Continuous wearable ECG hyperkalemia-risk screening in RAASi titration — the story so far

How to read this. This is a synthesis, not a new source of truth: every claim it cites carries the tag the candidate file gives it right now, copied not paraphrased, and every line quoted from a deck is quoted verbatim. See docs/narrative-spec.md.

The one-paragraph version

This is a continuous, adhesive-patch wearable ECG device that runs a deep-learning model between outpatient visits, scoring ECG-morphology patterns for hyperkalemia risk in adult CKD and heart-failure patients being initiated or up-titrated on a RAAS inhibitor, and pushing an alert that triggers — never replaces — a confirmatory lab draw. The regulatory shape is unsettled by design: the candidate's own frontmatter carries two live, unresolved routes, a 510(k) contingent on a predicate that has not been found and a De Novo route the file now judges at least as likely. The one real payer signal on record is CMS's own proposed RAPID coverage pathway — verified as to what it says, not as to whether this device would ever qualify or whether the notice survives its comment period. Every mechanism figure in the file traces to one retrospective, single-institution study read at abstract depth only, and the one assumption the entire clinical workflow depends on — whether a clinician will act on an ECG-only flag the way they act on a lab value — has been run through this repo's own adoption connector and come back structurally unanswerable.

The story so far

RAASi therapy improves outcomes in heart failure and CKD but raises serum potassium, so titration is gated on repeat venous draws between clinic visits [dossier: potassium-monitoring-ckd]. This candidate's mechanism replaces part of that gate with a continuous adhesive ECG patch that scores waveform morphology for hyperkalemia risk and pushes an alert triggering, never replacing, a confirmatory draw. The closest analog, AliveCor's KardiaK Software Platform, reportedly received FDA Breakthrough Device Designation in September 2018 [claim 1: unconfirmed], but no connector in this repo checks designation status directly, and FDA's own disclosure policy means a pre-authorization designation is structurally unconfirmable against any registry in the first place [claim 14: unconfirmed]. No 510(k) clearance or De Novo grant for KardiaK has been located [claim 2: unconfirmed] — a broadened, non-top-1-match search across the whole openFDA 510(k) and classification databases found zero devices, under any name, carrying a hyperkalemia, electrolyte, or potassium indication, and confirmed the one product-code family built for ECG-based AI "notification software" covers ten unrelated cardiac conditions and none of them [claim 25: unconfirmed]. That narrows the regulatory question specifically to a De Novo route with no existing predicate-code family to shorten it — a founding entrant, not a follow-on, if this device reaches FDA first. One further verified fact sits adjacent without resolving any of it: FDA's classification database records a regulation_number for that product-code family that does not correspond to any real, codified CFR section as of the 2026-08-19 eCFR issue [claim 26: verified] — real, but judged by the file's own Scorer too tangential to this device's own pathway question to move Regulatory pathway off blank.

CMS has published three successive instruments since 2021 trying to time Medicare coverage to FDA authorization for Breakthrough-designated devices — one finalized and repealed within ten months, one capacity-limited and now paused, and a third (RAPID) out for comment as of this file's generation, entering at the IDE-presubmission stage and excluding IVDs by name (seed context, not evidence). This candidate's non-IVD, screening-trigger framing reads that pathway's own design levers rather than following them by accident. The candidate's claims restate what the notice itself says, each carrying the Federal Register document's own number rather than a reference to the seed: RAPID requires "separately payable" status [claim 4: verified], is limited to Class II/III Breakthrough-designated devices entering at the IDE-presubmission stage [claim 5: verified], excludes IVDs by name [claim 6: verified], and is a notice with an open comment period, not a final rule, closing 2026-10-13 [claim 7: verified]. Those four are the only verified claims in the file that speak to this device's own situation at all — and all four describe the document's text, not whether this device qualifies for it.

Every clinical figure traces to one paper: a Mayo Clinic-affiliated deep-learning model trained on 1,576,581 ECGs from 449,380 patients and validated retrospectively on 61,965 stage-3-or-greater CKD patients across three Mayo sites, reporting 2-lead AUCs of 0.883 (Minnesota), 0.860 (Florida), and 0.853 (Arizona) [claim 17: unconfirmed] — figures the Verifier confirmed exactly against the source abstract, correcting an earlier claim's looser "approximately 0.88-0.89" framing [claim 3: unconfirmed]. At the model's own 90%-sensitivity operating point, specificity was only 54.7%-63.2% [claim 17: unconfirmed], against a validation-cohort hyperkalemia base rate of just 2.6%-4.8% [claim 18: unconfirmed] — a low-prevalence target that makes the false-alarm question real, not hypothetical: most flagged patients without true hyperkalemia would still need a confirmatory draw under this device's own intended-use framing. No non-Mayo, prospective, or external validation of the model has been located by any pass on this file.

The file is unusually direct about its own limit here: whether a clinician managing RAASi titration would act on an ECG-based hyperkalemia flag without an accompanying lab value, and order a confirmatory draw promptly, was run through this repo's own adoption connector and returned "unverifiable" — stays unconfirmed permanently unless a human attaches real research [claim 12: unconfirmed]. The nearest real-world proxy is what clinicians do when a lab value, not an ECG flag, comes back high: in a US HFrEF registry, RAAS-inhibitor discontinuation ran 12.7%-20.4% by drug class over a median 18 months, with CKD the only independent predictor of increased discontinuation risk [claim 20: unconfirmed]. The dossier frames the gap precisely: every study isolating hyperkalemia specifically as the discontinuation reason is non-US; the one US study reports a general discontinuation rate with CKD as a predictor rather than hyperkalemia isolated as the reason; and none of the studies, US or otherwise, involve an ECG-based flag at all — every one is a response to an actual lab value, the opposite of what this device asks a clinician to act on [dossier: potassium-monitoring-ckd].

What's solid

• CMS's RAPID coverage-pathway notice requires the device to be "separately payable" — billable to Medicare in its own right [claim 4: verified]
• The same notice limits RAPID eligibility to Class II and Class III Breakthrough-designated devices, entering at the IDE-presubmission stage [claim 5: verified]
• The same notice excludes IVD products from RAPID by name [claim 6: verified]
• The same notice is a notice with a comment period, not a final rule — comments closing 2026-10-13, nothing in it in force as of this candidate's generation [claim 7: verified]
• FDA's device/classification database's regulation_number field for the ECG-AI "notification software" product-code family (including QYE) does not correspond to any real, codified 21 CFR 870.2380 section as of the 2026-08-19 eCFR issue [claim 26: verified]

What's still open

• Whether AliveCor's KardiaK Breakthrough Device Designation (reported 2018) is real and current — structurally unconfirmable against any FDA registry until authorization happens, by FDA's own disclosure policy [claim 1: unconfirmed], [claim 14: unconfirmed]
• Whether any 510(k) or De Novo grant exists for KardiaK, or for any ECG-based hyperkalemia predicate under any company name — the broadened search found none, narrowing this device to a founding De Novo entrant [claim 2: unconfirmed], [claim 25: unconfirmed]
• Whether an FDA De Novo bulk-data file can even be fetched to check the De Novo question further — none exists at the one URL this repo knows to check [claim 15: unconfirmed]
• The Mayo Clinic mechanism evidence at full-text depth — abstract-only, single institution, no external validation located [claim 3: unconfirmed], [claim 17: unconfirmed], [claim 18: unconfirmed]
• The HF/CKD comorbidity prevalence and US RAASi prescription-prevalence figures the file originally cited — the underlying two papers were never re-identified across two research passes, though adjacent, better-sourced substitutes were found instead [claim 8: unconfirmed], [claim 9: unconfirmed], [claim 27: unconfirmed], [claim 28: unconfirmed], [claim 29: unconfirmed]
• No HCPCS code, payment rate, or coverage policy specific to this service has been identified; a live CMS file exists but this repo's fetch tool cannot currently ingest its ZIP format [claim 11: unconfirmed], [claim 16: unconfirmed]
• The core adoption assumption — whether a clinician would act on an ECG-only flag without a lab value — run through the adoption connector and returned structurally unanswerable [claim 12: unconfirmed]
• Freedom to operate: AliveCor may hold patent claims on ECG-morphology-based potassium inference; the claim line itself was found truncated in the file and no patent search could be run for lack of a configured API key [claim 13: unverified]
• Six deck-surfaced assumptions, appended unverified and never yet checked: the between-visit draw burden this device targets [claim 31: unverified]; guideline recognition, currently absent [claim 32: unverified]; patient/caregiver ability to wear the patch unsupervised [claim 33: unverified]; a reachable adoption share distinct from raw prevalence [claim 34: unverified]; a price or payment rate, established nowhere in the file [claim 35: unverified]; and whether any company besides AliveCor is pursuing this mechanism [claim 36: unverified]
• Three more deck-surfaced feasibility assumptions: whether the Mayo model's accuracy transfers to a continuously worn patch's signal [claim 37: unverified]; whether a bench validation on wearable-patch data exists before any patient wears the device [claim 38: unverified]; and whether the patch and reader would reliably operate through multi-day unsupervised wear [claim 39: unverified]
• Dossier context: capillary fingerstick sampling, the domain's other conservative alternative, has been repeatedly abandoned in adult inpatients — specimen rejection "as high as 30%" in an initial cohort and hemolysis running roughly 1.7x venous — real evidence the underlying monitoring burden is genuine and unsolved, not evidence anyone will adopt this device's particular fix [dossier: potassium-monitoring-ckd]

Where the three lenses agree — and where they don't

Desirability turns on the same clinician-behavior question the story-so-far already centers on — whether anyone acts differently on an ECG-only flag than they already do on a lab value.

"whether a prescriber managing RAASi titration would act on an ECG-based hyperkalemia flag — ordering a confirmatory draw promptly — without an accompanying lab value" — the load-bearing condition of the desirability deck (built 2026-08-30).

Viability turns on the reimbursement pathway this file has actually verified something about, and whether this specific device would ever qualify for it.

"whether CMS's proposed RAPID coverage pathway (document 2026-16368) is finalized after its comment period closes 2026-10-13, and whether this device would hold the Breakthrough designation and "separately payable" status RAPID requires" — the load-bearing condition of the viability deck (built 2026-08-30).

Feasibility turns on the one transfer question every other engineering condition sits downstream of — whether the Mayo model's accuracy survives leaving the clinical-grade capture it was built and validated on.

"whether the hyperkalemia-detection accuracy demonstrated on Mayo Clinic's retrospective, clinical-grade ECG captures survives the move to a continuously worn adhesive patch's signal" — the load-bearing condition of the feasibility deck (built 2026-08-30).

All three lenses agree on one thing without saying so directly: every one of them ends at a question no connector in this repo can resolve — an adoption behavior, a policy finalization plus a designation this repo cannot check, and a bench transfer nobody has run. Where they diverge is what would move first: viability's condition has a date already attached to it (2026-10-13); desirability's and feasibility's do not.

The load-bearing question

If the De Novo pathway clears, CMS finalizes something close to RAPID, and the Mayo model's accuracy survives the move to a continuously worn patch — does any of that change whether a clinician acts on an unconfirmed ECG flag the way the file's own evidence shows they act on a confirmed lab value, or does the device end up generating alerts into the same gap the dossier's non-ECG discontinuation data already describes?