A dose of insight: using new predictive models, insurers can identify members most likely to have health problems and the best ways to help keep them healthy.

As health plans turned away from the clinical micro-management that helped spawn the "managed care backlash" in the late 1990s, they moved to a disease management strategy, focusing on the small number of people who account for the majority of costs. This strategy has evolved into a sophisticated "population health management" strategy that targets behavioral, coaching and

case management support services to individuals at different levels of need. The key to any such clinical intervention strategy is accurate and early targeting. In order for clinical supports to be helpful, health benefits companies must identify the people most in need and stratify them to match them to the most appropriate support program.

New Models

Health benefits companies, through the rich claims information they collect to process transactions, have a perspective on each individual's health history unmatched elsewhere in the health system. These data, combined with other information about the individual that may help signal a change in future health risk, tan be analyzed to estimate the health trajectory of individuals and populations.

Health benefits companies have been using modeling for years to calculate future costs for insured groups and individuals. Using demographic data and medical and pharmacy claims, these early models analyzed historical relationships between the known data and a specific variable, such as overall health-care costs, to predict costs for the coming year. Conventional models have been successful in identifying high-risk individuals, or those who have experienced costly clinical episodes. Since these models rely on evidence of high service utilization, conventional models are retrospective. There are several limitations to conventional models, however. They generally assume linear relationships between current and future costs, independence of clinical condition variables, and normal error distributions. None of these assumptions tends to be true in health care. Today's high users of health care services are not necessarily tomorrow's high users, and a small change in health state today may mean a significant risk of future health burden. Moreover, individuals sick with diabetes, heart disease and other such conditions are often sick with multiple conditions, and there is a clinical relationship between their conditions that has an impact on their disease trajectory.

Assembling a group of research scientists and engineers from outside the health industry, Humana has developed new models for predicting future health needs (clinical events) at the individual level, including new approaches to assessing the impact of disease clusters and stage of disease on future health risk and illness burden. Disease clustering is a technique developed to account for the complex interaction of conditions when a person is facing multiple related diseases. An advanced diabetic, for example, may be dealing with cardiovascular disease, renal disease and peripheral vascular problems. Disease progression is a technique for assessing the stage of disease.

Humana's predictive models focus on early detection of illness or disease to reduce adverse events and help the member bring his or her health condition under control, whether that involves adherence to preventive routines or decision support that helps the member avoid unnecessary emergency room or hospital admissions.

When predicting the likelihood of illness for an individual health-plan member, that member is assigned to one of several risk groups based on disease condition, health-care utilization patterns and demographics. Each risk group, which is homogeneous to improve the overall predictive accuracy, has its own set of tailored models optimized for the population that falls into the group. Analysts synthesize the multiple predictions--for the risk groups the member belongs to--to create an estimate, called a severity score, of future health burden for each member.

Predicting Individual Behavior

The goal with predictive modeling is not simply to identify a pattern of utilization, but a pattern of behavior that tan be understood and tracked so that the health benefits company tan tailor and rime appropriate support services to make a difference in the member's health.

As observers learn more about which individuals engage in health improvement behaviors and what kind of intervention is most effective at influencing individual behaviors, they will be able to make additional types of predictions to help target clinical interventions through predictive modeling. These include predictions of who is likely to engage in a clinical support program (Humana assigns each member a "propensity score") and who among those who will engage in a program is likeliest to change behavior sufficiently to show some results (referred to as "behavioral elasticity"). Analysts combine all of these various scores-risk, propensity and elasticity-to develop a "clinical intervention appropriateness score. "The intention is to allocate individuals to specific intervention channels--messaging, health coaching services, disease management or case management--based on member classification, clinical program characteristics (channel capacity and return on investment) and consumer preferences.

Overcoming Data Limitations

Because many of the insured population barely use the health-care system, it's important to be able to find patterns and make predictions even on sparse data. In order to help people for whom there is no claims history, and to overcome the time lag for incoming claims, Humana has implemented a specially designed Health Risk Assessment that asks members questions about their health history and behavior that are most predictive of future health-care use. The HRA was developed with the help of simulations based on the company's data, and is tightly coupled to its predictive models. Adding the patient-reported data from the HRA will significantly improve the predictive power of the models and also will help the company to deliver services to individuals earlier in the plan year, even those without claims experience. Integrating the HRA with predictive modeling will help direct clinical resources and interventions to the right people at the right time.

Right Intervention, Right Time

Through predictive modeling, Humana hopes to get ahead of the curve by creating a system of integrated clinical support that links services appropriate to each individual's need at a moment when those services will most likely be effective. The goal is to make the clinical intervention personally relevant, maximize its impact on the member's health and improve outcomes for the member.

Today, successful predictive modeling is integrated with clinical management tools. Even the most sophisticated and accurate predictive models tan only influence rare through integration with health management.

Humana's clinical support services run the gamut from intensive case management services to personalized coaching services; outbound, automated phone calls with personally relevant health-care tips, and personalized messaging. An integrated approach to managing population health provides the tools to deliver support at various levels to help individuals manage the progression of disease and mitigate their own illness and financial burden.

Through all of these services, the goal is to influence the way people behave as they deal with health issues and interact with the health system. Predictive modeling is the tool that enables Humana to meet its members where they arc, and to deliver services that are personally relevant and timely. The better a health benefits company is able to target and personalize its services, the more likely it will be able to develop a productive, trustful, supportive partnership with its members and increase their satisfaction with their health benefits and with their health care.

Key Points

* New predictive models can assess the impact of disease clusters and stage of disease on future health risks.

* They also identify patterns of behavior so that the health benefits company can provide appropriate support services to make a difference in the member's health.

Contributor Dr. Jonathan T. Lord is chief innovation officer at Humana Inc., Louisville, Ky.

Learn More

Humana Health Plan Inc. A.M. Best Company # 68898 Distribution: Agent/brokers, direct to employees, direct

For ratings and other financial strength information about this company, visit www.ambest.com