In our last post, we deconstructed the shared savings model in a VBC, breaking down the financial mechanics that govern most value-based care arrangements. Understanding the rules of the game—the benchmarks, attribution, and quality gates—is essential. But knowing the rules is only half the battle. The other half is knowing where to act.
When you’re responsible for the health outcomes and costs of thousands of patients, how do you effectively manage their care to succeed? The answer lies in one of the most powerful tools in modern healthcare analytics: risk stratification.
Risk stratification is the data-driven process of identifying patients at different levels of health risk to proactively match them with the right level of care, at the right time. It’s the engine that transforms an organization from reactive to proactive. This post will break down what risk stratification in value-based care is, what fuels it, the common methodologies (with their challenges), and—most critically—how to turn its insights into action.
Why Risk Stratification is the Foundation of VBC
Traditionally, healthcare has been reactive. Resources are deployed when a patient presents with an illness. Risk stratification flips this model on its head. It recognizes the well-known principle that a small percentage of the patient population—often the 5% with complex, chronic conditions—drives a disproportionately high percentage of total healthcare costs.
By identifying who is in that 5%, as well as the 15-20% who are on the cusp of becoming high-risk, organizations can allocate their most valuable clinical resources—care managers, social workers, pharmacists—efficiently and effectively. Instead of treating everyone the same, you tailor your interventions to have the greatest impact, preventing costly adverse events before they happen.
The Data Inputs: Fueling the Risk Engine
An effective risk model is only as good as the data it’s built on. A comprehensive view of the patient is essential, requiring the integration of multiple data streams—though challenges like data silos and inconsistent quality often arise. Here’s what to include:
- Claims Data: This is the bedrock. It provides a rich history of diagnoses (what conditions a patient has), procedures, and utilization patterns (e.g., emergency department visits, hospital admissions).
- Clinical Data (EHRs): This adds vital context that claims data lacks. It includes lab results (like A1c levels for diabetics), vital signs (blood pressure), and active problem lists, which can signal how well a condition is being managed.
- Pharmacy Data: This stream reveals which medications have been prescribed and, more importantly, patterns of medication adherence. A patient not refilling their hypertension medication is a clear red flag that claims data alone would miss.
- Social Determinants of Health (SDOH): This is the next frontier of accurate risk modeling. Data related to factors like housing instability—shown to correlate with higher ER visit rates—food insecurity, and transportation challenges are proving to be powerful predictors of future health needs and costs. Incorporating SDOH data is a hallmark of advanced healthcare analytics strategies.
To overcome integration challenges, consider using standardized formats like HL7 FHIR to ensure data from disparate sources aligns seamlessly.
Common Methodologies: From Simple Rules to Predictive Power
Once the data is aggregated, how do you calculate risk? The methodologies range from straightforward to highly complex, each with its own hurdles:
- Rule-Based Models: This is the most basic approach. A simple rule might classify a patient as “high-risk” if they have three or more chronic conditions and have been admitted to the hospital twice in the last year. These models are transparent and easy to understand but often miss emerging risks in patients who don’t yet meet rigid criteria.
- Industry-Standard Predictive Models: More advanced systems use validated predictive models. The most well-known is Hierarchical Condition Categories (HCCs), a model used by Medicare to predict future healthcare expenditures based on a patient’s documented diagnoses. An HCC score offers a nuanced view of risk, though it presents a couple of challenges.
- The CMS HCC model relies heavily on historical data and may lag in capturing real-time changes.
- For the HHS-HCC model, which operates on a concurrent risk adjustment basis—meaning it uses data from the same year—relying on it early in the year could indeed miss diagnoses from previous months
- Custom Machine Learning Models: The most sophisticated approach involves building custom predictive models tailored to a specific population or outcome. For instance, an organization might predict 30-day readmission risk for heart failure patients or the likelihood of a high-cost ER visit in the next six months. These models require significant data and expertise, with the risk of overfitting if not carefully validated.
From Score to Action: The Most Critical Step
A risk score is just a number. Its value is only realized when it’s used to drive a clear, operational care management strategy. The most effective way to visualize and act on this is through the Population Risk Pyramid along with a Data Table which helps identify members in a tier and reasons why they are assigned to that tier.
This pyramid informs a tiered intervention strategy:
- Tier 1: High-Risk (The Top ~5%)
These patients, often with multiple complex chronic conditions, require intensive, high-touch support.- Intervention: Enrollment in complex case management programs, direct oversight by a dedicated care manager, frequent physician follow-ups, and coordinated care planning across all specialists.
- Tier 2: Rising-Risk (The Next ~15-20%)
This group may have a poorly controlled chronic disease or emerging risk factors. The goal here is prevention and stabilization.- Intervention: Targeted outreach from health coaches, medication adherence support, ensuring care gaps are closed (e.g., cancer screenings), and education on managing their specific conditions.
- Tier 3: Low-Risk (The Bottom ~75%)
This is the generally healthy population. The goal is health maintenance and wellness.- Intervention: Less intensive support through automated communication, digital engagement tools, wellness reminders, and ensuring they receive routine preventive care.
Conclusion: The Strategic Imperative of Knowing Your Population
Risk stratification in value-based care is far more than a technical exercise; it is the core operational strategy of any successful program. It provides the roadmap for where, when, and how to deploy clinical resources to make the greatest impact. By combining robust data with sound methodology and a clear operational plan, healthcare organizations can move from being reactive problem-solvers to proactive guardians of population health—a transition essential for thriving under any VBC model.
In our next article, we will be stepping into the workshop to develop a simple, yet robust rule-based model for risk stratification.
What risk stratification challenges have you encountered in your practice? Share your insights in the comments!