Value-Based Care Generates the Evidence Base Healthcare Needs

Two of healthcare's favorite buzzwords are tripping over each other, and almost no one will say so out loud. Evidence-based medicine (EBM) demands decades of evidence. Value-based care (VBC) demands immediate innovation and earlier interventions. The gap between them is where truly preventive medicine quietly stalls.

The conflict is structural. EBM asks what the evidence supports. VBC asks what investment will produce the best outcomes. They should be working together. They aren't because of the data deficiencies systemic to the industry.

EBM's standards for evidence — randomized clinical trials, slow-moving guidelines, retrospective consensus — were developed for a different era of medical innovation. They don't match the timescale VBC demands. Many proactive preventive interventions plainly do not have the trial data EBM demands, so they don't get recommended. This represses the volume of early interventions. The longitudinal evidence baseline never gets built. The cycle closes out the work VBC was designed to enable.

The chicken-and-egg problem in clinical evidence

Consider the case of subclinical heart failure screening. The current ACC/AHA guidelines don't recommend systematic screening for Stage B heart failure in at-risk populations. The reasoning is a familiar EBM catch-22: there isn't yet randomized trial evidence demonstrating that screening improves long-term outcomes, so the screening doesn't get recommended.

A study by physicians at CenterWell, published in JACC, illustrates what's lost. The team prospectively analyzed 63,512 seniors using claims data, identifying patients with at least two Stage A risk factors (diabetes, hypertension, vascular disease, metabolic syndrome) and offering them screening echocardiograms. Seventy-eight percent of the population qualified. Of those who qualified, 64% completed the echo. And 53% of those screened met criteria for Stage B heart failure — meaning roughly a quarter of the at-risk senior population had subclinical structural heart disease that would otherwise have gone undetected. The study also incidentally identified 610 ascending aortic aneurysms.¹

This is exactly the kind of evidence that should drive next-generation guidelines. But it is observational, and under the dominant EBM framework, observational evidence gets discounted relative to RCT data even when the observational study is well-designed, large, and prospective. The RCT that would prove the case will not happen, because no commercial sponsor exists to fund a multi-decade prevention trial on subclinical cardiac structural changes. The screening that would generate the longitudinal evidence base does not happen at scale, because the guidelines don't recommend it. The guidelines don't change, because the longitudinal evidence doesn't exist.

The cost of that delay is, unfortunately, revealed to many physicians through personal experiences. Practice patterns around preventive cardiac screening often update only after someone the clinician knows personally has a tragic episode that a preemptive screening might have caught; a sign that personal experience is outpacing published evidence, which is again a sign that the evidence system is too slow.

The same pattern shows up in less dramatic places. Clinical studies of alcohol consumption in Iran (where alcohol is criminally prohibited and severely stigmatized) produce documented prevalence rates approaching what gets recorded in clinical encounters in major U.S. cities.² The data is telling you what got documented, not what is actually happening. When clinical practice insists on evidence that depends on accurate documentation of patient behavior, and that documentation is systematically distorted by the very thing being studied, the resulting "evidence base" is at best a floor and at worst an artifact.

EBM doesn't need to be replaced, but the standards need to evolve and promote better data for longitudinal observation. Treating observational evidence at scale as inferior to non-existent RCT data isn't rigor. It is paralysis.

Why we don't have the longitudinal evidence we need

The economics of evidence generation helps to explain the gap. Randomized controlled trials are expensive, optimized for treatment effects rather than prevention effects, and structurally biased toward interventions someone can commercialize. The prevention studies that would test whether early intervention on metabolic dysfunction, cognitive trajectories, or subclinical cardiac changes improves outcomes over decades do not fit grant cycles, do not fit product launch timelines, and do not pay for themselves.

Real-world evidence frameworks are starting to expand what counts. The FDA has issued guidance on the use of real-world data in regulatory submissions. Registry-based research is growing. Observational studies using claims data at population scale are increasingly capable of producing findings the industry can't ignore. But the dominant clinical guidance still privileges RCT data, and the transition is slow.

This is where value-based care creates an opening the field hasn't fully recognized. Organizations operating under capitated arrangements have both the longitudinal data flow and the financial incentive to invest in prevention. They are also positioned to generate the observational evidence base the field has been waiting for. A VBC organization that systematically screens its at-risk population for subclinical disease, tracks progression over time, and documents outcomes is producing exactly the kind of large-N longitudinal evidence the RCT economy cannot. VBC isn't just a payment model. It's an evidence-generation model, if it is allowed to be.

The diagnostic cost paradox

The other thing standing in the way of preventive screening at scale is cost. The case for systematic Stage B heart failure screening, or systematic earlier-stage detection across any chronic condition, depends on the per-patient cost of the screening being low enough that the population-level math works. In a system experiencing rapid technological advancement, the cost of diagnostic services should be falling, but that's not the case across the board.

The first reason is that labor, not equipment, is the binding constraint on diagnostic capacity. Imaging volumes are growing 3–4% annually, while the workforce that operates the imaging system is contracting in the places that matter most. The radiology technologist vacancy rate has nearly tripled in three years, from 6.2% to 18.1%.³ These are the people who physically run the machines, position patients, and execute imaging protocols — work that doesn't yet have an AI-augmented productivity equivalent in the way that scan interpretation does. The MRI sits idle not because the equipment is constrained, but because there's no one available to operate or read it.

Where AI is genuinely closing the productivity gap is in scan interpretation. Radiology was a central example in our Intelligence First piece — the FDA had authorized 1,104 AI-enabled radiology devices through the end of 2025, the largest single domain of FDA AI authorizations. AI-augmented radiology genuinely improves throughput on the read side. The productivity gains, though, are not showing up as lower diagnostic prices for patients. They are accruing to providers and AI vendors. The market structure does not translate productivity into price competition.

Where consolidation has reduced that competition, the gap is starker. Hospital outpatient departments routinely charge 30% or more above freestanding imaging centers for identical procedures, justified by facility fees that bear no relationship to the underlying cost of the scan. Patients walking into a hospital are paying a premium for the building, not for the imaging, and they are paying it because they often have no way to know they could pay less down the street.

Hospital price transparency rules, in effect since 2021, were supposed to fix this. They've done something different. Four years in, the dominant pattern is price convergence rather than price reduction. High-end prices have fallen by an inflation-adjusted 6.3% annually while low-end prices have risen 3.4%.⁴ Variation across markets has narrowed; the average has crept up. The transparency law gave lower-paid providers leverage to negotiate up toward the market median without forcing high-paid providers down past it. Patients haven't yet seen the meaningful price reductions the law intended.

This is the second place where AI could do some useful work. The transparency data is structured poorly enough that no human can navigate it, but it is at least published. Pattern-matching across thousands of disclosed rates to surface the actual cheapest options for a given procedure in a given market is exactly the kind of pattern recognition AI excels at. The transparency law produced unusable data for the average patient to derive any real information. AI is the tool that can finally make sense of it.

The pattern across all of this is the same one we've described throughout the series. The technology exists. The data exists. The economic structures around the technology and the data prevent the gains from reaching patients.

Wearables, EMRs, and the data the system can't yet use

The forward-looking version of this problem is wearables, and the most recent data from the same group at CenterWell illustrates what becomes possible when value-based primary care deploys them at scale. Beginning in April 2025, CenterWell launched a cardiac arrhythmia screening program in its senior primary care clinics, providing continuous ambulatory ECG monitors to high-risk patients identified by diagnostic risk factor — uncontrolled diabetes, COPD, advanced CKD, pulmonary hypertension, and others. By September 2025, just five months in, 21,589 patients had completed 14 days of monitoring with a 99.2% compliance rate. Arrhythmias were detected in 84.5% of participants. More than a quarter of cases were considered urgent or time-sensitive.⁵

This is what the system looks like when the data substrate, the upstream care model, and the wearable integration all come together. None of it required an RCT. None of it required guideline change first. The VBC organization deployed the tool, captured the data, acted on the findings, and is now positioned to publish the longitudinal evidence base the field has been waiting for. The follow-up question — whether early detection improves outcomes and reduces hospitalizations — is exactly the kind of work that the screening program is structured to answer.

This is happening against a backdrop of broad wearable adoption. Forty-four percent of Americans own activity-tracking wearables — Apple Watches, Oura rings, Fitbits, Whoops, continuous glucose monitors, sleep trackers, and more.⁶ The data these devices generate is longitudinal, behavioral, and orders of magnitude richer than what claims and EHR encounters can capture. It is exactly the substrate genuine upstream care depends on.

The technical integration problem is actually getting solved. Validic launched native Epic and Oracle Health wearable integration at CES 2025, supporting 350+ device models from Apple, Fitbit, Garmin, Google, Oura, Samsung, and Withings. Academic medical centers are running production integrations. A 2026 study in Frontiers in Digital Health found that of 843 patients who connected consumer wearables to a major academic medical center's EHR, 82% remained engaged at three months.⁷ The data is starting to flow.

The harder problem is what to do with it once it arrives. A survey of physician attitudes toward smartwatch cardiovascular data found that clinicians broadly agree the data is valuable, but cite two primary barriers to actually using it: the absence of billing codes for reviewing wearable data, and the lack of time during the fifteen-minute clinical encounter to do anything with it.⁸

AI is well-positioned to relieve both barriers. Pre-processing wearable data streams into actionable clinical summaries means the physician's fifteen minutes aren't spent data-mining. The billing codes catching up is a policy problem, but the time-savings problem is largely solved with the correct pre-visit processes in place.

The CenterWell arrhythmia program shows what this looks like in production. Continuous ambulatory monitoring devices in this category use AI-augmented interpretation as standard practice; without it, reviewing 14 days of continuous ECG data per patient across 21,589 patients in five months would be impossible at the cardiologist or radiologist staffing levels the diagnostic system currently supports. The screening works at that scale because the technology category is built around an AI-first review workflow, with urgent and time-sensitive findings surfaced for human clinical confirmation.

The connection back to the previous piece is direct. The wellness industry isn't only providing the upstream health conversation the clinical encounter doesn't have time for. It's also generating health data that the clinical system receives but cannot yet reimburse for, cannot yet plan for, and does not yet have the evidence base to act on. The system has ceded both halves of the upstream work — the conversation and the data — to actors operating outside the clinical economy. The CenterWell program is a glimpse of what reclaiming that work looks like.

EBM and VBC working together

The synthesis the series has been building toward sits here.

VBC should be expanding the evidence base EBM operates on, not fighting against it. The capitated economics and longitudinal data flows of VBC organizations are exactly what large-N observational evidence generation requires. EBM should be a real-time loop fed by that evidence, not a slow-moving anchor that delays preventive work for decades while the RCT system fails to deliver studies that won't be commercially funded. The Kaizen frame matters here — continuous improvement requires continuous evidence, which requires continuous data, which requires continuous engagement. Each piece of the series points at one of those requirements.

If you want to talk about where your organization fits, where your data infrastructure is and isn't ready, or what generating publishable evidence from your VBC operations could actually look like, we'd like to talk.