Oxehealth Vital Signs

Performance Metrics

Clinical Evidence

The evidence splits into two questions. Does the camera measure pulse and breathing rate accurately? And does using it make a ward safer or cheaper? The first has a clear answer. The second does not. Accuracy rests on the FDA submissions. The De Novo study (DEN200019, granted March 2021) was an observational, uncontrolled comparison in 60 adults aged 18 to 81, recruited so that skin type, body-mass index and age matched US census data, with at least six people per pair of Fitzpatrick classes. The comparator was the SOMNOscreen contact device. The primary endpoint was the root-mean-square difference (RMSD) across all measurements. Pulse RMSD was 1.81 beats per minute (one-sided 97.5 per cent confidence bound 2.19) against an objective of 3. Breathing RMSD was 1.17 breaths per minute (bound 1.33) against an objective of 2. The study was not powered for subgroups. Under a protocol the FDA calls a worst case, with a reading demanded at fixed times during every activity, 58 per cent (51 to 65) of in-label pulse attempts and 73 per cent (68 to 79) of breathing attempts produced a displayed value. The device says nothing the rest of the time. The August 2025 clearance (K243687) added a second validation: 30 subjects, including adolescents aged 12 to 17, monitored against an FDA-cleared finger pulse oximeter and a capnogram annotated by a physician blinded to both outputs, across realistic activities and metronome breathing up to 39 breaths a minute. The mean absolute error was significantly below 3 beats and 2 breaths per minute (p at or below 0.05), and every subgroup by age, sex, Fitzpatrick type and BMI stayed within those limits. Neither study is peer reviewed and neither reports the actual K243687 error values. One peer-reviewed accuracy study exists. Lewis and colleagues (Thorax 2024) ran Oxevision in six single rooms, two on respiratory wards at the Royal Brompton Hospital and four on an orthopaedic ward at the John Radcliffe, for 2,571 hours in 27 patients (10 respiratory, median age 58; 17 orthopaedic, median age 81). Nurses were blinded to the camera and it was not used for care. Of 102 paired pulse readings, the nurse value was 3.91 beats per minute higher on average, with limits of agreement from minus 10.5 to plus 18.3. Of 154 paired breathing rates, nurse values clustered at 16 to 18 and bore no relation to the camera (limits of agreement minus 8.39 to plus 6.57), running high at low rates and low at high rates. The camera's breathing rate was checked against a blinded reviewer counting breaths on 80 video clips from 10 patients: all within 3 breaths per minute, root-mean-square error 1.31. There was no contact reference for pulse, so the study shows how nurses and the camera differ, not which is right. The authors say the system's clinical use remains to be established in a trial with outcomes such as unplanned ICU admission. Oxehealth funded the study and the staff posts on it and employed or paid four of the eight authors. The outcome literature is a different matter. Ndebele and colleagues (Journal of Mental Health 2024) reported a non-randomised before-and-after evaluation over 24 months on one female and one male acute ward at Coventry and Warwickshire Partnership NHS Trust: bedroom self-harm incidents fell 44 per cent and bedroom ligature incidents 48 per cent. Two of the four authors are Oxehealth staff. A 2026 editorial in the same journal by a King's College London biostatistician (Chua, "a tale of two rates") revisits how those incident rates were calculated; its text is paywalled and was not read for this entry. The five-trust extension by Kekic and colleagues (Journal of Psychiatric and Mental Health Nursing 2024), which reported five before-and-after studies across 12 wards and pooled the four with control groups into a 38.9 per cent relative reduction, was retracted in August 2024. All four authors were Oxehealth employees and had not declared this as a competing interest; the editor stated the paper would have been rejected without review had they done so. The economic model built on those studies (Buckley, Malcolm and Hanlon, PLOS Digital Health 2024) was retracted in July 2026. The editors found inconsistent data sources, unjustified inputs, a possible selection bias that the competing-interest handling did not address, and that the authors, who worked for a consultancy funded by Oxehealth, had never seen the full clinical data because Oxehealth ran the analysis. Two earlier cost models from the same consultancy (Journal of Medical Economics 2022) stand, but rest on the same single-trust before-and-after data. An independent systematic review of surveillance technology on mental-health wards (Griffiths and colleagues, BMC Medicine 2024) covered nine vision-based monitoring studies, rated half of all included studies low quality, and concluded that quantitative findings on self-harm, violence and cost were inconsistent or weak and the evidence insufficient. The earliest report (Barrera and colleagues, Evidence-Based Mental Health 2020) found that remote observations agreed with in-person ones over about 755 patient nights with no untoward incident, and described itself as preliminary. It declared no competing interests; a correction in June 2026 added that two of its authors are Oxehealth employees. A 2022 npj Digital Medicine study of camera monitoring in 15 post-operative intensive-care patients came from the Oxford group that founded the company, but used a research system and is not evidence for the cleared device. The same applies to the Oxevision Sleep Device, a sibling clearance. AurisMark reasoning: a De Novo grant and four 510(k) clearances in the United States, a Class IIa MDR registration in EUDAMED, a Class IIa MHRA registration in the UK and published evidence on file meet the one-mark bar. A second mark is withheld by editorial decision (28 September 2026). The prospective evidence for the measurement claim is thin and vendor-controlled: two FDA validations of 60 and 30 subjects, and one vendor-funded ward comparison with no contact reference for pulse. The outcome evidence has failed: both multi-site outcome papers are retracted, an independent systematic review finds the evidence insufficient, and a coroner's report describes the system being used in place of in-person checks. Re-evaluate for a second mark on an independent prospective study of the cleared device, with the retractions resolved.

Inside the Auris+ Listing

Regulatory Approvals

Safety Record

No recall, enforcement action or MAUDE adverse-event report naming Oxehealth or Oxevision, and none under product code QME, was found in the openFDA device databases queried directly on 14 September 2026. The FDA's identified risk is delayed or incorrect treatment from an erroneous reading or a misread one; the mitigations are the special controls and labelling. The labelling excludes patients with known arrhythmias, because the device reports pulse rate rather than heart rate, warns that accuracy may fall above 110 beats per minute, which the De Novo cohort did not include, and lists bright light, showers and clothing as conditions that block a reading. The clinical hazard is the reading that never comes: in the De Novo study roughly four in ten in-label pulse attempts returned nothing, and the indication makes periodic in-person checks a condition of use. Patient acceptability is contested in the peer-reviewed record: interviews at Broadmoor (BMC Psychiatry 2023) found the system acceptable when used for deterioration and safety but raised privacy and "big brother" concerns, and a 2026 letter from survivor researchers linked to the Stop Oxevision campaign argues that surveillance is not safety. Two records concern how the system is used rather than what it measures. A coroner's Prevention of Future Deaths report (Morgan-Rose Hart, Essex, 19 December 2023, to Essex Partnership University NHS Foundation Trust) found that Oxevision alerts had been muted or reset without the correct procedure, that a bathroom alert was reset and the patient left unattended for about 50 minutes, and raised the concern that staff may have used Oxevision instead of face-to-face observations rather than alongside them. The Broadmoor interviews record patients describing a death in bed that the system had not detected or prevented. Neither is a report of a measurement fault; both bear on the condition of use that in-person checks continue.

Intended Use & Indications

Per the FDA indications for use (K243687, the widest current wording): The Vital Signs device is intended for noninvasive spot measurement of pulse rate and breathing rate when the subject is still. It is software assessing video footage from a fixed- installation solution for use within single occupancy rooms within hospitals, general care and secured environments with professional healthcare oversight and where a framework exists which mandates periodic checks by a trained professional to ensure subject safety. The system is intended for use by appropriately trained staff and should not be used by untrained users. The Vital Signs device is indicated for use on subjects 12 years of age or older who do not require critical care or continuous vital signs monitoring. The device is not intended to be the sole method of checking the physical health of a subject. The De Novo wording (DEN200019), carried unchanged into K211906, K220899 and K251200, says "estimated breathing rate (chest wall movements)", adds "with a duty of care" and limits use to those 18 years of age or older. Prescription use.