How can we match the right resources and interventions to patient needs? Better risk stratification

Sept. 11, 2017 / By L. Gordon Moore, MD

With the unrelenting pressure to reduce unnecessary costs in healthcare delivery and despite (or maybe because of) the lack of clear direction from the federal government, Medicaid plans across the U.S. are increasingly holding healthcare delivery systems to account for quality and cost outcomes.

As a consequence, managed care plans and health systems entering into value-based contracts search for ways to identify and mitigate factors that negatively impact the people and population for whom they are responsible.

One logical way to start is to identify people with gaps in care delivery: Unmet preventive or chronic condition care needs. This is a fine start as there are many people with gaps and the work is logical and appealing to clinicians addressing each patient and their needs. Disease-focused clinical registries can be useful when tracking multiple factors over time.

While gap closure is excellent and appropriate work, it has some limitations as a strategy to improve population health outcomes and reduce unnecessary costs. For example, people are not all the same and may need very different types and intensities of assistance. Figure 1 is from a study of people with diabetes, demonstrating a striking difference in hospitalization rate based on severity of the disease and presence of comorbidities. A more nuanced segmentation strategy makes it possible to match resources to intensity and severity of needs.

Figure 1:  Rates of hospital admission per 1,000 people with diabetes per year[i]

When resources or time are limited, it is logical to consider ways to stratify people by the intensity of their conditions, as this makes it possible to focus limited resources on those with the greatest needs and over time expand the focus to population segments with fewer needs. Hierarchical risk stratification models like 3M’s Clinical Risk Groups are useful tools enabling population segmentation based on total illness burden and the probability of potentially preventable events.  Figure 2 segments a population by aggregating illness burden groups (aggregated Clinical Risk Groups in this example), demonstrating that people with chronic conditions—both simple and complex—represent the highest total as well as potentially preventable medical expenditures.

Figure 2: Aggregated Clinical Risk Group population segments total medical expenditure (horizontal axis) and potentially preventable expenditure (vertical axis)[ii]

Medicaid plans are becoming increasingly aware of non-medical factors that degrade clinical outcomes and add to healthcare costs. Information on these non-medical factors provides an opportunity to further nuance a segmentation strategy. In a study of Midwest Medicaid plan’s population, we found that the population is far from homogeneous. In addition to the highly variable medical illness burden, this population could be split into those who reported difficulty paying for essential needs and those that did not report difficulty paying for essential needs. 

Figure 3 shows the rate of emergency department and hospitalization use per 100 people per year, separating those who report being financially stable versus not stable. The “number of vital questions” refers to work we published demonstrating the link between certain responses and prospective probability of unfortunate outcomes.[iii]

Figure 3[iv]

The bottom line: Segmentation strategies can gain increasing sophistication and nuance by using methodologies that understand total illness burden and can identify non-medical factors that predict important health outcomes. Medicaid plans and provider systems in value-based purchasing arrangements may better match resource and interventions to patient needs when using more sophisticated segmentation strategies.

L. Gordon Moore, MD, is senior medical director for Populations and Payment Solutions at 3M Health Information Systems.

[i] Bernstein, Richard H. “New Arrows in the Quiver for Targeting Care Management: High-Risk versus High-Opportunity Case Identification.” The Journal of Ambulatory Care Management 30, no. 1 (March 2007): 39–51

[ii] Analysis of adjudicated health plan claims data set, unpublished.  3M Health Information Systems, Inc.

[iii] Wasson, John H., Laura Soloway, L. Gordon Moore, Paul Labrec, and Lynn Ho. “Development of a Care Guidance Index Based on What Matters to Patients.” Quality of Life Research: An International Journal of Quality of Life Aspects of Treatment, Care and Rehabilitation, April 11, 2017. doi:10.1007/s11136-017-1573-x.

[iv] Analysis of ~20k people receiving Medicaid in a Midwest state, comparing responses on a health risk assessment tool to self-reported ED & hospital use.  Unpublished 3M HIS.