Long-term condition (LTC) management is a major driver of clinical workload in primary care. Traditional recall processes are often manual, inconsistent and slow, contributing to delays in care and inequalities in patient engagement. In 2025, multiple UK primary care sites implemented an agentic AI-enabled recall system to streamline asthma reviews, reduce unwarranted variation and prioritise patients based on clinical need.To evaluate the impact of an AI-enabled digital recall pathway on: Patient engagement and completion of LTC reviews Administrative workload and appointment demand Consistency of clinical coding and review quality Equity of access across age groups Operational efficiency and capacity releaseA retrospective observational study was conducted using operational data from primary care sites deploying a digital asthma recall pathway. The AI system generated structured questionnaires, analysed responses using condition-specific clinical logic, and produced draft clinical summaries for clinician verification. Outcomes included digital uptake, response rates, appointment utilisation, coding consistency and workflow impact. Descriptive statistics were applied.Among more than 3,000 asthma patients invited digitally, immediate uptake reached 68%, with overall questionnaire completion exceeding 80%. Engagement increased by 39% among patients <25, a cohort with low participation in traditional reviews. Manual call volumes fell by 91%, releasing substantial administrative capacity. Clinical appointment demand reduced by 41%, with all patients still receiving appropriate clinical review. Time from first invite to completion of annual and medication reviews reduced by 46%. Coding quality standardised across all sites, with 0 coding errors and generating high-quality structured data for population-health management.AI-enabled recall significantly improved efficiency, patient engagement and clinical consistency. The shift from manual to automated review pathways freed capacity for urgent care and complex multimorbidity management. Standardised coding improved data reliability and strengthened the foundation for proactive population-health planning. Variation between sites highlighted the importance of workflow mapping and iterative optimisation.Agentic AI-enabled recall offers a scalable model for improving LTC management in primary care systems. By enhancing engagement, reducing workload and standardising clinical processes, this approach strengthens safety, equity and population-health capability. This model demonstrates a scalable approach to strengthening primary care delivery and may inform international efforts to modernise primary care.

Primary care clinicians face high documentation burden, averaging more than 13 hours per week of administrative tasks, contributing to burnout, reduced job satisfaction and after-hours workload. Ambient AI-enabled scribing has emerged as a potential solution to improve efficiency, reduce cognitive load and enhance consultation quality. The programme assessed implementation of ambient AI technology in General Practice to evaluate its effectiveness, safety and patient acceptability.To assess the impact of ambient AI scribing on: Documentation time during and after consultations Clinician wellbeing Quality and accuracy of clinical notes Patient experience and consultation rapport Scalability and safety in routine practiceA mixed-methods observational evaluation was conducted across multiple primary care sites serving approximately 500,000 patients. Quantitative data included utilisation metrics, consultation time saved, documentation time reduction, wellbeing indicators and safety outcomes. Qualitative data included clinician and patient feedback. Evaluation encompassed thousands of AI-enabled consultations during pilot and scaled rollout phases.Across 47 clinicians completing 2,879 AI-enabled consultations in the initial phase, documentation burden reduced substantially: 51% reduction in in-consultation documentation time 61% reduction in documentation completed after hours 63% reported completing clinics faster These effects persisted during wider rollout, with 1.2 million minutes of consultations recorded and approximately 17,000 consultations transcribed each month, saving 530 hours of clinical time monthly. Wellbeing outcomes included: 45% higher satisfaction with work–life balance 58% reduction in documentation-related stress Quality and safety indicators strengthened: note-quality satisfaction increased by 38% confidence in accuracy improved by 33% no safety incidents reported Patient experience was positive, with clinicians reporting improved rapport and patients describing feeling more listened to.Ambient AI scribing substantially reduced administrative workload and improved both clinician wellbeing and documentation quality. Adoption varied between users, highlighting the importance of training and cultural readiness. Improvements were observed across clinicians with varying digital confidence, indicating strong potential for broad adoption. Safety remained high through clinician-in-the-loop verification and structured governance.Ambient AI scribing is a safe, scalable and high-impact intervention that reduces documentation burden and enhances quality of care. This model demonstrates a scalable approach to support workforce sustainability in primary care and may inform international efforts to modernise primary care.

Primary care systems internationally face increasing pressure from rising demand, constrained workforce capacity and growing administrative burden. Traditional telephone and face-to-face access routes struggle to manage high volumes effectively, contributing to delays and unmet need. In 2024 we started widespread implementation of AI-enabled digital triage systems to support more timely, structured and equitable access. Digital inclusion support was provided to ensure equitable access across diverse population groups.To evaluate the impact of an AI-enabled triage system on: 1. Patient demand patterns and automation rates 2. Response times to patient requests 3. Proportion of requests managed within clinically appropriate timeframes 4. Operational efficiency across participating sitesA retrospective observational study was conducted across multiple primary care sites serving approximately 500,000 patients, using routinely collected operational data. All patient-initiated requests submitted through the AI-enabled triage system over a 12-month period were included. Requests were categorised as medical or administrative. Outcomes included automation rate (%requests resolved without human intervention), mean clinician response time, and appointment allocation patterns. Descriptive statistics were used to analyse trends across the study period.A total of 425,500 patient requests (73.6% were medical, 17.1% administrative, 9.3% othe) were processed. Automated clinical decisioning resolved 40% of medical requests, with some sites reaching automation rates >90%. For requests requiring human review, the mean response time from the clinical inbox was 48mins. Across all request types, 76% patients were offered or received appropriate care within one week. Medical requests required an average of 3m42s for completion by clinicians across all pathways. The system enabled redistribution of workload based on urgency and ensured consistent triage outcomes across sites.The introduction of AI-enabled triage demonstrated substantial potential to improve access, reduce administrative burden and improve consistency within primary care. Automation contributed to significant workload reduction, while maintaining rapid response times. The consistency of overall outcomes across varied practice environments suggests strong potential for transferability to other primary care systems.AI-enabled triage systems can enhance responsiveness, consistency and efficiency in large primary care settings. This model demonstrates a scalable approach to strengthening primary care digital delivery and may inform international efforts to modernise primary care.

Artificial intelligence (AI) is experiencing massive growth in today's world. It is gradually becoming established in healthcare. Administrative tasks such as collecting patient histories can be burdensome and tedious for practitioners. However, these tasks are essential for proper patient follow-up. The use of automated tools such as AI could improve the daily lives of doctors if it proves to be effective.Is AI at least as effective as practitioners at gathering patient histories?Prospective study comparing medical histories taken by private general practitioners and by AI during the same initial consultation with a new patient. Practitioners were contacted via social media and professional associations. Medical histories were received anonymously by the investigator. Nabla software was used to standardize the use of artificial intelligence. As this software has a free trial version, no funding was required. A request was made to the local ethics committee, which issued a favorable opinion.204 practitioners submitted a total of 942 patient files with medical histories completed by themselves and by artificial intelligence. With regard to pediatric populations (<18 years of age), no significant differences were found other than in family and allergic histories. The 19-35 and 36-65 age groups showed a difference in medical, surgical, obstetric, and allergy histories, as well as family histories. The 66+ age group showed the same result, with the addition of a difference in long-term illnesses. Artificial intelligence systematically found at least as many, if not more, medical histories than the practitioner.Artificial intelligence is effective at collecting patient histories. It is even superior in many areas, particularly when it comes to collecting family and obstetric histories or non-drug allergies. This tool could save us valuable time during these initial interviews. Better medical record keeping is an essential part of follow-up and prevention. Strength: prospective, anonymous study across different regions, large sample size Weakness: recruitment bias due to voluntary participation, possible reporting bias, Nabla software may have intrinsic collection biasesEven though AI seems promising, we must approach it with caution, respecting ethical rules and keeping the patient's interests in mind.

Large language models tools are increasingly used by clinicians to access medical knowledge, interpret guidelines, and support decision making. Yet most available tools are general purpose, insufficiently validated, not transparent in their sources, and not adapted to national clinical contexts. This raises concerns for safety, trust, and applicability in primary care. There is a growing need for sovereign, clinically aligned AI tools designed specifically for healthcare professionals and rigorously validated. XXX was developed in this context, combining generative AI with retrieval augmented generation across >100 curated French medical sources (HAS, ANSM, scientific societies). Preliminary benchmarking showed XXX achieving high scores (>80/100) and outperforming ChatGPT-5 on the 2023 national medical exam (ECNi) by +5 points, supporting further structured evaluation. The present validation study is conducted with an independent team of clinical researchers from two French university hospitals. The primary objective is to estimate the proportion of XXX answers rated pertinent and exhaustive (Likert ≥4/5) by independent expert physicians. Secondary objectives assess partially pertinent answers (Likert =3), erroneous or off-topic answers (Likert ≤2), severity of hallucinations, inter rater agreement, and performance variations across query types and clinical domains. This retrospective observational diagnostic study uses anonymised real-world questions asked by healthcare professionals using XXX. A stratified random sample will ensure representation of diverse primary care contexts. Retrospective answers will be blindly assessed by two independent experts physicians using standardised Likert scales (third reviewer in case of disagreement). Analyses will include descriptive statistics, score distributions, Cohen’s κ for inter-rater agreement, and exploratory analyses to identify factors associated with lower performance. The target sample (≈150-250 queries) ensures 5-10% precision for the primary estimate. Expert assessments and full analysis are currently being made, and will be completed by April 2026. Results including primary and secondary outcomes and exploratory analyses will be available for presentation at WONCA Europe 2026. This study will provide the first structured clinical validation of a French, retrieval-based conversational AI dedicated to healthcare professionals and aligned with national medical practice. Findings will inform clinicians and policymakers on the opportunities and limits of dedicated medical converational AI tools based on LLMs in primary care.

The use of clinical decision-support systems (CDSS), such as clinical decision rules, algorithms, or machine learning (ML) based applications has gained attention in recent years. However, their adoption and effectiveness may vary across different healthcare systems and settings.This study aims to describe and compare the current use of various decision-support and prediction tools in both homecare and long-term care for older people across health professionals from seven different countriesThis study analysed the survey of a CDSS pilot study in a purposive sample of health professionals working with older adults with complex chronic conditions from seven different countries. The survey included participants’ general background information, their current usage of decision support tools, and attitudes on the potential benefits of CDSS.A total of 151 professionals participated in the pilot study. Most participants were physicians (56.3%) or nurses (37.7%). Half of the participants were from primary care or homecare facilities, while the other half worked in long-term care. Our results show significant variation in the adoption and use of decision-support tools across samples from the seven countries. Most currently used CDSS were for diagnostic purposes or concerned guideline implementation, not aimed at prognostic information. In contrast, prognostic tools were most frequently mentioned by respondents as being valuable improvements to clinical practice.While some country samples reported well-integrated digital health infrastructures and higher CDSS adoption rates, others still face challenges in implementing these. However, we found multiple examples of emerging tools. Our findings highlight the need for improvement of current CDSS and development and implementation of particularly predictive CDSS.This study collected the experiences with CDSS of healthcare professionals in a purposive sample across seven countries.  We found that CDSS usage varied across countries with a limited uptake in general, and especially for predictive CDSS. Participants mentioned multiple examples of CDSS they used, mostly for diagnostic rather than predictive purposes. Nevertheless, many professionals reported that predictive tools informing them about patients’ health and intervention outcomes would be most beneficial. This highlights the need for further implementation of current systems and the development of more predictive CDSS.

Primary care systems worldwide, including in Slovenia, face increasing workload due to rising numbers of consultations, diagnostic demands, and administrative tasks, while the healthcare workforce continues to decline. These pressures reduce patient access, increase waiting times, and contribute to burnout among healthcare professionals. Artificial intelligence (AI)–supported self-triage tools have emerged as a potential digital solution to optimize patient flow, reduce unnecessary visits, and improve safety and efficiency in family medicine.To summarize current evidence on AI-supported self-triage tools, evaluate their potential role in primary care, identify benefits and limitations, and explore considerations for potential implementation in Slovenian family medicine settings.A narrative review of published studies, feasibility assessments, and performance analyses of AI-based symptom checkers and triage tools (including ODISSEE, MayaMD, CAREPOI, QUI, Babylon, and Infermedica VTCR) was conducted. Key dimensions were extracted: diagnostic/triage accuracy, safety, usability, user characteristics, integration challenges, and implementation requirements.AI self-triage tools demonstrate promising but variable performance, with reported triage accuracy ranging from 69% to 97%, high sensitivity for identifying urgent conditions, and strong user acceptability. Some tools show comparable or superior triage accuracy to clinicians in controlled scenarios, while others reveal safety concerns and inconsistent diagnostic reliability. Studies indicate reductions in unnecessary in-person visits (up to 12.5%) and improved patient self-management (up to 25%). Users tend to be younger, more educated, and digitally literate. Successful implementation depends on effective integration with existing IT systems, leadership engagement, and clinician trainingAI-assisted self-triage offers opportunities to reduce primary care workload, enhance early detection of urgent conditions, and empower patients through structured symptom reporting. However, challenges remain regarding diagnostic precision, population-specific reliability, over-triage tendencies, and the need for regulatory and safety frameworks. Slovenian primary care, characterized by increasing demand and workforce shortages, could benefit from such tools, but requires robust clinical validation, national guidelines, and adequate digital infrastructure.AI-based self-triage tools hold substantial potential to support Slovenian primary care by improving efficiency, safety, and patient experience. Before national-level deployment, pilot testing, workforce training, and rigorous evaluation of accuracy, safety, and usability are essential to ensure that such systems complement—rather than replace—the irreplaceable clinical judgement of family physicians.