Five percent of patients incur nearly 50 percent of United States’ health care costs, and there is growing evidence that investing resources in these individuals can improve care while decreasing costs. In kind, provider organizations are increasingly adopting high-risk care management, a strategy that relies on coordinated outpatient care to reduce costly emergency department (ED) visits and inpatient admissions. However, complex care management programs are costly in themselves, so it is important to select patients who are not only high risk but are also most likely to benefit from such programs.
We now have a wealth of clinical and financial data coupled with an expanding list of analytical solutions to facilitate these choices. However, to best identify the right patients for targeted population health interventions, we need to consider other data sources as well, particularly provider insights and patient-reported data.
The Limits Of Big Data
Big data analytics have been quite successful in predicting high-risk patients. Most current models use some combination of claims-based data, demographics, prior acute care utilization, and high-risk conditions or medications to predict future health care costs and utilization patterns. These tools are essential for stratifying a large patient population, homing in on those patients who are at risk and may benefit from interventions. However, they tend to over-represent patients with transiently high costs due to a catastrophic event. And they may miss social and other factors that play important roles in shaping health outcomes.
Looking beyond the clinical conditions that appear in patient charts, we find people with bills to pay, children to raise, and jobs to keep. They may live in houses with stairs that they can no longer climb confidently or have trouble making medical appointments because they are struggling with addiction. In short, patients’ lives—and their relationships with the health care system—are messy. And until analytics can more effectively capture and weigh these psychosocial data elements, such tools may not always untangle the mess enough to allow providers to arrive at the best decisions for their patients.
Consider, for example, an elderly man with diabetes and congestive heart failure who is frequently admitted to the hospital with complications of these chronic diseases. At face value, he seems to be the perfect candidate for care management. Yet, a care manager’s ability to help reduce the frequency of his hospital visits depends on what precipitates these visits. He may have poor eyesight and have trouble differentiating his pills; he may not have transportation to get to his outpatient cardiology and primary care appointments; or he may be doing everything right and simply have a complex disease that is difficult to manage even with expertise from a team of well-coordinated providers.
In addition, is he willing and able to engage with a care manager? Or is he the sort of person who only wants to hear from his doctor? How might his prior experiences and his personality factor in to the appropriate intervention? What about the role of trust?
Employing A Hybrid Approach
The optimal approach to selecting the right high-risk patients for care management involves a combination of analytics and human insight. Organizations can start with a population health review using claims and electronic health record (HER) data to identify high-risk patients. Primary care physicians and other clinicians could then review the list to refine its results.
Clinicians often know details about their patients that are not captured in commonly used risk algorithms: their living situation, social support networks, and ability to follow through with personal health goals. Leveraging provider insights can allow an organization to select patients for complex care management who will truly benefit from additional services or home-based interventions, and may in turn save providers time by helping to manage some of their most complex patients.
Evidence suggests that provider instincts are reliable: a recent study compared the ability of physicians to identify complex patients with two commonly used analytic methods and found that patients identified as complex by providers had higher rates of hospitalization and emergency department visits compared to more traditional analytic methods.
Another study found that primary care providers performed better than algorithms at identifying patients with increasing hospitalization rates over time — these are the very patients who would be most likely to benefit from early intervention.
Factoring In Patient Insights
Finally, it makes sense to look to our patients for guidance. Patients understand their health challenges and their readiness to engage with care managers in behavior change better than anyone else. Quantitative patient-reported data—such as health literacy scores and the ability to manage activities of daily living—should be an integral component of any risk stratification system. Provider organizations could use assessments of patient activation, physical function, and social needs such as lack of food or housing.
Studies suggest that patient-reported data are helpful adjuncts to prediction tools. A study by Mayo Clinic researchers found that heart failure patients who reported low physical function had higher rates of hospitalizations and ED visits than their peers. Another study among Medicare beneficiaries found that using three patient-reported questions (self-reported health status, having a bothersome health condition, and needing help with one or more activities of daily living) could predict hospital admission and identify the costliest patients.
When the results of these questions were added to a commonly used risk algorithm, its predictive power increased. Finally, a recent study using the Patient Activation Measure reached a similar conclusion: measuring disease self-management may help identify patients at greatest need of additional care interventions.
By asking patients about their ability to manage chronic disease, we stand a better chance of identifying which patients are most in need of additional services. Patient reported outcome measure instruments have been costly to administer, but this will be less of an issue as we transition to electronic health records and more routinely capture patient-reported measures in daily practice. Going forward, we need to determine how patient-reported data can best contribute to prediction tools and to design and validate instruments that can reliably connect patient data to actionable outcomes.
Provider intuition and patient self-knowledge are valuable additions to the data bytes collected from medical claims and clinical information systems. Today, capturing and applying this information is largely an art. Going forward, analytics will need to adapt so that this valuable but nuanced information can more readily and effectively blend with clinical and financial data, allowing provider organizations to better care for the highest risk patients.