Predicting Severe Diabetes Complications Using Administrative Claims Data in Maryland

Why It Matters

The model demonstrates that claims‑based risk stratification can dramatically improve identification of patients at imminent risk, enabling targeted care‑management that could curb costly hospitalizations. Its superior performance over HCC scores validates the value of disease‑specific predictive analytics for public‑health programs.

Key Takeaways

• Model uses 219 risk factors from claims and SDOH data.
• Top 10% risk scores captured 56.9% of severe events.
• Outperformed CMS HCC scores by 19.4 percentage points.
• Mean risk score predicts 1.24% monthly complication likelihood.
• Scalable to other states and payer populations.

Pulse Analysis

Diabetes remains a leading driver of U.S. health‑care spending, with 2022 costs estimated at $413 billion. Traditional risk‑adjustment tools, such as CMS Hierarchical Condition Category scores, offer broad population insights but lack the granularity to pinpoint imminent complications. Leveraging the ubiquity of Medicare claims and integrating publicly sourced social‑determinant metrics, Maryland’s new model fills this gap, providing a month‑ahead view of severe type 2 diabetes events. By focusing on administrative data, the approach sidesteps the need for costly lab results while still capturing utilization patterns that signal deteriorating health.

The model’s architecture—219 candidate variables distilled to 95 significant predictors through rigorous stepwise selection—delivers a concise risk score that translates directly into actionable intelligence for primary‑care teams. During its first production cycle, the top decile of patients identified by the score accounted for nearly 57% of all severe diabetes complications, a stark improvement over the 37.5% captured by HCC scores. This concentration effect means care managers can allocate limited resources to a small, high‑risk cohort, potentially averting hospital admissions that average several thousand dollars each. Moreover, the model’s integration into Maryland’s health information exchange (CRISP) ensures real‑time access for clinicians, fostering a data‑driven workflow that complements, rather than replaces, clinical judgment.

Beyond Maryland, the model offers a template for other states and payers seeking cost‑effective population health tools. Its reliance on standardized claims data makes replication feasible across Medicare, Medicaid, and private insurers, while the inclusion of social‑determinant factors aligns with emerging value‑based care incentives. As policymakers prioritize preventive strategies to curb chronic‑disease expenditures, such disease‑specific predictive analytics could become a cornerstone of accountable care initiatives, driving both improved outcomes and measurable savings.