Proponents see it as the step beyond disease management. Why wait until a patient becomes part of a chronically ill group?
The 80/20 rule, or some variation of it, often pops up in financial matters. In philanthropy, it holds that 80 percent of a not-for-profit organization’s funds will be contributed by 20 percent of its supporters. Yet, perusing the list of top contributors to the college’s annual fund drive won’t necessarily help you pinpoint the people inclined to cough up seven-figure contributions.
In fact, the guy who’s ready to drop a few million bucks in your lap might not even be on the annual fund list — but he’s out there.
Likewise, a simple chart review may not help an HMO predict which members will suffer catastrophic (and expensive) events, but those patients are out there, too. In many cases, the clues are waiting for the astute fund-raiser or health plan to discover.
In addition, intuition or decades of on-the-job experience might not be sufficient to unearth the clues. Humans, alas, seem to be overly confident in their decision-making skills.
That’s where predictive modeling comes in. The models help people grapple with mountains of data and make decisions about what to do, or not to do, to maximize outcomes.
Such models are, of course, mathematical. They range from relatively simple linear equations to sophisticated software that uses neural networks to accommodate intricate nonlinear relationships. In other words, the working definition of predictive modeling is imprecise, varying from vendor to vendor.
However, in one version or another, predictive modeling seems to be cropping up everywhere now. It’s actually late to come to health care, having long been used in financial services, meteorology, and air traffic control.
“Predictive modeling is hot, hot, hot,” says Al Lewis, executive director of the Disease Management Purchasing Consortium. “We’ve done more bids in predictive modeling this year than in any other category.”
American Healthways, the nation’s largest provider of disease management services, likes to use the phrase “care enhancement” to differentiate its strategy and comprehensive service offering from plain old, pure “disease management.”
The company’s president, Richard Rakowski, says the overarching premise behind care enhancement is that the market is evolving to fill in the gaps in health care, including using predictive modeling to better identify patients who need additional health care resources.
“Disease management has made some inroads, but now we have to expand to a broader population with broader tools as we shift from a focus on utilization management to a focus on outcomes improvement,” he says. “Utilization management has run its course.”
$40 billion market
Whereas disease management programs tend to focus on patients already diagnosed with specific high-cost conditions such as heart failure, diabetes, and asthma in an attempt to control utilization, predictive modeling often casts a broader net, focusing on patient populations and outcomes, not a handful of diseases.
The market still has a need for disease-specific products, Rakowski says, but now it’s time to look at whole populations. Instead of just focusing on chronically ill patients with comorbid conditions and chronically ill patients who have yet to develop comorbid conditions, this broader approach encompasses patients who are at risk for chronic illness as well as those who are not currently at risk.
American Healthways says it has identified 30 “impact conditions” that affect people’s lives but can be prevented, or at least ameliorated, by timely interventions. American Healthways calculates that these 30 conditions accounted for half of the $1.4 trillion spent on direct medical expenses in the United States last year.
Rakowski figures that predictive modeling, combined with care enhancement programs, could save 20 percent of these costs, or $140 billion. If 70 percent of the savings is kept by health plans, that leaves $42 billion in fees to be garnered by American Healthways and its competitors. No wonder predictive modeling is hot, hot, hot.
Of course, there’s the all-important matter of demonstrating that those fees are justified.
American Healthways is so cognizant of the need to validate the outcomes attributed to predictive modeling that it is underwriting an Outcomes Verification Program at Johns Hopkins University, involving the schools of medicine, nursing, and public health.
American Healthways is providing $6.2 million in cash and stock over five years to support the program, whose first assignment will be to evaluate the company’s predictive modeling products. In time, other companies’ products will be evaluated, too.
“True predictive modeling needs to be validated on data sets and reportable performance measures, such as sensitivity and specificity,” says Jeffrey Rice, MD, American Healthways’ executive vice president.
Rice says positive financial results from the implementation of a predictive modeling program generally can be seen in 12 to 24 months.
The approach to predictive modeling at Active Health Management (AHM), based in New York City, is different from most others, says the company’s CEO and president, Lonny Reisman, MD. “We make predictions about individual patients on the basis of extrapolations from the clinical literature,” he says. “We’re applying existing knowledge, not creating new knowledge.”
AHM maintains a panel of 50 specialists nationwide to review the medical literature. The specialists convene quarterly to decide which published findings should be incorporated into the AHM software.
“Within days, we can identify the patients who would benefit from newly published information,” says Reisman. This process vastly accelerates the time needed for a new finding to move from the research bench into widespread clinical application — commonly thought to require 7 to 10 years.
For example, AHM was quick to make use of the results of Heart Outcomes Prevention Evaluation (HOPE) soon after they were published last year in Lancet and the New England Journal of Medicine.
The study showed that an ACE inhibitor, ramipril, is beneficial in a broad range of patients who are at high risk for cardiovascular events but who lack evidence of left ventricular systolic dysfunction or heart failure. The benefits observed were in addition to those achieved via proven secondary prevention measures, such as aspirin, beta blockers, and lipid-lowering agents.
Not living up to potential
In the course of working with an employer coalition interested in improving the quality of health care, Reisman discovered for himself that HMOs weren’t living up to their promise and potential, especially with respect to prevention and caring for the chronically ill.
This became apparent, he says, when quality was assessed via measures that were more clinically robust than the standards used by the National Committee for Quality Assurance and HEDIS.
Reisman also observed that HMOs have collected lots of potentially useful data. Unfortunately, the data tend to be siloed, and an HMO’s goal usually is to reduce costs within each silo. Reisman saw the possibility of integrating the data from all the silos and then applying evidence-based standards to the data.
“This is a concurrent dynamic approach. It’s not just data mining. We’re coupling aggregated data with advances in medical knowledge.”
AHM strives to identify high-risk patients who otherwise might fall through the cracks. This occurs in two ways. First, by breaking down silos and aggregating the data, AHM can help physicians reduce the medical errors that result from communication breakdowns among the many physicians who may be treating a single patient.
Second, AHM helps physicians bridge the knowledge gap resulting from the difficulty every physician has in keeping up with the advances reported in the literature, be they the use of beta blockers to treat heart failure (which until relatively recently was believed to be contraindicated) or the newest biologic agents for treating rheumatoid arthritis.
Once a patient at risk has been identified, AHM generally has provided a specific recommendation to the health plan, which passes it on to the physician. Some of AHM’s newer contracts call for AHM to communicate directly with physicians.
“Predictive modeling significantly enhances the possibility of doing good,” Reisman says. “We’re saving lives and we’re saving money.”
Its product doesn’t focus on utilization, but what Reisman calls “bread-and-butter therapies.” One difference between it and the traditional approach of case management for potentially high-cost patients is its application of evidence-based standards. “Most other models suggest that high-risk cohorts be offered case management services, which is fine, provided there’s a mechanism to ensure that clinical aspects of care are addressed.”
Reisman says AHM’s product has been best received by large employers who are frustrated by what they perceive as managed care’s failure to deliver fully on its promise. MCOs, he says, are open to the idea of predictive modeling, but with reservations: Some are concerned it may be intrusive, despite the fact that physicians generally appreciate the input because of its clinical value. Others claim they are developing similar systems themselves.
The federal government also has expressed interest, and AHM has two pilot programs under way with the Federal Employee Health Benefits program and Medicaid.
Finding high-risk candidates
In Wisconsin, Wausau Benefits, a third-party administrator serving commercial populations under age 65, has adopted AHM software and its own proprietary algorithms to develop a “high-tech, high-touch” predictive modeling program, a component of its comprehensive service encompassing utilization management, case management, disease management, and population health services.
Elaine Mischler, MD, Wausau Benefits’ medical director, says the company uses computer software — the high-tech aspect — to identify those patients who can be helped by disease management in eight conditions: asthma, COPD, congestive heart failure, coronary artery disease, depression, diabetes, hypertension, and low-back pain.
Wausau studied its 400,000 lives to find a critical marker for each condition that could place patients in high-cost and low-cost cohorts. Although these markers are proprietary, Mischler revealed the key markers for two conditions: an HbA1c test for patients with diabetes, and the lack of ER visits for patients with asthma.
These markers pointed the way to significant savings: $90 less per member per month for diabetes patients who had the test versus those who did not, and $69 less PMPM for asthma patients who had no ER visits versus those who did. This analysis was done on the population before institution of any disease management programs.
The intent of the program is to help Wausau’s customers move patients from the high-cost cohort to the low-cost cohort — and keep them there. “We want these patients to be optimally well despite their illness,” Mischler says.
That’s where the high-touch aspect of case management comes in — a team of 30 experienced nurses doing utilization management, case management, and disease management over the phone.
“Twenty percent of the patients need ‘high touch’ — something more than a brochure about a condition or drug information mailed to them,” she says.
Wausau is so confident that its predictive modeling product will generate meaningful savings that it will put its fee at risk for a customer with 5,000 employees or more. Because of the difficulties of making predictions on smaller numbers of employees, Wausau will discuss putting a portion of the fee at risk for smaller customers.
A question of ethics?
Ultimately, predictive modeling is aimed at improving outcomes and reducing costs by helping clinicians — over the short term, to recover from bad decisions, be they errors of commission or omission — and, over the long term, to make better decisions.
Although some physicians may object to predictive modeling on the grounds that it intrudes on their professional autonomy, another school of thought holds that, in the presence of a valid predictive model, it’s unethical for a practitioner not to use it. A long history of predictive modeling in disparate fields demonstrates its superiority over mere clinical experience and intuition.
Predictive modeling vs. the experts
In each of the studies from disparate fields listed below, predictions based on statistical predictive rules (SPRs) were more reliable than predictions by experts.
|Prediction Studied||Experts||Date Published|
|Success of electroshock therapy||Medical and psychological hospital staff||1941|
|Academic performance||Admissions officers at selective colleges, medical schools, law schools, graduate school in psychology||1957, 1971, 2000|
|Diagnosing patients as neurotic or psychotic||Clinical psychologists||1968|
|Risk of SIDS in newborns||Clinicians||1975, 1977, 1985|
|Credit risk||Bank officers||1983|
|Presence, location, and cause of brain damage||Experienced clinicians, prominent neuropsychologist||1983|
|Progressive brain dysfunction||Clinicians||1984|
|Quality of vintage of red Bordeaux||Wine tasters||1995|
|SOURCE: TROUT JD, BISHOP M.: 50 YEARS OF SUCCESSFUL PREDICTIVE MODELING SHOULD BE ENOUGH: LESSONS FOR PHILOSOPHY OF SCIENCE.
Available at «http://hypatia.ss.uci.edu/lps/psa2k/fifty-years.pdf».
Paul Lendner ist ein praktizierender Experte im Bereich Gesundheit, Medizin und Fitness. Er schreibt bereits seit über 5 Jahren für das Managed Care Mag. Mit seinen Artikeln, die einen einzigartigen Expertenstatus nachweißen, liefert er unseren Lesern nicht nur Mehrwert, sondern auch Hilfestellung bei ihren Problemen.