idea-017 · 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.
The chair: engineering. The people who would have to build a prescription bi-level ventilator with a back-up rate, a reusable heated transcutaneous CO2 sensor a layperson applies in a bedroom, and the software that turns an overnight trace into a settings proposal a physician signs — and then produce the evidence an FDA reviewer asks for.
How to read this: every slide is a condition, not a conclusion. A bracketed claim reference points into knowledge-base/candidates/idea-017.md with the tag it carries there; a no-claim marker means nothing in the file speaks to this. See docs/deck-spec.md.
First build. Four of this deck's conditions had nothing in the file behind them and are now write-backs (claims 37-40). Note the asymmetry this deck starts from: the two classification facts are verified, and nothing at all in the file is a measured engineering quantity.
Would have to be true: The device would have to deliver bi-level positive airway pressure with a fixed back-up rate at high-intensity settings, and concurrently record overnight transcutaneous CO2 from an ear or chest sensor, store both traces together, and compute a proposed settings change against a practitioner-set CO2 target — with no setting changing without practitioner approval.
Where it stands: The quantities are available for the ventilation half and absent for the recording half. The coverage-derived floor is settled, and the file's own figure for it was wrong: the claim that NCD 240.9 contemplates IPAP above 20 cm H2O is [claim 5: refuted] — the NCD's own definition reads "a minimum IPAP >= 15 cm H2O and backup respiratory rate of at least 14 breaths per minute", the back-up-rate half is verbatim correct, and the string "20 cm H2O" does not occur anywhere in the section. The design target should instead be read off trial practice, which is higher: HOT-HMV's median settings were IPAP 24 cm H2O (IQR 22-26), EPAP 4, back-up rate 14 [claim 15: verified]. So the pressure envelope the hardware must cover is roughly 15 to the high twenties, and the engineering consequence of the refutation is that the coverage bar is lower than assumed while the clinical bar is higher. For the recording half there is no specification of any kind in the file — no accuracy target, no drift budget, no recording duration, no sampling interval [no claim]. The intended-use statement is explicit that the titration is advisory: the device "presents that proposal to the treating practitioner for approval; no setting changes without practitioner approval", and that narrowing is a deliberate lever to stay on the 510(k) route rather than a detail.
What would settle it: For the ventilation envelope, the NCD text already read this pass plus the HOT-HMV methods for the full settings protocol. For the recording half, a requirements specification written against the FDA special-controls guidance for PcCO2 monitors — which no connector can reach and which a human must attach (see Slide 3).
If it's false: If the device cannot hold the upper end of the clinical pressure range, it delivers a coverage-compliant therapy that is not the therapy HOT-HMV tested, and the mechanism argument on Slide 2 no longer transfers.
Would have to be true: Titrating home NIV against an overnight transcutaneous CO2 trace would have to be the thing that produced HOT-HMV's admission-free-survival benefit — and the home-recorded tcCO2 signal would have to track arterial PaCO2 closely enough, across a whole night, to be a titration target rather than a trend indicator.
Where it stands: The outcome is verified and the mechanism underneath it is not. HOT-HMV extended median admission-free survival to 4.3 months versus 1.4 months for home oxygen alone, adjusted hazard ratio 0.49 (95% CI 0.31-0.77), with 12-month mortality not significantly different [claim 15: verified]. But that the trial titrated to overnight transcutaneous CO2 is not anchored, and the file is explicit that this is a real source silence rather than a tooling gap: the full structured abstract was retrieved and read, it gives design, population, interventions and median settings, and it says nothing about transcutaneous CO2 or about what the titration target was — the protocol lives in a methods section behind a paywall no connector here can reach [claim 16: unconfirmed]. The signal-fidelity half runs into a constraint from the regulation itself, which is the more useful finding: 21 CFR 868.2480 identifies a cutaneous CO2 monitor as intended "to monitor relative changes in a hemodynamically stable patient's cutaneous carbon dioxide tension as an adjunct to arterial carbon dioxide tension measurement" [claim 17: verified] — narrower than using it as the titration input, which the file records as a pre-submission question rather than a database fact. Whether the home-recorded signal is good enough for the intended role is now on record as this deck's write-back [claim 37: unverified].
What would settle it: Open the JAMA methods section for PMID 28528348 — free to a human, paywalled to every connector here, and named in the candidate's own Scores table as the item that is "free and fast". For the signal, published tcCO2-versus-arterial agreement data in adults during sleep, plus a bench and in-clinic agreement study on this candidate's own sensor.
If it's false: The device is a loop copied from a trial that used a different loop. The mechanism factor loses its only substantive claim, and the whole site-of-care argument — that this is the HOT-HMV titration performed at home — becomes a new clinical hypothesis requiring its own trial rather than a relocation of a proven one. This is this deck's load-bearing condition; see Slide 9.
Would have to be true: FDA would have to let the combined device onto a 510(k) — a ventilator claim inside the existing MNS space plus a recording claim inside the existing LKD space, with the novel function confined to software that recommends rather than acts — rather than reading the combination as a new device type requiring De Novo.
Where it stands: Both halves are established Class II categories and the combination has no predicate. Continuous ventilators are classified at 21 CFR 868.5895, Class II (performance standards), product code MNS "Ventilator, Continuous, Non-Life-Supporting", and the Philips BiPAP AVAPS is cleared under it as K102465, decided 2011-03-24 [claim 18: verified]. Cutaneous CO2 monitors are Class II (special controls) at 21 CFR 868.2480, product code LKD, review panel AN, with the special control being FDA's PcCO2/PcO2 special-controls guidance [claim 17: verified] — and the guidance's own text was never opened, because no connector reaches FDA guidance and fda.gov gates automated fetches, so what it requires is unread. The predicate problem is stated precisely in the file: K102465's Indications for Use cover noninvasive ventilatory support for OSA and respiratory insufficiency in hospital or home, and name neither CRF consequent to COPD nor any CO2-recording or titration-recommendation function — "so the combined-function half of this candidate has no predicate indication here" [claim 18: verified]. No cleared combined device was identified, and that is an absence-of-evidence result rather than white space [claim 19: unconfirmed]. The device_class: field carries De Novo under 21 CFR 860.200 as a live third route. Whether FDA accepts the 510(k) without a clinical study is this deck's write-back [claim 38: unverified].
What would settle it: An FDA pre-submission meeting on the combined-function question — a human act, and the only thing that settles it — plus a human attaching the PcCO2 special-controls guidance document so its testing requirements can be read into the build spec. Neither is reachable from this repo.
If it's false: The programme moves to De Novo with a clinical study, which lengthens the calendar past anything the capital-intensity factor currently assumes (see the viability deck's Slide 4). Note that the advisory-titration narrowing is the one design lever that keeps this route open: the moment the software adjusts pressure automatically, the file's own rationale says the device becomes a closed-loop physiologic control system with a life-relevant output, and route 3 is the expected destination.
Would have to be true: The reusable sensor head — a heated electrochemical element held against ear or chest skin — would have to hold calibration across an unattended eight-hour home recording, without a mid-session recalibration or membrane change, at a skin-heating temperature that causes no thermal injury to elderly COPD patients who cannot easily reposition themselves.
Where it stands: The physics is named in the regulation and the tolerances are named nowhere. 21 CFR 868.2480's identification describes the device as "a noninvasive heated sensor and a pH-sensitive glass electrode placed on a patient's skin" [claim 17: verified] — so heating and a glass electrode are inherent to the device type rather than a design choice, and both are drift and safety surfaces. Nothing in the file carries a calibration interval, a drift figure, a membrane-change schedule, a site-rotation requirement or a skin-temperature limit [no claim]; the proposition is now on record [claim 39: unverified]. The one commercial datapoint that bounds the problem is that the incumbents' cleared systems are attended or supervised: SenTec's cleared indication reads "in hospitals, hospital-type facilities, intra-hospital transport environments, and — if under clinical supervision — home environments" [claim 24: unconfirmed]. Unsupervised overnight home use is outside every cleared indication named in this file.
What would settle it: A bench drift study on the candidate's own sensor over an eight-to-ten-hour run against a reference gas, plus the skin-temperature and contact-time limits in the PcCO2 special-controls guidance and the applicable thermal-safety standard — the guidance needs a human to attach it (Slide 3), and the bench work is the candidate's own to do.
If it's false: The recording either needs a clinician present, which reinstates the facility event the whole economic argument exists to remove, or it needs the patient to intervene mid-night, which collides with the usability condition on the desirability deck's Slide 6. Either way the "home" in "home titration" stops being true.
Would have to be true: Before a patient is involved, a bench rig would have to show that the titration-recommendation software never proposes a settings change from a bad trace — that sensor detachment, poor skin contact and artefact are detected and surfaced as a rejection rather than silently averaged into a number a physician signs.
Where it stands: Nothing in the file speaks to the software at all — no algorithm description, no failure-mode analysis, no software safety level, no alarm behaviour [no claim]. The proposition is now on record [claim 40: unverified]. What sharpens the requirement is the approval architecture the intended use depends on: the practitioner "sets and approves every ventilation setting", which makes the physician the mitigating control for every software error — and a mitigating control that is handed a plausible-looking recommendation computed from a detached sensor is not a control. The adjacent hardware condition is [claim 39: unverified] (Slide 4), and the signal-validity condition is [claim 37: unverified] (Slide 2); this slide is about what the software does when those two fail, which is the case they will actually be in.
What would settle it: A bench protocol run against recorded traces with injected fault modes — detachment, drift, contact loss, gaps — measuring the fraction the algorithm rejects and the fraction on which it computes a proposal anyway. This is the candidate's own bench work and is the first entry in a minimum viable bench-evidence list; nothing in this repo greenlights anything downstream of it.
If it's false: A physician approves a pressure change derived from an artefact, in a population where the therapy is delivered unattended at IPAP in the twenties overnight. That is the clinical-risk failure mode of this device, and it would be the substance of the Mechanism & clinical risk factor rather than the trial result currently standing in for it.
Would have to be true: The product would have to be manufacturable as a durable rental asset that survives repeated redeployment between patients, with a consumable sensor element supplied reliably enough that a DME supplier can keep a patient recording — and it would have to be set up and serviced in a patient's home by a supplier's respiratory therapist, not by the manufacturer.
Where it stands: The materials risk in this exact device class is documented and verified, which is unusual and worth taking seriously: Philips issued a voluntary recall notification on 2021-06-14 covering ventilators, CPAP and bi-level devices over polyester-based polyurethane sound-abatement foam that "may degrade into particles which may enter the device's air pathway and be ingested", confirmed across six openFDA recall records spanning exactly those three categories, all still "Open, Classified" [claim 22: verified] — so sound-abatement material selection and its degradation pathway is a named, precedented design constraint for anyone building a home bi-level device, not a hypothetical. The channel reality is an unresearched assumption: Medicare RADs are dispensed and billed by DME suppliers, and whether suppliers will stock, set up and service a device carrying an extra consumable sensor inside the rental allowable has had no supplier economics researched behind it [claim 28: unconfirmed]. The home-use human factors — whether a patient or caregiver can place the sensor unaided — is stated on the desirability deck's Slide 6 and referenced rather than restated here [claim 31: unverified]. Nothing in the file carries a sterilisation requirement, a shelf life for the consumable, or a supply-chain source for the sensor element [no claim].
What would settle it: A human-factors validation study in real homes with the target population, which a 510(k) for a home-use device would require in any case; plus a materials and biocompatibility plan for the air pathway and the skin-contact surfaces, written against the special-controls guidance and the Philips recall as a known failure precedent.
If it's false: If suppliers will not service the sensor, the recording channel stops being used within weeks of setup regardless of how well it works — the same failure the viability deck's Slide 7 reaches from the money side.
Would have to be true: The design would have to be practisable without a licence — in particular the volume-assured/target-ventilation control approach and the sensor head's construction, heating and calibration scheme would have to be either free to practise or designable around without giving up the function.
Where it stands: Unexamined, not clear — and structurally unexaminable from this repo. No patent search was performed, no patent number, family or assignee list is named anywhere in this candidate, and the file states in terms that the FTO field is unexamined rather than clear [claim 25: unconfirmed]. That claim names the two risk surfaces precisely, and they are exactly the two things this deck's Slides 1 and 4 depend on: (a) volume-assured / target-ventilation control algorithms of the kind marketed as iVAPS (ResMed) and AVAPS (Philips Respironics), and (b) transcutaneous CO2 sensor head construction, heating and calibration. The block is structural: patent takes a specific patent number and there is no patent search connector in this repo, so an assignee-portfolio or subject-matter search cannot be run here at all, with or without an API key, and inventing a plausible-looking number to have something to query would be fabrication. The expectation that the ~1990 foundational bi-level-with-back-up-rate patents have expired is inference from the modality's age; the file notes that even a favourable lookup would be a database fact about dates and never a legal FTO conclusion, since term extensions, continuations and later improvement patents are outside what the connector sees [claim 26: unconfirmed]. Whether the integration itself is patentable — the offensive side of the same question — is stated on the viability deck's Slide 5 [claim 35: unverified].
What would settle it: A PatentsView or EPO OPS assignee-portfolio and subject-matter search against ResMed, Philips Respironics, Sentec AG, Radiometer Medical and Perimed AB, followed by a draft FTO memo for legal. The file already hands a searcher the entry points: K161492 (ResMed platform, iVAPS named in the IFU), K102465 (Philips), and product code LKD for the sensor surface.
If it's false: FTO blocked is the rubric's only automatic kill, applying regardless of every other score — and it is currently the factor with the least evidence behind it anywhere in this file. If the block lands on surface (a), the workaround is a compromise of the therapy; if it lands on surface (b), the sensor is bought rather than built, which changes the cost structure the viability deck's Slide 7 depends on.
Every condition above with nothing verified behind it — the no-claim markers, plus the unverified claims this deck itself put on record. Read this slide first.
The pattern is worth naming: the three verified claims this deck leans on (claims 17, 18 and
wrong for somebody else. Not one measured engineering quantity about this candidate exists anywhere in the file, which is what a feasibility deck at stage: verified looks like when the verification work so far has been regulatory rather than physical.
If only one thing from this chair could be checked: whether the noninvasive ventilation in HOT-HMV was in fact titrated to overnight transcutaneous CO2 [claim 16: unconfirmed].
This one rather than the others because it is the premise the device is built on and the file says so in its own words — the whole product is that trial's titration loop moved from a laboratory into a bedroom. If the loop being copied is not the loop that produced the 4.3-versus-1.4-month result, then every engineering condition below it is work toward an unproven clinical hypothesis: the sensor accuracy target on Slide 2 has no benchmark, the bench evidence on Slide 5 has no endpoint to validate against, and the regulatory argument on Slide 3 loses the analog it leans on. It is also, of every unknown on this deck, the only one that is free and fast: the answer is in a methods section a human can open today, and it is blocked here solely because JAMA full text is paywalled to every connector in this repo.
Naming it is not a recommendation, a gate, or a kill.