David Vanness, assistant professor of population health sciences at the University of Wisconsin-Madison, shows a slide from a statistical model in his office. Vanness develops mathematical models to assess the effectiveness of different health care treatments.
Madison - What is the maximum dose of a drug that will limit side effects to an acceptable level? How many patients would benefit from a treatment and how many would be harmed? Would the potential benefit of screening everyone for a disease outweigh the potential harm to people who receive unneeded care because of inaccurate results?
These are common questions in modern medicine. Yet answering them often would require clinical trials that span decades.That's not practical or even feasible. Instead, researchers are turning to computer models, meshing math and biology to develop a better understanding of the effectiveness of diagnostic tests and treatments.David Vanness at the University of Wisconsin School of Medicine and Public Health is one of those researchers.Vanness, an economist, spends a good part of his days writing intricate equations designed to gauge what is known - and what isn't known - about how well a diagnostic test or treatment works."My goal is to help take the research that's already being done about new medical technologies and help decision-makers determine whether they are valuable to patients and to society," he said.
The amount of information generated by the advances in medicine often is too overwhelming and too complex to analyze and understand without computer models.
It's an esoteric field - the domain of people with a gift for math - but the work of Vanness and other researchers could one day help ensure that patients receive the most effective care.
The goal is to give doctors better information on which tests, drugs and procedures work best for specific patients. Too often, that information simply doesn't exist, particularly for new treatments."We are always playing catch-up," Vanness said. "The joke in technology assessment is it's always too early to assess a technology until it's too late."
Vanness' expertise is building Bayesian models - a type of statistical analysis used to determine probability and measure uncertainty.
Advances in medicine often start with imperfect evidence and build on that. Bayesian models are a way of gauging how much confidence you can have in research findings, such as whether patients or a specific group of patients would benefit from a treatment.
"That's really what my job is about - trying to find out what the data can tell us that's of value and trying to highlight where it fails and why it fails," he said.
Making sense of data
Computer models enable researchers to integrate and analyze data from different sources, such as clinical trials, medical records and insurance claims.
"You've got to find something to string it all together," Vanness said, "and you use mathematical models and computers to do that."
That, in turn, could help doctors and scientists answer questions about the effectiveness of different treatments more quickly.
If you want to know where to focus your research energies and resources," Vanness said, "then you have to know what you don't know."
The work brings together an array of disciplines - medicine, applied math, biostatistics, epidemiology, economics, engineering.
"It's a grab bag," he said.
On Wall Street, Vanness would be known as a "quant," and he could have plied his skills building mathematical models to help predict whether an investment would go up or down.
Models are widely used in weighing and analyzing potential benefits and harms from a treatment or diagnostic test.
Researchers use models, for example, to help determine whether patients should be screened for a disease, estimating how many patients would benefit by detecting the disease early and how many would be harmed by additional tests or unnecessary treatments when the initial results are inaccurate.
"Modeling is a way to help you weigh those trade-offs and understand how important they are," said Douglas Owens, a senior investigator at VA Palo Alto Health Care System and a professor of medicine at Stanford University.
An example is the controversial recommendation that screening most women in their 40s for breast cancer would result in more harm - because of the risk of complications from biopsies when mammograms incorrectly indicate a possible tumor - than the overall benefit of detecting cancer in some women.
The recommendation sparked a firestorm of criticism. And without question, some women are alive today because they were screened in their 40s. But that doesn't undercut the panel's conclusions that the harms overall outweighed the benefits.
A bigger role ahead
Models have inherent limitations. They are only as good as the data and assumptions on which they are based. They also can make dumb - or wildly varying - predictions if the researcher misses something in the data.
But models are being used more cautiously, Vanness said, and in the past decade they have gotten better at measuring uncertainty.
"But there's still a lot to be learned," he said.
At the same time, models are certain to have a larger role in research, in part because of advances in computer processing power and new statistical tools for analyzing and interpreting data and assessing uncertainty.
"The kind of things we do routinely now would have been very difficult to do 20 years ago," said Owens, the VA Palo Alto investigator.
Models also hold the potential for helping doctors and scientists determine the effectiveness of different treatments more quickly.
For example, more information will become available from electronic health records in the coming decade, such as how patients respond to a specific treatment, and models can help analyze that data.
They also could become an indispensable tool as more is learned about the link between someone's genetic makeup and disease and how he or she may respond to a specific treatment.
An additional advantage is they can be revised and updated when new information becomes available.
The ultimate goal is to provide better information to doctors and patients. That also could help make better use of the money spent on health care.
"We really are looking to direct resources toward things that work the best," said Pamela McMahon, associate director of the Institute for Technology Assessment at Massachusetts General Hospital.
Lowering health costs
Roughly half of the growth in medical spending is from new technology. Yet, whether a new test or treatment results in better outcomes for patients, or is more effective than less costly options, often isn't known.
Determining what treatments help patients will be essential if the United States and other countries are to slow the rise in health care spending.
Doctors, economists and policy analysts have long called for more research on comparative effectiveness - research that looks at what works best for specific patients.
The health care reform law allocated a total of $560 million through 2013 and more than $500 million a year beginning in 2014 on comparative-effectiveness research.
In addition, the Recovery Act of 2009 allocated $1.1 billion for the research.
The initiative will face its share of challenges. The research can be immensely complex and rarely yields straightforward answers. And ways must be found to find answers more quickly.
That will require new ways of analyzing data and new research methods.
"So much data gets collected in the clinical trials that are conducted that could have the keys to all sorts of treatment choices," Vanness said. "We just don't look at it."
"The hardest part about what I do sometimes is just this feeling that the answer is out there," he added, "and it's locked in the database somewhere or it's locked in the experiences of patients who just simply aren't being asked, 'Hey, how are you doing?'
"If we could just find efficient ways to capture and analyze that data, we could have a much healthier population."
Reporter Guy Boulton conducted research for this story during a fellowship funded by the Kaiser Family Foundation, a nonprofit, nonpartisan health policy research organization with offices in Menlo Park, Calif., and Washington, D.C.
By Guy Boulton of the Journal Sentinel
No comments:
Post a Comment