Food Insecurity Identification Modeling for Medicare Enrollees Using Administrative Data

Why It Matters

Accurate, data‑driven identification of food‑insecure Medicare members allows managed‑care organizations to focus interventions where they are needed most, improving health equity and potentially lowering overall costs.

Key Takeaways

• Model predicts food insecurity with AUC 0.82 across 462k Medicare members
• Prior social‑need documentation and dual Medicare‑Medicaid enrollment are top predictors
• Chronic conditions showed little impact on food‑insecurity risk in the model
• State‑level random effects reveal significant geographic variation in risk
• Scalable mixed‑effects approach enables targeted outreach without universal screening

Pulse Analysis

Food insecurity has risen to the forefront of health‑related social needs, especially among older adults on fixed incomes. Traditional universal screening in Medicare Advantage plans strains resources and often yields incomplete data, prompting many payers to explore predictive analytics. By leveraging existing claims, health risk assessments, and publicly available Social Vulnerability Index scores, predictive models can flag high‑risk members before they present with acute health crises, aligning with CMS’s push for proactive social‑determinant interventions.

The Elevance Health study demonstrates that a parsimonious hierarchical generalized linear mixed model can reliably predict food‑insecurity risk across a large, diverse cohort. Using 10‑fold cross‑validation, the model’s AUC of 0.82 indicates strong discrimination, while the inclusion of random intercepts for Medicare market states captures state‑level variation often missed in flat models. Notably, the presence of multiple prior social‑need codes and dual Medicare‑Medicaid enrollment drove the largest effect sizes, dwarfing the influence of chronic conditions such as diabetes or cardiovascular disease. This finding underscores that non‑clinical factors are primary levers for identifying vulnerable beneficiaries.

For managed‑care organizations, the practical implication is clear: integrating such a model into care‑management workflows can streamline outreach, prioritize limited social‑service resources, and improve referral efficiency to programs like SNAP or meal delivery. Moreover, the model’s scalability—relying on data already captured in administrative systems—reduces the need for costly, organization‑wide screening initiatives. As health systems increasingly tie value‑based payments to social‑determinant outcomes, data‑driven risk stratification will become a cornerstone of equitable, cost‑effective care delivery.