May 8, 2025
Standard risk scores look good on paper, but too often they don't translate into meaningful action.
If you’ve spent time in the trenches of value-based care, you’ve probably seen the limitations of standard risk scores. They look good on paper — color-coded dashboards, tidy numbers — but too often, they don’t translate into meaningful action.
That’s where a deep-dive into proper segmentation gives us the edge.
Instead of just labeling patients based on how sick they are, segmentation lets you group them by the kind of support they actually need. Done well, it gives your team clear priorities, smarter workflows, and a way to make interventions stick.
Here’s how to build a segmentation strategy that works — not just in theory, but in the day-to-day reality of value-based care.
Avoidable admissions and readmissions aren’t just expensive — they could be signals that something upstream broke down. In value-based care, every unnecessary ER visit or preventable bounce-back reflects a missed opportunity.
The problem? Most segmentation efforts lump very different patients into the same “high-risk” bucket. A patient with poorly managed heart failure and no caregiver is nothing like a young adult struggling with depression and chronic no-shows — but a generic risk model may treat them the same.
When we treat everyone alike, our care plans start to feel generic, our outreach misses the mark, and our teams burn out chasing leads that go nowhere.
Smart segmentation corrects this. It gives us a clearer view of who needs what kind of help — and how we can deliver it.
So how do we move from generic risk scores to a smarter segmentation strategy? Start here.
Don’t start by tossing every chronic condition into one category. Segment based on how those conditions behave over time and what their care trajectories look like.
For example, CHF patients often need tight follow-up for med titration after discharge. ESRD patients might need more focus on transportation or home health support. Grouping them just because they’re high risk misses the point.
Once you’ve got a clinical foundation, layer in social and behavioral risk. Things like food insecurity, lack of a caregiver, or low health literacy often drive readmissions more than a diagnosis ever could.
We’ve had real success identifying patients likely to be isolated post-discharge — those who don’t have anyone at home, who miss home health visits, or who never pick up their meds. That’s where a well-timed call or home visit can change the story.
Utilization patterns matter. Does the patient frequently visit the ED? Do they miss follow-up appointments? Avoid care altogether? These behaviors offer clues and insights into expected future patterns.
Also pay attention to where the patient goes after discharge. There’s a big difference between someone going home with no services and someone transferring to a skilled nursing facility. Segmenting by disposition helps tailor follow-up plans.
Let’s be honest: many segmentation efforts start in Excel, using weighted scores that blend comorbidities, age, and recent utilization. It’s a fine place to start — but don’t stop there.
Here are a few advanced methods we’ve seen work:
Here’s what this could look like in practice:
Medicaid
Using Latent Class Analysis, you might identify subgroups at high risk for care coordination failure — such as patients with frequent ED use, low health literacy, and unstable housing. These segments could then be targeted with mobile care teams and SDOH navigation support to improve continuity of care and reduce acute events.
Medicare & MA
A hierarchical clustering approach could identify patients with multiple comorbidities who also have low activation scores and medication adherence issues. For this group, strategies like pharmacy reconciliation, nurse-led check-ins, and proactive behavioral coaching can help reduce exacerbations and readmissions
.
Commercial Populations
Using K-means clustering or decision trees, you might uncover a segment of young adults with high rates of mental health-related readmissions and substance use. For this group, digital mental health navigation tools, peer support programs, and ongoing engagement platforms could be deployed. Another high-cost commercial segment — pregnant individuals with unmanaged gestational diabetes or limited prenatal care access — can be segmented and supported with targeted digital health coaching and community-based prenatal resources.
A robust data foundation is essential. This means integrating EHR data, claims, pharmacy data, SDOH feeds, and more. Timeliness matters — data that is weeks or months old may no longer reflect a patient’s current risk level.
Common pitfalls to avoid include siloed data sources, overfitting models to narrow datasets, and insufficient validation. It’s also important to constantly evaluate your models in the context of your organization’s unique population and workflows.
Emerging tools, like Flatiron’s Medical Language Model, can pull deeper insights from provider notes and patient messages — potentially flagging rising-risk patients earlier based on communication patterns. This may be better explored in follow-up tools and steps.
Insights only matter if they can be acted on. Assign tailored interventions to each segment. For example:
Make sure segmentation maps to real workflows, staff roles, and resources. Then, establish and track KPIs such as intervention uptake, readmission rates, patient activation levels, and ROI. Predictive models should be monitored for drift over time.
Segmentation isn’t just a side project or a one-time model build. It’s a core capability for any serious value-based care organization. Patients evolve. Care models evolve. So your segmentation strategy should too. Revisit it regularly. Bring clinical, analytics, and operational leaders into alignment. And above all — make sure segments lead to real-world action.
Smart segmentation isn’t just data work — it’s care work. And it’s one of the few levers we have to truly reduce avoidable utilization in a lasting way.
Let’s move past generic risk labels and start building care strategies that actually meet patients where they are.
