The Department of Health and Social Care has announced that it is developing a world-first artificial intelligence (AI) early warning system to automatically identify safety concerns across the NHS. In this blog, Patient Safety Learning reflects on this announcement and how this might work.
Within the NHS there are an increasing discussions about the potential opportunities that AI can present to improve the efficiency and effectiveness of healthcare treatment and outcomes. Recent examples of this include:
- Using time-saving AI-enabled ambient scribing products that can automatically listen and capture conversations.[1]
- Using AI tools that assist with diagnosis and decision making.[2]
- Applying new AI systems that can analyse data, such as patient and family feedback and experiences, revealing new insights which can be used to improve services.[3]
In the most recent development, the Department of Health and Social Care has this week announced plans to create a new AI early warning system to flag safety issues in real time and trigger inspections. In the announcement it states that:
“The new safety warning system, being developed as part of the government’s 10 Year Health Plan, will rapidly analyse healthcare data and ring the alarm bell on emerging safety issues. Work on rolling out the system is already underway. A new Maternity Outcomes Signal System will launch across NHS trusts from November, using near real-time data to flag higher than expected rates of stillbirth, neonatal death and brain injury.”[4]
The Department says that this new Maternity Outcomes Signal System will use near real-time data to flag higher than expected rates of stillbirth, neonatal death and brain injury. While it has an initial focus on maternity, the Department has indicated that when fully implemented this could be applied to hospital databases across other types of injuries and incidents.
Patient Safety Learning welcomes this announcement. While significant efforts have been made to collate useful patient safety data, too often organisations, both at a local and national level, remain often data-rich and information poor. Any technological developments that can be applied to reveal the systemic causes of patient safety failure in a consistent and compelling way are to be commended. However, we do have a number of questions stemming from today’s announcement and about this new Maternity Outcomes Signal System.
How will it operate in practice?
Today’s announcement leaves open several practical questions about how the Maternity Outcomes Signal System will function, including:
- What data will this AI model be trained on? This will be critical to informing how it performs in practice. This can also affect bias and data, potentially overrepresenting certain groups or viewpoints, which can exacerbate and entrench health inequalities.
- Will these data include a patient’s clinical records? If so, how will patients provide consent for their data to be used in this way? Indeed, how will patient, family and carer insights be sought to inform such analysis? And how will staff-reported data, including whistleblowing testimonies, be captured?
- How much medical terminology will this new AI understand? Does it ‘get’ subtle semantic differences? This issue is currently being debated around the accuracy of AI-enabled ambient scribing software in taking patient records.
- How will alerts in the system operation? How will they be raised and to who? Will alerts be graded in severity?
How will alerts be acted on?
A key problem in patient safety remains “the implementation gap”; the difference between what we know improves patient safety and what is done in practice.[5] We already have findings from a wide range of different investigations and inquiries, with accompanying safety actions and recommendations; however, putting these into practice remains difficult. As we noted in our report Mind the implementation gap: The persistence of avoidable harm in the NHS, this is in part because we often lack systematic approaches to implementing and evaluating safety recommendations.[6]
If a new AI early warning system is to be effective, it must be matched by the appropriate mechanisms to act on its findings and subsequently check that these actions result in improvement. The initial announcement states that where concerns are raised “the Care Quality Commission (CQC) will deploy specialist inspection teams as soon as possible to investigate and take swift action”. It’s interesting that the immediate response is not for the organisation itself to undertake an investigation but for this immediately to be escalated to CQC, the systems regulator. We would like to see greater clarity on the criteria for triggering an investigation, but also:
- What these investigations look like in practice. For instance, how might they consider not only systemic issues but, where appropriate, matters of staff competence that could be of concern to professional regulators.
- What the threshold for investigations will be and the capacity being planned at CQC to meet the demand.
- How implementation of the changes they may mandate are responded to (are they observations, recommendations for directors?) and how will responses be monitored and evaluated.
- What steps are put in place if the situation does not improve following these investigations.
How will learning be shared?
We would also like to see greater detail on how learning will be shared from this new system. Also is it anticipated that its findings and alerts could connect to existing reporting and investigation frameworks, such as the Learn from Patient Safety Events (LfPSE) service, the Medicines and Health products Regulatory Agency (MHRA) Yellow Card Scheme and the Patient Safety Incident Response Framework (PSIRF).
Where safety alerts flag issues for the CQC to respond, the problems identified and action taken to address them may be applicable elsewhere in the NHS, the independent sector and across the UK. There needs to be mechanisms in place to enable the sharing of this data.
Concluding thoughts
In summary, we welcome this statement of commitment to patient safety and the use of emerging technologies to ensure healthcare better proactively assess risk and respond to it to prevent avoidable harm. As with all innovative initiatives, there will be many issues to work through to turn concept into practical reality. Patient Safety Learning looks forward to hearing more about this journey and would welcome the opportunity to directly engage and support this development, including using the insights and expertise from its extensive patient safety networks of clinicians, patient safety experts and patient safety partners.
References
[1] NHS England. Guidance on the use of AI-enabled ambient scribing products in health and care settings, 27 April 2025. https://www.england.nhs.uk/long-read/guidance-on-the-use-of-ai-enabled-ambient-scribing-products-in-health-and-care-settings/
[2] Sunny Deo. One size does not fit all. How AI and better data can help us embrace complexity in diagnosis and treatment. Patient Safety Learning, 6 May 2025. https://www.pslhub.org/learn/patient-safety-in-health-and-care/diagnosis/one-size-does-not-fit-all-how-ai-and-better-data-can-help-us-embrace-complexity-in-diagnosis-and-treatment-r13090/
[3] Ben Kenyon. From pain to progress: How NHS trusts are tackling the complaints crisis with AI. Patient Safety Learning, 23 June 2025. https://www.pslhub.org/learn/commissioning-service-provision-and-innovation-in-health-and-care/digital-health-and-care-service-provision/288_artificial-intelligence/from-pain-to-progress-how-nhs-trusts-are-tackling-the-complaints-crisis-with-ai-r13266/
[4] Department of Health and Social Care. World-first AI system to warn of NHS patient safety concerns, 30 June 2025. https://www.gov.uk/government/news/world-first-ai-system-to-warn-of-nhs-patient-safety-concerns
[5] Suzette Woodward. Patient safety: closing the implementation gap. The King's Fund, 30 August 2016. https://www.kingsfund.org.uk/insight-and-analysis/blogs/patient-safety-closing-implementation-gap
[6] Patient Safety Learning. Mind the implementation gap: The persistence of avoidable harm in the NHS, 7 April 2022. https://www.patientsafetylearning.org/blog/mind-the-implementation-gap-the-persistence-of-avoidable-harm-in-the-nhs
