Using Predictive Analytics to Address Social Determinants of Health
Bob, Emily, Sarah, and Bill develop type 2 diabetes at age 49. By their 50th birthdays, Emily and Bob have gotten their blood sugar under control. But Bill spends his birthday in the hospital with diabetic-related foot ulcer, and Sarah learns that she has developed kidney disease from uncontrolled diabetes.
This disparity in outcomes was not caused by a difference in clinical care. All the treating physicians followed evidence-based protocols. But Bill and Sarah are dealing with stressful social, economic, and environmental problems—so-called social determinants of health—that have been shown to increase the risk of medical events. Bill recently experienced a divorce and moved across country to take on a 60-hour-a-week job. Meanwhile, Sarah had to quit high school as a young woman to help support her family, and decades of low-paying jobs have landed her in debt with little extra money for her medications.
In recent years, health care stakeholders have come to realize that focusing solely on reducing medical risk through preventive and chronic care management will not produce desired population-level results. Research shows that clinical care only has a 20% influence on the length and quality of life among members of a community. In comparison, 50% is tied to social determinants of health, including poverty, unstable housing, high-stress life events, and a lack of education or social support. These factors can interfere with a person’s ability to engage in healthy behaviors and can cause chronic stress, which biologically increases disease risk.
Recognizing the importance of social determinants of health, many health plans are expanding their care management capabilities to help members overcome nonmedical barriers. For instance, care managers might help Sarah obtain free or discounted diabetes medication and assist Bill in managing his stress or joining a support group. However, since care management resources are limited, health plans should target those members who need these services the most. This requires a sophisticated predictive modeling approach that uses the most relevant data.
Predicting Social Determinants of Health
Predictive analytics involves sorting and weighing various types of data using an algorithmic model to forecast future outcomes. Health plans have traditionally used claims data to stratify members and identify higher-risk members in need of case management, such as those who frequently visit the emergency department (ED) or use a lot of other high-cost resources. From an analytical perspective, this is a fairly simple data exercise.
More difficult is pinpointing those members who currently do not use many health services but have a high likelihood of having a major medical event in the near future. Identifying these imminent super-users in time to intervene and change their health trajectory holds great promise for not only improving outcomes and quality of life but lowering costs.
Claims data alone does not typically contain enough clues about members’ lives to adequately stratify which ones have an elevated health risk due to various social determinants of health. If only claims data was used, Bob, Emily, Bill, and Sarah may all have been identified as needing care management since they were all recently diagnosed with diabetes. Additional data would be needed to find evidence that Bill and Sarah are struggling with socioeconomic stressors and, ideally, to pinpoint the specific stressors that Bill and Sarah need assistance managing.
Combining claims data with other types of data paints a detailed picture of a member’s life situation. Relevant data includes:
• US census data, which provides granular details about a member’s community—down to a 3-block area—including the average income and the density of housing.
• Marketing or consumer data, including credit card transactions, aggregated by various companies (eg, LexisNexis, Acxiom).
• Community rankings, such as the Gallup-Healthways Well-Being Index, that provide additional information about the neighborhood a member lives in.
After data mining for evidence of medical and behavioral issues, social determinants of health, and other relevant issues, the algorithm assigns a risk score to each member, which helps care managers pinpoint who most needs their services. Critically, analytics tools can pinpoint specific social determinants of health that a member may be struggling with and auto-populates a care plan with recommendations for care managers to follow up on. For instance, marketing data may reveal that a member does not own a car and also lives far from a health care provider, which may mean the member has transportation challenges. The auto-populated care plan will highlight that this member may need help arranging a ride to physician appointments.
Refining Predictive Modeling
Predicting health outcomes via data analytics is still a developing science. Critical questions are still being worked out, including what are the key socioeconomic risk factors (eg, a member’s income, level of indebtedness) that increase a member’s risk of a near-term medical event and what are the most relevant sources of data to use to uncover these risk factors. The best way to determine the answers to these questions is to continually run, update, and refine the predictive model using various types of data. Today, marketing data is often the most useful information source to use for identifying socioeconomic risk factors, particularly impactful ones that care managers can address, such as unstable housing, food insecurity, and lack of transportation.
Targeting Care Management
Not all care management is equal. A key factor to success is whether the member is engaged, or willing to work with the care manager. Health plans that use a targeted approach informed by predictive analytics have increased engagement and retention rates as high as 25% and 50%, respectively, compared to the industry average of 10% to 20%.