 |
| Michael L. Millenson |
It was Archimedes who famously said, “Give me a place to stand and a lever long enough, and I will move the world.” For those who work in the world of predictive medicine, a swelling stream of digitized data represents the lever they hope will prove powerful enough to move health care in a radically new direction.
At the “micro” level, predictive medicine is all about genomics, proteomics (the set of proteins within a particular cell type) and individual biomarkers of diseases with well-defined precursors, such as some breast cancer. Less glamorous, but far more advanced in actual application, is the “macro” level of prediction. Practitioners use readily available health information from all members of a population to forecast its future health needs and health care costs. The aim is to uncover what groups of individuals are likely to get sick, their risk for specific diseases (typically chronic conditions, such as diabetes) and the likely treatment costs they will incur. In contrast to some of the predictions based on genetic predisposition, the macro predictors focus on the near future, a time frame of keen interest to everyone from practicing clinicians to insurance underwriters.
Prevent High-Cost Interventions
Like the witches in Macbeth, whose mysterious boiling cauldron is fed with a weird jumble of ingredients, predictive modelers also concoct their own recipe consisting of various risk adjustment calculations meant to mime the messiness of real-life variation. While eye of newt and toe of frog were not actually mentioned at a recent joint meeting of the Society of Actuaries and the Disease Management Association of America, repeated references to the “receiver operating curve,” “r-squared” and “run-out” seemed equally esoteric to the uninitiated.
Depending on your point of view, the forecasts are terrible or terrific. Arlene Ash, a pioneering mathematical researcher as well as co-founder of the firm DxCG, remembers when models could predict no more than 6 percent of any one individual’s variation in total cost. Today, that number has risen to as much as 15 percent.
Not impressed? Then take a step backward. When current predictive models are applied to a population of any notable size, the ability to forecast which group of patients will use 10, 20 or even 100 times the resources of another group soars as high as 98 percent. Put differently, while you may not be able to say much about any particular tree, you can say quite a bit about different parts of the forest.
Moreover, when the data on particular patient subgroups are subjected to further scrutiny, one can engage in “case picking,” that is, finding patients for whom timely intervention may prevent the progression from low-cost to high-cost illness. It’s high blood pressure without the heart attack, asthma without the emergency department visit.
IT Can Help or Harm
If the law of large numbers gives one lever to predictive modelers, an even more powerful extension is provided by the stream of digitized data emerging from the growth in health information technology. Pharmacy data and lab results are becoming more comprehensive and more readily available. Electronic medical records capture greater clinical detail, while personal health records offer the potential of using patients’ own perceptions of their health status as a regularly updated data point. Remote monitoring devices comprise a kind of electronic diary. And in the clinical realm, expanding micro predictions may one day lead to synthesizing the discoveries of geneticists with the acumen of actuaries.
The good news is that predictive medicine can make care more personalized and more effective, yet still help lower overall costs. After all, early interventions that preserve health are far less costly than later interventions meant to restore it.
The bad news is that personalization can cut two ways. Health plans and physician groups that can identify more expensive patient subgroups may be tempted to make themselves unattractive to those subgroups. Moreover, patients concerned about their privacy may not be assuaged by assurances that those who stir ever-more-personal information into a rapidly expanding data stream have nothing but benign intentions. Stolen laptops and other recent security lapses have given rise to plenty of toil and trouble in the IT arena.
Hospitals have a special role to play in this brave new world. Cast a net over any hospital service area, and you’ll find that residents often stay in the community even when their job or health insurance carrier changes. That stability gives hospitals the opportunity--and the obligation--to leverage the power of predictive modeling to improve community health. Clinical and financial considerations can be harmonized, rather than pitted against each other.
You don’t need a crystal ball to see that a sense of responsibility for all segments of the population, from the very sick to the robustly healthy, will be critical to the hospital mission of the future.
Michael L. Millenson, based in Highland Park, Ill., is an author, consultant and a visiting scholar at Northwestern University’s Kellogg School of Management in Evanston, Ill. He also is a regular contributor to Most Wired OnLine.
GIVE US YOUR COMMENTS!
HHNMostWired welcomes your comment on this article. E-mail your comments to hhn@healthforum.com, fax them to Most Wired Magazine Editor at (312) 422-4500, or mail them to Editor, Most Wired Magazine, Health Forum, One North Franklin, Chicago, IL 60606.
If you would like a FREE Subscription to Most Wired OnLine, please click here to register.
This article first appeared on August 8, 2006 in HHN's Magazine online site.
To respond to this article, please click here.
