Why hybrids are hot: the value of hybrid predictive models, which allow employers and health plans to analyze claims and health-risk assessment data, outweigh the benefits of using either model by itself, proponents of hybridized models argue.
Or, "We think you may be at risk for heart disease, and we can provide health coaching to help you avoid that." Would you consider it your lucky day?
In the name of controlling health care costs, these kinds of calls can indeed be fortunate events.
As the discipline of disease management enters a new, more mature era, its ability to identify just who needs to receive such preventive contact is entering a new phase as well.
Predictive modeling, the name given to techniques used to determine which employees or health plan members are most likely to incur the highest medical costs in the near or medium term, is becoming more advanced.
Cutting-edge modelers at disease management firms are combining survey information provided by individual health-risk assessments with historic and real-time health care utilization data to predict which patients will most likely get sick, and how much they are likely to cost, a year or more into the future.
"Traditionally, our field has used just medical and pharmacy claims to make these predictions, but now we are starting to simultaneously incorporate self-report information and other types of data, which dramatically improves predictive value," says Derek Newell, vice president for commercial accounts at Life-Masters, a disease-management company in Irvine, Calif.
And yet, many firms that consult in disease management remain highly sophisticated in primarily one or the other of these areas, providing predictive modeling through analysis of claims or self-reported health-risk assessment data.
American Healthways, a large disease-management company in Nashville, Tenn., uses claims data to determine which individuals in commercial health plans and employer groups may be entering disease-management categories.
With more than 1.5 million lives under management, the company can look at groups of people based on their use of medical care, determine which are probably sick, and identify those who are likely to cost the health plan more--or less--in the next fiscal year.
"Even after a triggering event, there is value in catching these individuals, because they can get phenomenally more expensive--especially if they don't get the kind of coaching and other management that helps to assure that they receive appropriate standards of care," says Carter Coberley, vice president of informatics at American Healthways.
With the claims approach, modelers have the advantage of being able to reassess groups as quickly as usage data arrives. And the frequency of this will depend on the contract they have with each client.
"We can receive data as often as the health plan can provide it," says Heath Shackleford, a spokesman for American Healthways. "Most commonly, we get it monthly or quarterly. But for some accounts we even get data that refreshes nightly."
This allows the disease management program to record new events in a person's health status. But for the most part, the approach identifies persons who have already entered a disease category and who are likely to be expensive, even with intervention.
"The fact is that these individuals are already sick," says Julie Meek, CEO of The Haelan Group, which uses an approach based on the health-risk assessment model. "Predictive modeling ought to be a way to find employees before they actually generate claims and utilize health care."
Haelan, based in Indianapolis, serves large and midsize self-insured employers. The firm surveys employees, collecting information on each subject's medical history, family, diet and lifestyle, self-perceived risks and other areas. The group then processes the data using algorithms that it has developed.
The predictive models that emerge are aimed at finding employees who are a high risk but are not yet costing the health plan much money; who don't yet belong in any disease-management category, but are headed there.
"Employers are also realizing that they can't just go after the 3 percent of employees most at risk, because then they end up actively coaching only the sickest 1 percent of their patients," says Meek. "You won't get a good ROI on the effort unless you go after the top 10 percent of at-risk individuals."
She adds: "That's the sweet spot in predictability and impactibility that allows you to ultimately manage the 6 percent to 7 percent of your population who are or will be your high utilizers."
But finding this group requires gathering data voluntarily from employees. (The specifics of the information for individuals must be kept from employers for privacy reasons.)
Most often, employer programs will try to gather the information at open enrollment times, and here incentives have made big differences in participation.
"Carrots" offered to employees for completing health-risk assessments may include lowering the coming year's premiums, cash or bonus checks, or copay waivers. As a result, participation rates of 80 percent or more are not unusual.
TWO SOURCES A BETTER BET
But the programs collecting self-report data generally only do so once per year. What happens if there is a change in a person's health status during the year?
Leading modelers are increasingly turning to the power of predictive systems that use both types of data--self-report and usage data.
Some smaller, more agile companies are realizing the potential of this combined approach and taking it to the bank by buying competitors with the expertise.
To stay at the forefront of such advances, for example, LifeMasters recently acquired Medical Scientist Inc., a San Francisco-based leader in predictive modeling.
"If we get a population of 30,000, say, from Aetna or some other large carrier or self-insurer, they think they know who to worry about," says Newell. "But we can help them truly stratify people in terms of types of interventions needed, and improve their ability to predict which people will be catastrophically expensive."
He notes that customers often ask the company to do a claims analysis first, to "find low-hanging fruit and predict cost drivers." As a second phase, they may then begin to look at health-risk assessment data on subsets of the population.
Like other large disease management providers, LifeMasters can also deliver nurse-based coaching, consulting and education, which can include apprising the individual's physician.
Once the patient is profiled, responses can be updated quickly, particularly if the client shares new claims rapidly. If the person is in contact with a health coach and provides updates in person, then too, the data can be updated quickly. American Healthways, which runs one of the largest disease-management call centers, uses this approach to collect self-report data.
"We perform a routine health-risk assessment on anyone who has already been flagged," says Coberley. American Healthways also offers a health-risk assessment package that anyone in their managed populations can complete over the Web.
Additionally, disease managers may incorporate actual test-result data in their predictions. This requires patient permission and may involve soliciting copies of original lab forms from the patient. And in the not-too-distant future, modelers may also weave other types and categories of information into their overall rubric (see "Digging for Other Sources of Data").
"You need to plug in different models to get the people flying below the radar. It's a combination that works best," says Michael Haffney, president of Agency Associates Inc., in Indianapolis.
Haffney is an example of the kind of benefits broker who is recommending predictive modeling to clients.
A CDHP ALTERNATIVE
Both claims and health-risk assessment data, used alone, can lead to steps that keep people healthy and medical costs low.
But proponents of the more multifaceted modeling maintain that employers and health plans have not gotten enough bang for the buck out of conventional disease management, and the hybridized predictive modeling is part of the solution.
"There's tremendous confusion in the marketplace even about the term 'predictive modeling,' which has been used by insurers in a pure actuarial sense, to predict costs for the coming year and to adjust risk. But our modeling is a newer approach that predicts high-care users this year and beyond. It's simple to understand but complex to do," says Meek.
In general, large carriers deserve credit for expanding the use of disease management in health care; however, they have not been as limber in embracing new advances. American Healthways, which serves primarily insurance clients, is cautious in revealing its interest in self-report modeling.
"I can't comment on whether we are doing it, but I can say that there is potential to couple data. To maximize benefits you need to be doing both," says Coberley. "We are looking for ways to link different types of data and synergize them to produce the greatest return on investment. Currently, we can mine that information in our database from the health-risk assessments that are completed, and we are studying it. It's a high priority in research and development. Time will bear out to what extent we can combine these approaches."
But turnover is high in health plans and, with new members arriving each year, not knowing who might end up being expensive is a liability.
"Traditionally insurers have been good groupers but not good predictors," explains Newell. "It's exciting that employers, who often have a longer-term relationship with individuals, are the ones pushing the envelope on this."
Cutting-edge predictive modelers may exaggerate when they refer to their science as a form of artificial intelligence, but it certainly delivers intelligence on health plan members, sometimes to their surprise.
When disease management programs contact individuals for the first time, they don't generally ask for personal information, but try to confirm if there are reasons for collecting more information and then work gently toward getting it and having subjects complete an assessment.
"There's no legal issue with calling members based on their claims data, but there's a personal issue," says Newell. "We get different types of reactions but only a minority of them are defensive about how we got their information. You have to gain their trust."
Says Meek, "Relying just on claims data is like driving while looking only through your rear-and side-view mirrors. To look out the front windshield you need to be using a prospective model. Combine that with good telephonic coaching and you can really start to overcome decisional and behavioral barriers in people's lives [that affect their medical costs]."
"Consumer-directed health plans are getting all the press right now, and everyone thinks it's the way to save money, but we're still waiting for more research on that," says Haffney, the insurance agent.
"With health-risk assessments, predictive modeling and health coaching, we already have a large body of knowledge that there is a return on investment. I'm presenting that to every, even partially self-funded, client that I have. I consider it a new role for brokers."
RELATED ARTICLE: Digging for other sources of data.
What other types of data might modelers use? And which might they find statistically significant in predicting which employees will cost a health plan the most? How about a person's annual review in the workplace, performance appraisals, attendance at work, job type and reported job satisfaction?
Some modelers are already in the formative stages of wrapping this kind of human resources data into the calculus they use to help their clients identify subsets of employees who will demand the most from a health plan.
(Using data traditionally found in a human resources department will enhance the need to use coding that ensures no breach of individual privacy by disease managers or employers looking at results.)
Modelers may also use age, race, zip codes, census data, and other demographic information--in theory, anything that improves their precision in pointing to those who have or will enter a disease-management category. The specifics of the information are used for prediction only, not for management; whereas, claims and health risk assessment data are used for both.
"This is the next level of predictive modeling in the employer market--taking a comprehensive look [at what is known about populations], possibly with the intent of designing benefits packages around key chronic diseases that are driving cost," says Newell.--Russ Allen
RUSS ALLEN, a Pennsylvania-based writer, is a frequent contributor of health care-related stories to Risk & Insurance[R]. He can be reached at risk email@example.com.
|Printer friendly Cite/link Email Feedback|
|Publication:||Risk & Insurance|
|Date:||Jun 1, 2005|
|Previous Article:||Expanding the reach of RMIS: in today's world, risk managers need information they can act on, and a risk management information system that can...|
|Next Article:||Dwellers of the valley of the rifts: "buyer beware" need not apply henceforth, proclaims Marsh's new CEO. Yet Aon's chairman beseeches buyers to...|