Oxford Medical Simulation

An authored, standardized simulation library is the product, with AI layered into it. The company's August 2025 article says the core of each scenario is standardized so learners see the same condition, while natural language processing and large language models enable voice control and open-ended conversation with virtual patients; one communication scenario has a family member, Natalie, respond to the learner's word choice and intent. The platform page names AI-driven communication as one of three pillars beside clinical reasoning and procedural mastery, and the privacy notice says voice input is transcribed by a speech-to-text service such as Microsoft Azure.

What an institution buys is the library of clinical scenarios across nursing, medicine and allied health, delivered in VR or on screen, with objective action-by-action scoring, debriefs, educator analytics and tools to adapt or build scenarios. That stands without the models; the generated dialogue adds open conversation to part of it. Speech recognition and generated dialogue are learned functions found on the company's own pages.

Scoring is automated and explained, educators see the results, and the override step is missing. The platform page describes objective, standardized scoring of every action mapped to set skills, debriefs after each scenario, educator analytics across individuals and cohorts, and autonomous practice that needs no facilitator; the company's AI article says generated output must be checked for accuracy and its limitations monitored. The standardized core of each scenario bounds what the generated dialogue can change.

What is not published is whether an educator can override an automated competency result, whether a learner can appeal one, and what happens when a learner's spoken input falls outside what the scenario expects. In health systems the results support onboarding and competency programs, so the review step matters.

The approach and one supplier are named, without versions or update practice. The company says scenarios keep a standardized core, with natural language processing and a large language model enabling voice control and open-ended dialogue with virtual patients, and the privacy notice names Microsoft Azure speech-to-text as the service that transcribes voice input. The language model and its provider are not named, nor how it is constrained to each scenario's clinical facts, and no model version, change log or update cadence is published.

The boundary between authored and generated content is partly described (the standardized core versus open dialogue), which tells a buyer where editorial review applies.

The supply chain is named in substance, with the language model provider missing. The privacy notice lists the businesses that help provide the service: AWS for cloud storage, Microsoft Azure for speech-to-text in voice-controlled scenarios, Google, Zoom, Sybill AI, Fireflies.ai and Seismic for meetings and sales, Intercom for the help desk, SendGrid for email and HubSpot for marketing. It states the terms for voice data: recordings are stored on AWS for no more than two weeks, sent encrypted to the speech-to-text service, which keeps no copy, and then deleted.

The provider of the large language model behind open dialogue is not named, and nothing states what it receives or keeps.

Published and presented research on the product is summarized on the company's research page, with sources named. It reports Hulfish (2024), a study of medical residents in internal medicine, emergency medicine, family medicine and pediatrics that found no significant differences in technical skills, non-technical skills or knowledge retention between manikin-based simulation and OMS; Brown (2022), where junior nursing students' learning experience was similar and nurse practitioner students' consistently better than face-to-face simulation on CLECS 2.0 and SET-M; Brown (2023) on usability; Bradley (2024) and Godbold and Cairnduff (2024) on learner preference and confidence; an NHS case study (2023) reporting nearly double the learners and a 74 percent reduction in staffing time and estate costs; and a York Health Economics Consortium analysis (2022) of cost per use.

These are the company's own summaries rather than the full papers, and none measures whether training on the platform changes care of real patients.

Retention is stated specifically and training use only for voice. The Global Privacy and Biometric Notice says voice recordings made when learners use voice control are stored on AWS, sent encrypted to a speech-to-text service such as Microsoft Azure that keeps no copy, kept no longer than two weeks, used in that time only to check speech recognition accuracy and resolve support issues, and then permanently deleted; the company creates no biometric identifiers from them. After a customer contract ends, personal data may be kept for up to 12 months and is then deleted or anonymized. The company says it never asks for health data and does not sell personal data.

What is not stated is whether learner conversation text, actions and scores are used to train or tune the language models behind the dialogue, or what the language model provider receives and keeps. Safety engineering, meaning how a clinically wrong or inappropriate generated reply is caught, is not published. No patient information is processed; the patients are simulated.

A detailed privacy notice covers the service, and no business associate status or scope position on patient data is stated. The Global Privacy and Biometric Notice covers the websites, the web app and the simulation services, sets out the company's role as processor for customer organizations and as controller for some data, and says the company never asks users for health data about themselves.

It does not address protected health information: the patients are simulated and nothing suggests real patient data enters the platform, but a statement to that effect, with the scope explained, is what a buyer would rely on.

Recognized certifications are named in the company's own material without the artifact. The privacy notice lists ISO 27001 certification among its security measures and says internal processes follow Cyber Essentials Plus and ISO 27001; the platform page lists ISO 27001 compliance with SSO. No certificate, certifying body, scope or expiry is published and there is no trust portal.

As educational simulation software it makes no device claim, and none applies; nothing it sells is positioned for patient care. Neither the website terms nor the product pages state an intended purpose or device position. What governs its use is education regulation: state boards of nursing decide how much simulation may substitute for clinical hours, and programmatic accreditors in the United States and the United Kingdom assess how simulation is used; the company maps scoring to set skills so programs can evidence competencies.

Education regulation governs the product, and it is not a device pathway.

The company states principles and publishes no evaluation. Its August 2025 article on AI in simulation says AI is not a set-and-forget tool, that output must be double checked for clinical accuracy, and that its limitations need close monitoring. Nothing describes how the scenario library or the generated dialogue is reviewed for stereotyped portrayal of patients, whether speech recognition works equally well across accents and dialects, or whether automated scoring differs across groups of learners.

Those are the two risks here: authored portrayals carry their authors' assumptions (though, unlike generated ones, they can be reviewed once and stay fixed), and automated feedback on communication can penalize dialect or cultural style.

Scoring is explained and debriefed, and nothing stands behind the output. The platform page says every action is objectively ranked, tracked and explained, with scoring protocols mapped to set skills, and that debriefs after each scenario tell learners where to focus; educators see individual and cohort results. That lets a learner or educator trace a score to the actions behind it. The website terms of use make no promise of accuracy and limit liability; any commercial terms sit in customer agreements that are not published.

The boundary between authored and generated content matters for errors: the company says the core of each scenario is standardized, and generated dialogue is used for open conversation. Nothing is published on how generated responses are checked, what happens when a learner says something the scenario did not anticipate, or how a learner or educator challenges a score.

No clinical record integration exists: charting happens inside the simulation. The systems that matter here are the buyer's own, learning management and student information systems for schools and learning or credentialing systems for health systems. The platform page lists SSO access and LMS integration and says the platform integrates with existing systems, and scoring is mapped to set skills; no learning management system, standard such as LTI, or HR or credentialing platform is named.

Residency is stated by customer country, with the processing path partly open. The privacy notice says information is generally stored in the country where the customer organization is based, with Canadian customers' data stored in the United States; that some information may be transferred abroad for specific purposes under approved contracts; and that AWS provides cloud storage. The company operates as Oxford Medical Simulation Inc. in Somerville, Massachusetts and Oxford Medical Simulation Ltd., registered in England and Wales.

What is not stated is which AWS regions serve which countries, where speech-to-text and dialogue processing run, and what stays on a headset or laptop between sessions, since learners train on institutional and personal devices.

No price is published; the company publishes realized cost from a study instead. Its research and ROI pages cite a York Health Economics Consortium case study (2022) finding OMS cost between $2.22 and $14.38 per use against $28.38 to $394.95 for physical simulation, and a return on investment of 120 to 3,000 percent. Those figures describe what studied institutions spent per use, which depends on how much they used the platform, not what a new customer would pay. No license price, unit of charge or tier is published, every path leads to a demo, and whether VR headsets are included is not stated; the platform runs on headsets or on Mac and Windows computers.

How the product is delivered is clear, and so is the cost range seen in studied deployments.

Coverage is named across professions and settings, with studies behind part of it. The site offers the platform to nursing programs, medical schools, physician assistant and allied health programs and graduate medical education, and to health systems for onboarding and transition to practice, quality and safety, and risk and compliance, in VR or on screen. The research page's studies cover nursing students, nurse practitioner students and medical residents in internal medicine, emergency medicine, family medicine and pediatrics.

Nothing is published for allied health, physician assistant or health system onboarding outcomes, and breadth of the library is not depth in any one specialty.