New study shows Medical Home Network’s AI risk prediction model incorporates social determinants of health and identifies more high-risk patients

Medical Home Network (MHN) today announced its risk stratification model, which incorporates social determinants of health (SDOH) and uses AI, can more accurately identify high- and rising-risk members than traditional models, according to new research published in the peer-reviewed American Journal of Managed Care. This allows providers to focus on patients with the most immediate needs and deliver whole-person care.

“Traditional risk models primarily rely on lagged claims data to make predictions,” said study author Todd Burkard, MHN vice president, Data Analytics. “Using AI, we can leverage more timely data sources such as health risk assessments, ADT, and care management activity into a more actionable prediction.”

The study, “Improving Risk Stratification Using AI and Social Determinants of Health,” highlights new evidence that a combination of claims information, demographics and real-time data on behavior, social determinants of health, and admissions, discharges and transfers (ADT) can improve identification of members with the highest future medical spending. Because past utilization doesn’t always predict future utilization, real-time SDOH and ADT data provides greater insight into which members are likely to need more resources going forward, improving resource allocation and empowering care teams to provide more efficient, whole-person care.

“People in under-resourced communities face a variety of challenges including social factors, and it can be difficult for providers to predict how those challenges will affect their health. MHN’s dynamic risk model gives providers a better chance to devote resources and care to patients with the most pressing needs and deliver whole person care,” said study author Cheryl Lulias, MHN president and CEO.

According to the study, an AI-based model that included non-traditional, real-time data sources identified 41% more of the highest-risk members than a standard model that used historical claims and demographic data alone.

“Identification of future high utilizers is the first step toward improved care coordination, better health outcomes and more controlled medical spending,” said first author Nathan Carroll, PhD, associate professor at Virginia Commonwealth University. “This research shows investment in data infrastructure can pay off for care management programs.”

The study included spending data from 61,850 Medicaid members continuously enrolled in MHN’s Accountable Care Organization between May 2018 and April 2019. Researchers compared healthcare spending for members with risk scores in the top 5% of MHN’s AI model to those in the top 5% of the traditional Chronic Illness and Disability Payment System (CDPS) model. The average spend per member in the top 5% of risk scores for the AI model was $14,349 compared to $11,808 in the traditional model.

Read the full study in The American Journal of Managed Care: https://doi.org/10.37765/ajmc.2022.89261