idea-011 · feasibility deck

Every slide states a condition that would have to be true, then reports where it stands using the candidate file's own claim and its own confidence tag. A deck never upgrades a tag, invents a number, or recommends anything, and there is no ask slide.

Continuous wearable ECG hyperkalemia-risk screening — feasibility

The chair: engineering. The people who have to turn a single-institution retrospective algorithm into a continuously worn patch, get it through a regulatory pathway that may not exist yet, and produce evidence a reviewer will ask for.

How to read this: every slide is a condition, not a conclusion. A bracketed claim reference points into knowledge-base/candidates/idea-011.md — the number is the claim's position in its ## Claims list, the tag is copied from it and never adjusted here. A [no claim] marker means nothing in the file speaks to this. This is the first deck built for this candidate. See docs/deck-spec.md.


Slide 1 — What the device actually has to do

Would have to be true: An adhesive patch would have to continuously record ECG waveform data between clinic visits, score it on-device or in the cloud for hyperkalemia-risk pattern (e.g. T-wave morphology), and push an alert prompting a confirmatory lab or point-of-care potassium draw — never a stand-alone diagnostic claim.

Where it stands: This is the candidate's own deliberate design choice, stated in frontmatter rather than as a checkable claim: a screening-trigger framing, chosen specifically to avoid a stand-alone diagnostic-replacement claim's harder evidence bar — [no claim] backs the framing itself. No sensitivity, specificity, or alert-latency target is specified anywhere in the file as a build requirement — [no claim]. The nearest performance figures are Mayo Clinic's own retrospective results: AUC 0.853-0.883 across three validation sites, specificity 54.7%-63.2% at a 90%-sensitivity operating point [claim 17: unconfirmed] — a result, not a spec.

What would settle it: A clinician-defined target operating point (sensitivity/specificity trade-off) for this intended use, which no source in this file has set; whether FDA would accept "screening adjunct" as materially lower-burden than a stand-alone diagnostic claim is explicitly unconfirmed and named in the file as exactly the kind of question Factor 1 verification should resolve.

If it's false: Without an agreed target operating point, the mechanism slide below has nothing to be measured against.


Slide 2 — The mechanism that has to hold

Would have to be true: ECG-morphology changes (e.g. T-wave features) would have to discriminate hyperkalemia risk at a clinically useful accuracy — and that accuracy would have to be established at more than abstract depth.

Where it stands: The best evidence on record is a single Mayo Clinic-affiliated deep-learning model, trained on 1,576,581 ECGs from 449,380 patients and validated retrospectively on 61,965 stage ≥3 CKD patients across three Mayo sites, reporting 2-lead AUCs of 0.883 (Minnesota), 0.860 (Florida), 0.853 (Arizona) [claim 17: unconfirmed] — figures the Verifier confirmed match the source abstract exactly, correcting an earlier claim's imprecise "approximately 0.88-0.89" range [claim 3: unconfirmed]. The base rate the model was validated against was itself low: hyperkalemia prevalence 2.6%-4.8% in those same cohorts [claim 18: unconfirmed] — a low-prevalence target is a harder one for any screening test to hold specificity against. Both claims stay unconfirmed, and the file is explicit about why: a literature-detail read returns a structured abstract, "a lead to read, not evidence for a claim," so full-text methodology and any external (non-Mayo) validation remain unread.

What would settle it: A full-text read of PMID 30942845 for methodology, any 4-lead result, and external validation; a search for any prospective or non-Mayo multi-site validation published since 2019 — none has been located by any pass on this file so far.

If it's false: If the mechanism does not discriminate risk usefully outside a single health system's retrospective data, nothing downstream — the build spec, the bench plan, the regulatory pathway — has a working device to attach to.


Slide 3 — What the pathway forces you to build

Would have to be true: Whichever regulatory route applies would have to define a concrete evidence and standards package — and the candidate's own frontmatter carries two live, unresolved routes rather than one.

Where it stands: A 510(k) route is contingent on a cleared predicate existing, and none has been found: no clearance or De Novo grant for KardiaK, the closest analog, has been located [claim 2: unconfirmed]. A broadened, non-top-1-match openFDA search confirms this more specifically — zero devices under any name carry a hyperkalemia, electrolyte, or potassium indication, and all ten product codes in the one classification family built for ECG-based AI "notification software" cover unrelated cardiac conditions [claim 25: unconfirmed] (mixed finding). That narrows the practical build target to De Novo, with no existing predicate-code family to shorten the special-controls package — a harder, more specific engineering target than "510(k) or De Novo" in the abstract. A separately verified but tangential finding is that this product family's own regulation_number field does not correspond to a codified CFR section [claim 26: verified] — a registry-metadata quirk, not a build requirement.

What would settle it: FDA pre-submission (Q-Sub) engagement once a De Novo path is committed to, to define the special-controls performance battery this device would need to meet — a route with no shortcut available given claim 25's finding.

If it's false: If a 510(k) predicate surfaces later, the build package shrinks substantially; if it doesn't, the special-controls package has to be built from nothing, which is the more expensive and slower of the file's own two named routes.


Slide 4 — The hardest unknown

Would have to be true: The hyperkalemia-detection accuracy demonstrated on Mayo Clinic's retrospective, single-institution, clinical-grade 2-lead ECG captures would have to transfer to a continuously worn adhesive patch's signal — despite motion artifact, multi-day electrode contact degradation, and a lead placement that may differ from a bedside capture.

Where it stands: No claim in the file addresses this transfer question at all — the source evidence [claim 17: unconfirmed] is entirely short, clinical-grade, single-timepoint ECG captures, not continuous ambulatory wear. This deck's own write-back names the gap directly: whether the demonstrated AUC and specificity survive the move to a continuously-worn patch's signal quality is unestablished anywhere in this file [claim 37: unverified].

What would settle it: A bench comparison of algorithm performance on wearable-patch ECG signal (with realistic motion artifact and multi-day wear degradation) against the same algorithm's performance on Mayo's original clinical-grade captures.

If it's false: The device's core mechanism claim degrades exactly where it matters most — in the unsupervised, multi-day, ambulatory setting this candidate is designed for — and no amount of regulatory or reimbursement progress compensates for that.


Slide 5 — Bench evidence before anything lives

Would have to be true: A bench or retrospective validation would have to show the model holds its discrimination accuracy on continuously streamed wearable-patch waveform data — before any patient wears the device.

Where it stands: No such validation exists in the file — this deck's write-back records the gap as a standalone condition [claim 38: unverified], directly downstream of Slide 4's transfer question. The device would also have to survive multi-day continuous wear physically (battery, skin adhesion, wireless streaming) in an unsupervised home setting, which is likewise unaddressed by any existing claim [claim 39: unverified].

What would settle it: The bench validation named above, run against a target operating point that Slide 1 notes does not currently exist either; a multi-day wear-reliability bench test.

If it's false: The mechanism and the hardware both fail before an animal or patient is ever involved — which is the cheapest possible place for this candidate to fail, and this file currently has zero evidence ruling either failure out.


Slide 6 — Making it, and using it

Would have to be true: The patch, reader, and streaming pipeline would have to be manufacturable at scale, correctly applied by an untrained patient or caregiver, and reliably operate for days at a time in a home rather than a clinic.

Where it stands: No claim addresses manufacturing, supply chain, or the home-use reality — [no claim] on all three. The multi-day wear-reliability condition named on Slide 5 [claim 39: unverified] and the patient/caregiver application question named on the desirability deck [claim 33: unverified] both bear directly here; this slide references rather than restates them, per the deck-spec's guidance on shared conditions.

What would settle it: A contract-manufacturing quote against a defined device spec (which Slide 1 notes does not yet exist); a usability study with real patients and caregivers, not technicians.

If it's false: A device that performs on the bench but cannot be reliably worn or applied by its actual intended user never generates the flag it was built to generate.


Slide 7 — What FTO forbids

Would have to be true: The device's design — patch form factor, lead configuration, and algorithm — would have to be buildable around whatever AliveCor (or any other holder) has patented in ECG-morphology-based potassium inference.

Where it stands: Genuinely open. AliveCor holds Breakthrough Device Designation on the closest competing mechanism and may hold patent claims on ECG-morphology-based potassium inference, but the claim naming this was found truncated in the file, no specific patent number was ever identified, and no lookup could be run — the patent connector requires a specific patent ID, and this environment's PATENTSVIEW_API_KEY is not configured [claim 13: unverified].

What would settle it: A configured API key and a real assignee search for AliveCor and adjacent ECG-hyperkalemia patent families, down to specific patent numbers and their claim scope.

If it's false: Automatic kill under the Stage 5 gate, regardless of every other score — the one condition on this deck with that property.


Slide 8 — Where this deck outruns the file

Every condition above with nothing verified behind it. This candidate carries one verified claim relevant to this lens [claim 26: verified], and it is tangential — a registry-metadata discrepancy in an unrelated product-code family, not a fact about this device's own mechanism, build spec, or pathway.


Slide 9 — The load-bearing condition

If only one thing from this chair could be checked: 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 [claim 37: unverified].

Every other feasibility question on this deck assumes the mechanism holds outside the exact conditions it was measured in. The regulatory build spec (Slide 3), the bench-evidence plan (Slide 5), and the manufacturing question (Slide 6) are all downstream of an algorithm that works on this device's actual signal — not the short, clinical-grade captures the only available evidence [claim 17: unconfirmed] was built and validated on. This is also the cheapest of this deck's open questions to test: a bench comparison, not an animal study or a regulatory submission, and it is the one result that would tell engineering whether anything else here is worth funding.

Naming it is not a recommendation, a gate, or a kill.