Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Oxevision camera monitoring on NHS mental health wards

The evaluation was written by the seller: a bedroom monitor no one independent checked

A camera on the wall of a psychiatric bedroom measures your pulse and your breathing without touching you, and flags when you leave the room or stay too long in the bathroom. It is sold as safety. Modeled on Oxevision on National Health Service (NHS) mental health wards. The trust that bought it leaned on an evaluation to say it worked. The catch is who wrote that evaluation: the company selling the camera authored the business case, shaped the survey, and guarded the report as the bedrock of its sales drive. So the one check that was supposed to sit outside the vendor — the independent word on whether the thing is safe — was the vendor's all along. No independent error rate for the alerts was ever published. The person in the bed was told the camera was normal practice and that consent would not be required, and when she asked for it off, it stayed on. And when the scrutiny finally came, the company changed its name. Your move is not to buy a sharper camera. It is to get the boundary and the evaluation out of the seller's hands.

Stylized model of a documented deploymentBehavioral-health & crisis triage

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Oxevision-class bedroom vital-signs monitor on a psychiatric ward network: 6 components and 12 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 2 assumed · 4 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

  • assumed

    This models the continuous-bedroom-monitoring pattern documented in the Oxevision case file — not a reconstruction of the actual device. The defining choice is that the one edge meant to sit outside the vendor's control, the independent evaluation of the tool, is the edge the vendor authored: the internal clinical evaluation was vendor-shaped (vendor-authored business case, survey questions described as leading, the Early Insight report guarded as the bedrock of its sales drive), so the independent model check is drawn empty and the vendor rebrand from Oxehealth to LIO mid-inquiry sits as a governance event on the vendor-hosted boundary.

  • baseline

    The load-bearing pathways are the vendor-capture loop (the retained data feeding the vendor's own evaluation, and the bedroom video, derived data, and observation record all egressing to the vendor-hosted platform) and the observation-substitution pressure on the nursing link. Structurally the governing levers protect the independent evaluation and the vendor boundary, not the monitor's raw alert accuracy — for which no independent error or false-alert rate is even published.

  • baseline

    The consent and authority edge is bypassed at insertion: the earlier-era SOP told patients that consent for Oxevision would not be required, and the ombudsman found even the 2025 SOP lets a nurse-in-charge or doctor override a capacitous patient's refusal on clinically-safe grounds with multidisciplinary-team approval, so the person in the bed has no working turn-off pathway. This is carried as a documented pathway on the served-patient boundary and in the SOP-supervision check, never computed as a patient outcome.

  • baseline

    The over-reliance question is authored honestly: the substitution drift is drawn as a documented pressure (the 2020 business case monetized fewer one-to-one observations and faster rounds while policy insisted the system complements and does not replace observation), but in the one adjudicated case the ombudsman found on balance, via observation charts, nurse-adviser review and key-card data, that in-person observations did continue and did not uphold the over-reliance limb. The nursing correction edge is therefore real; the peer drift is a pull, not an adjudicated finding.

  • baseline

    The audit trail is lossy by design, drawn as a second inactive absence (a record-against-record check): raw unblurred footage auto-deletes at twenty-four hours, door key-card logs drop entries for departed staff, and the ombudsman found the care plans contained no record of Oxevision use at all, so how the system was actually used cannot be reconstructed. write-gate (durable logging) and oversight-cadence (periodic audit and the standard operating procedure (SOP) re-review the ombudsman ordered) are its institutional response.

  • assumed

    Served patients are not in the dynamics, and no clinical, distress, or safety outcome for any patient is computed here — a mandatory boundary for a behavioral-health cell. The documented iatrogenic exposure (the ombudsman found the surveillance exacerbated the patient's anxiety, vulnerability and fear, and that she changed her behaviour to protect herself from the technology) is recorded in the case file and measured outside a diagram like this one; vendor outcome-reduction claims are marketing figures, disputed by a campaign-linked meta-analysis reporting no statistically significant association with four of five patient-safety measures, and the linkage of the technology to inpatient deaths is contested campaign and journalistic reporting, not an adjudicated causal finding.

What this example does not show

  • Served patients are not modeled here, and no clinical, distress, or safety outcome for any patient is computed from anything in this diagram; the Lab reads institutional propagation only. The documented iatrogenic harm — the ombudsman found the surveillance exacerbated the patient's anxiety, vulnerability and fear, and that she changed her behaviour to protect herself from the technology — is recorded in the case file and measured outside a diagram like this one.
  • The complaint was partly upheld: the ombudsman upheld the consent, information, turn-off and record-keeping failures, but on the over-reliance question found on balance that in-person observations did continue and did not uphold that limb. Nothing here should be read as an adjudicated finding that the camera replaced human care.
  • No independent error or false-alert rate for the monitor is published. The vendor's outcome-reduction figures are marketing claims, disputed by a campaign-linked meta-analysis reporting no statistically significant association between the system and four of five patient-safety measures; the linkage of the technology to inpatient deaths is contested campaign and journalistic reporting, not an adjudicated causal finding; and the adoption share is contested (vendor 50 percent of National Health Service (NHS) mental health trusts, legal submissions about 40 percent, an advocacy count of 25 trusts).
  • A safe starting baseline is a property of this model, not a safety promise for any real deployment. The internal clinical evaluation that shaped the roll-out was obtained under freedom-of-information law and shared with journalists, not leaked, and remains unpublished in full; its contents are cited via that reporting, and the defining point is that no fully independent effectiveness evaluation of the tool exists in the public record.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

  • 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.

    empirical
    • Government Parliamentary and Health Service Ombudsman, Final report on complaint C-2118934 about Essex Partnership University NHS Foundation Trust (Oxevision) (2026) https://stopoxevision.com/wp-content/uploads/2026/04/Final-Ombudsman-Report-Miss-B-1-1.pdf
    • Investigative Williamson, NHS Trust Spent Millions on Controversial Spy Camera Tech Despite Damning Internal Report (Novara Media, 2026) https://novaramedia.com/2026/01/15/nhs-trust-spent-millions-on-controversial-spy-camera-tech-despite-damning-internal-report/
    • Investigative Williamson, Creepy Bedroom Surveillance Tech a Clear Legal Risk for NHS Trusts (Novara Media, 2026) https://novaramedia.com/2026/06/23/creepy-bedroom-surveillance-tech-a-clear-legal-risk-for-nhs-trusts/
    • Advocacy National Survivor User Network, NHS Trust forced to admit potential misuse of Oxevision (now LIO) (2025) https://www.nsun.org.uk/news/nhs-trust-forced-to-admit-potential-misuse-of-oxevision-now-lio/
    • Advocacy Stop Oxevision, campaign website and resources page (2026) https://stopoxevision.com/resources/

  • The ombudsman's report on the case (decision 27 March 2026) found the trust did not seek or revisit the patient's consent for the bedroom monitoring, did not turn the camera off when she asked, gave her no information about it, and kept no record of how staff used it, and that even the trust's revised 2025 procedure still permits overriding a capacitous patient's refusal on clinically-safe grounds with multidisciplinary-team approval; on the separate question of over-reliance the ombudsman found on balance, cross-referencing observation charts, a nurse-adviser review, and door key-card data, that in-person observations had continued and did not uphold that part of the complaint.

    empirical
    • Government Parliamentary and Health Service Ombudsman, Final report on complaint C-2118934 about Essex Partnership University NHS Foundation Trust (Oxevision) (2026) https://stopoxevision.com/wp-content/uploads/2026/04/Final-Ombudsman-Report-Miss-B-1-1.pdf

Where this connects

Institutional pressures in this domain

  • Workload surge — Demand outruns staffing; per-case attention shrinks and review becomes triage.
  • Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.
  • Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Deadline pressure — Statutory or managerial timeliness rules reward fast approval of machine output over slow disagreement.
  • Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.

All of them in context on the Behavioral-health & crisis triage domain page.

Levers available here and the patterns behind them

Documented case histories