Gaining key business insights ensures that the right people receive the right healthcare now and in the future.
With the financial decline of 2008 adversely affecting most sectors of the economy, the healthcare industry is feeling the wake of the economic turbulence. Health plan profitability is down, largely due to investment losses, and employers are aggressively seeking ways to control costs. Across the industry, organizations are being challenged to do more with less, and eliminate programs that cannot demonstrate a solid return on investment.
Gaining key business insights ensures that the right people receive the right healthcare now and in the future.
Matthew McGinnis, senior director of Healthways’ Center for Health Research, saw the challenge even before last year’s economic downturn. “Eventually, we might have fewer resources available relative to our membership,” said McGinnis last April. “So, the ability to zero-in on the right people at the right time will be even more critical in the next 10 years than it was in the last 10 years.”
Of course, the key to achieving that goal is having the right insight into the business to guide decision making. Business analytics, and particularly predictive modeling, provide the opportunity to be proactive in investment evaluation, detection of opportunities for savings, and in becoming more intelligent about processes across a host of subject areas housed inside a typical health plan.
Predictive Modeling for Health Management
By far the most common place to find predictive modeling solutions is in the field of health management. By health management, we mean any area that focuses on intervening into the healthcare delivery process to ensure a more desirable clinical outcome for members. Condition management, disease management, care and case management, and wellness programs all fall under this umbrella of activities.
Typically, predictive modeling in health management predicts the total medical costs for a member over the next year. This type of predictive model is so common, in fact, that many health plan professionals believe that this is solely what is meant by “predictive modeling.” Predictive modeling, however, has a long and successful story beyond health management, and even healthcare. Grocers and big box retailers use predictive modeling to anticipate inventory needs at stores before existing stock is depleted. Web commerce sites use predictive modeling to show customers recommendations for products they might be interested in given their previous searches and purchases. These broader applications of predictive analytics offer tangible examples of how health plans can derive additional benefits through the thoughtful application of advanced analytical routines in other areas of the business.
One of the simple ways an organization can use predictive modeling in new ways for health management is to change the target of what is being predicted. Instead of predicting costs, program management teams can create predictions around behavior that is targeted toward the program in question. For example, case management programs that focus on high-risk cases may be primarily interested in averting hospitalizations. Predictive models are able to predict whether someone is likely to need an inpatient hospital stay over the next three months. Similarly, programs designed to manage the appropriate use of emergency departments (ED) can build a predictive model focusing on ED utilization as the target.
For more information
on SAS predictive
Modeling that aligns the predicted target with the intended management program is imperative to showing health management ROI. Moving program management “upstream” to avoid “events” is crucial, since traditional programs that simply focus on coordinating care once a medical event has already occurred have shown little evidence of reducing hospitalizations or costs. For example, results from the “Medicare Coordination of Care Demonstration” indicated that only two out of 15 programs that participated between April 2002 and June 2005 showed statistically significant reductions in hospitalizations. That is the result one might expect from simple random variability given the sample size. In addition, none of the 15 programs showed a net cost savings, according to the report.
Rather than focus on coordinating care after the fact, predictive modeling allows management programs to avoid adverse events altogether. Further, predictive modeling can also help target individuals who are most likely to be receptive to a program, avoiding wasted time and costs on members who will be less responsive. “We want to develop predictive models that not only identify and classify patients who are at risk,” says Adam Hobgood, director of statistics at Healthways’ Center for Health Research, “but also anticipate who is at the highest risk for specific diseases and complications, and then determine which of those are most likely to comply with recommended standards of care.”
Predictive Modeling for Actuary and Pricing
Health management activities are not the only areas that can benefit from an infusion of predictive analytics. Professionals responsible for underwriting and actuarial activities inside health plans are also prime beneficiaries.
Today’s healthcare climate produces an interesting tension for actuarial departments. The advent of consumer directed health plans and health savings accounts (HSA) has brought consumerism to the forefront of new health products. More than 6 million Americans now have HSAs and the asset rate is growing faster than the rate during the first years of individual retirement accounts. Two Blue plans, Highmark and Blue Cross Blue Shield of South Carolina, have opened shopping mall outlets to sell health insurance products in a direct retail environment.
One of the simple ways an organization can use predictive modeling in new ways for health management is to change the target of what is being predicted.
With this movement towards individual purchasers comes an increased risk of statistical arbitrage for an insurer, meaning the products being sold are mispriced to the long-run detriment of the insurer. Individuals are almost always more aware of their own health risks than their employer. So when an employer purchases healthcare coverage for its entire population of employees, both the insurer and employer lack visibility into individual employee needs and costs, so they employ statistical pooling methods to feel reasonably safe about the cost risk that is being transferred from the employer to insurer.
However, something different happens when an individual purchases insurance; the game changes. Event risks are insurable only insomuch as the risk can be fairly considered uncontrollable and random. In this case, events not controllable for an employer purchasing health insurance become very controllable for an individual purchaser. An individual can choose to engage in any number of activities linked to poor health outcomes and is, obviously, better positioned to know whether they will engage in the behavior than is an insurer.
This is where predictive modeling can help lower the risk of statistical arbitrage. While individuals certainly have better knowledge of their own behavior, they are not likely to have firmly quantified the healthcare implications of those factors. The insurer can develop predictive models to gain a better understanding of the relative risks, and then use demographic and supplementary information to segment the application pool into appropriate pricing strata.
These predictive modeling applications (capturing the interplay between information, risk and decision rights) become even more important when considering the evolving role of government intervention in health reimbursement and private health insurance markets.
While it is speculative to say what specific action the federal government will take towards healthcare reform, it should be noted that since the late 1990s predictive modeling has been a key reimbursement tool used by the Centers for Medicare and Medicaid Services (CMS). The Balanced Budget Amendment of 1997 put into motion several programs by which CMS uses predictive models to determine reimbursement rates to both health plans (diagnosis cost group models for Medicare Part C) and providers (ambulatory payment classifications for hospital outpatient services).
It is entirely possible that future healthcare reform will also utilize predictive reimbursement strategies in other sectors. In such instances, it is not only imperative to have a keen understanding of the impact of the reimbursement methodology to a plan’s revenue stream, but also to build a proactive tier of predictive models to anticipate where enrolled members are likely to fall in the next year.
Insights related to populations movement from one reimbursement strata to another (aka, migration), as well as the impact of members entering and leaving a plan (aka, churning), can be simulated ahead of time using predictive models, allowing plans to proactively manage their population.
The heart of any health plan is the ability to correctly pay claims in a timely manner. Health plans that struggle with their operational activities face an uphill battle in the market. Predictive modeling increasingly is being used to help organizations spot issues inside their operations.
Sometimes these issues are related to fraud. By calculating a ratio of actual-to-expected costs in a given year and then identifying extreme outliers (e.g., geography, provider and other demographic factors), organizations can surface potential problems more quickly and comprehensively than through the application of simple rules engines. A similar approach was recently used to detect a fraudulent scheme where unnecessary outpatient surgeries were being performed on volunteers who, in return, were being given a cash payment.
Using predictive models to look for outliers may not always lead to fraud, however. HBF Group, western Australia’s largest provider of health insurance, uses predictive models to generate accurate reports that highlight and pinpoint aberrant behavior from within the company’s membership and suppliers, such as health service providers. Once these reports are generated, they are passed on to relevant HBF business units that undertake any further investigative or corrective action that may be required.
“In the vast majority of cases,” says Mark Pereira, manager of Business Information Services at HBF, “the aberrant behavior we identify is actually caused by clerical errors, rather than intentional fraud. But the fact remains that with the models in place, our organization is able to act quickly, and then work with members and providers to identify and rectify any particular issues.”
Bringing It All Together
Duplicative work is a waste. With all of these potential benefits available for predictive modeling across organizations, smart organizations invest in the creation of enterprise centers of excellence related to advanced analytics. These centers can serve many functions; for example, information aggregation, enterprise information models and shared software, training, and best practices.
In the earlier example, a predictive model to predict ED visits can pinpoint access of care issues. That same model can also be utilized to measure, for example, whether a client’s ED use is within acceptable norms given the health risks of the population. That would be of interest to the employer and customer reporting units. Further, the model can be used to look for aberrant behavior in members who are fabricating musculoskeletal issues in EDs in order to be prescribed narcotics. A Special Investigations unit would be interested in such cases.
Using predictive analytics in the ways mentioned here can deepen a plan’s insight about its customers, help choreograph the interaction with the customer by engaging only in activities that have high return potential, and continuously improve a plan’s market performance by learning, improving, and optimizing resource investments.