Shifting behavioral health upstream with predictive analytics

As the pandemic fades away, opportunities to solve widespread behavioral health challenges — ranging from opioid use disorder and loneliness to acute mental health episodes in young people — remain a top priority for progressive healthcare organizations.

Recently, physician organizations, such as the American Academy of Pediatrics  and public officials, including the Surgeon General  and the Biden-Harris administration , have recommended many new initiatives and recommendations, including expanding access to peer support and adding capacity for mental health crisis response in high-need communities.

Increasing access to treatment and support is essential for the millions of people and families affected by a mental health condition or substance use disorder (SUD). That alone may not be enough, however. Reversing the current trends will require shifting behavioral healthcare upstream and increasing the focus on early intervention and prevention.

Predictive analytics are a powerful tool in this effort. Predictive models that tap into large datasets to identify high-risk individuals enable providers to proactively reach out to at-risk patients and families — getting ahead of behavioral health needs before they arise.
 

Joining forces (and data)

Predictive models are only as good as the data that feeds them. Whole health is the product of a wide range of medical, behavioral, and social factors, so a model that focuses on only one (or even two) of those drivers is certain to have some blind spots. Incomplete data creates an incomplete picture of people and populations.

Unfortunately, after more than a decade of government-led efforts to promote interoperability in health data , siloed data remain a challenge. From EHR data to utilization and claims data, the systems and data used in physical and behavioral health remain largely separate, with little standardization, interoperability, or communication between them. What's more, those systems are also typically disconnected from the systems and data related to SUD treatment and interactions with social service agencies, the criminal justice system, and more.

Integrating data is key to predictive analytics. In the same way that integrating medical, behavioral health, and pharmacy benefits gives healthcare organizations a fuller picture of an individual’s whole health, bringing diverse data into a single model creates new opportunities to identify signals and patterns that provide a deeper understanding of behavioral health risk factors.

The Connecticut Behavioral Health Partnership  (CBHP), which manages the behavioral healthcare for more than 800,000 Medicaid beneficiaries in the state, illustrates the groundwork needed for effective predictive analytics. Founded in 2006, the CBHP brings together multiple state agencies, including the Department of Children and Families, the Housing Management Information System, the Department of Mental Health and Addiction Services, and the Department of Corrections. (Carelon Behavioral Health, formerly known as Beacon Health Options, has served as the administrative services organization for the CBHP since its inception.)

The CBHP membership has built a shared database  with more than 100 variables spanning demographics, utilization, and cost of care broken out by medical, behavioral, dental, and pharmacy. Carelon and the CBHP subsequently built predictive models that mine this rich data for insights.

For example, the CBHP has used these models to analyze data from adolescents and young adults who have received behavioral health treatment. Researchers ran several tests to identify which demographic, clinical, and utilization factors are most closely associated with discontinuing treatment, which is common among young people transitioning to adulthood. In the preliminary findings , a model that zoomed in on 18 separate variables correctly predicted with 80% accuracy which individuals would stay engaged in treatment.
 

Improving outcomes with proactive outreach

Leveraging insights from predictive models enables care teams to improve outcomes and lower healthcare costs by minimizing avoidable emergency department visits and hospital stays.

For example, our suicide prevention program for children and adolescents  uses a predictive model that looks for patterns in diagnosis and claims data linked to an increase in suicide risk. That can include behavioral health conditions, such as depression and substance use, as well as physical symptoms, such as recurring headaches.

When these predictive models identify an individual at risk, a case manager reaches out to provide resources and support to help them manage their mental health. Case managers also connect the individual to peer wellness and recovery specialists, people who have experienced mental health challenges.

More than 4,000 young people have engaged in the program since 2018. Suicidal events have dropped more than 20% among those with commercial coverage, and behavioral health spending has decreased by 30% per member per month.
 

Creating a virtuous cycle

Predictive analytics have the potential to change the trajectory of behavioral health in the U.S. Realizing that potential depends on advanced integration and collaboration among providers, health plans, state and local agencies — as well as generating actionable insights for providers.

When diverse partners work side by side and integrate their data in the service of predictive analytics, it not only improves outcomes in the short term but also creates a virtuous cycle for the long term. As more data enters the system, providers and health plans collaboratively optimize the predictive models and identify new opportunities to steadily move behavioral health services upstream and, ultimately, improve the health of our healthcare system.