Benchmarking Artificial Intelligence vs General Practitioners Decision-Making in Same-Day Appointments Triage: A Mixed-Methods Study in UK Primary Care

Background Artificial intelligence (AI) is increasingly used to support clinical decision-making, particularly in primary care triage. However, few studies have benchmarked AI triage tools against general practitioner (GP) assessments in real-world settings. This study evaluated the agreement between an AI-enabled triage tool (Visiba Triage) and GP urgency ratings for same-day appointment requests. Secondary aims included assessing perceptions of safety, accuracy and usability from both clinician and patient perspectives.

A mixed-methods study was conducted using data from patients requesting SDA between January and June 2024. Urgency scores generated by the Visiba Triage AI tool based on a modified Manchester Triage System were compared to GP-assigned ratings using Spearman’s rank correlation and Cohen’s kappa. Ordinal logistic regression assessed associations between demographics and patient satisfaction. Thematic analysis of interviews with eight GPs explored perceptions of the AI tool’s performance.

A total of 649 participants were included in this study. The majority were females and of White ethnicity. There was a strong correlation between AI and GP urgency ratings (ρ=0.796, p 0.001), with 83.7% categorical agreement across eight urgency levels (κ 0.69, p 0.001). The AI system demonstrated safety-conscious design, with a greater likelihood of over-triage whilst rarely under-triaging. No cases deemed non-urgent by AI were later reclassified as emergencies by GPs. Qualitative findings supported the quantitative results, highlighting perceived accuracy and safety. Current limitations include suboptimal integration with patient medical records. Patient satisfaction varied significantly by age, with older adults (60+) reporting lower satisfaction (aOR 0.25, 95% CI 0.12-0.52).

This study demonstrates that AI-enabled triage can closely mirror clinical judgement in a primary care setting, offering a safe, scalable solution to manage demand for same-day care. Safe adoption of AI triage tools in healthcare should include real-world assessment and benchmarking against consensus clinician judgement in real-time. Key Takeaways AI-enabled triage tools can achieve substantial agreement with GP urgency assessments, with 84% categorical concordance and no observed cases of significant under-triage. The AI model demonstrated a safety-conscious design, favouring over-triage to reduce patient safety risks, especially in emergency scenarios. Older adults reported significantly lower satisfaction with AI triage, highlighting the need to address digital literacy and inclusion when implementing such tools. GPs expressed high confidence in AI performance at acuity extremes, particularly for self-care and emergency cases, though noted contextual limitations without EHR integration. This real-world study highlights the potential of AI triage to enhance clinical efficiency, particularly in managing same-day appointment demand in overstretched systems like the NHS. Ongoing clinician oversight remains essential to mitigate AI limitations in complex cases and ensure equitable, safe deployment at scale. Competing Interest Statement KL, AP and HG are employees of Visiba. RR is an employee of WRMP. The other authors did not declare any interests. Funding Statement This research was funded by the Invention for Innovation (i4i) Programme I4i Funding At the Speed of Translation (FAST) NIHR207305.. Austen El-Osta is grateful for support from the National Institute for Health and Care Research (NIHR) Applied Research Collaboration NorthWest London. The views expressed in this article are those of the authors and not necessarily those of the NIHR or Department of Health and Social Care. Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: The study received ethics approval from Imperial College Research Ethics Committee (ICREC #7106912). Participants consented to take part before the start of interviews. Participants were free to withdraw from the interview at any time. Interview data was pseudonymised. The interviews were transcribed with the principle of anonymity in mind and transcriptions were not outsourced, therefore no confidentiality agreements were required. All data generated or analysed during this study are included in this published article. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Data Availability The data that support the findings of this study are available from the corresponding author, AEO, upon reasonable request