All news

Your reflection seems spot on. It would be really interesting to understand which balancing measures are being considered as part of the review of Martha’s Rule, particularly to explore any unintended consequences alongside the intended benefits.

Thank you for the reflection. This gave me a lot to think about. One thought that stayed with me was whether off-mandate requests might actually be a signal of unmet needs elsewhere in the system. If patients and families repeatedly choose Martha's Rule for concerns it wasn't designed to address, perhaps that reflects a perception that it is the pathway most likely to result in someone listening and responding. That feels like an important insight, not just about Martha's Rule, but about the wider governance environment in which it sits.

Many healthcare organisations are introducing Artificial Intelligence (AI) into work systems that were not designed for high levels of automation. In practice, this often means that new tools are added locally while team structures, responsibilities, interfaces, and escalation routes remain unchanged. The result can be more cognitive load, fragmented coordination, and unclear ownership rather than safer care. A safer approach is to redesign the work system itself: the teams, tasks, technologies, and governance that shape day-to-day clinical work. One practical step is to model team structures and work processes explicitly so that roles, interfaces, risks, and control measures are visible, traceable, and reusable. This can also provide a structured basis for Large Language Models (LLMs) to support governance tasks such as incident summaries, handover drafts, and training scenarios, all under human oversight. In healthcare, the question is not whether Artificial Intelligence will be used, but whether it will be introduced into work systems that are already difficult for staff to coordinate safely. When a new tool is added without clarifying responsibilities, interfaces, or decision authority, the likely result is not transformation but cognitive load: more interruptions, more workarounds, and less clarity about who should do what, when, and on what basis. That matters for patient safety because poorly aligned work systems can weaken handovers, delay escalation, and increase cognitive burden in already demanding settings. For that reason, I believe the starting point should be the work system, not the tool. Team organisation needs to be designed around the real demands of care: the patient journey, the coordination load between professions, the need for escalation, and the information required to act safely. In practical terms, that means defining clear team boundaries, explicit interface agreements between teams, and reliable modes of collaboration for shared problem-solving and specialist support. This makes accountability more transparent and reduces the risk that important tasks fall through the cracks. A useful next step is to model team structures explicitly. Using, e.g., Sparx Systems Enterprise Architect and SysML (Systems Modelling Language), it is possible to describe not only system structure but also behaviour, requirements, and team interfaces consistently. In a healthcare context, that can include teams, roles, responsibilities, decision points, escalation routes, handover dependencies, risks, and existing control measures. The value is not modelling for its own sake. The value is that operational knowledge becomes structured, reviewable and reusable across governance, training and redesign work. Once this information is modelled in a disciplined way, there is also a plausible route to safer use of Large Language Models. Rather than asking an LLM to generate advice from unstructured discussion alone, organisations can use structured models as a controlled knowledge base, the "Single-Source-Of-Truth". Under human review, that can support practical outputs such as incident summaries, handover drafts, training cases, draft requirements, and options for redesign. The important point is that the model provides consistency and traceability: users can see which role, task, interface, or risk the output is based on. In a patient safety setting, that is far more defensible than relying on text-based documents alone. This last point is an informed systems-engineering inference from structured modelling practice. My view is that healthcare organisations will get more value from AI when they first make their work systems visible. If we want safer care, we need to design for human-AI collaboration in the same disciplined way that we design for staffing, escalation, and accountability. Models will not replace professional judgment, but they can make coordination, governance, and learning more reliable. That is where the patient safety opportunity lies.

AI in Healthcare requires regulation by the MHRA as a medical device. I don't think that the department of Health has the capacity.