Objectives: The Massachusetts Medicaid and Children's Health Insurance Program, MassHealth, offers comprehensive Senior Care Options (SCO) plans to its Medicare-eligible members 65 years and older. Historically, MassHealth has paid a fixed per-person capitation rate for any "nursing home-certifiable" SCO member despite considerable heterogeneity of need. Our objective was to develop a model to predict long-term services and supports (LTSS) costs for community-dwelling SCO members.
Study design: Concurrent predictive modeling.
Methods: We studied nursing home-certifiable SCO members who were enrolled for at least 183 days during 2016-2017 and used linear models to predict annual cost of community-based LTSS from demographic, medical, social determinants of health, and functional characteristics. We evaluated model performance using predictive performance (R2) and predictive ratios (observed costs divided by predicted costs) for various vulnerable subgroups.
Results: The modeling population included 35,259 enrollees. Mean (SD) annualized LTSS cost was $14,071 ($13,174). Functional status (ie, activities of daily living [ADLs] and instrumental ADLs) accounted for most of the variability in community LTSS cost (R2 = 18.4%) explainable by available variables. The Massachusetts SCO (MA-SCO) model (R2 = 21.6%) predicts accurately for several high-cost, vulnerable subgroups. Compared with fixed per-member capitation payments for all, the MA-SCO model reduces, for example, the payment to one plan by 28% and increases that to another by 35%.
Conclusions: Predictive models using administrative data and functional status information can appropriately allocate payments for subgroups of members with LTSS needs that differ substantially from average. Calibrating payment to need mitigates incentives for population skimming and promotes the sustainability of mission-oriented organizations.