In 2020, four researchers at Duke and ATI asked a blunt question — are physical therapists ready to deliver value-based care for musculoskeletal pain? Their answer was honest and uncomfortable: not yet. Not because the models are wrong, but because the tools, the evidence standards, the shared language, and even the working definition of harm don’t exist. Six years on, the models have multiplied — bundles, at-risk MSK, upstream physical therapy — and the missing pieces are still missing.
For decades, the safety question in musculoskeletal care was whether a clinician might miss something serious. The evidence is that physical therapists clear that bar — no worse than anyone else at catching the red flag. Value-based care forces a harder, less comfortable definition: harm is also the unwarranted visit, the avoidable image, the escalation to a surgery the patient didn’t need, the inevitable care merely delayed.
And here the paper does something rare — it turns the lens on its own side. Getting to care sooner is not a free win. The authors cite evidence that patients who saw a physical therapist first sometimes had higher downstream hospitalization, not lower. Early is only valuable if it’s the right care for this patient. Which raises the only question that matters:
Who actually benefits — and how would you know before you treat them?
The paper’s central shift is from diagnostic to prognostic decision-making. Today, treatment happens because a diagnosis is “amenable” to it — regardless of whether the patient will actually improve. Value-based care asks the harder thing: treat by the probability of benefit, and refer when the odds are low. Everyone in the field nods at this.
Then the paper admits the problem out loud: the prognostic tools to do it “don’t exist,” have “marginal predictive accuracy,” aren’t validated, and aren’t in the EHR. That admission is the whole story, and it points at something the paper doesn’t quite name. You cannot build a tool that predicts who will benefit without a longitudinal record of who got what and what happened next. A prognosis is a function of accumulated outcomes. No record, no prognosis — and no prognosis, no value-based decision.
The record isn’t downstream of the model. It’s upstream of all of them.
The authors are just as hard on the research. Randomized trials, they argue, can’t answer value questions — their tight controls destroy the real-world generalizability that policy needs. The observational studies that dominate the field are riddled with selection bias. What they call for is a different substrate: pragmatic trials embedded in real care, causal-inference methods, rigorous real-world data.
Read that as a spec and it describes an artifact: longitudinal, real-world, rigorously captured, attributable. That artifact is an attested outcome record — one where a named clinician signs what happened, the provenance is tracked, and the patient carries it forward. The same record that lets you make a prognostic decision for one patient is the evidence base that proves value across thousands. It was never a billing byproduct. It is the evidence layer the whole enterprise runs on.
And the signature on it does one specific job. It is not there to make the data more accurate — a machine checks a fact better than a tired human does. It is there so the record is accountable: so there is an identifiable person who stands behind what it says, which is precisely what a payer, a regulator, and a court require before they will act on it.
Build that, and the prognosis, the evidence, and the value follow. Skip it, and you have another model with nothing underneath.
That’s the layer I build — the attested, owned outcome record that a prognosis, an evidence base, and a payer can all stand on. Not a new care model. The thing the care models have been missing. Read the fuller thesis →
Blaine Warkentine, MD · SolvingHealth · Boulder, CO · Get in touch →
A clinical-translator’s read of a published Perspective — not clinical, policy, or investment advice. The argument extends the authors’ physical-therapy framing to the broader musculoskeletal episode; the characterizations of harm, prognostic decision making, and research standards are drawn from the source above.