Onboard Placement Confirmation for Operator-Independent Emergency Intubation

1. Clinical Need

Emergency tracheal intubation fails on the first attempt in roughly a third of prehospital cases, and every repeat attempt is associated with increased morbidity. Prehospital first-pass failure is “nearly 35%” (Haggerty et al., Science Translational Medicine, 2025), against a pooled prehospital first-pass success range of 59% to 98% (Hayes-Bradley et al., British Journal of Anaesthesia, 2024). That 39-point spread is not explained by equipment. It is explained by who is holding it.

The Dutch prehospital record shows why. Across 3,632 intubations, switching from direct to video laryngoscopy moved ambulance nurses from 45.5% to 64.8% first-pass and HEMS physicians from 85.9% to 86.9% (Maissan et al., European Journal of Trauma and Emergency Surgery, 2022). Instrumentation substitutes for experience only where experience is absent, which is the prehospital case.

The second and more dangerous failure is unrecognised oesophageal intubation, and it is a separate problem from placement. In that same series, upgrading the laryngoscope did not reliably resolve confirmation: it coincided with a higher rate in the traumatic arrest subgroup and a lower one in the medical arrest subgroup. Both are subgroup rates whose denominators the paper does not publish, and the paper's own conclusion attributes the traumatic-arrest reversal specifically to tubes placed by ambulance nurses before HEMS arrival, the least experienced operator class and also the class an operator-independent device targets (section 3).

The confirmation standard itself carries a 3% false-positive rate, meaning the tube is in the oesophagus and capnography reports the trachea, measured across 2,192 emergency intubations, in a registry series where 4% of attempts ended in accidental oesophageal placement and 10% occurred in nontraumatic cardiac arrest (Li, Journal of Emergency Medicine, 2001; full figures and scope qualifiers in section 3). Current American Heart Association guidance recommends continuous waveform capnography to confirm and monitor tube placement, including during CPR, so the residual problem is physiological rather than procedural: in low-flow states exhaled carbon dioxide can be low or absent, so the modality guidance mandates is least informative exactly where the event rate is highest. The unmet need is not a better way to place the tube. It is a way for an operator who cannot interpret a waveform to know, in that population, whether the tube is in the airway.

2. State of the Art

Three research programs define the frontier, and they do not overlap.

Soft robotic self-guided placement (Elliot Hawkes, Mechanical Engineering, UC Santa Barbara, with David Haggerty; clinical co-investigators C. J. Winckler and D. A. Wampler at UT Health San Antonio, D. R. Drover at Stanford Anesthesiology, A. Mahajan and J. P. Zevallos at UPMC). The Hawkes group works on everting “vine” robots, which extend by pressurised eversion of a thin-walled tube so the tip advances without the body sliding against surrounding tissue. In the airway this removes the need for a glottic view: the device follows the anatomical path rather than being aimed along a line of sight. The methodology is deliberately non-computational. There is no sensing and no control loop; the morphology of the everting tube plus a shaped introducer does the work perception would otherwise do. That choice is what makes the device cheap, sterilisable and shelf-stable, and it is the origin of the gap this brief addresses.

Teleoperated robotic intubation (Thomas Hemmerling, Anesthesia, McGill University). The Kepler Intubation System put a video laryngoscope on a Kinova JACO arm under joystick control and produced the only in-human robotic intubation series in the literature. It is fourteen years old and unsuperseded in humans, and its methodology, a human driving a conventional laryngoscope remotely, is what the soft robotic work bypasses.

Confirmation as a diagnostic problem (Hao-Chang Chou and Wan-Ching Lien, Emergency Medicine, National Taiwan University Hospital; and the emergency airway registry tradition represented by Li at Harvard Medical School). The tracheal rapid ultrasound exam reads the tracheal air column directly through a transducer over the suprasternal notch, at 98.2% accuracy against capnography. Its limitation is the operator: a trained sonographer and a free hand, neither of which a single responder running a resuscitation has.

Operator-independent placement now works. Operator-independent confirmation does not exist. The gap is visible in the 2025 study's own protocol: success was “verified by passing the Ambu aScope bronchoscope through the core of the ETT to provide a confirming view of tracheal rings,” an instrument held by investigators, not by the user, and not carried on an ambulance.

3. Foundational Research

Haggerty DA, Cazzoli JR, Wayne MA, Winckler CJ, Wampler DA, Jarvis JL, et al., Hawkes EW (2025). “A soft robotic device for rapid and self-guided intubation.” Science Translational Medicine 17(815):eads7681. DOI 10.1126/scitranslmed.ads7681. PMID 40929248. A tip-everting soft overtube, 12 mm in diameter from 0.005 inch TPU film with a polycarbonate insert overmolded in lubricious silicone, everts along the airway and carries a soft endotracheal tube to the trachea without a glottic view. Evaluation ran in two stages: one expert user across 50 sequential intubations in a Laerdal Airway Management Trainer and 10 tests in a single fresh-frozen cadaver bust (cephala and torso, dissected at the shoulders and nipple line, UC Irvine Willed Body Program); then eight EMS providers, four paramedics and four EMTs, using the device and a video laryngoscope on eight unembalmed adult cadavers after a 3-minute device video, a 5-minute video-laryngoscopy video and one practice attempt with each on a training cadaver. Expert results were 100% success at mean duration under 8 seconds. In the provider study (Table 2, all users, all airways), first-pass success was 86.9% for the device against 62.5% for video laryngoscopy, odds ratio 4.0, 95% CI 0.92 to 17, P = 0.055, which is not statistically significant; overall success 95.7% against 91.7% (P = 0.576); mean attempts 1.1 against 1.6 (P = 0.062); and overall duration 20.8 seconds against 44.4 seconds (P = 0.008), the only significant whole-cohort result. In the difficult-airway subgroup, 14 trials per arm, first-pass success was 92.9% against 35.7%, odds ratio 23.4, 95% CI 2.23 to 236, P = 0.002, with overall duration 23.7 against 64.4 seconds (P = 0.012) and attempts 1.1 against 1.8 (P = 0.008). The authors describe the study as preliminary and state that no power analysis was used in sample size determination. We take this device class as the given substrate and state its evidence precisely, because the difficult-airway subgroup, not the all-airways first-pass number, is where the effect is established. Two facts are load-bearing for our program: the device is explicitly non-electronic, and success was adjudicated by an investigator passing a bronchoscope through the tube.

Li J (2001). “Capnography alone is imperfect for endotracheal tube placement confirmation during emergency intubation.” Journal of Emergency Medicine 20(3):223-229. DOI 10.1016/s0736-4679(00)00318-8. PMID 11267809. A meta-analysis of experimental capnography trials in emergency populations, paired with an analysis of consecutive National Emergency Airway Registry intubations. Across 2,192 intubations, capnography showed sensitivity 93% (95% CI 92 to 94%) and specificity 97% (95% CI 93 to 99%), a 7% false-negative and 3% false-positive rate, numbers needed to harm 14, 33 and 10. Of 4,602 NEAR intubations, 4% were accidental oesophageal placements and 10% occurred in nontraumatic cardiac arrest, a setting in which capnography was not recommended at the time of writing. Two qualifiers travel with this paper and we carry them. Its accuracy figures treat tracheal placement as the positive class, the opposite convention to the operating point we specify in section 6, so the two are not directly comparable. And its cardiac-arrest statement reflects 2001 practice: current American Heart Association guidance recommends continuous waveform capnography for confirmation including during CPR. What survives is physiological. Low pulmonary blood flow suppresses exhaled carbon dioxide, so the mandated modality is least informative where the event rate is highest, which is the design premise for perfusion-independent channels.

Maissan I, van Lieshout E, de Jong T, van Vledder M, Houmes RJ, den Hartog D, Stolker RJ (2022). “The impact of video laryngoscopy on the first-pass success rate of prehospital endotracheal intubation in The Netherlands: a retrospective observational study.” European Journal of Trauma and Emergency Surgery 48(5):4205-4213. DOI 10.1007/s00068-022-01962-7. PMID 35362731. A retrospective analysis of 3,632 prehospital intubations stratified by provider and laryngoscope type, with unrecognised oesophageal intubation recorded on HEMS arrival. First-pass success was 45.5% (391 of 859) and 64.8% (125 of 193) for ambulance nurses with direct and video laryngoscopy, 57.6% (34 of 59) and 77.2% (125 of 162) for HEMS nurses, 85.9% (790 of 920) and 86.9% (1,251 of 1,439) for HEMS physicians. Unrecognised oesophageal intubation was 30.6% versus 37.5% in the traumatic circulatory arrest subgroup and 20% versus 0% in the medical cardiac arrest subgroup. These are subgroup rates, not rates over the 3,632-intubation series, and the abstract does not publish the numerator and denominator for each cell; the 20% and 0% figures in particular imply small counts. The paper's conclusion further restricts the traumatic-arrest reversal to tubes placed by ambulance nurses before HEMS arrival. This makes our central claim testable while bounding it: within one dataset, for the least experienced operator class, an instrument change improved placement and did not reliably improve confirmation. The reversal between subgroups is a real limit on interpretation and is why our validation stratifies by arrest type.

Prekker ME, Driver BE, Trent SA, et al.; DEVICE Investigators (2023). “Video versus Direct Laryngoscopy for Tracheal Intubation of Critically Ill Adults.” New England Journal of Medicine 389(5):418-429. DOI 10.1056/NEJMoa2301601. PMID 37326325. A multicenter randomised trial across 17 emergency departments and intensive care units. Among 1,417 patients, 91.5% intubated by an emergency medicine resident or critical care fellow, first-attempt success was 85.1% (600 of 705) versus 70.8% (504 of 712), absolute risk difference 14.3 percentage points (95% CI 9.9 to 18.7, P < 0.001), stopped early for efficacy. Severe complications were 21.4% versus 20.9% (absolute risk difference 0.5 percentage points, 95% CI −3.9 to 4.9), and safety outcomes including oesophageal intubation were similar. Funded by the U.S. Department of Defense (NCT05239195). We adopt the 70.8% direct-laryngoscopy arm, not the 85.1% video arm, as our animal-work go/no-go floor: it is the honest comparator for operators with less training than a critical care fellow.

Chou HC, Tseng WP, Wang CH, Ma MH, Wang HP, Huang PC, et al. (2011). “Tracheal rapid ultrasound exam (T.R.U.E.) for confirming endotracheal tube placement during emergency intubation.” Resuscitation 82(10):1279-1284. DOI 10.1016/j.resuscitation.2011.05.016. PMID 21684668. A prospective observational study during emergency intubation, with a transducer placed transversely over the suprasternal notch and quantitative waveform capnography as criterion standard. Among 112 patients, 17 (15.2%) with oesophageal intubations, overall accuracy was 98.2% (95% CI 93.7 to 99.5), kappa 0.93, sensitivity 98.9%, specificity 94.1%, median operating time 9.0 seconds (IQR 6.0 to 14.0). This establishes both that a non-capnographic signal can adjudicate placement at high accuracy in under ten seconds, and the constraint we must remove: it requires a trained sonographer holding a probe.

Hayes-Bradley C, McCreery M, Delorenzo A, Bendall J, Lewis A, Bowles KA (2024). “Predictive and protective factors for failing first pass intubation in prehospital rapid sequence intubation: an aetiology and risk systematic review with meta-analysis.” British Journal of Anaesthesia 132(5):918-935. DOI 10.1016/j.bja.2024.02.004. PMID 38508943. A PRISMA systematic review with meta-analysis searching MEDLINE, EMBASE, CINAHL and the Cochrane Library to 3 March 2023 (PROSPERO CRD42022353609), covering first-pass success in prehospital rapid sequence intubation on live patients. First-pass success ranged 59% to 98%. Failure was predicted by age under one year, blood or fluid in the airway, restricted jaw or neck movement, trauma, nighttime procedures, distorted face and upper-airway anatomy, and equipment issues; protective factors were an experienced intubator, adequate training, certain video laryngoscopes, inclined stretcher positioning, a bougie and laryngeal manoeuvres. The authors note heterogeneity limits stronger conclusions. This paper defines the subgroups our validation matrix must cover, and every one of them is absent from the cadaver evidence above.

Hemmerling TM, Taddei R, Wehbe M, Zaouter C, Cyr S, Morse J (2012). “First robotic tracheal intubations in humans using the Kepler intubation system.” British Journal of Anaesthesia 108(6):1011-1016. DOI 10.1093/bja/aes034. PMID 22457483. A pilot study in 12 consented surgical patients of a joystick-teleoperated Kinova JACO arm driving a Pentax AWS video laryngoscope. Success in 11 of 12 (91%) at median total time 93 seconds (IQR 87 to 109, range 76 to 153), the failure caused by laryngoscope fogging. Still the only in-human robotic intubation dataset. It quantifies the cost of teleoperation, 93 seconds against 21 for the passive device in cadavers, and shows the stall has been in sensing and control rather than mechanics.

4. Competitive Landscape

One company was identified in the exact space. The search covered company sites, the FDA 510(k) and classification databases, NIH RePORTER and the published literature, and did not cover SBIR.gov, DoD CDMRP or USAMRDC records, BARDA, or private-market databases, so this is a bounded count rather than a certainty. Vine Medical, Inc. (Santa Barbara, California) is the UC Santa Barbara spinout founded by the first author of the 2025 paper and holds US Patent 12,280,208, “Vine Robot Tracheal Intubation Device.” Two of the paper's authors are employed by Vine Medical with an ownership stake and four more hold an ownership stake, per the competing-interests statement; the work was supported in part by the National Science Foundation. No FDA clearance exists and no private funding round could be verified, but the resource asymmetry should not be understated: the paper's funding statement records National Science Foundation award 1944816, National Science Foundation SBIR award 2305627, a David and Lucile Packard Foundation award (2020-71-383), and U.S. Army Medical Research Acquisition Activity award HT9425-23-1-0872. The group already holds both an NSF SBIR and an Army medical research award. We state this plainly because it bounds what we propose: the guidance mechanism is patented and being commercialised by its inventors with federal support, so a program proposing to rebuild it would be redundant and encumbered.

Three programs are adjacent, each attacking one failure factor rather than operator dependence. IVOS Medical is developing a video laryngoscope sleeve with integrated continuous suction for the bloody-airway case, on $2.14 million of NIH NHLBI SBIR funding. SmartAirway LLC is developing an articulating stylet on one NHLBI SBIR Phase I award. Spiro Robotics has clinically studied a joystick-driven robotic flexible intubation scope and states it is not cleared for commercial use. None is autonomous and none addresses confirmation.

Commercial activity does not solve the research problem. Every effort above improves the instrument in the operator's hands. None produces a living-airway dataset for operator-independent placement, and none produces an adjudication method that works without a trained interpreter. The research questions are open regardless of what the companies ship.

5. Addressable Scope and Public Benefit

Bottom-up. The National EMS Information System recorded 76,123 patients intubated out of hospital in 2021, 68,027 in 2020 and 83,022 in 2019 (Huebinger et al., Annals of Emergency Medicine, 2023;82(6):763-765). Two limits travel with this number: NEMSIS is a voluntary registry that does not include every EMS agency, so the counts are a floor rather than a census, and the research letter's full text is paywalled, so the per-year counts are as reported for that letter rather than read from its own tables. Failed attempts consume devices, and the right multiplier is this product's own measured attempt ratio of 1.1 per success rather than video laryngoscopy's 1.6, because units of this product are consumed at this product's rate; that gives roughly 84,000 single-use units per year on the registry floor. (The 1.1 against 1.6 difference is itself P = 0.062 and the study capped attempts at three, so 1.1 is a point estimate rather than a settled consumption rate.) Unit price is a stated assumption, not a sourced figure: single-use airway adjuncts and disposable video-laryngoscope blades sell in the tens to low hundreds of dollars, so the band used is $150 to $400, giving a prehospital-only revenue range of $12.6 million to $33.6 million per year. This is a 100%-share ceiling, not a forecast: it assumes every prehospital intubation in the registry uses one of these devices and that no competing product takes share.

This is below $100 million and we flag it. A prehospital-only disposable is not a venture-scale business at registry-floor volumes, across the whole price band. For a funder the consequence is the opposite of discouraging: it is the structural reason the private market has not produced the confirmation layer, because the segment that needs it most cannot pay for it alone. Public funding is the mechanism that fits, and the return is measured against the 4% oesophageal intubation rate in the National Emergency Airway Registry rather than in device revenue.

Serviceable scope. At TRL 4 to 5 with no living-airway data, the realistic first segment is ALS-capable ground EMS agencies in structured evaluation plus emergency departments benchmarking against DEVICE. At the $250 midpoint against the 84,000-unit floor that is about $21 million per year, expanding with in-hospital volume this analysis does not separately source and therefore does not claim.

Top-down cross-check (publisher summary figures; the report page returns 403 to direct retrieval). Grand View Research, Airway Management Devices Market Size, Share and Trends Analysis Report, 2026-2033, puts the global market at USD 2.64 billion in 2025, reaching USD 4.26 billion by 2033 at a 6.10% CAGR over 2026 to 2033, with North America at 38.54% revenue share in 2025 and infraglottic devices the largest segment at 33.0%. North American infraglottic devices therefore run near USD 335 million, so the bottom-up prehospital figure is 3.8% to 10.0% of that segment. One assumption is load-bearing there: it multiplies a regional revenue share by a global segment share and reads the product as a regional segment size, which holds only if the North American airway-device mix matches the global mix. Subject to that, the two methods agree on order of magnitude.

Reimbursement. Clinician work is already captured under CPT 31500, “Intubation, endotracheal, emergency procedure,” which per CPT and NCCI guidance covers emergency and not elective intubation, so no new procedure code is needed. Prehospital use is bundled into advanced life support emergency ground transport (HCPCS A0427 and A0433 families) and absorbed by the agency; in-hospital use is absorbed into the facility payment. Translation runs through agency and hospital procurement on a demonstrated-safety argument, which is what a funded preclinical program produces.

6. Research Gap and HHA Contribution

What has not been done, precisely. No self-guiding airway device carries any sensing; the 2025 device is explicitly non-electronic. No operator-independent placement has been shown in a living airway; every result is mannequin or cadaver. No automated adjudication exists that does not require a trained interpreter of a capnography waveform or an ultrasound image. And the subgroups that drive prehospital failure, infants, bloody airways, restricted neck movement and distorted anatomy, appear in none of the device literature.

The technical gap. Between a device that places a tube in a cadaver at 87% first-pass and a deployable system lies one missing signal path: the device must tell its user, with no trained interpretation step, whether the tube is in the trachea, in a patient whose pulmonary blood flow may be near zero and in whom the incumbent modality is least reliable. Chou et al. show a non-capnographic physical signal carries enough information to adjudicate at 98.2% accuracy in under ten seconds; Li et al. show the incumbent is 3% wrong in the lethal direction and not recommended in 10% of the population. Nothing closes the loop between those two facts and an unattended device.

Proposed methodology. An instrumented introducer collar integrated into the everting overtube's existing polycarbonate insert, plus a locked classifier that emits a placement verdict.

• Sensing and state space. Four synchronised channels on a 10-second rolling window: mainstream capnometry at 100 Hz; airway pressure at 100 Hz; four-electrode bioimpedance, 50 kHz carrier sampled at 200 Hz, which separates the air-filled trachea from the collapsed, mucosa-apposed oesophagus independently of gas exchange; and acoustic reflectometry, a 200 Hz to 8 kHz chirp sampled at 44.1 kHz, returning cross-sectional area against distance and so separating a cartilage-supported lumen from a compliant one. A three-axis IMU at 100 Hz gives motion context for artifact rejection. The bioimpedance and acoustic channels do not depend on pulmonary blood flow, which is why they are the centre of the design rather than an addition to capnography.
• Algorithm. A 1D temporal convolutional network with dilated causal convolutions and gated attention pooling over the four channels, trained with focal loss for the class imbalance inherent to a 4% event rate, wrapped in split conformal prediction to give a calibrated abstain region rather than an uncalibrated softmax score. It is benchmarked head to head against two baselines it must beat before adoption: gradient-boosted trees on handcrafted waveform features, and a fixed capnography threshold rule representing current practice. A learned model that cannot beat XGBoost on handcrafted features does not belong in a safety-critical device.
• Action space and safety bounds. The device actuates nothing. Its output is the three-valued verdict {tracheal, oesophageal, indeterminate}, as an indicator state and an audible alert. The binding safety rule: indeterminate and oesophageal drive the same user action, withdraw and reattempt, so the device can never instruct a user to retain a tube it failed to confirm. Alert latency budget 10 seconds from tube seating, set against the 9.0 second median of the ultrasound comparator.
• Objective and measurement. Maximise sensitivity for oesophageal placement at a fixed false-positive rate of no more than 1% on tracheal placements, measured against direct endoscopic visualisation as reference standard on bench and in animal. Overall accuracy is explicitly not the objective: it is dominated by the majority class and hides the error that kills.
• Pre-training environment and validation model. A hardware-in-the-loop bench pairing an instrumented airway trainer with a driven lung simulator, across a matrix including a low-pulmonary-flow condition reproducing cardiac arrest, instilled blood, simulated secretions and three tube sizes. Validation in swine (Sus scrofa), chosen because porcine tracheal and oesophageal dimensions and mucosal properties approximate the adult human airway and it is the established model for airway device evaluation; deliberate oesophageal placements supply the positive class.

Why HHA is positioned for this. The gap spans three disciplines. Hass Dhia owns the clinical reference standard, the airway anatomy and acute physiology behind the sensing choice, and the stratified validation design. Haedar Hadi owns the temporal convolutional architecture, the benchmark against classical baselines, and the evaluation methodology that keeps the operating point honest. Ahmed owns design for manufacturability from month one rather than at the end. Section 9 maps each to deliverables.

Why the originating groups have not closed it. Three reasons, none a criticism. The Hawkes group is a mechanical engineering laboratory whose contribution is precisely that the device needs no electronics; a sensing and classification stack is a different discipline and would dilute the result that makes the device manufacturable. Vine Medical has a commercialisation mandate and a capital constraint, and its fastest route to revenue is clearing a mechanical device against existing airway predicates, whereas an automated diagnostic claim adds years. Chou and colleagues work in an emergency department where a trained sonographer is available by definition, so the operator-independence constraint does not bind their work. Each behaves rationally within its own incentives, and the gap sits between all three.

7. Comparable Funded Projects

Federal money is already committed to this clinical failure, and all of it funds a better tool for a trained operator. All figures were read from the NIH RePORTER v2 API.

Gabriel Punsalan, IVOS Medical, Inc., NIH NHLBI SBIR. 1R43HL164325-01 ($247,334, FY2022), 2R44HL164325-02 ($824,627, FY2024), 5R44HL164325-03 ($926,634, FY2025) and 5R44HL164325-04 ($144,919, FY2026), totalling $2,143,514 for a video laryngoscope sleeve with integrated continuous suction. The closest comparable: a small company, an emergency airway device, and a Phase I to Phase II progression at NHLBI. It targets the bloody-airway failure predictor from Hayes-Bradley, one of the subgroups in our validation matrix, so the programs are complementary rather than competing.

Rick Rutgers, SmartAirway LLC, NIH NHLBI SBIR Phase I. 1R43HL174223-01A1, “A Novel Articulating Stylet for Improving First-Pass Intubation Success,” $313,365, FY2025. A 2025 Phase I award on first-pass success establishes both the current Phase I award size at this institute and that NHLBI is accepting new entrants on this exact outcome measure.

Ari Moskowitz, Albert Einstein College of Medicine, NIH NHLBI. 1R61HL162980-01, “Hospital Airway Resuscitation Trial,” $788,728, FY2022. A phased-innovation award on airway management during resuscitation, the clinical context in which our low-pulmonary-flow condition sits, showing NHLBI funds airway work at strategy level as well as device level.

Robyn Wing, Rhode Island Hospital, NIH NIGMS. P20GM139664, a feasibility trial of an implementation bundle for the National Emergency Airway Registry for Pediatric Emergency Medicine pre-intubation checklist, $261,688 in FY2024 with continuations in FY2025 and FY2026. Paediatric airways are the highest-risk subgroup in Hayes-Bradley, and the registry our outcome measures would report into is itself federally supported. The DEVICE trial was funded by the U.S. Department of Defense (NCT05239195).

The instructive result is the absence. A RePORTER search for awards whose title or abstract contains both “intubation” and “robotic” across FY2022 to FY2026 returns exactly one award, a robotic enteroscope for small-bowel access (1R43DK143858-01A1, Drive Medical, NIDDK, $315,747, FY2025), not an airway device. That absence carries a matched positive control: the identical advanced_text_search surface over the same fiscal years returns 223 awards for “intubation” alone, including both IVOS Medical and SmartAirway, and adding “robotic” collapses it to that single enteroscope. NIH RePORTER records no robotic or autonomous intubation project in five fiscal years while recording $3,507,295 across the four awards itemised above (the Wing figure is its FY2024 amount only) on the same clinical failure. The claim is scoped to NIH RePORTER and does not cover DoD or other agency records. Our proposal extends work the agency already funds into the half of the problem it has funded no one to solve.

8. Opportunity Assessment

TRL 3, composite. A system takes the TRL of its least mature required component, not an average, and the components here are far apart. Operator-independent mechanical placement is TRL 5: a working prototype validated in a relevant environment by the intended end-user population, on the section 3 evidence. The integrated confirmation module is TRL 3: the constituent modalities are mature as standalone monitoring, but no integration into a self-guiding device has been built or published, so the system as a whole is a proof-of-concept. Advancing requires a living airway and a confirmation channel that is not a surgically cut tracheal port.

The research questions, stated as questions.

• Can bioimpedance and acoustic reflectometry discriminate tracheal from oesophageal placement at a false-positive rate at or below 1% when pulmonary blood flow is near zero? This is the question the incumbent modality cannot answer, and it is the pass or fail of the program. It is answered on the bench, in the low-flow arm, before any animal work.
• Does a passive everting device retain its cadaver first-pass performance in a living airway with muscle tone, laryngospasm and active secretions? Answered in swine against a prespecified floor of the 70.8% direct-laryngoscopy arm of DEVICE.
• Does a learned temporal model beat gradient-boosted trees on handcrafted features by enough to justify the regulatory cost of a learned model? If not, the correct answer is to ship the simpler model, and the benchmark is designed so that outcome is visible rather than buried.

Risks and mitigations. Technical: the four channels may be individually insufficient in the low-flow state and fusion may not recover the loss, mitigated by making the bench low-flow arm a hard gate at month six, before animal spend. Translational: unit cost may exceed the EMS consumable price point, which at a sub-$100 million market would make the segment unreachable, mitigated by design for manufacturability from the first prototype. Competitive: Vine Medical may integrate confirmation itself, mitigated by a module that is device-agnostic and applies equally to bougies, supraglottic airways and conventional intubation.

Regulatory pathway. The two halves separate, and the separation is favourable. Verified against the FDA classification database, the mechanical device sits near existing airway classes (rigid laryngoscopes CCW, Class I, 21 CFR 868.5540; tracheal tube stylets BSR, Class I, 21 CFR 868.5790; tracheal tubes BTR, Class II, 21 CFR 868.5730), with direct precedent for a blind intubation guide in the Augustine Intubation Guide (K882659, 1988; K954446, 1995) and an active class including the Prodol Meditec Airtraq (K121378, 2012), the Venner APA Oxy Blade (K191602, 2019) and Cook's Aintree Intubation Catheter and Arndt Airway Exchange Catheter (K162729, 2017). The confirmation module, which this program targets, has the capnograph as its nearest product code rather than a predicate (CCK, Class II, 21 CFR 868.1400): outputting a placement verdict rather than a waveform is a different intended use with different technological characteristics, so substantial equivalence is unlikely to hold and the route is De Novo. The codes above are cited as the nearest existing classification, not as a 510(k) path. The locked-versus-adaptive decision determines the timeline and we make it explicitly: locked. The classifier is frozen at submission, with a Predetermined Change Control Plan filed alongside under FDA's marketing-submission framework for AI-enabled device software functions, specifying the anticipated modifications, the methods to implement and validate them, and the impact assessment, so the adaptive path is pre-authorised rather than needing a fresh submission. That De Novo plus PCCP also creates a two to three year barrier to entry in a category whose mechanical device is publicly described and easy to copy, making the regulatory work a moat as well as a cost. Human-subjects work requires IRB approval and, for a first-in-human protocol in a population that cannot consent, an exception from informed consent under 21 CFR 50.24 with community consultation, scheduled in parallel with the animal work.

First six months. Build the collar and the bench, collect the four-channel corpus across the perturbation matrix including the low-flow arm, train and benchmark the three models, and gate on the first research question before committing animal spend.

9. Team Fit

Hass Dhia, Principal Investigator. MS Biomedical Sciences with a medical school background including service as an anatomy teaching assistant, and an AI infrastructure architect. Maps to: clinical problem identification and the framing of confirmation as a separable failure mode; the anatomical and physiological basis of the sensing design, specifically why a cartilage-supported tracheal lumen and a collapsed mucosa-apposed oesophagus are separable by impedance and acoustic cross-section when they are not separable by gas exchange; the choice of endoscopic visualisation as reference standard; the perturbation matrix and stratification by arrest type; and the overall AI system architecture.

Haedar Hadi, Co-Investigator. MS Computer Science, Boston University, Information Systems focus, with a background in cloud and database architecture. Maps to: the temporal convolutional model and its conformal calibration layer; the evaluation methodology that fixes the operating point at a 1% false-positive rate rather than reporting overall accuracy; the benchmark design putting the learned model head to head against gradient-boosted trees and a threshold rule; and the compute and data infrastructure for a multi-channel waveform corpus with the provenance tracking a regulatory submission requires.

Ahmed, Director of Manufacturing, Key Team Member. Maps to: design for manufacturability of the instrumented collar within the existing overmolded introducer geometry; tolerance analysis across the electrode and transducer placements the classifier's calibration depends on; sterilisation compatibility and shelf stability under ambulance temperature cycling; quality systems and process validation against a single-use consumable cost target.

The lab-to-production bridge. Most research proposals end at “it works in the lab.” This proposal includes explicit DFM milestones at every phase, ensuring that prototype decisions consider production scaling, tolerance analysis and quality systems from day one. This addresses the valley of death between TRL 4-5 prototypes and TRL 7+ deployable systems, the gap where most funded research stalls. It matters more than usual here: at a prehospital market below $100 million, a device that cannot be built at an agency consumable price point has no deployment path regardless of how well it classifies, and sensor placement tolerance is not a packaging detail but an input to the classifier's calibration.

What the team does not have, and how the award closes it. HHA operates no animal facility and employs no veterinary surgical staff, so the swine work is budgeted as a subcontract to an AAALAC-accredited large-animal facility, HHA retaining protocol design and data analysis. HHA employs no regulatory affairs staff, so the pre-submission, De Novo strategy and PCCP drafting are budgeted as a consultancy engagement. Neither gap is on the critical path for the month-six bench gate, which is deliberate: the first tranche buys the answer that decides whether the rest is worth funding. HHA is not proposing to out-build a funded startup on its own patented mechanism, but to supply the computational and manufacturing layer that mechanism needs and that neither its inventors nor its commercialiser is incentivised to build.

10. Recommended Next Steps

Target programs, in priority order. NIH National Heart, Lung, and Blood Institute SBIR is primary, on the strength of four IVOS Medical awards and one SmartAirway award at this exact clinical problem. NIH National Institute of Biomedical Imaging and Bioengineering is secondary for the sensing and classifier work, squarely within its remit. The U.S. Army Medical Research and Development Command Combat Casualty Care Research Program is third. The Department of Defense funded the DEVICE trial, and operator-independent airway management far forward is a stated military need, but we note against our own interest that USAMRAA has already funded the incumbent group's device (HT9425-23-1-0872), so this door is not open ground and the approach should lead with the confirmation layer rather than with placement. An NIH R21 is the alternative instrument if the work is framed as method development.

Funding range. Based on the comparable awards in section 7, a Phase I request of $250,000 to $400,000 is calibrated to the institute's recent behaviour (IVOS Phase I at $247,334 in FY2022; SmartAirway Phase I at $313,365 in FY2025), and a Phase II follow-on of $825,000 to $1,500,000 matches the IVOS Phase II at $824,627. The month-six bench gate is deliberately sized to fit inside a Phase I.

The sequencing point is that the manufacturing track and the regulatory track both start in month one. Neither waits for the science, because in this category both of them, not the science, decide whether a working prototype ever reaches an ambulance.