Fiona Stevenson

Background: In the United Kingdom, first contact practitioners (FCPs) provide direct access to musculoskeletal (MSK) assessment and management in primary care. This means FCPs have to assess patients with no prior medical screening, which assumes reliance on developed skills and experience. The pool of appropriately experienced physiotherapists is limited, and less experienced clinicians have been recruited. A 2026 national evaluation found substantial variation in FCP service delivery, with data collection and governance among its structural causes. The Primary Care Individual Social Norms MSK Data Dashboard (PRISM) dashboard gives FCPs feedback on their clinical activity relative to their peers to support reflection toward improving MSK care. Objective: This study aimed to co-design the metrics, filters, and guidance content of the PRISM dashboard and guidebook to reflect evidence-based practice and the real-world context of FCP roles, and apply a critical realist approach to identify the mechanisms and conditions shaping its design. Methods: This qualitative co-design study used a modified nominal group process across 2 online workshops in June and August 2025, with 10 purposively sampled participants: clinical FCPs, national physiotherapy and data leaders, and a patient representative. Two analytic strands were applied to the same data. The consensus process produced the metric set and filters. Critical realist thematic analysis of the transcripts using NVivo (version 12 Plus; QSR International), with line-by-line coding using inductive and a priori codes, identified the conditions shaping how those metrics would need to be interpreted. Results: Participants reached consensus on a core metric set describing the breadth of the FCP role, separating variation generated by clinical decisions from variation generated by structure such as local diagnostic access, prescribing rights, and referral pathways. Contextual filters were co-designed at patient, clinician, and service level to keep structure visible. Supervision emerged as a condition surrounding the dashboard rather than a component, since the same comparison prompts reflection where role clarity and psychological safety are present and defensiveness where they are not. A tool dependent on supervision would work least well where support is weakest. This means that an option for stand-alone use is a design requirement. Three data integrity risks were identified: recording reliability, data gaming, and clinician burden. The guidebook enables these conditions to be transferred into practice, with separate guidance for FCPs and supervisors and a 4-step reflection structure. Conclusions: Expert co-design produced a metric set describing the breadth of the role while accommodating variation between services and acknowledging the conditions under which PRISM can function as intended. This paper moves from the national position that FCP services collect data to a specification of the conditions under which collecting data produces learning. Mapping of the metrics against National Institute for Health and Care Excellence MSK guidelines and quality indicators is reported in a companion paper.

Background: AI is increasingly prevalent within health care; however, its implementation within the hospital setting may come with its own limitations, given the high-pressure and sensitive environment. Barriers and facilitators to the implementation of AI within hospitals have been largely explored quantitatively, and there is a scarcity of qualitative research on the implementation of AI tools to improve air quality. Objective: This study aims to qualitatively explore, through a normalization process theory (NPT) lens, the anticipated barriers and facilitators to the successful implementation of an AI tool to guide the placement of mitigation strategies to reduce airborne transmission of infections within the hospital setting. Specifically, we sought to (1) understand the day-to-day work of key hospital staff involved in reducing airborne transmission of infections, (2) identify potential barriers to the implementation of an AI tool to guide the placement of mitigation strategies in the hospital setting, and (3) recognize necessary conditions for the successful implementation of the AI tool within the hospital setting. Methods: We conducted ethnographic observations (n=4) and semistructured interviews (n=18) with hospital staff following preliminary one-on-one meetings and group discussions involving patients, members of the public, national leaders in ventilation working within academic and industry environments, and hospital staff (n=70). Data were collected between October 2025 and May 2026 in the United Kingdom. Interview transcripts, field notes, and summaries from preliminary wider engagement work activities were analyzed in NVivo (Lumivero; version 12) following codebook thematic analysis to facilitate confirmatory triangulation. Findings were organized according to the core constructs of NPT. Results: Hospital staff understanding of the AI tool and their preparedness to invest time in its implementation varied across those consulted. The anticipated work required by hospital staff to successfully implement the AI tool and its recommended mitigation strategies was identified as a significant challenge for its scalability. Differences in the ways of working among trusts and the competition for limited resources (collective action), and issues associated with hospitals’ physical estate (coherence) were seen as particularly problematic. The need for local champions (cognitive participation) was raised as a facilitator, as well as expert validation and the assessment of the AI tool’s effectiveness (reflexive monitoring). Conclusions: This preimplementation study suggests that the successful implementation of an AI tool to improve air quality in the hospital setting depends not only on the effectiveness demonstrated by the AI tool but also on the way trial outcomes are conveyed to hospital staff. Hospital staff anticipate an initial resistance to the AI tool’s implementation. Further, the impact of the AI tool and suggested mitigation strategies in workflows, costs, and staff and patient outcomes may be decisive.

There is a growing use of artificial intelligence (AI) in healthcare settings, but challenges exist with AI adoption and its long-term use. This umbrella review funded by the NIHR (NIHR205439), aimed to identify the facilitators and barriers of AI implementation within hospitals and was registered on PROSPERO. Five databases (MEDLINE, HMIC, CINAHL Plus, Web of Science and Cochrane Reviews) were searched in January 2025 and re-run in February 2026, 1184 articles were screened, with 20 included. The inclusion criteria encompassed reviews implementing AI within hospital settings. The quality of the data were assessed using the ROBIS checklist and data were synthesised using the NASSS (Nonadoption, Abandonment, and challenges to the Scale-up, Spread, and Sustainability) framework. We found AI implementation was affected by: whether an AI solution as an intervention had been externally validated to ensure generalisability across different settings; evidence that the AI solution brings measurable gains; the level of trust in and understanding of the AI solution among hospital staff; the budgets and resources available to onboard the AI solution, train staff, and maintain the solution; the need for national policies on funding and regulating AI solutions. These factors affected the adoption, spread, scalability, and sustainability of AI implementation and should be considered for future implementation.

BACKGROUND First Contact Practitioners (FCPs) are increasingly embedded in UK primary care to provide direct access to musculoskeletal (MSK) assessment and management, but face significant variation in governance, scope of practice, and supervisory support. Real-world data feedback that helps clinicians reflect on their practice against evidence-based standards could support safe, high-quality care, but no validated tool of this kind currently exists for this relatively new and structurally complex role. OBJECTIVE To co-design the PRISM (Primary Care Individual Social Norms MSK) dashboard and accompanying guidance document with FCP experts, and to map the resulting metrics against national clinical guidelines and quality indicator frameworks. METHODS We conducted a qualitative co-design study using a modified Nominal Group Process across two structured workshops with 10 purposively sampled participants, including practising FCPs, physiotherapy leaders, and a patient and public involvement representative. Consensus-based ranking identified candidate metrics, quality indicators, and contextual filters, while critical realist thematic analysis of workshop data surfaced the structural and governance conditions shaping how PRISM should function. Following the workshops, the co-designed metric set was independently mapped against 15 NICE musculoskeletal guidelines and the Braybrooke et al. primary care MSK quality indicator framework. RESULTS Participants reached consensus on a core metric set, three categories of contextual filters, and content priorities for an accompanying guidebook. When mapped against the NICE guidelines, all 15 were at least partially supported by PRISM, with 10 directly supported and none unaddressed. Against the 11 Braybrooke quality indicator themes, all were at least partially supported, including 6 of 9 process and structural themes directly supported. Exercise prescription and self-management advice emerged as the most evidence-significant metrics across both frameworks. Participants identified data quality, supervisory variation, and the risk of social norms feedback being misread as performance judgement as key implementation risks, leading to the dashboard being designed to function both within structured clinical supervision and as a standalone reflective practice tool. CONCLUSIONS Co-design with FCP experts produced a dashboard metric set that maps credibly against national evidence-based quality standards, demonstrating that structured co-design with clinical experts is an effective and replicable method for grounding digital health tools in evidence for new and structurally complex clinical roles. The critical realist analysis identified governance variation and supervisory quality as preconditions for safe implementation, which will be tested empirically in an upcoming feasibility trial. CLINICALTRIAL not applicable.

Involving end-users in the development of an AI tool is an important facilitator to its implementation. Usability testing was therefore conducted with a prototype user interface of an Artificial Intelligence driven air-Safety Tool (AISaT) to capture the perspectives and user experiences of AISaT from 10 staff members across two hospitals working within estates, infection prevention and control, and clinical areas, to inform the development of next iterations of AISaT. The perspectives shared could be grouped under improvements to the understand-ability; content; navigation; visibility; usability; workflow; ownership; and frequency of use of the tool. There were key areas that can and will be easily improved within AISaT, however there were areas that required a deeper level of critical reflection, such as incorporating data on more existing variables in a room (i.e., existing ventilation) and determining who is infected, and the level of breathing.

BACKGROUND Artificial intelligence (AI) is increasingly prevalent within healthcare, however, its implementation within the hospital setting may come with its own limitations given the high pressured and sensitive environment. Only a few studies have qualitatively explored barriers and facilitators to the implementation of AI within hospitals and there is a scarcity of qualitative research on the implementation of AI guided tools to improve air quality. OBJECTIVE This study aims to qualitatively explore, through a Normalisation Process Theory (NPT) lens, the anticipated barriers and facilitators to the successful implementation of an AI tool to guide the placement of mitigation strategies to reduce airborne transmission of infections within the hospital setting. Our objectives were to: a) Understand the day-to-day work of key hospital staff involved in reducing airborne transmission of infections; b) Identify potential barriers to the implementation of an AI tool to guide the placement of mitigation strategies in the hospital setting; and c) recognise necessary conditions for the successful implementation of the AI tool within the hospital setting. METHODS We conducted ethnographic observations (n=4) and semi-structured interviews (n=9) with hospital staff following preliminary and wider stakeholder engagement work involving patients, members of the public, national leaders in ventilation working within academic and industry environments, and hospital staff (n=70). Interview transcripts, fieldnotes and summaries from preliminary wider engagement work activities were analysed in NVivo 12, to facilitate triangulation through comparisons between data. Findings were organised according to the core constructs of NPT. RESULTS Hospital staff understanding of the AI guided tool and their preparedness to invest time in its implementation was variable across hospital staff at this pre-implementation stage (coherence and cognitive participation). The work hospital staff will need to do to successfully implement the AI guided tool and its recommended mitigation strategies, is expected to be a significant challenge for the implementation of the AI guided tool at scale given the differences in the ways of working among Trusts and the competition for limited resources (collective action). The effectiveness of the AI guided tool will need to be proved and anticipated concerns around its future refinement addressed (reflexive monitoring). CONCLUSIONS This pre-implementation study suggests that the successful implementation of an AI guided tool advising on the placement of mitigation strategies to improve air quality in the hospital setting depends not only on the effectiveness demonstrated by the AI tool at trial stage but the way outcomes are shared with relevant stakeholders within the hospital setting. While initial resistance by hospital staff is expected, the impact of the AI tool and suggested mitigation strategies in workflows, costs and staff and patient outcomes, together with the way this information is conveyed to hospital staff may be decisive.

Background: Dementia in minority ethnic groups is on the rise. South Asian people represent 9.3% of the UK population and are the largest minority ethnic (ME) group. Most dementia care in South Asian communities is provided at home by family carers, yet there is low uptake of care planning and post-diagnostic support services. This review explored factors influencing care planning and post-diagnostic support for South Asian people with dementia. Methods: Systematic review and thematic synthesis of qualitative studies exploring care planning and post-diagnostic support experiences and views of South Asian people with dementia, family carers, and health and social care professionals (HSCPs). There were no limits on country or date of publication. Six databases were searched (inception - June 2024) and two reviewers independently screened and quality-appraised studies. Reflexive thematic analysis was used to generate themes, which were then mapped to the Socioecological Model. The review protocol was registered on PROSPERO [Registration number: CRD42023404125]. Results: From 3165 studies found, 2069 were screened (after de-duplication) and 24 studies were included for thematic synthesis. Key influential factors were mapped to one of four levels of a modified version of the Socio-ecological Model (SEM): (1) Individual and community level factors - cultural duty for home care, stigma and misinformation, support networks; (2) Interpersonal level factors - language barriers and person- and family-centred care; (3) Organisational and systems level factors - lack of integrated support, cultural competence, system pressures and institutional discrimination; (4) Structural level factors - socioeconomic and policy considerations. Conclusions: Recommendations to improve dementia care planning and post-diagnostic support for South Asian communities include (1) Reframing narratives around dementia and help-seeking through culturally-tailored community interventions; (2) Culturally competent, person- and family-centred care; (3) Holistic and integrated support beyond clinical care; (4) Equitable partnership working with South Asian communities to co-produce dementia services.

Objective: We aimed to determine whether UCLP-PRIMROSE (a care innovation to reduce physical health inequalities for people with severe mental illness) could be set up in the current UK National Health Service (NHS) context and identify the processes, barriers and facilitators to implementation. Design: We employed a convergent mixed methods approach, combining interviews, ethnographic site visits and the collection of meeting notes and uptake data for core model components. Interview transcripts were analysed using reflexive thematic analysis, and all qualitative data, including interview transcripts, were analysed using the Consolidated Framework for Implementation Research. Qualitative work and insights from implementation uptake frequencies were integrated using Normalisation Process Theory. Setting: We evaluated implementation in Yorkshire and three London boroughs, mainly within general practices. Participants: We conducted interviews with 39 staff members who were implementing and/or delivering UCLP-PRIMROSE. Intervention: UCLP-PRIMROSE is an integrated evidence-based care pathway developed to reduce cardiovascular disease risk and mental health relapse in people with severe mental illness. Results: Adaptation and delivery varied in completeness and consistency across 24 general practices and their wider care teams. Factors outside the implementation teams’ influence challenged the embedding of UCLP-PRIMROSE. Factors included the impact from the immaturity of NHS integrated care systems, unintended consequences of the incentivised NHS severe mental illness physical health check and limited capacity for implementing in a system facing resourcing challenges. Drivers of successful implementation included staff being aligned with the values of the UCLP-PRIMROSE model and system leaders acting as champions. Supportive foundational processes acted as facilitators: these included protected and prioritised time for reflection, learning and problem solving. Conclusions: Implementation of UCLP-PRIMROSE was moderately successful in a relatively short period of time. At the end of the research, all teams wanted to sustain delivery. However, further pathway simplification and additional resources are required to spread UCLP-PRIMROSE beyond early pockets of good practice.

Background: By July 2020, large numbers of post-COVID patients were experiencing symptoms for weeks or months, but traditional National Health Service models of rehabilitation service delivery could not meet demand. Objectives: Design and deploy a digital health intervention to provide digitally delivered, remotely supported rehabilitation to long COVID patients on complicated and evolving pathways. Methods: The multidisciplinary team combined established research methods based on engineering and computer science (considering safety, stability and user requirements) with those based on biomedical and health service research (considering effectiveness and population impact). Qualitative data comprised recordings of meetings between study team members and clinicians and semistructured interviews with clinician and patient users. Quantitative data comprised referral, registration and usage rates; demographic and clinical characteristics of patients; and patient-reported outcome measures. Results: We created a modifiable digital health intervention, ‘Living With COVID Recovery ^TM developed by Living With Ltd’, London, UK, that continues to be used by National Health Service trusts. The digital health intervention included integration into a clinical pathway, a clinician-facing dashboard, two-way messaging and a patient-facing app with information and evidence-based treatments. We aimed to register 1000 users. By study completion on 20 December 2022, there were 9781 patients invited, of whom 7679 (78.5%) had registered, at 33 National Health Service clinics. Limitations: Data came from patients at long COVID clinics, however data were unlikely to be representative of people with long COVID. We could not observe clinics under lockdown and had limited access to patient digital health intervention users or to people not engaging with the digital health intervention. Patient user data were incomplete, with inconsistent patient-reported outcome measure and other questionnaire data completion and no data on initial severity of disease, vaccination status, comorbidities or other individual circumstances. Conclusions: Long COVID can be extremely debilitating, comparable to stage IV lung cancer in relation to fatigue and health-related quality of life. Care and rehabilitation should address the management of fatigue and reflect the impact of social disadvantage on symptom severity. With sufficient resources, a digital health intervention can be developed quickly and effectively using agile methodology and bringing together a genuinely multidisciplinary team, including, importantly, an industry partner. Digital health intervention product design and deployment are both important in getting National Health Service trusts, healthcare professionals and patients to engage with a digital health intervention. Projects should work closely with all user groups. Lockdown and the unmet need of a new patient group encouraged those who might otherwise have been reluctant to try a digital health intervention. Many patients and clinics accepted this digital remote support, which helped patients feel cared for while reducing strain on health services. This may encourage acceptance of other digital health intervention, although medical record integration remains a deterrent to clinics. Future work: This research focused on the development, deployment and evaluation of a digitally enabled rehabilitation programme for long COVID. Clinical effectiveness will be assessed within the Symptoms, Trajectory, Inequalities and Management: Understanding Long-COVID to Address and Transform Existing Integrated Care Pathways (UCL, London, UK) study. Funding: This synopsis presents independent research funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme as award number NIHR132243.

Background: Digital health interventions can support health-related knowledge transfer, for example through websites or mobile applications, and may reduce health inequalities by making health care available, where access is difficult, and by translating content to overcome language barriers. However, digital health intervention can also increase health inequalities due to the digital divide. To reach digitally excluded populations, design and delivery mechanisms need to specifically address this issue. This review was conducted during the evolving COVID-19 pandemic and informed the rapid design, deployment and evaluation of a post-COVID-19 rehabilitation digital health intervention: ‘Living with COVID Recovery’ (LWCR). LWCR needed to be engaging and usable for patients and to avoid exacerbating health inequalities. LWCR was introduced as a service into 33 NHS clinics, was used by 7679 patients, and evaluation ran from August 2020 to December 2022. Objective: To identify evidence-based digital health intervention design and deployment features conducive to mitigating the digital divide. Methods: Cochrane Library, Epistemonikos, National Institute for Health and Care Excellence Evidence, PROSPERO, PubMed (with MEDLINE and Europe PMC) and Turning Research into Practice; OpenGrey and Google Scholar were searched for primary research studies published in English from 1 October 2011 to 1 October 2021. Setting and population: Adults who were likely to be affected by the digital divide, including older age, minority ethnic groups, lower income/education level and in any healthcare setting. Interventions: Any digital health intervention with features of design and/or deployment intended to enable access and engagement by the population of focus. Comparators: Any or none. Outcome measures: Any related to participants’ access and/or use of digital health intervention and/or change in digital skills and confidence. Analysis: Data from studies that met the inclusion criteria were extracted, narratively synthesised and thematically analysed. Results: Twenty-two papers met the inclusion criteria. Digital health interventions evaluated included telehealth, text message interventions, virtual assistants, self-management programmes and decision aids. Design themes included: Co-development with end-users, user testing through iterative design cycles, digital health interventions that also helped improve digital skills and digital health literacy, tailoring for low literacy through animations, pictures, videos and writing for low reading ages; virtual assistants to collect information from patients and guide the use of a digital health intervention. Deployment themes included: Free devices and data, or signposting to sources of cheap/free devices and Wi-Fi, text message interventions, providing ‘human support’, providing tailored digital skills education as part of the intervention and enabling peer/family support. Limitations: Our search extended to late 2021, and there has been a massive increase in the literature following the pandemic. However, as our review was undertaken to inform the LWCR digital health intervention design and deployment, we have reported the results that informed this work. The studies included in the review were heterogeneous, so generalisability may be limited. Few randomised controlled trials assessed the digital health intervention’s impact on digital health skills by using validated measures. Conclusions: Using the design and deployment findings described above when developing digital health interventions may help overcome the digital divide. Beyond informing the LWCR digital health intervention development, the review findings have wider implications for the equitable design, delivery and evaluation of digital health interventions. Funding: This article presents independent research funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme as award number NIHR132243.