AI Lightning Talks 1
Tracks
Room 2
| Tuesday, December 1, 2026 |
| 3:15 PM - 4:15 PM |
| Concurrent Room 1 |
Overview
10-Minute AI Lightning Talks
Details
1. Dr Adeola Bamgboje-Ayodele
2. Dr Bethany Van Dort
3. Siyun (Ruby) Wang
4. Dr Yue Wu
5. Xuheng Yu
6. Sin Hung (Timothy) Yu
7. Prof Melissa Baysari
Speaker
Dr Adeola Bamgboje-ayodele
University Of Sydney
Factors shaping clinician trust in AI‑driven clinical decision support
Abstract
Background and aim
Artificial intelligence (AI)–driven clinical decision support systems (CDSS) have the potential to improve decision‑making in emergency department (ED) settings, particularly for complex and time‑critical presentations such as paediatric respiratory infections [1]. Despite advances in explainable and causal AI approaches, clinician acceptance and sustained use of AI‑enabled CDSS remain limited, with lack of trust frequently cited as a key barrier [2]. As AI‑based CDSS are increasingly introduced into acute care environments, there is an urgent need to understand how clinician trust is established and how it evolves over time in real‑world clinical contexts. The aim of this study was to explore how ED clinicians conceptualise initial trust in, and ongoing trust with, explainable AI‑driven CDSS for paediatric respiratory diagnosis support.
Methods
We conducted a qualitative study involving semi‑structured interviews with 21 ED clinicians recruited across Australia. Interview data were analysed using inductive thematic analysis. Two researchers independently coded the data, with consensus reached through iterative discussions.
Results
Findings indicate that initial trust is shaped by perceptions of model accuracy and validation, provenance and credibility of the tool, explainability of recommendations, endorsement by trusted clinical leaders, and individual clinician attitudes toward AI. Trust over time is influenced by observed patient outcomes, opportunities for audit and feedback, familiarity gained through repeated use, and perceived system performance and usability. Poor usability and technical instability were reported to erode trust, even when clinical outputs were viewed as reasonable.
Discussion and implications
Understanding how trust is established and evolves provides critical insights for the design and implementation of trustworthy AI in emergency care. Designing AI‑driven CDSS that highlights local validation, transparent provenance, layered explanations, and robust usability may support sustained clinician engagement and safe integration into ED workflows.
References
1. Kareemi H, et al. DOI:10.1111/acem.15099
2. Tun HM, et al. DOI:10.2196/69678
Artificial intelligence (AI)–driven clinical decision support systems (CDSS) have the potential to improve decision‑making in emergency department (ED) settings, particularly for complex and time‑critical presentations such as paediatric respiratory infections [1]. Despite advances in explainable and causal AI approaches, clinician acceptance and sustained use of AI‑enabled CDSS remain limited, with lack of trust frequently cited as a key barrier [2]. As AI‑based CDSS are increasingly introduced into acute care environments, there is an urgent need to understand how clinician trust is established and how it evolves over time in real‑world clinical contexts. The aim of this study was to explore how ED clinicians conceptualise initial trust in, and ongoing trust with, explainable AI‑driven CDSS for paediatric respiratory diagnosis support.
Methods
We conducted a qualitative study involving semi‑structured interviews with 21 ED clinicians recruited across Australia. Interview data were analysed using inductive thematic analysis. Two researchers independently coded the data, with consensus reached through iterative discussions.
Results
Findings indicate that initial trust is shaped by perceptions of model accuracy and validation, provenance and credibility of the tool, explainability of recommendations, endorsement by trusted clinical leaders, and individual clinician attitudes toward AI. Trust over time is influenced by observed patient outcomes, opportunities for audit and feedback, familiarity gained through repeated use, and perceived system performance and usability. Poor usability and technical instability were reported to erode trust, even when clinical outputs were viewed as reasonable.
Discussion and implications
Understanding how trust is established and evolves provides critical insights for the design and implementation of trustworthy AI in emergency care. Designing AI‑driven CDSS that highlights local validation, transparent provenance, layered explanations, and robust usability may support sustained clinician engagement and safe integration into ED workflows.
References
1. Kareemi H, et al. DOI:10.1111/acem.15099
2. Tun HM, et al. DOI:10.2196/69678
Biography
Dr Adeola Bamgboje-Ayodele is the Research Lead for the Discipline of Design, and a Lecturer in Design and Innovation at The University of Sydney's School of Architecture, Design and Planning. Prior to this, she was a Research Fellow in Human Factors within the Faculty of Medicine and Health at the University of Sydney where she led a research stream on consumer-facing digital health tools and virtual care technologies. Adeola previously conducted postdoctoral research at the Ingham Institute of Applied Medical Research where she worked on designing and evaluating web portals for people living with and beyond cancer. Her research to date has focused on digital health intervention design, development, evaluation, and implementation to benefit clinicians, patients, caregivers, and health service organisations.
Her work has resulted in a commercialised digital health technology, redesigned systems in Australia's first virtual hospital, changes in policies in New South Wales (NSW), and changes in work practices in NSW hospitals. Her current research interests include applications of inclusive design principles in technologies used by priority populations including older people and people who are culturally and linguistically diverse; user trust in artificial intelligence (AI-driven) tools; and the futures of patient-work.
She is an EMCR co-chair for the International Ergonomics Association's Healthcare Ergonomics Technical Committee, a leadership team member of the Sydney Health Partner's Virtual Care and Geriatric Medicine Clinical Advisory Groups (CAGs) and an Associate Editor for JMIR Human Factors.
Dr Bethany Van Dort
The University of Sydney
Health Professionals’ Perceptions of Explainable AI in Clinical Decision Support
Abstract
Aim: To understand health professionals' perceptions of Explainable AI (XAI) used for clinical decision support systems (CDSS) in hospital settings.
Methods: A scoping review was conducted to synthesise health professionals’ perceptions. MEDLINE, Embase, and Web of Science were searched. Studies were included if they reported qualitative findings on health professionals’ perceptions of XAI-enabled CDSS used in hospital settings. Study characteristics and perceptions were extracted and inductively coded.
Results: Sixteen studies were identified. Most studies focused on tools used for predicting clinical outcomes. Health professionals reported using the explanations in the CDSS to validate model outputs and wanted systems to also include actionable information. Ten studies focused on perceptions of specific explanation techniques, eight of which reported that participants had difficulties interpreting the explanations. Regarding design, participants preferred explanations that included concise, high-level information with further details available on demand. They preferred simple plots for quick interpretation, clear visual indicators such as colour coding, and the aggregation of similar features. Participants also wanted clinical context information presented alongside explanations.
Discussion/ Implications: This review found that explanation techniques and the design of explanations both influence how health professionals interpret CDSS outputs. Explanations are incorporated into systems to improve understanding. Interestingly, this review found health professionals valued explanations not only because they helped them validate outputs, but also for the actionable insights they could provide. However, most XAI systems were not designed for this purpose, which suggests systems without recommended actions may not be perceived as useful. Participants also found some explanations difficult to understand, and reported design aspects that could aid interpretation. If not designed to meet user needs, XAI explanations in CDSS may not improve understanding or may increase cognitive load in high-pressure clinical settings.
Methods: A scoping review was conducted to synthesise health professionals’ perceptions. MEDLINE, Embase, and Web of Science were searched. Studies were included if they reported qualitative findings on health professionals’ perceptions of XAI-enabled CDSS used in hospital settings. Study characteristics and perceptions were extracted and inductively coded.
Results: Sixteen studies were identified. Most studies focused on tools used for predicting clinical outcomes. Health professionals reported using the explanations in the CDSS to validate model outputs and wanted systems to also include actionable information. Ten studies focused on perceptions of specific explanation techniques, eight of which reported that participants had difficulties interpreting the explanations. Regarding design, participants preferred explanations that included concise, high-level information with further details available on demand. They preferred simple plots for quick interpretation, clear visual indicators such as colour coding, and the aggregation of similar features. Participants also wanted clinical context information presented alongside explanations.
Discussion/ Implications: This review found that explanation techniques and the design of explanations both influence how health professionals interpret CDSS outputs. Explanations are incorporated into systems to improve understanding. Interestingly, this review found health professionals valued explanations not only because they helped them validate outputs, but also for the actionable insights they could provide. However, most XAI systems were not designed for this purpose, which suggests systems without recommended actions may not be perceived as useful. Participants also found some explanations difficult to understand, and reported design aspects that could aid interpretation. If not designed to meet user needs, XAI explanations in CDSS may not improve understanding or may increase cognitive load in high-pressure clinical settings.
Biography
Bethany Van Dort is a research fellow in the Digital Health Human Factors Group at the University of Sydney. Her research uses mixed-methods approaches to understand how digital technologies influence clinician workflow, decision-making, and patient safety. Bethany completed her PhD at the University of Sydney, focusing on antimicrobial stewardship in hospitals and the interplay between technology, people, and organisations. Following this, she undertook postdoctoral research at Amsterdam University Medical Centre in the Department of Medical Informatics, focusing on human-centred explainable AI. Bethany is currently working as part of the NHMRC Centre of Research Excellence on Human Factors and AI-enabled Clinical Decision Support. Her research aims to generate practical, evidence-based insights that inform system design, governance, and policy, ultimately contributing to safer and more effective healthcare systems.
Mrs Siyun (Ruby) Wang
University Of Sydney
Interface Features Shaping Clinician Trust and Acceptance of AI‑Enabled CDSS
Abstract
Background
AI-enabled clinical decision support systems (AI-CDSS) are increasingly deployed for diagnosis, risk assessment, and acute clinical decision-making, yet algorithmic performance alone does not ensure appropriate clinician trust or use [3]. Clinicians do not interact with AI models directly; instead, they interpret outputs through interface elements that structure visibility, timing, and explanation [2, 3]. As a result, interface design shapes how clinicians judge system reliability, adapt trust, and decide whether to rely on AI-CDSS recommendations [1]. However, it remains unclear which interface design features influence clinician trust and engagement. This scoping review aimed to: (1) identify interface design features in AI-CDSS; (2) examine how these features influence clinicians' trust; and (3) explore associations between trust‑related interface features and clinicians' acceptance and use.
Methods
PubMed, Embase, and Scopus were searched using four concepts: clinical decision support, explainable artificial intelligence, hospitals, and trust-related outcomes. Two reviewers completed title/abstract and full text screening in Covidence, resolving disagreements by discussion. Data were synthesised narratively.
Results
Of 925 studies retrieved from the databases, thirteen studies met eligibility criteria. Interface features included output presentation, such as output displays, alerts, dashboards, explanations, and recommendations. Across studies, interface design shaped clinicians' ability to integrate AI-CDSS outputs into cognitive workflows, supporting sense-making, prioritisation, and time-pressured decision-making. Interface features influenced trust and acceptance through their interaction with clinical tasks, contexts, and workflow demands.
Conclusion
Interface design is a determinant of how clinicians develop trust in and decide whether to rely on AI-CDSS. However, current reporting of interface features remains inconsistent and underspecified, limiting cross-study comparison. Future research should describe interface design features systematically and examine how design choices shape trust and acceptance to support human-centred and safety-focused AI-CDSS design aligned with clinical work.
References [1] Bussone et al. https://doi.org/10.1109/ICHI.2015.26 [2] Rosenbacke et al. https://doi.org/10.2196/53207 [3] Tun et al. https://doi.org/10.2196/69678
AI-enabled clinical decision support systems (AI-CDSS) are increasingly deployed for diagnosis, risk assessment, and acute clinical decision-making, yet algorithmic performance alone does not ensure appropriate clinician trust or use [3]. Clinicians do not interact with AI models directly; instead, they interpret outputs through interface elements that structure visibility, timing, and explanation [2, 3]. As a result, interface design shapes how clinicians judge system reliability, adapt trust, and decide whether to rely on AI-CDSS recommendations [1]. However, it remains unclear which interface design features influence clinician trust and engagement. This scoping review aimed to: (1) identify interface design features in AI-CDSS; (2) examine how these features influence clinicians' trust; and (3) explore associations between trust‑related interface features and clinicians' acceptance and use.
Methods
PubMed, Embase, and Scopus were searched using four concepts: clinical decision support, explainable artificial intelligence, hospitals, and trust-related outcomes. Two reviewers completed title/abstract and full text screening in Covidence, resolving disagreements by discussion. Data were synthesised narratively.
Results
Of 925 studies retrieved from the databases, thirteen studies met eligibility criteria. Interface features included output presentation, such as output displays, alerts, dashboards, explanations, and recommendations. Across studies, interface design shaped clinicians' ability to integrate AI-CDSS outputs into cognitive workflows, supporting sense-making, prioritisation, and time-pressured decision-making. Interface features influenced trust and acceptance through their interaction with clinical tasks, contexts, and workflow demands.
Conclusion
Interface design is a determinant of how clinicians develop trust in and decide whether to rely on AI-CDSS. However, current reporting of interface features remains inconsistent and underspecified, limiting cross-study comparison. Future research should describe interface design features systematically and examine how design choices shape trust and acceptance to support human-centred and safety-focused AI-CDSS design aligned with clinical work.
References [1] Bussone et al. https://doi.org/10.1109/ICHI.2015.26 [2] Rosenbacke et al. https://doi.org/10.2196/53207 [3] Tun et al. https://doi.org/10.2196/69678
Biography
Siyun Wang is a Master's student undertaking a research course at the University of Sydney, specialising in interactive design and AI in healthcare. Her current research focuses on interface features shaping clinician trust and acceptance of AI-enabled clinical decision support systems.
Xuheng Yu is a co-author and Master's student at the University of Sydney, with research interests in interactive design and AI in healthcare.
Adeola Bamgboje-Ayodele is a Lecturer at the University of Sydney and project supervisor for this research, with expertise in interactive design and AI in healthcare.
Dr Yue Wu
The University of Sydney
Designing xAI user-interface to support pneumonia diagnosis in emergency departments
Abstract
Aim: Distinguishing bacterial from viral pneumonia represents a major diagnostic challenge for clinicians in paediatric emergency departments (ED). We developed an explainable artificial intelligence (xAI) model to deliver transparent diagnostic predictions on the causative pathogen for pneumonia. We sought to apply human factors (HF) principles to translate our xAI-model into a clinical decision support (CDS) tool for ED management of children with suspected pneumonia.
Methods: We surveyed clinicians (n=99) from the Paediatric Research in Emergency Departments International Collaborative (PREDICT) research network to identify design preferences. Semi-structured interviews were conducted (n=21) to gain an in-depth understanding of antibiotic prescribing decision-making. We facilitated a workshop with ED doctors (n=4) to validate understanding of existing workflows and discuss hypothetical workflows. An interactive wireframe was created and feedback sought from five clinicians with diverse ED experiences.
Results: Synergising information gathered via different methods, most participants agreed the xAI-tool should provide information on the likely causative pathogen and how this compared with the baseline, i.e., the cohort of interest on average. Feedback from the wireframe evaluation highlighted the importance of describing the baseline cohort clearly and succinctly in practice. Clinicians indicated a preference for receiving predictions tied to explicit management actions. There were diverse perspectives on how the tool should be used by junior versus senior clinicians. Preferred degree of prediction explanation remains an important area of further research.
Implications: HF principles were used to guide the design of an xAI-based CDS tool for managing children with suspected pneumonia in ED. Our mixed methods approach combined Likert-scale questionnaire, qualitative interview, group workshop and 1-on-1 evaluation to gain complementary insights. We show how a cross-disciplinary team of HF researchers, clinicians, AI developers and user-interface designers can work together to embed complex analytical solutions into high-pressure work environment to meaningfully address clinical needs.
Methods: We surveyed clinicians (n=99) from the Paediatric Research in Emergency Departments International Collaborative (PREDICT) research network to identify design preferences. Semi-structured interviews were conducted (n=21) to gain an in-depth understanding of antibiotic prescribing decision-making. We facilitated a workshop with ED doctors (n=4) to validate understanding of existing workflows and discuss hypothetical workflows. An interactive wireframe was created and feedback sought from five clinicians with diverse ED experiences.
Results: Synergising information gathered via different methods, most participants agreed the xAI-tool should provide information on the likely causative pathogen and how this compared with the baseline, i.e., the cohort of interest on average. Feedback from the wireframe evaluation highlighted the importance of describing the baseline cohort clearly and succinctly in practice. Clinicians indicated a preference for receiving predictions tied to explicit management actions. There were diverse perspectives on how the tool should be used by junior versus senior clinicians. Preferred degree of prediction explanation remains an important area of further research.
Implications: HF principles were used to guide the design of an xAI-based CDS tool for managing children with suspected pneumonia in ED. Our mixed methods approach combined Likert-scale questionnaire, qualitative interview, group workshop and 1-on-1 evaluation to gain complementary insights. We show how a cross-disciplinary team of HF researchers, clinicians, AI developers and user-interface designers can work together to embed complex analytical solutions into high-pressure work environment to meaningfully address clinical needs.
Biography
Dr Yue Wu is a Senior Research Fellow at Health and Clinical Analytics Lab, The University of Sydney. With a transdisciplinary background in mathematics and microbial evolution, she applies advanced modelling to support decision-making across clinical care, public health, and clinical trials. Yue has published in leading journals including Artificial Intelligence in Medicine, PLOS Computational Biology, and Communications Medicine, and currently supervises four PhD students.
Yue's expertise centres on causal modelling, Bayesian networks (BN), and explainable AI (xAI), spanning the full lifecycle from model development using domain knowledge and data to deployment of decision-support tools in real-world settings. Yue has extensive experience supporting decision-making for diverse stakeholders, including clinicians managing serious infections, COVID-19 responses, kidney trials, cerebral palsy, and vaccine safety monitoring. Working closely with the Sydney Children’s Hospitals Network and the Learning Health Unit, Yue has led the development of digital dashboards that transform routinely collected electronic medical record data into actionable insights, enabling clinicians to monitor care, identify gaps, and improve practice.
Yue's work has increasingly focused on establishing scalable research infrastructure and cross-disciplinary methods. She leads interdisciplinary collaborations with AI scientists, human factors researchers, and health economists to bridge critical implementation gaps, advancing methods for transporting and updating xAI models, communicating predictions and uncertainty, and measuring clinical impact. As President (2025) of the Bayesian Network Modelling Association, Yue promotes transparency, reproducibility, and ethical use of modelling in healthcare.
Mr Xuheng Yu
The University of Sydney
Understanding Dynamics of Clinician Trust in AI‑Driven Clinical Decision Support
Abstract
Aim.
Artificial intelligence-driven clinical decision support systems (AI-CDSS) are increasingly used in hospital diagnosis, risk assessment and treatment support, but their safe use depends on how clinicians develop and adapt trust within complex clinical and organizational contexts. Existing reviews conceptualise trust or acceptance as a static outcome, leaving how trust develops or changes during real‑world use unclear [1–3]. This scoping review synthesised evidence on how clinician trust in AI CDSS changes over time in hospital settings.
Methods.
PubMed, Embase, and Scopus were searched using four concept blocks: clinical decision support, explainable artificial intelligence, hospital and trust-related outcomes. Two reviewers completed title/abstract and full text screening in Covidence, resolving disagreements by discussion. Data were synthesised narratively.
Results.
Of 925 records retrieved from databases, fifteen studies were included. The evidence base was heterogeneous, and methodologically diverse. Few studies examined clinician trust directly; instead, trust was commonly inferred through related constructs such as confidence, reliance, perceived safety, transparency, adoption, or non‑use. As a result, explicit temporal evidence on trust was limited. Few studies examined changes in trust over time, but many identified trust related responses at early implementation stages, including pre-implementation and early use. Across these studies, trust appeared to begin forming before full implementation and to be adjusted during early exposure and initial use. Trust remained conditional on multiple interacting factors, including system performance, explainability, workflow integration, and clinical context. While trust and system use were related, usage did not necessarily indicate appropriate or sustained trust.
Discussion.
There is limited direct evidence on evolution of trust in current research on AI-CDSS. Greater methodological attention to timing, exposure duration, and longitudinal trust‑related outcomes is needed to capture how clinicians’ interactions with these systems develop across implementation lifecycle.
References.
[1] Tucci et al. https://doi.org/10.21037/jmai-21-25
[2] Rosenbacke et al. https://doi.org/10.2196/53207
[3] Hua et al. https://doi.org/10.1016/j.artmed.2023.102698
Artificial intelligence-driven clinical decision support systems (AI-CDSS) are increasingly used in hospital diagnosis, risk assessment and treatment support, but their safe use depends on how clinicians develop and adapt trust within complex clinical and organizational contexts. Existing reviews conceptualise trust or acceptance as a static outcome, leaving how trust develops or changes during real‑world use unclear [1–3]. This scoping review synthesised evidence on how clinician trust in AI CDSS changes over time in hospital settings.
Methods.
PubMed, Embase, and Scopus were searched using four concept blocks: clinical decision support, explainable artificial intelligence, hospital and trust-related outcomes. Two reviewers completed title/abstract and full text screening in Covidence, resolving disagreements by discussion. Data were synthesised narratively.
Results.
Of 925 records retrieved from databases, fifteen studies were included. The evidence base was heterogeneous, and methodologically diverse. Few studies examined clinician trust directly; instead, trust was commonly inferred through related constructs such as confidence, reliance, perceived safety, transparency, adoption, or non‑use. As a result, explicit temporal evidence on trust was limited. Few studies examined changes in trust over time, but many identified trust related responses at early implementation stages, including pre-implementation and early use. Across these studies, trust appeared to begin forming before full implementation and to be adjusted during early exposure and initial use. Trust remained conditional on multiple interacting factors, including system performance, explainability, workflow integration, and clinical context. While trust and system use were related, usage did not necessarily indicate appropriate or sustained trust.
Discussion.
There is limited direct evidence on evolution of trust in current research on AI-CDSS. Greater methodological attention to timing, exposure duration, and longitudinal trust‑related outcomes is needed to capture how clinicians’ interactions with these systems develop across implementation lifecycle.
References.
[1] Tucci et al. https://doi.org/10.21037/jmai-21-25
[2] Rosenbacke et al. https://doi.org/10.2196/53207
[3] Hua et al. https://doi.org/10.1016/j.artmed.2023.102698
Biography
Xuheng Yu is a Master of Interaction Design and Electronic Arts student at the University of Sydney. His research interests sit at the intersection of human-computer interaction, health technology, explainable artificial intelligence, and human factors. His current capstone research examines how clinician trust in AI-driven clinical decision support systems develops and changes across different stages of implementation and use.
Mr Sin Hung (Timothy) Yu
University Of Sydney
Digital scribe adoption in clinical practice - a scoping review
Abstract
Background: Digital scribes use automatic speech recognition to transcribe a clinician-patient conversation and use generative artificial intelligence (AI) technology to summarise the transcript into relevant documentation templates.
Aim: To survey and summarise recent literature published on the factors affecting the adoption of digital scribes, and the impacts of digital scribes on patients, clinicians, and health organisations.
Methods: Following the PRISMA-ScR guideline, MEDLINE, CINAHL, Web of Science, SCOPUS, and EMBASE were searched for original research papers or case studies that described and evaluated the implementation or use of digital scribes in real-life healthcare settings. Themes and sub-themes were mapped to the updated Consolidated Framework for Implementation Research (CFIR) and the CFIR Outcomes Addendum.
Results: Of the 4772 studies screened, 29 were included in this review. Digital scribes were generally regarded as acceptable (n=11) and usable (n=8), but concerns over the accuracy of notes generated by the scribes were also reported (n=9). Impacts of digital scribes included improved clinician wellbeing (n=12), reduction in documentation burden (n=18), and improved patient-clinician interaction (n=10). Limited studies examined the impact of digital scribes on broader health organisation outcomes, such as cost and productivity (n=3).
Discussion: The fast pace at which digital scribes are being adopted warrants regular updated reviews. Further studies are required to compare the additional benefits of digital scribes over and above human scribes, if any, the role of patient trust in the adoption of digital scribes, and how digital scribes perform in various clinical settings.
Conclusion: Digital scribes could reduce clinicians’ documentation burden, but their adoption was limited by accuracy concerns. While patient satisfaction with digital scribes appears positive, this should be a focus for future studies.
Aim: To survey and summarise recent literature published on the factors affecting the adoption of digital scribes, and the impacts of digital scribes on patients, clinicians, and health organisations.
Methods: Following the PRISMA-ScR guideline, MEDLINE, CINAHL, Web of Science, SCOPUS, and EMBASE were searched for original research papers or case studies that described and evaluated the implementation or use of digital scribes in real-life healthcare settings. Themes and sub-themes were mapped to the updated Consolidated Framework for Implementation Research (CFIR) and the CFIR Outcomes Addendum.
Results: Of the 4772 studies screened, 29 were included in this review. Digital scribes were generally regarded as acceptable (n=11) and usable (n=8), but concerns over the accuracy of notes generated by the scribes were also reported (n=9). Impacts of digital scribes included improved clinician wellbeing (n=12), reduction in documentation burden (n=18), and improved patient-clinician interaction (n=10). Limited studies examined the impact of digital scribes on broader health organisation outcomes, such as cost and productivity (n=3).
Discussion: The fast pace at which digital scribes are being adopted warrants regular updated reviews. Further studies are required to compare the additional benefits of digital scribes over and above human scribes, if any, the role of patient trust in the adoption of digital scribes, and how digital scribes perform in various clinical settings.
Conclusion: Digital scribes could reduce clinicians’ documentation burden, but their adoption was limited by accuracy concerns. While patient satisfaction with digital scribes appears positive, this should be a focus for future studies.
Biography
Timothy Yu is a PhD candidate in the Digital Health Human Factors Group, Faculty of Medicine and Health, University of Sydney. His PhD project aims to evaluate the implementation of AI digital scribes in public hospital settings in collaboration with the Sydney Local Health District. He is a registered nurse and a health service manager with experience working in Radiology, Public Health, and Health Informatics. Tim is passionate about how innovative digital health technology can be safely implemented in clinical practices, and how these technologies impact patients, and the day-to-day work of clinicians and the health organisation.
Prof Melissa Baysari
Professor Of Human Factors
The University Of Sydney
Optimising AI-enabled clinical decision support using a HFE approach
Abstract
Context and Aim
Many basic clinical decision support (CDS) systems suffer from poor design and implementation, and consequently are worked around, misused, or abandoned by clinician users. Human factors and ergonomics (HFE) principles have been applied to CDS to improve its design and implementation, but application of HFE to Artificial Intelligence (AI)-enabled CDS is in its infancy. AI-enabled CDS leverages advanced data analytics, machine learning, and considers a broader context (i.e. patient data) than basic CDS, presenting potential to transform care delivery. However, there are many risks associated with AI use, and effective CDS relies on AI systems being well-designed, well-understood by end-users, aligned with clinician workflow, and able to function within broader sociotechnical healthcare systems.
We have established a new NHMRC “Centre of Research Excellence (CRE) in Human Factors and AI-enabled CDS”. The aim of this presentation is to outline the objectives for our new CRE, key members, and opportunities for future collaboration and participation.
Approach
Our CRE will:
1. develop a HFE framework and toolkit to support design, implementation, evaluation and ongoing use of safe and effective AI-enabled CDS;
2. apply our HFE framework and toolkit to existing and emerging AI-enabled CDS to ensure it is safe and effective;
3. evaluate and refine our HFE framework and toolkit; and
Discussion and Implications
Our CRE will ensuring AI-enabled CDS systems are designed to fit the capabilities, limitations, needs and preferences of users, and the broader healthcare systems in which they will be deployed.
Our ultimate goal is to establish a national centre for expertise and guidance for healthcare organisations adopting AI-enabled CDS. We will future-proof our HFE approach and ensure it is sustained post-CRE via the building of clinician and early career researcher capacity in health services research and HFE.
Many basic clinical decision support (CDS) systems suffer from poor design and implementation, and consequently are worked around, misused, or abandoned by clinician users. Human factors and ergonomics (HFE) principles have been applied to CDS to improve its design and implementation, but application of HFE to Artificial Intelligence (AI)-enabled CDS is in its infancy. AI-enabled CDS leverages advanced data analytics, machine learning, and considers a broader context (i.e. patient data) than basic CDS, presenting potential to transform care delivery. However, there are many risks associated with AI use, and effective CDS relies on AI systems being well-designed, well-understood by end-users, aligned with clinician workflow, and able to function within broader sociotechnical healthcare systems.
We have established a new NHMRC “Centre of Research Excellence (CRE) in Human Factors and AI-enabled CDS”. The aim of this presentation is to outline the objectives for our new CRE, key members, and opportunities for future collaboration and participation.
Approach
Our CRE will:
1. develop a HFE framework and toolkit to support design, implementation, evaluation and ongoing use of safe and effective AI-enabled CDS;
2. apply our HFE framework and toolkit to existing and emerging AI-enabled CDS to ensure it is safe and effective;
3. evaluate and refine our HFE framework and toolkit; and
Discussion and Implications
Our CRE will ensuring AI-enabled CDS systems are designed to fit the capabilities, limitations, needs and preferences of users, and the broader healthcare systems in which they will be deployed.
Our ultimate goal is to establish a national centre for expertise and guidance for healthcare organisations adopting AI-enabled CDS. We will future-proof our HFE approach and ensure it is sustained post-CRE via the building of clinician and early career researcher capacity in health services research and HFE.
Biography
Professor Melissa Baysari leads the Digital Health Human Factors Group in the Sydney Nursing School at the University of Sydney. Melissa’s work seeks to understand and improve the interactions between healthcare providers, health technologies and organisations, and she is particularly interested in evaluating and ensuring the usability and safety of health technologies, and minimising unintended consequences following implementation. She has published widely in the areas of patient safety, electronic medical record systems and decision support, she has secured over AUD $35Mill in research funding, and her research has resulted in a number of significant changes being made to clinical information systems, as well as to hospital policy and work practices.
