Brian McMillan

Aims: NHS England adopted ‘HeLP‐Diabetes’, a web‐based diabetes self‐management and education service, effective in a clinical trial and implemented it across England as ‘Healthy Living for people with type 2 diabetes’ (Healthy Living). This paper aims to narratively synthesise findings from a programme of mixed‐methods evaluations undertaken alongside the national implementation. Methods: Interviews with programme users and NHS staff, content analysis of the website, analysis of anonymised large datasets and calculation of the costs and health benefits. Results: Service users and health professionals welcomed Healthy Living and made suggestions for improvement. The web content had fidelity to HeLP‐Diabetes and was suitable for supporting self‐management. Similar to other online programmes, there was variation in uptake and retention, and many people spent a short time engaging. Despite these challenges, Healthy Living registrants had modestly lower blood glucose at one year (by −1 mmol/mol (95% CI: −2, −1); −0.1% (95% CI: −0.2%; −0.1%)) compared with usual care; results were better in those using the programme more, and it was cost‐effective. Conclusions: We present the first logic model of a UK diabetes self‐management and education programme. The implications for UK diabetes care include a need for clarity on eligibility for different services, the importance of expanding access to people from minority ethnicities and areas of highest deprivation, and the need to improve retention. A future research agenda includes achieving consensus on a roadmap through diabetes services, how and why people choose (or not) to participate in self‐management services, and which components of these services are most strongly associated with positive health benefits.

Aims: The National Health Service (NHS) offers Healthy Living , an online Diabetes Self‐Management Education and Support service, to help people Self Manage Type 2 Diabetes, with modest 1‐year reductions in HbA1c levels, systolic blood pressure and body mass index. We aim to assess the cost‐effectiveness of Healthy Living compared with usual care. Methods: A patient‐level economic simulation model (UKPDS‐OM2.1) was used to conduct an economic analysis from the NHS perspective over a lifetime horizon, discounted at 3.5%. The model translated risk factor reductions to long‐term cost and quality‐adjusted life year (QALY) estimates. Additional analyses estimated intervention and medication cost consequences. Results: The UKPDS‐OM2.1 predicted a modest improvement in health (0.016 QALYs) and a small reduction in lifetime healthcare costs (−£4 per person). Offsetting with intervention costs (£7 per user) and increased medication costs (£212 per user) yielded a net cost of £216 per user. The cost‐per‐QALY of £13,192 was below the lower NICE threshold (£25,000/QALY), indicating that the programme is cost‐effective. Sensitivity analyses largely demonstrated robustness, but the finding was sensitive to structural assumptions about disease progression post‐intervention with the inclusion of medication costs. Conclusion: Healthy Living appears to be a low‐cost, online intervention for self managing type 2 diabetes, generating modest health benefits that are likely to be considered cost‐effective use of NHS resources. While there is some uncertainty in the cost‐effectiveness and the modelling is based on effectiveness estimated from observational data, improvements in uptake and adherence has the potential to improve the value for money to the NHS.

Summary: Background: Phenotypic age acceleration (PhenoAgeAccel), derived from PhenoAge, and MetaboHealth are composite exposures of biological ageing and metabolic health associated with dementia-related outcomes. Whether these associations are causal and reflect the exposures, constituent biomarkers, or both remains unclear. Methods: This study included UK Biobank participants of White British genetic ancestry. MetaboHealth was derived from nuclear magnetic resonance (NMR) metabolomics and PhenoAgeAccel from clinical biomarkers and chronological age. Genome-wide association studies (GWAS) were conducted for MetaboHealth (n=272 568) and PhenoAgeAccel (n=274 077). Independent genome-wide significant variants were used as genetic instruments in two-sample Mendelian randomisation (MR) with FinnGen all-cause dementia summary statistics. Inverse-variance weighting was the primary MR method. Causal network analysis estimated relationships among constituent biomarkers and dementia. Findings: GWAS identified 126 and 141 independent genome-wide significant variants for MetaboHealth and PhenoAgeAccel, of which 109 and 141 were retained as genetic instruments. MR found no evidence of a causal effect of genetically predicted MetaboHealth (per unit: OR 0·83, 95% CI 0·49–1·42; p=0·51) or PhenoAgeAccel (per year: OR 0·99, 95% CI 0·95–1·02; p=0·44) on all-cause dementia, with consistent findings across sensitivity analyses and robust MR methods. Lower lymphocyte percentage and higher NMR-derived glucose had direct relationships with dementia in the joint constituent-biomarker network. Interpretation: MR provided no evidence that either composite exposure causally influenced dementia. The network prioritised lymphocyte percentage and NMR-derived glucose, supporting examination of composite exposures alongside their constituent biomarkers. Funding: NIHR, UKRI, MRC, UK Dementia Research Institute, Innovate UK, and European Union. Full funding details are provided in the acknowledgements. Research in context: Evidence before this study: We searched PubMed from database inception to July 11, 2026, without language restrictions, using combinations of the terms “dementia”, “Alzheimer’s disease”, “vascular dementia”, “biological ageing”, “phenotypic age”, “PhenoAge”, “PhenoAgeAccel”, “MetaboHealth”, “metabolomics”, “genome-wide association study”, “Mendelian randomisation”, and “causal network”. Previous GWAS characterised the genetic architecture of PhenoAgeAccel and MetaboHealth. Longitudinal studies linked PhenoAgeAccel, derived from PhenoAge, to incident all-cause, young-onset, and late-onset dementia, dementia subtypes, cognition, and brain structure; MetaboHealth was associated with poorer cognitive performance, 10-year cognitive decline, and reduced functional independence. Related PhenoAge-based measures have also been studied in relation to modifiable factors and intervention response. Genetic causal evidence remained limited: one two-sample MR study found no evidence that PhenoAgeAccel affected Alzheimer’s disease or vascular dementia, while previous causal-discovery research placed Phenotypic Age within a dementia network without examining its constituent biomarkers. Added value of this study: To our knowledge, this is the first study to genetically evaluate PhenoAgeAccel and MetaboHealth as composite exposures and then deconstruct their constituent biomarker relationships using causal network analysis in dementia. It also provides the first large-scale GWAS of MetaboHealth using the complete Phase 3 UK Biobank Nightingale dataset. The expanded GWAS identified additional loci and enlarged the available genetic instrument sets, while the network analysis resolved the constituent biomarker structure underlying both composite exposures. Implications of all available evidence: For causal investigation, analysing PhenoAgeAccel and MetaboHealth alongside their constituent biomarkers reveals relationships obscured within the composite exposures. This does not preclude their use for prediction or risk stratification. Lymphocyte percentage and NMR-derived glucose therefore warrant further investigation as potential indicators of immune and metabolic pathways relevant to dementia.

Objective: To examine UK general practitioners’ (GPs) adoption of ambient artificial intelligence (AI) scribes and to assess user-reported error rates, workflow impact and consent practices in primary care. Methods: We conducted a nationwide online mixed-methods survey of GPs recruited via Doctors.net.uk . Detailed analyses of use, errors and workflow impact focused on current users of ambient AI scribes. Results: In August 2025, of 1003 respondents, 14% (n=141) reported current use of ambient AI scribes, 39% (n=396) intended to adopt them soon and 46% (n=466) had no plans to use them. Among users (n=141), Heidi Health predominated (86%). Most reported efficiency gains: 80% (n=112) reported reduced time spent on documentation and 70% (n=99) reduced cognitive load. Documentation quality was judged positively, with 55% (n=78) rating outputs as better than standard notes. Errors were common but usually minor: 32% (n=45) reported errors often/always, including 14% (n=20) with significant-to-critical implications. Errors were most frequent in multiparty consultations (38%), complex histories (35%) and non-English encounters (31%). Consent practices varied: 63% (n=89) routinely sought consent, with ≤10% of patients declining. Free-text responses (21% of users) highlighted benefits for workflow, alongside concerns about accuracy, ethics and system integration. Discussion: Findings suggest that ambient AI scribes deliver meaningful efficiency gains and improved perceived documentation quality, but introduce non-trivial risks related to accuracy, equity and medicolegal accountability. The uneven performance in complex and multilingual consultations raises particular concerns about potential exacerbation of healthcare disparities. Conclusion: Ambient AI scribes are already in use across UK primary care. Proactive regulation, consistent consent practices and independent evaluation, including patient perspectives, are urgently needed to ensure safe, equitable and sustainable implementation.

Background: The prevalence of type 2 diabetes (T2DM) is increasing rapidly in the UK and worldwide and is linked to adverse outcomes including premature death. Previous studies have shown that developing self-management skills in this population can lead to health improvements. The National Health Service in England has implemented Healthy Living (HL), an online Diabetes Self-Management Education and Support (DSMES) intervention, offering information about T2DM and help with adopting healthy behaviours. Aims: To examine the uptake and retention of people living with T2DM registered with Healthy Living and how its use is associated with changes in 1-year clinical outcomes (effectiveness) compared with people with T2DM who did not register (controls). Methods: Anonymised linked patient-level Healthy Living and National Diabetes Audit (NDA) data were used to identify adults with T2DM in England. Multivariable logistic regression models identified predictors of participation in Healthy Living (uptake). Using 1-to-5 case-control propensity score matching (on age, sex, baseline HbA1c, body mass index (BMI), blood pressure (BP), cholesterol, ethnicity, deprivation) and multivariable linear and logistic models, we examined how Healthy Living use was related to 1-year HbA1c, BMI, BP, new insulin use, and completion of eight care processes recommended for people with diabetes (effectiveness). Several sensitivity and sub-group analyses were conducted to assess the robustness of the findings. Results: A total of 21,820 people with T2DM activated a Healthy Living account. Compared with non-participants, account activators (cases) were more likely to be female (OR: 1.91; 95%CI: 1.85, 1.96), less likely to be Asian (OR: 0.35; 95% CI: 0.33, 0.37) or Black (OR: 0.56; 95% CI: 0.52, 0.60) compared with white people. Assessing effectiveness, 4,940 Healthy Living cases were matched to 24,685 NDA controls. Compared with controls, at 1-year, Healthy Living cases had lower HbA1c (by −1.3 mmol/mol (95% CI: −1.7, −0.8) or −0.1% (95% CI: −0.2, −0.1)); BMI (−0.2 kg/m ² ; 95% CI: −0.3, −0.1), systolic BP (−1.2 mmHg; 95% CI: −1.6, −0.7), and diastolic BP (−0.6 mmHg; 95% CI: −0.9, −0.3); higher odds of completing care processes (OR: 1.6; 95% CI: 1.5, 1.8), but non-significant for insulin use (OR: 1.0; 95% CI: 0.8, 1.2). The results of the sensitivity and sub-group analyses were consistent with the main findings. Conclusions: People who activated a Healthy Living account experienced, on average, moderate health benefits compared with non-participants. These benefits would be expected to reduce risks for diabetes-related complications at a population level.

Objective: Meaningful assessments of how large language models (LLMs) incorporate clinical guidelines require large-scale testing over many queries. Here, we evaluate the prevalence of clinical guideline omissions and hallucinations in a large sample of diagnostic LLM outputs. Methods: We used simulated case vignettes and zero-shot prompting to generate diagnostic outputs and rationales from GPT-4.1 and DeepSeek-V3. English case vignettes were created for hypercholesterolaemia and type-2 diabetes mellitus. Each vignette contained identical medical information, while sociodemographic characteristics varied in terms of sex, ethnicity and location. We calculated the prevalence of existing and hallucinated clinical guidelines in LLM outputs across disease, LLM and sociodemographic characteristics. Results: We analysed a total of 12 197 LLM outputs, which quantifies three hazard areas: omissions (up to 97% for DeepSeek-V3 and 46% for GPT-4.1), hallucinations (up to 9%) and inconsistencies (guideline citation rate ranging from 0% to 78.39% across sociodemographic vignettes). Omission and hallucination rates were generally similar across vignettes with different sex or ethnicity data, yet were particularly sensitive to patient location. Discussion: This study highlights significant variability in clinical guideline prediction across two different diseases, three different sociodemographic variables and two LLMs, even when the LLMs were instructed by identical prompts, establishing clinical guideline prediction in LLM outputs as a stochastic event. Conclusion: The stochastic nature of LLMs creates a unique challenge for evidence generation and clinical deployment. Being able to measure and capture this stochasticity within high-quality research designs will be a prerequisite to advancing the responsible deployment of LLMs in healthcare.

Aims: The study aimed to assess feasibility, acceptability and safety of an ILED in a randomised controlled trial of ILED vs. best NHS care healthy diet and exercise advice in women with Gestational Diabetes Mellitus (GDM) with body mass index ≥27.5 kg/m ² (≥25 kg/m ² in high‐risk minority ethnic groups) in the last trimester of pregnancy. Methods: Participants were randomised (1:1) to receive an ILED (1000 kcal on two non‐consecutive days of the week with 5 days/week of best NHS care healthy diet and exercise advice) or best NHS care healthy diet and exercise advice 7 days/week from recruitment (24–30 weeks gestation) until delivery (BNC). Primary outcomes were uptake, retention and adherence to the interventions. Maternal and neonatal safety outcomes were explored, and secondary outcomes assessed fidelity of the delivery of the intervention and completeness of study data across all trial participants. Results: Participants were randomised to ILED ( n = 14) or BNC ( n = 12) and 73% ( n = 19; ILED n = 10, BNC n = 9) received their allocated intervention. There was low study retention (58%; ILED n = 7; BNC n = 8) and low adherence to the ILED median (IQR) 37% (8%–50%) out of prescribed days; however, the study achieved a feasible uptake rate (31%; n = 26). There were no adverse events specifically related to the ILED. Conclusion: Our findings suggest that ILED may not be feasible in the third trimester in people with GDM. Future research should explore alternative interventions and interventions earlier in pregnancy for the prevention and management of GDM. Trial registration number : NCT05344066

Background Mental health conditions are a major global challenge with rising demand for accessible, effective, scalable treatments. Digital peer support interventions are a promising way to extend support beyond clinical settings, but their effectiveness requires comprehensive evaluation. Objective To evaluate the effectiveness of digital peer support interventions in improving clinical symptoms, functioning, and treatment engagement among individuals with mental health conditions in outpatient settings. Study selection and analysis: We conducted a systematic review and random-effects meta-analysis of controlled interventional studies. Five databases (MEDLINE, CENTRAL, Embase, PsycINFO) were searched up to January 2025. Studies evaluated digital peer support via online platforms, mobile apps or digital communities for people aged ≥16 years with mental health conditions. Outcomes included clinical symptoms (depression, anxiety), functioning (quality of life, social functioning) and treatment engagement. Risk of bias was assessed using Cochrane Risk of Bias 2.0 for randomised controlled trials and ROBINS-I for non-randomised studies. Certainty of evidence was assessed using GRADE (Grading of Recommendations Assessment, Development and Evaluation). Findings: 29 studies including 5825 participants were included. Digital peer support was associated with small-to-moderate improvements in symptoms of depression (standardised mean difference (SMD) −0.28; 95% CI −0.42 to −0.14) and anxiety (SMD −0.47; 95% CI −0.68 to −0.27). Functional outcomes improved modestly: social functioning (SMD 0.18; 95% CI 0.07 to 0.29), quality of life (SMD 0.14; 95% CI 0.02 to 0.26), patient activation (SMD 0.39; 95% CI 0.23 to 0.55) and personal recovery (SMD 0.23; 95% CI 0.11 to 0.35). No significant effects were observed for treatment engagement or satisfaction. Preliminary evidence suggested sustained benefits for depression, anxiety and social functioning. Conclusions and clinical implications: Digital peer support offers modest improvements in symptoms and functioning for individuals with mental health conditions and may be considered as an adjunct to usual care to enhance engagement and provide accessible support between clinical contacts. Key priorities include establishing optimal intervention models, clarifying longer-term benefits, and ensuring these approaches can be delivered safely and sustainably within routine outpatient services. PROSPERO registration number: CRD42023445194.

Background: Post‐traumatic stress disorder (PTSD) affects approximately 1 in 20 individuals, with primary care playing a crucial role in early identification and management. However, PTSD is frequently underdiagnosed in primary care settings. The COVID‐19 pandemic has exacerbated the incidence of PTSD, leading to an increased demand for accessible mental health services. While NICE guidelines recommend trauma‐focused therapies, access remains limited, underscoring the need for alternative strategies to support management within primary care. Aim: To optimise and refine through co‐production the PTSD Hub app, an existing digital resource, to better support individuals with PTSD in primary care settings, improving usability, clinician engagement, and patient outcomes. Design & Setting: A co‐production approach was employed with stakeholders, including people with lived experience of PTSD, informal carers, and primary care professionals. Workshops were held online to gather insights on how to enhance the PTSD Hub app for better integration into primary care. Method: Two Nominal Group Technique workshops were conducted to identify key barriers to PTSD care in primary care and generate solutions for improving the app. Participants ranked and prioritised features based on relevance to primary care. Results: Twenty‐six participants attended the workshops, representing a diverse mix of stakeholders. Key barriers identified included stigma, clinician time constraints, and limited access to trauma‐focused therapies. Solutions included improving symptom tracking, enhancing customisation features, and raising clinician awareness. Conclusion: The study highlights the potential of the PTSD Hub app in improving and supporting PTSD management in primary care. Future research should evaluate the app's effectiveness in real‐world primary care settings. Patient and Public Contribution: Patients and members of the public were actively involved throughout this study using a co‐production approach. A lived experience advisory panel of six individuals with PTSD provided input at all stages, including shaping the study design, reviewing participant materials, and advising on workshop facilitation. Additionally, 26 participants, comprising individuals with lived experience of PTSD, informal carers, and healthcare professionals, contributed to two Nominal Group Technique workshops. A lived experience research partner (NC) was an integral part of the research team, attended all team meetings, co‐facilitated workshop sessions and contributed to design, analysis and manuscript preparation.