UCLA

Background: Patients who use a large volume of health services, especially costly acute care, are known as high utilizers. An increasing number of programs target high utilizers, with a goal of improving health and reducing disproportionate use. These real-world programs often enroll heterogeneous populations with varying health needs and utilization histories. To appropriately evaluate programs that target high utilizers, there is a need to understand enrollee heterogeneity and how program effects vary across subpopulations. Objectives: Using data from California Whole Person Care (WPC), which provided cross-sector care coordination spanning health and social services to high-utilizing Medicaid patients, I explored: (1) what health needs characterized enrollee subpopulations, (2) what utilization trajectories prior to WPC enrollment characterized enrollee subpopulations, and (3) what were differential program effects for classes defined based on health needs and utilization. Methods: I used WPC enrollment data, and Medicaid enrollment and claims. I used latent class analysis (LCA) to identify classes based on health needs, and group-based trajectory modeling (GBTM) to identify classes based on pre-enrollment utilization trajectories. I used difference-in-difference analysis to evaluate impacts of care coordination across classes. Results: LCA identified five classes, the largest consisting of enrollees with low overall needs (32.0%), high physical health needs (27.5%), and high behavioral health needs (26.3%). GBTM identified two classes: “Moderate-to-High” utilization (18.6%), and “Low” utilization (81.4%). High behavioral health need was associated with high utilization. Compared to program-assigned target populations, analysis classified more enrollees as having high health needs, and fewer as high utilizers. When LCA and GMM classes were cross-tabulated, all had significantly decreased adjusted rates of ED visits after WPC participation, and all but one had decreased hospitalizations. Enrollees with high behavioral health needs had significantly larger decreases in hospitalizations compared to several classes. Implications: Governments and health care organizations should consider enrollee heterogeneity when developing cross-sector care coordination or other interventions to reduce utilization. Classification methods including LCA and GBTM can inform intervention tailoring and evaluation. Though care coordination is promising for enrollees with many backgrounds, programs should anticipate larger impacts for some subpopulations, and provide additional resources or innovative strategies accordingly.