Would you buy a product that promised that 60 percent of the time it works every time? Maybe for caricature news anchors like Ron Burgundy, there is no question that a method (exotic cologne) with this type of track record (for attracting women) would be a good investment. But what if that rate was found to be true for a surgery that cost tens of thousands of dollars and might save your life—with a small risk of serious complications? Or a relatively cheap allergy medication that was prone to causing headaches?

Such calculations—sometimes more eloquently phrased—are a mainstay of health care, especially as providers, patients and insurers face soaring costs and shrinking budgets.

But most of the current assessments of costs, risks and likely benefits are averages from a broad population. So any given individual—or subset of individuals—could have vastly different outcomes. And for those who have looked deeply at the data, those instances of great success or devastating side effects are not always random—a pattern that indicates that averages are not only superficial, but also can be hazardous.

"The choice that maximizes the population's health," write the authors of a new essay on the subject, "is not always the same as the best choice for a specific individual." And that is where our current health care calculations often fall short, says John Ioannidis, of the Stanford School of Medicine's Prevention Research Center and co-author of anew paper, published online July 12 in PLoS Medicine.

The medical field is awash in contingencies—every patient his or her own universe of variables. So when a new treatment is under evaluation, researchers often assess it in terms of how many high-quality days it adds to patients' lives on average—and the price is averaged out across that extra time. So an arthritis drug might run on average $4,849 per every extra year of "best quality life" (i.e. quality-adjusted life year, or QALY) that it delivers for an average patient. These reckonings allow for easy comparisons of a variety of different drugs and procedures. But of course this flat dollar value fails to acknowledge that some groups of people might see far greater benefits from the medication, whereas other groups might suffer worse outcomes than if they took a comparatively higher-priced medication—or perhaps than if they took no medications at all. Knowing those odds for specific patients or types of patients at least could save time, anguish and money wasted on ineffective interventions.

Some researchers have been calling for better individualized data for years. Richard Kravitz, a professor and co-vice chair of research at the University of California, Davis Division of General Medicine, co-authored a paper outlining the hazards of averages, several years ago. "It should seem obvious that treatment effects are not necessarily the same for everyone," he and his co-authors wrote in the 2004 article. "What may not be so obvious is that misapplying averages can cause harm, by either giving patients treatments that do not help or denying patients treatments that would help them."

A complex computation might not be able to dial up your own personal chances of having success with—or complications from—any given treatment any time too soon. But Ioannidis and his colleagues hope that with better information and analysis, providers and insurers can help to optimize the application of the interventions that are already out there.

Deficient in data
In this awkward era of trying to cut medical costs while improving care, researchers are often faced with a paucity of data on which to base their calculations.

Right now, the cost-effectiveness of a handful of interventions has been parsed for different subpopulations of patients. But Kravitz says that "it's really hard to get reliable information" for subpopulations. In fact, it's even hard to get reliable information on the general population. Big clinical trials are expensive, time-consuming and complicated. So setting them up to examine different subgroups of people—with enough in each group to make findings statistically significant—is a challenge. But Kravitz considers it a crucial step to the world of more personalized and more efficient treatment.

Ioannidis agrees. "I see this as a mainstream research effort," he says.

These sorts of tailored prescriptions would be a boon to those who do not fit in the comfortable center of the bell-curve distribution for any given intervention. But, says Kravitz, it could encounter resistance within the existing healthcare machinery. "It will drive the payers crazy," he says. For insurance companies and government programs that are accustomed to streamlined averages, this new model "breaks apart the single rule that applies to everybody." In the end, however, it could save money by helping to select the intervention that will have the most benefit delivered at the lowest cost for a certain profile, say, a middle-aged diabetic Asian-American man who is at high risk for a stroke—not just the general population.

Comments