TEMPO lands while wearable endpoints still wobble

On April 24, 2026, FDA’s Digital Health Center of Excellence launched TEMPO — Technology-Enabled Meaningful Patient Outcomes for Digital Health Devices — in coordination with CMMI’s ACCESS chronic-care model. Manufacturers admitted to the pilot can request enforcement discretion on certain requirements when their device is offered to ACCESS participants; in exchange, they must collect real-world performance data and share it with FDA. Statements of interest opened January 2; follow-up information requests to selected candidates went out in March. The DHCoE hub scopes eligibility to four ACCESS clinical areas: early cardio-kidney-metabolic, cardio-kidney-metabolic, musculoskeletal, and behavioral health, with conditions limited to those listed in the Federal Register notice.

The structure is novel. FDA and CMS are simultaneously shaping the regulatory and reimbursement pathway for the same device, with RWE as the connective tissue. For sponsors of digital therapeutics and SaMD, that collapses two evidence conversations that used to happen serially.

The counter-current

Four months before TEMPO launched, FDA updated its General Wellness Policy to let consumer devices “estimate, infer, or output” blood-pressure and blood-glucose readings without authorization, provided claims are framed as wellness. Per STAT, Oura, Samsung, and crowdfunded entrants like Pin Pulse have already shipped BP/glucose features under that exemption. So the same agency now demanding real-world performance data under TEMPO let an entire consumer category bypass authorization a quarter earlier. Whatever evidentiary floor TEMPO establishes will be read against that contrast.

The endpoints aren’t ready for the ask

The harder problem is that the analytic foundations for wearable-derived endpoints are still visibly unsettled, and TEMPO will surface it. Eric Topol’s VO2 max critique is the cleanest case study: wearable-imputed VO2 max from Apple Watch, Garmin, and Fitbit runs a mean absolute percentage error of 7–16% against laboratory metabolic-cart measurement, systematically underestimating fit individuals and overestimating unfit ones, with added optical-heart-rate bias in people of color. Over 99% of the fitness–mortality evidence in a 2024 meta-analysis derives from MET-based cardiorespiratory fitness, not VO2 max. A widely deployed consumer digital biomarker, mapped onto an outcomes literature that doesn’t actually validate it.

The methodology side is catching up, slowly. A 2025 Journal of Biopharmaceutical Statistics paper frames autocorrelation, irregular sampling, device-dropout missingness, and ICH E9(R1)-aligned estimands for high-frequency signals as the open problems for continuous-monitoring submissions — CGMs, accelerometers, ambulatory BP. A new Statistics in Medicine paper proposes a maximum-likelihood scalar-on-function quantile regression with a subject-specific, time-varying validity indicator that separates structural zeros — non-wear, connectivity dropouts — from intrinsic near-zero activity. Applied to childhood obesity accelerometry, the corrected step counts align with criterion energy-expenditure measurements. Useful, but not yet validated in a regulatory submission, and software scalability isn’t discussed.

Then there’s provenance. Dexcom warned that G7 lots destined for destruction were stolen in transit and are circulating through unauthorized dealers. Any TEMPO submission relying on CGM data — and the cardio-kidney-metabolic use area effectively guarantees some will — now has a chain-of-custody question on the audit trail. Real-world performance data inherits real-world supply chains.

What it means for the evidence package

TEMPO is, functionally, an RWE-collection pilot bolted onto a CMS payment model, and the “meaningful patient outcomes” language signals that COAs will sit at the center of accepted submissions. Biometrics teams should expect to defend three things in parallel: the measurement properties of the device-derived endpoint against a credible reference (the VO2 max problem), the analysis plan for continuous-monitoring data under E9(R1) (the estimand problem), and the device provenance and firmware version controls feeding the dataset (the Dexcom problem). None of those are solved by a vendor dashboard, however real-time.

The pilot is too narrow to set precedent on its own — four use areas, limited conditions, voluntary. But it’s the first concrete artifact of how FDA wants the digital-health evidence conversation framed, and the contrast with the wellness exemption tells you the agency is sorting devices into two regulatory tiers in real time. Sponsors who treat TEMPO submissions as a wellness-tier exercise will find out which tier they’re in the hard way.

Protocol read: TEMPO is the first place FDA will see, in writing, what sponsors actually think a defensible digital-endpoint evidence package looks like. The methodology gaps are real, named in the literature, and not yet closed — the pilot will expose which sponsors have been reading the JBS papers and which have been reading the press releases.

What to do now:

• Inventory active and planned digital-endpoint studies against the four ACCESS use areas; flag which would be TEMPO-eligible and which need the analytic gaps closed before submission.
• Pre-specify validity indicators and non-wear handling in SAPs for wearable-derived endpoints — cite the E9(R1) estimand framing explicitly rather than letting it sit in an appendix.
• Add device provenance, lot, and firmware version to the data-management plan for any trial using consumer-grade or third-party-sourced sensors; the chain-of-custody question is now on the table.