Domain Atlas / Behavioral-health & crisis triage
Oxevision camera monitoring on NHS mental health wards
Work with this case in the PAN Lab ↗
In March 2026 the UK Parliamentary and Health Service Ombudsman partly upheld a complaint that an NHS mental health trust installed camera-based, contact-free bedroom monitoring on a psychiatric ward without seeking a patient's consent, gave her no information about it, and did not switch it off when she asked; the case documentation and investigative reporting describe an internal clinical evaluation that the vendor is reported to have authored the business case for and shaped, a rebrand of the vendor during a statutory inquiry, and an open data-protection investigation, while the tool's own outcome-reduction figures are vendor claims contested by a campaign-linked meta-analysis and its adoption share across NHS mental health trusts is reported only as a contested range.[5]
What happened
Oxevision is a wall-mounted, infrared-sensitive camera unit placed in a patient's bedroom on an NHS mental health inpatient ward, built by Oxehealth Ltd, a spin-out of Oxford University biomedical engineering. It measures pulse and breathing rate without contact and detects events — a patient leaving the room, remaining in the bathroom for more than three minutes, or lingering at the doorway for more than three minutes — pushing colour-coded alerts to ward-base monitors watched by nursing staff. Under the standard operating procedure examined by the ombudsman it records pixelated video during red alerts, gives staff a ten-second clear image on demand for a vital-signs check, retains that raw unblurred footage for twenty-four hours before automatic deletion, and records no sound. Trusts deploy it under their own local procedures; Essex Partnership University NHS Foundation Trust (EPUT) was an early adopter, and its executive operational committee recommended roll-out in 2020 after a trial on four wards. It is in live use across a contested share of England's NHS mental health trusts — the vendor claims 50 percent, legal submissions put it at about 40 percent (roughly 20 of 50 trusts), and the survivor network NSUN counts 25 active trusts — so the figure is best read as a range with attribution rather than a settled number.
On 27 March 2026 the Parliamentary and Health Service Ombudsman (PHSO) partly upheld a complaint (case C-2118934) brought by a former inpatient, anonymised in the report as Miss B, against EPUT. The ombudsman found that during her 2021 to 2022 admission the trust did not seek her consent for Oxevision or revisit it, did not switch the camera off when she asked, gave her no information or resources about it, and kept no record of how staff used it in her care. It ordered the trust to pay her 925 pounds, send an acknowledgment letter, produce an action plan, and re-review the standard operating procedure, and it found the surveillance had exacerbated her feelings of anxiety, vulnerability and fear, and that she had changed her behaviour to try to protect herself from the technology. On one limb the complaint did not succeed: on the question of over-reliance, the ombudsman found on balance — cross-referencing observation charts, a nurse-adviser review, and door key-card data — that staff had continued to carry out in-person observations, and did not uphold that part. The partly-upheld structure matters and should not be flattened into a finding that the camera replaced human care.
Two features of the trust procedure drew specific criticism. The earlier-era procedure instructed staff to tell patients that Oxevision monitoring is part of normal practice and that consent for its use will not be required — a documented bypass of the consent and authority step. The ombudsman found even the revised 2025 procedure still departs from General Medical Council standards and NHS England's February 2025 principles for using digital technologies in mental health inpatient care, because a capacitous patient's refusal can be overridden by a nurse-in-charge or doctor decision, with multidisciplinary-team approval, on clinically-safe grounds — meaning the trust may keep filming a refusing, capacitous patient. Separate reporting describes systemic patient-facing information failures: patients told images would be blurred when clear footage existed, told cameras were mandatory trust policy when no such policy existed, and a peer-reviewed critique reporting that staff reassured patients the camera was not recording them all the time.
The defining governance fact is who controlled the evidence. A February 2023 internal clinical evaluation by Oxford Health NHS Foundation Trust — obtained by the campaign group Stop Oxevision under a freedom-of-information request and shared exclusively with Novara Media in January 2026, and still not published in full — recorded that the vendor had authored the trust's business case, asked survey questions the report described as leading, and guarded the Early Insight evaluation area as the bedrock of its sales and marketing drive; the report concluded that successful business relationships are built on trust, and that this was not evident. Oxford Health nonetheless spent about 2.4 million pounds between November 2021 and September 2024 against estimated staffing savings of about 709 thousand pounds. The vendor's own outcome claims — reductions of 44 percent in self-harm, 48 percent in falls, 37 percent in assaults and 71 percent in one-to-one observations, alongside incident-detection counts on EPUT wards — are marketing figures, disputed by a campaign-linked meta-analysis reporting no statistically significant association between the system and four of five measures of patient safety, and a 2024 review that Novara reports found insufficient evidence for the promised safety and cost benefits. Novara also reports the vendor is planning to roll out a whole-hospital patient-monitoring platform in 2026.
The scrutiny has escalated on several fronts, and the vendor's identity has shifted in the middle of it. The statutory Lampard Inquiry, opened on 9 September 2024 into the deaths of more than 2,000 people under Essex mental health inpatient care, took Stop Oxevision's evidence privately on 14 May 2025 after postponing its public Oxevision evidence, and heard EPUT and vendor evidence in October 2025; at the inquiry EPUT admitted there is the possibility of misuse across all trusts using the tool, including malicious staff access to live bedroom feeds, and the vendor's patient-safety head accepted that round-the-clock filming could constitute a very significant invasion of privacy. In August 2025, mid-inquiry, the vendor rebranded from Oxehealth to LIO or LIO Health; NSUN reports the parent company and possibly the product were renamed, and the oxehealth.com domain now redirects to liohealth.com. On 18 May 2026 the law firm Bindmans, acting for Stop Oxevision, asked the Information Commissioner's Office for an expedited data-protection investigation into consent and transparency, opaque trust-vendor data arrangements, the classification of pixelated video and derived datasets, and retention and secondary uses without a clear lawful basis; Novara Media reported on 23 June 2026 that the ICO had opened an investigation, quoting a Bindmans solicitor that there is a clear legal risk that other NHS trusts using the tool in similar ways could face comparable complaints. Campaigners and bereaved families have linked the technology to at least four inpatient deaths; that linkage is journalistic and campaign reporting, not an adjudicated causal finding, and is carried here as contested. As of mid-2026 there has been no national suspension.
The sociotechnical reading
Most surveillance cases in the Atlas turn on what the tool does to the person it watches. This one turns on who was allowed to say whether the tool works. Oxevision inserts a continuous computer-vision sensing node into the staff-patient observation link inside a psychiatric bedroom: it reads the room, measures vital signs without contact, and pushes alerts to the ward base. On the system map, the interesting edge is not the alert. It is the evaluation edge — the independent check that is supposed to sit outside the vendor and tell the trust whether the thing is safe. In this case that edge was captured. The company selling the camera authored the trust's business case, shaped the survey the evaluation ran on, and guarded the Early Insight report as the bedrock of its sales drive. The one link that was meant to be the vendor's counterweight was the vendor's all along, and no independent error or false-alert rate for the monitor was ever published. When the governance record that vouches for a tool is written by the party selling it, that record is not evidence; it is sales collateral wearing an evidence costume.
That captured evaluation edge is joined by two others that make the shape distinctive. The first is the consent and authority edge, bypassed at the point of insertion: the procedure told patients the camera was normal practice and that consent would not be required, and even after revision it lets staff keep filming a capacitous patient who has refused. The person in the bed — the one node with the strongest claim to control the feed — has no working turn-off pathway, and their authority is subordinated to staff discretion at both eras of the procedure. The second is the boundary: the bedroom video, the derived data, and now the observation record itself all live on the vendor's platform, because the trust records observations electronically inside the Oxevision system, so the sensing vendor also hosts the compliance record. The audit trail is lossy by design — raw footage auto-deletes in a day, access logs drop departed staff, and the ombudsman found the care plans held no record of the monitoring at all — which means the one place you would look to reconstruct how the system was used is thin by construction and vendor-hosted by default. And when the scrutiny finally arrived, the vendor changed its name. A rebrand mid-inquiry is a governance event on the vendor node: it resets nothing about the captured evaluation or the boundary, but it can reset the clock in the public record if no one is tracking the tool and its data across the rename.
The distinct lesson the Atlas draws here is that a vendor-authored evaluation is not a weak check but a captured one, and the two fail differently. A missing check can be commissioned; a captured check first has to be pried out of the seller's hands before anything independent can take its place — which is why the governing levers point at the boundary and the evidence, not the camera. Buying a sharper monitor is the false move: the containment was never the alert accuracy, and the harm is not a missed vital sign but an intimate surveillance the ombudsman found worsened the very anxiety it was meant to keep safe. What actually moves this system is a vendor gate that secures independent test access and a data-processing agreement or self-hosting to close the model's path out of the boundary; an oversight cadence that carries the procedure re-review the ombudsman ordered and re-checks the regime against national consent standards on a rhythm; a write gate and durable logging so the record of what happened is not the vendor's to keep or lose; and a standing vigilant channel that keeps the in-person observation alive as the camera scales, since a tool sold to reduce human observation invites exactly the substitution the business case priced in. The honest boundary throughout, and a hard rule for a behavioral-health cell: served patients are not modeled here, no clinical, distress, or safety outcome is computed from anything in this reading, the complaint was only partly upheld — the over-reliance limb was not — and every vendor effectiveness figure remains a contested marketing claim, countered by the independent evidence, until an evaluation the vendor did not write says otherwise.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.