful implications, says Brownstein: Limiting airline travel could give health care providers a crucial head start on controlling a flu pandemic.
Monitoring drug safety
CHIP researchers have been translating their biosurveillance methods and applying them to a sister field—pharmacovigilance, a burgeoning area of research that monitors the safety of drugs once they go on the market. "A drug enters a population just like a virus does," says Brownstein. "So we look at the pattern of that drug's effects in a way that's similar to studying the spread of disease."The recent study on Vioxx shows how population health monitoring can potentially spare thousands from unforeseen harmful or even fatal drug side effects by identifying events that were missed during the drug approval process. Mandl, Brownstein and Isaac Kohane, MD, PhD, director of CHIP, hypothesized that heart attacks linked to Vioxx could have been detected much earlier. Analyzing two major Boston hospitals' records from the past 10 years, they found a clear spike in heart attacks, nearly a 20 percent jump, just eight months after Vioxx was introduced—and more than four years before it was pulled from the market. "It's amazing we saw an effect this clear," Brownstein says. "But it's even more amazing that nobody noticed."
"The problem was that no one was looking," says Mandl. "People haven't taken the surveillance approach to health. But this finding suggests there's a real need to monitor the health of populations broadly and look at the impact of therapies."
Mandl and his team are already looking at other drugs, like Avandia, approved to treat type 2 diabetes, to see if they can detect effects that might have been missed during the drug's testing phase, or pick up later problems not reported by drug companies. "The issue is that drug approvals are based on small trials that are focused on efficacy, not safety," says Mandl. "Drugs behave differently in large numbers than they do in small tests, and those rare side effects do happen." In this way, CHIP is playing the role of ombudsman to the pharmaceutical industry—a role that is sorely needed, according to Brownstein. "Our hope is that we can build a system to look at all drugs and adverse events and learn how to find the needles in haystacks," he says.
Taking it to the streets
Taking pharmacovigilance in a different direction, Brownstein recently looked for early signs of abuse of prescription opioid pain killers like Oxycontin. "In Boston, opioid abuse seems to be a huge problem, so we've been adapting our outbreak surveillance systems to look for clusters of abuse," he says. "It's really about having a hunch that there could be some kind of link out there, and then creating a system to figure it out."In this case, Brownstein looked at media data on news events, such as when and where a pharmacy was robbed, a dealer busted or a doctor reported to be over-prescribing, and pitted these reports against data from clinical sources such as EDs, poison control centers and drug treatment facilities. He has found that an increase in media reports actually preceded clinical outcomes, such as overdoses, reported by these sources. "We find these news events to be early indications of problems in the community," he says. "It's fascinating that the media actually act a sentinel information source."
Pushing the limits
While CHIP researchers can readily mine clinical data from patients seen at Children's, as well as certain public health databases, they've come up against limits. Data sharing between hospitals or research institutions doesn't exist, in part because of government regulations designed to protect patients' privacy. "Data may be shared temporarily for joint research efforts for multi-institutional collaborations, and after September 11, we were able to get more hospitals to share ED data for biosurveillance," says Mandl. "But those are exceptions and represent very narrow slices of data sets. There's surprisingly little information sharing going on."Since patients' clinical information isn't shared among institutions, many researchers are hoping that the advent of electronic personal health records will allow them to gather data directly from individuals. More than a decade ago, CHIP researchers pioneered a secure, lifelong, Web-based personal medical record system, now called Indivo, that turns control of medical records over to patients, who can choose whether to grant access to institutions, clinicians and researchers. Indivo served as the model for more recent large-scale systems developed by Microsoft and Google, and large employers like Wal-Mart, Intel and AT&T are now implementing the CHIP software for their employees.
It may seem counterintuitive, but Kohane and Mandl believe that putting patients in the driver's seat would make them more inclined to share their information for the public good, thus allowing far more data produced during routine medical care to be used in public health surveillance and research.
"This would be an important step forward for medicine," says Mandl. "We could look at the impact of our therapies on health in a highly responsible and comprehensive way. For the first time since information has been stored electronically, we could begin to use that information and see benefits in the not-too-distant future. The public assumes that we've been keeping better track than we have been. As a profession, we have the opportunity to live up to those expectations."