Ian A. Scott

We conducted an exploratory mixed‐methods study of clinician attitudes and editing behaviours towards artificial intelligence (AI)‐generated discharge summaries. Five clinicians reviewed three summaries and indicated what changes they would make and why, answered a questionnaire, engaged in semi‐structured interviews and viewed screen mock‐ups of additional editing features. All clinicians expressed positive sentiments and saw AI as saving time. Edits were driven by professional pride and etiquette, most of which were not corrections of factual errors but optimisation of structure, coherence, conciseness, reasoning and tone. Tension arose between desire for personalised content versus potential benefits of standardisation. Various novel editing features were seen as expediting summary review.

AI-enabled clinical decision support systems (CDSS) have the potential to support the detection of patient deterioration, but their value depends on outputs that fit clinical workflows, support reasoning, and minimise alert fatigue. Little is known about clinician perspectives on Multi-Output Early Warning Systems (MO-EWS), which provide simultaneous predictions (e.g., overall deterioration, organ- or diagnosis-specific). This study explores clinician views on potential future MO-EWS and offers user-centred recommendations that may inform the design, usability, and adoption. The qualitative method of reflexive thematic analysis (RTA) was used to examine clinicians’ perceptions of existing Early Warning Systems (EWS) and envisioned MO-EWS tools for predicting clinical deterioration. Between January and June 2025, 22 clinicians, 11 doctors and 11 nurses, from Australian hospitals with experience in managing clinical deterioration participated in semi-structured interviews using an interview guide. Interview transcripts were analysed to generate codes and themes. Themes were mapped to system components (user interface (UI), Artificial Intelligence (AI) model, and clinical workflow) to develop component-level recommendations and assess their feasibility. Participants found existing EWS sometimes distracting and unsafe. They envisioned AI-based systems that provide multiple, contextualised outputs (e.g., causes of deterioration, organ dysfunction, clinical severity) as layered, adaptable tools that could support reasoning, prioritisation, and communication. While many UI improvements were perceived as achievable with current technologies, we noted that model development is more complex and that workflow integration would likely depend on infrastructure, policies, and clinician preferences. Clinicians in this study perceived potential benefits in MO-EWS that provide richer information when predicting clinical deterioration in individual patients. However, we also identified several design and implementation challenges. The resulting component-level recommendations represent provisional design hypotheses that require future prototyping and rigorous contextual evaluation before their impact on clinical reasoning or system adoption can be determined.

Background: Current guidelines favour preoperative withholding of angiotensin‐axis blockers (AABs) in patients without hypertension or systolic heart failure to decrease peri‐operative cardiorenal risk, although evidence of such effects is lacking. Objective: To assess rates of postoperative cardiovascular events with continuing or withholding AABs before elective, noncardiac surgery. Methods: A single‐center, single‐blind randomized controlled trial was conducted in a tertiary adult hospital. Consecutive patients assessed at a preoperative admission clinic and receiving AABs at a stable dose for at least 28 days were enrolled and randomized in a 1:1 ratio to continue or withhold AABs for 24 h preoperatively. The primary outcome was a composite of major adverse cardiovascular events (MACE) at 5 days postoperatively. Secondary outcomes included intraoperative hypotension, acute kidney injury (AKI), and length of stay. Results: Among 667 randomized patients with 29 postrandomization exclusions, 325 continuing AABs (control), and 313 having AABs withheld (intervention) were included in the final analysis. Mean age was 66 (SD+/−10) years; 64% were male. Primary outcome occurred in 9 of 325 (2.8%) control and 9 of 313 (2.9%) intervention patients (0.1% increase [95% confidence interval (CI): −2.5% to 2.8%]; p = 1.00); intraoperative hypotension occurred in 204 (63.5%) and 190 (60.7%) (2.8% decrease [95% CI: −5.5% to 9.6%]; p = 0.63) and AKI occurred in 24 (7.4%) and 15 (4.8%) (2.6% decrease [95% CI −1.1% to 6.3%]; p = 0.19). No significant between‐group differences were seen for secondary outcomes or subgroup analyses. Conclusions: In this study of patients with low cardiac risk undergoing elective, noncardiac surgery, withholding AABs for 24 h preoperatively did not reduce MACE, intraoperative hypotension or AKI.

Objectives: Assess effects of an ambient artificial intelligence (AI) scribe on documentation time, note quality, clinician and patient experience in clinician–patient encounters. Methods: Prospective, multisite, mixed-methods, before-and-after cohort study of an ambient voice to text AI scribe generating clinical notes conducted in four Australian hospitals involving 58 clinicians comprising doctors (n=39), allied health professionals (n=14) and nurses (n=5) with 621 and 338 patients undergoing non-scribe and scribe encounters, respectively. Results: Median total encounter time per encounter overall significantly reduced by 2.0 min (p<0.001); larger reductions for mental health (8.5 min; p=0.001) and allied health (11.8 min; p<0.001) encounters. Note quality averaged ≥3.8 on 5-point Likert scale (1=not at all; 5=extremely good) across 9 quality axes. Scribe use significantly improved 14 of 16 dimensions of professional fulfilment and burnout, with most clinicians reporting the scribe improved documentation efficiency (80.0%) and note quality (57.2%). Most patients reported better encounter quality (64.1%), clinicians listening more (75.2%), providing more explanation (74.5%) and being more empathetic (71.2%). Conclusions: Using an ambient AI scribe for clinician-patient encounters reduces documentation burden, enhances clinician fulfilment and satisfaction and improves quality of interactions for patients. Effects of longer-term scribe use on professional efficiency and well-being, and patient care and outcomes warrant investigation.

Introduction: Risk prediction tools (RPTs) use clinical data to estimate the likelihood of adverse events such as medication harm and institute preventive strategies. However, their adoption in hospitals can be limited by implementation challenges, workflow integration barriers, and technical constraints. Objectives: This study aimed to determine clinicians' and informaticians' perspectives on the requirements, factors influencing adoption of digital RPTs for medication harm in hospitals, and to identify effective implementation strategies. Methods: Semi‐structured interviews were conducted with clinicians and informaticians in three Australian digital hospitals within the Metro South Hospital and Health Service in 2025. Themes were inductively identified and deductively mapped to the Integrated Promoting Action on Research Implementation in Health Services (i‐PARIHS) framework. Results: Twenty‐two interviews were conducted with six doctors, eleven pharmacists, and five informaticians. Five themes were identified, reflecting key influences on the implementation of digital RPTs in hospitals, including clinician trust, perceived utility, and organisational readiness. Desired features such as automation, feedback functionality, and workflow compatibility were seen as essential to adoption. While enablers included comprehensive feedback mechanisms and adequate resource planning, barriers included poor integration with existing systems and uncertainty around tool reliability. Conclusion: Stakeholders recognised the value of digital RPTs in supporting medication safety. Understanding their perceptions and experiences is necessary for addressing barriers and enhancing tool acceptability. This study highlighted desirable tool features and organisational factors that influence successful implementation. These findings can inform strategies for effective implementation, adoption, and sustained use of RPTs in hospital settings.

The Royal Australasian College of Physicians (RACP) has recently issued a position statement titled ‘Using artificial intelligence in clinical practice’, which provides a framework for the safe and ethical integration of artificial intelligence (AI) into physician practice. The statement outlines core principles for implementing, interacting with and monitoring AI tools to ensure they enhance rather than replace clinical judgement, protect patient safety and privacy and abide by ethical and regulatory standards. It identifies important benefits, such as increased administrative efficiency and improved support for clinical decision‐making, while cautioning against risks such as algorithmic bias, data breaches and medicolegal liabilities. The statement also offers enabling strategies to overcome various barriers such as biased or inaccurate tools, professional and public resistance to the use of AI and adverse environmental impacts. The RACP advocates for physicians to engage in AI development and implementation using an agile human‐in‐the‐loop approach that prioritises patient safety, clinical governance, professional training and collaboration with stakeholders.

BACKGROUND: Safe and effective antidepressant deprescribing strategies are needed in general practice where most antidepressant prescribing occurs. METHODS: We conducted a pragmatic, cluster-randomised controlled trial in general practice to test invitation to general practitioner (GP) review combined with resources to inform shared decision-making and guide hyperbolic tapering for stopping antidepressants compared to usual care. Adults taking antidepressants for longer than 12 months were recruited from 26 Australian GP practices between March 2023 and November 2024, irrespective of their intention to stop or depression or anxiety symptom scores. The primary outcome was cessation at 12 months. Secondary outcomes included cessation at 6 months, and > 75% dose reduction and depression, anxiety and withdrawal symptom scores at 6 and 12 months. RESULTS: Overall, 483 patients were randomised. Mean age was 50 years; 73% were women; mean duration of antidepressant use was 14.1 years. Cessation at 12 months was observed in 32 of 215 (14.9%) intervention and 16 of 187 (8.6%) usual care patients (odds ratio (OR) = 1.95 [95%CI, 1.00 to 3.81]; p=0.050). Cessation at 6 months was observed in 11.7% intervention vs 4.8% usual care (OR = 2.68; 95%CI, 1.18 to 6.05), and > 75% dose reduction at 12 months in 19.6% intervention vs 9.9% usual care (OR = 2.28; 95%CI, 1.20 to 4.31). Symptom scores were similar between groups. No adverse events were attributable to the intervention. CONCLUSIONS: In general practice, invitation to GP antidepressant review combined with information and guidance on hyperbolic tapering can support cessation or dose reduction without causing adverse effects or relapse. Absolute cessation rates were modest but still meaningful given the high prevalence of long-term antidepressant use. TRIAL REGISTRATION: ANZCT registry identifier, ACTRN12622001379707p. WHAT IS ALREADY KNOWN ON THIS TOPIC: Long-term (>12 months) antidepressant use is common in primary care, despite being potentially unnecessary and associated with increased risk of falls, sexual dysfunction, weight gain and diabetes. Lack of recognition and management of antidepressant withdrawal symptoms and fear of relapse are important drivers of long-term use. Safe and effective antidepressant deprescribing strategies are needed in primary care where most antidepressant prescribing occurs. WHAT THIS STUDY ADDS: The RELEASE intervention comprised invitation to GP antidepressant review combined with patient-centred resources to inform shared decision-making and guide hyperbolic tapering. The intervention increased cessation or dose reduction, without increasing adverse effects including relapse, in a challenging population with mean duration of antidepressant therapy of 14.1 years and no intention to stop.

Artificial intelligence (AI)‐enabled risk prediction models are being increasingly used in hospital practice to predict patient risk of adverse events, aimed at giving early warning of impending events and facilitating preventive intervention. However, evidence of beneficial impact varies which may be due to clinicians not finding them useful because of suboptimal predictive accuracy (effectiveness), burden of false alerts (efficiency) or narrow prediction windows (utility). Current model performance measures do not capture the interdependency of these three dimensions. In this commentary, using sepsis risk prediction as an example, we present methods that assist clinicians to: assess the suitability of particular models and decide ideal decision thresholds; optimise model performance; and configure and deliver alerts to frontline clinicians.

Diagnostic error, defined as missed, wrong, or delayed diagnoses or those not communicated to patients, is common, affecting 5–10 % of hospital admissions and clinic visits. Such errors cause patient harm in up to 1 in 100 of such encounters and account for 10 % of all hospital deaths and serious adverse events. About 80 % of diagnostic errors are potentially preventable, most resulting from flaws in clinician reasoning in formulating and testing diagnostic hypotheses. The advent of artificial intelligence (AI), and large language models (LLMs) in particular, has attracted great interest in how these technologies can reduce diagnostic error within the context of bedside or clinic consultations. This narrative review aims to provide practising clinicians with a comprehensible analysis of where AI and LLMs are currently positioned in assisting diagnostic performance in clinician-patient encounters based on contemporary state-of-the-art research. It attempts to answer seven questions relevant to clinician understanding and adoption of AI/LLMs. It concludes that AI tools have matured to the extent that they can improve diagnostic decision-making of clinicians and can assist institutions in increasing diagnostic safety. The rapid development of LLMs and ongoing release of new versions necessitate continuous monitoring of their evolving diagnostic capabilities. Importantly, a balanced approach is required where LLMs work to complement, rather than replace, the nuanced diagnostic reasoning of clinicians.