We were reviewing a 74-year-old patient who lived two hours from the nearest major medical center. On the screen, his automated risk score was reassuringly green. Pulling from a sparse medical record and infrequent acute visits, the model flagged him as low risk and routed him to the default, low-touch outpatient pathway.
The man actually sitting in the exam room was actively deteriorating.
His data footprint was dangerously thin. But the system treated that thin record as reassuring rather than uncertain. Long stretches with little usable information looked like long stretches without trouble. The math was executed perfectly. The output was fiction.
The Power Behind Enterprise EHR Software for Large Healthcare Systems
Enterprise EHR boosts scalability, interoperability, and governance for large healthcare systems.
Overriding that score meant working around an automated pathway and reconstructing a fragmented history by hand. All that just to establish a reality the record had never captured.
What exactly is a clinician supposed to do when the measure is technically valid, but clinically wrong?
What value-based care got right
Value-based care fixed something that healthcare badly needed fixed. Fee-for-service rewarded activity. It never required that activity to deliver better outcomes. Value-based care forced systems to answer a far harder question: did the care actually work, and at what cost? Sajeed Chowdhury recently made the cost side of that case well.
From Blind Spots to Better Outcomes: How SNF Data Transparency Strengthens LTSS
What if health plans could identify member decline before an avoidable hospitalization occurs? Real-time clinical visibility into long-stay SNF members uncovers risk earlier and drives better outcomes.
Value-based care changed the rules. Outcomes, utilization, and accountability now had to be measured over time. For builders, that absolutely matters. Strong validation is not the same thing as value. It just isn’t. For a health-tech product to succeed and scale, it has to improve outcomes in the real world.
Value-based care was real progress.
But population-level accountability has a ceiling.
The average is only as good as the cohort
Value-based care works because it makes populations measurable. It has to. You can’t create accountability at scale without grouping people and comparing outcomes. That’s not a flaw. It’s the architecture.
Yet every cohort is an abstraction, and every abstraction depends on who was included in the data.
Martin and colleagues showed the problem clearly. Their risk scores were built largely from European-ancestry data. They performed best in the populations the data represented well. That makes sense. But as representation dropped, performance did too. That problem travels. A model can look strong overall and still be weak for people its data barely describe.
And therein lies the failure behind the green score in the exam room. Thin data do not equal low risk. Sometimes they just describe the patient badly.
A better average can still hide the person the average got wrong.
The unit of value is already moving
Lakshmi Halasyamani makes the patient-level point: value-based care has to work in the reality of that person’s life. She gets there by asking better questions. Precision tools get there by changing the instrument itself.
ArteraAI’s prostate cancer test helps identify which patients are likely to benefit from adding hormone therapy to radiation. In validation, only about one-third of patients were flagged as likely to benefit. That’s the point. The guideline defines the group. The tool tries to identify the person inside it.
The shift is showing up elsewhere too. Signatera starts with the individual patient’s tumor. Tempus is pushing the “care pathway” toward the person in front of the clinician.
Different products. Same directional shift.
Precision inherits the problem
These tools make predictions about individuals, but they still learn from populations. Artera’s evidence is strong, but even a strong product can be less reliable for people the underlying data barely represent.
The lesson from the green score carries forward: confidence is not the same thing as fit.
Precision inherits the gap it was supposed to close.
Value-based care makes precision auditable
A model can be confidently wrong. Outcomes are what catch it. Value-based care spent more than a decade building that exact discipline.
That’s the audit. Did the recommendation still hold up once it reached real patients?
When a product doesn’t hold up in practice, clinicians lose trust and start working around it. When it does hold up, it earns trust, gets adopted, and has a chance to scale.
Precision tools make the prediction. Value-based care tests it against reality.
The individual is the test
The man behind the green score didn’t need a better average. He needed a system that could tell the difference between a low-risk patient and one the data barely describe.
That’s the standard health-tech products now have to meet. Not just whether outcomes improve overall. Who benefits? Who doesn’t? And when is confidence misplaced?
A product doesn’t scale because it performs well on average. It scales because it holds up even for the patient the average misses.
Photo: MirageC, Getty Images
Mitchell Frye, PhD, AuD, MBA, is a health technology strategist and clinician-scientist focused on medtech, clinical AI, regulated care delivery, and real-world adoption. He leads a multidisciplinary rehabilitation section in the U.S. Department of Veterans Affairs health system and brings NIH-supported translational research experience in implantable devices, sensory neuroscience, neuroimmunology, and biofilms. His work focuses on the operating gap between biomedical evidence, clinical workflows, and commercial scale. He was previously an Assistant Professor at Rush University Medical Center and completed postdoctoral training at UT Southwestern and UT Dallas.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.