Medicare Advantage plans need a practical risk stratification pipeline to score members by clinical and financial risk, then rank them so care management and documentation teams prioritize outreach. The goal is a defensible, reproducible system that directs finite resources to improve outcomes and revenue.
The Core Architecture
Risk stratification combines scoring and segmentation. Each member gets a risk signal, then gets bucketed into tiers like rising-risk, high-risk, or catastrophic. The common mistake is relying on a single machine learning model. Instead, a layered approach works better: a stable, explainable base using RAF and chronic conditions, plus optional predictive overlays. Explainability is critical because care managers and auditors won't trust a black box.
Building the Feature Record and Scoring
Start with a member feature record that includes age, HCC codes, RAF score, care gaps, and utilization data like emergency department visits and inpatient stays. The RAF, calculated under CMS-HCC V28, serves as the defensible risk anchor. Everything else supplements that signal.
A simple scoring function can combine RAF with weighted utilization metrics. For example, add 0.15 times ED visits and 0.30 times inpatient stays to the RAF. Then assign tiers based on thresholds: catastrophic for scores 3.0 and above, high for 1.8 and above, rising for 1.0 and above, and stable for below 1.0. Keep weights transparent and tunable so operational teams trust the ranking.
Making Rising-Risk Actionable
The rising-risk tier often drives the most return on investment. These members are trending toward high cost but still have open documentation and care gaps. Surface their specific gaps—like overdue labs or undocumented chronic conditions—so outreach has a concrete target, not just a vague label.
Closing the Loop with Documentation
Stratification must feed back into documentation to be fully effective. When a rising-risk member has a suspected but undocumented chronic condition, that's both a care opportunity and a chance to improve RAF accuracy. The pipeline should emit those as work items for care managers.
Rather than rebuilding risk models from scratch, plans can ground their pipeline on a maintained population health risk stratification engine that handles HCC mapping and V28 weighting logic. That avoids fiddly, version-sensitive, audit-relevant work. For the operational framing, see Population Health Risk Stratification for MA Plans.