2D - "AI Innovation and Applications in Public Health"
Tracks
Track 4
Case studies on AI use in public health
Educating our workforce for AI
Enabling infrastructure for AI (sharing data, federated learning, etc.)
Equity and Ethics (privacy, accessibility, information bias, confidentiality and security)
Projects and programs harnessing AI for public health benefits
| Wednesday, November 11, 2026 |
| 1:30 PM - 3:00 PM |
Speaker
Mrs Jodie Fitzpatrick
Senior Health Promotion Officer
Prevention Strategy Branch, Population Health Division, Queensland Health
CareConverse: Using AI Simulation to Strengthen Antenatal Care Behaviour Change Communication Skills
Abstract
Background and Aim
Antenatal care clinicians often lack formal training in behaviour change techniques, such as motivational interviewing and brief intervention, despite these skills being essential for addressing sensitive health behaviours during pregnancy including smoking, poor nutrition, physical inactivity and alcohol use. This gap in knowledge and confidence can result in missed opportunities to support positive behaviour change, contributing to adverse maternal and infant outcomes and increased risk of chronic disease.
Methods and Analysis
CareConverse is an AI-enabled training program designed to enhance clinicians’ person-centred communication and evidence-based behaviour change skills. Co-designed with Clinical Midwife Consultants and subject matter experts, the program features four customised clinical scenarios addressing smoking cessation, nutrition and physical activity, immunisation, and alcohol consumption during pregnancy. Each scenario includes pre-brief educational videos, AI-simulated conversations with virtual clients, real-time feedback, and structured reflection activities. Evaluation, guided by the RE-AIM framework, assesses the program’s impact on clinician knowledge, confidence, and practice.
Outcomes
A six-month evaluation identified 505 registrants, with 47 (9%) engaging in the training and completing an average of 1.3 out of 4 activities. Despite low engagement, feedback was positive: 73% of participants agreed the simulated environment provided a safe space for practising difficult conversations, 84% reported increased confidence in addressing sensitive topics, and 81% noted significant improvements in their understanding of behaviour change principles.
Conclusion and Future actions
These findings highlight the potential of AI-driven simulations to enhance professional development by offering scalable, flexible and practical training in communication skills. However, barriers to engagement, including limited time during working hours, lack of access to private, well-equipped spaces, and challenges in adapting to AI tools, must be addressed to maximise program uptake and impact. CareConverse demonstrates a promising approach to addressing critical gaps in clinician skill development, ultimately supporting the delivery of person-centred antenatal care.
Antenatal care clinicians often lack formal training in behaviour change techniques, such as motivational interviewing and brief intervention, despite these skills being essential for addressing sensitive health behaviours during pregnancy including smoking, poor nutrition, physical inactivity and alcohol use. This gap in knowledge and confidence can result in missed opportunities to support positive behaviour change, contributing to adverse maternal and infant outcomes and increased risk of chronic disease.
Methods and Analysis
CareConverse is an AI-enabled training program designed to enhance clinicians’ person-centred communication and evidence-based behaviour change skills. Co-designed with Clinical Midwife Consultants and subject matter experts, the program features four customised clinical scenarios addressing smoking cessation, nutrition and physical activity, immunisation, and alcohol consumption during pregnancy. Each scenario includes pre-brief educational videos, AI-simulated conversations with virtual clients, real-time feedback, and structured reflection activities. Evaluation, guided by the RE-AIM framework, assesses the program’s impact on clinician knowledge, confidence, and practice.
Outcomes
A six-month evaluation identified 505 registrants, with 47 (9%) engaging in the training and completing an average of 1.3 out of 4 activities. Despite low engagement, feedback was positive: 73% of participants agreed the simulated environment provided a safe space for practising difficult conversations, 84% reported increased confidence in addressing sensitive topics, and 81% noted significant improvements in their understanding of behaviour change principles.
Conclusion and Future actions
These findings highlight the potential of AI-driven simulations to enhance professional development by offering scalable, flexible and practical training in communication skills. However, barriers to engagement, including limited time during working hours, lack of access to private, well-equipped spaces, and challenges in adapting to AI tools, must be addressed to maximise program uptake and impact. CareConverse demonstrates a promising approach to addressing critical gaps in clinician skill development, ultimately supporting the delivery of person-centred antenatal care.
Biography
Jodie Fitzpatrick is a dedicated Senior Health Promotion Officer with Prevention Strategy Branch, Queensland Health bringing extensive experience and passion to the field of public health. With a strong focus on chronic disease prevention, Jodie is committed to improving health outcomes through evidence-based strategies and innovative approaches.
Her professional interests centre on maternal and child health, where she works to support healthier pregnancies and early childhood development as a foundation for lifelong wellbeing. Jodie’s expertise spans program development, stakeholder engagement, and the implementation of initiatives that address key public health challenges.
Jodie is particularly driven by the opportunity to empower individuals and communities to make informed health choices, reduce risk factors, and enhance quality of life. Her work reflects a deep commitment to creating sustainable, population-level health improvements.
Dr. Edmund Kanmiki
Research Fellow
The University Of Queensland
Artificial Intelligence-Driven Personalised Nutrition Programs: A Systematic Review of Real-World Implementation Effectiveness
Abstract
Introduction: Personalised nutrition is widely recommended for the prevention and management of diet-related chronic conditions, including obesity, type 2 diabetes, and cardiovascular diseases. However, shortages of dietitians and nutritionists, costs, and logistical barriers hinder widespread implementation, particularly in low-resource settings. While AI technologies have the potential to revolutionise personalised nutrition, there is limited evidence of their effectiveness in real-world settings. This systematic review evaluates the implementation of AI-powered personalised nutrition interventions and identifies operational strategies for enhancing their efficacy in real-world settings.
Methods: Following PRISMA guidelines, we searched five databases (PubMed, Embase, Web of Science, CINAHL, and Scopus) on October 30, 2025. Eligible studies included those evaluating AI-driven personalised nutrition interventions in real-world human populations. Methodological quality and risk of bias were evaluated using the National Institutes of Health quality appraisal tool for intervention studies. Implementation outcomes were synthesised through the RE-AIM framework, focusing on the Reach, Effectiveness, Adoption, Implementation, and Maintenance of the interventions.
Results: The systematic search identified 11,152 records, of which 15 met the final inclusion criteria. All included studies were published within the last 5 years (2019–2025). Geographically, the evidence base was heavily skewed: 13 studies originated from high- or upper-middle-income countries, whereas only two were conducted in low-income settings. Furthermore, two-thirds involved small cohorts of fewer than 100 participants. While studies generally reported positive effectiveness and high initial adoption, all programs were still in pilot phases. None demonstrated integration into routine healthcare delivery. Finally, although program fidelity was high, there was a deficit in the reporting of associated intervention costs and strategies for long-term sustainability.
Conclusion: This nascent field has the potential to be highly effective in improving dietary behaviours and clinical biomarkers. However, the generally small sample sizes and concentration in wealthier countries limit generalizability and raise equity concerns. Future research should priorities larger, diverse populations and strategies for sustainable and equitable access.
Biography
Dr. Edmund W. Kanmiki is a public health researcher with a strong interest in health equity. He is currently a Research Fellow at the UQ Poche Centre for Indigenous Health. His current research is focused on the prevention and management of non-communicable diseases among Indigenous people and vulnerable populations in developing countries, as well as the application of emerging technologies to improve healthcare.
He previously held research roles at the University of Ghana and the Navrongo Health Research Centre. He is a recipient of the Mastercard Scholarship, Elsevier Atlas Award, and an RSTMH early-career grant, Dr. Kanmiki has published extensively to inform health policy programs. His research has been presented in several global fora and featured in prominent media, including The Conversation.
Ms Sarah Cunningham
Medical Student
Universty Of Queensland
Towards a gold standard for evaluating AI-generated health information: a scoping review
Abstract
Background and Aim
Large language models (LLMs) are increasingly used by consumers to obtain health information. However, variability and uncertainty in the quality, accuracy and safety of LLM-generated content may pose risks when used to inform health decisions. Despite an increasing number of studies assessing LLM-generated health information, evaluation methods remain highly heterogeneous, limiting the comparability, interpretability and reproducibility of findings. To date, no studies have comprehensively or systematically mapped and critically examined the tools used to evaluate LLM-generated consumer health information. This scoping review aims to systematically map existing evaluation tools, characterising their structural methodologies, quality domains, and validation status.
Methods and Analysis
Following PRISMA-ScR guidelines, this OSF-registered review includes a comprehensive search across seven databases (including PubMed and Scopus) for terms related to LLM interfaces, health information, and quality. Eligible studies include primary research that develops, validates, compares or applies a structured assessment tool to evaluate the quality of LLM-generated health information, where prompts are consumer-derived or explicitly consumer-simulated. A two-tiered narrative synthesis will examine: (1) studies using LLM quality assessment tools; and (2) studies reporting formal psychometric validation of those tools. Data will be extracted on bibliographics, validation methods, quality dimensions, and clinical context.
Outcomes
The review will provide a comprehensive map of existing tools used to assess LLM-generated consumer health information. The review will identify strengths, limitations and inconsistencies across current approaches, highlight tools with the strongest evidence of methodological rigour, and identify gaps in the current evaluation landscape.
Conclusion and Future Actions
Consensus regarding the use of standardised and validated tools for evaluating LLM-generated consumer health information are urgently needed to support safe and effective AI use in healthcare. Findings will inform future research, clinical governance and policy development, and provide a foundation for more consistent, transparent and robust LLM-output evaluation approaches.
Large language models (LLMs) are increasingly used by consumers to obtain health information. However, variability and uncertainty in the quality, accuracy and safety of LLM-generated content may pose risks when used to inform health decisions. Despite an increasing number of studies assessing LLM-generated health information, evaluation methods remain highly heterogeneous, limiting the comparability, interpretability and reproducibility of findings. To date, no studies have comprehensively or systematically mapped and critically examined the tools used to evaluate LLM-generated consumer health information. This scoping review aims to systematically map existing evaluation tools, characterising their structural methodologies, quality domains, and validation status.
Methods and Analysis
Following PRISMA-ScR guidelines, this OSF-registered review includes a comprehensive search across seven databases (including PubMed and Scopus) for terms related to LLM interfaces, health information, and quality. Eligible studies include primary research that develops, validates, compares or applies a structured assessment tool to evaluate the quality of LLM-generated health information, where prompts are consumer-derived or explicitly consumer-simulated. A two-tiered narrative synthesis will examine: (1) studies using LLM quality assessment tools; and (2) studies reporting formal psychometric validation of those tools. Data will be extracted on bibliographics, validation methods, quality dimensions, and clinical context.
Outcomes
The review will provide a comprehensive map of existing tools used to assess LLM-generated consumer health information. The review will identify strengths, limitations and inconsistencies across current approaches, highlight tools with the strongest evidence of methodological rigour, and identify gaps in the current evaluation landscape.
Conclusion and Future Actions
Consensus regarding the use of standardised and validated tools for evaluating LLM-generated consumer health information are urgently needed to support safe and effective AI use in healthcare. Findings will inform future research, clinical governance and policy development, and provide a foundation for more consistent, transparent and robust LLM-output evaluation approaches.
Biography
Sarah Cunningham is a third-year medical student at The University of Queensland, holding a Master of Public Health, and a Graduate Diploma in Clinical Epidemiology. Her professional background bridges academic research and public health practice, combining strong foundations in systematic review methodology and clinical teaching with applied experience in infectious disease surveillance and data management across Queensland Health. Sarah has a proven background in digital health interventions, having co-authored a successfully funded Wellcome Trust grant for a novel heat-health early warning system. She is passionate about leveraging clinical informatics, big data, and emerging technologies to address structural health inequities and strengthen systemic public health responses. She is motivated to continue her work in justice health to evaluate immunisation attrition across jurisdictional divides.
Miss Bianca Lau-goodchild
Research Assistant
University of Melbourne Climate CATCH Lab
Petabytes to Policy: Using AI to Turn Big Data into Climate-Health Action
Abstract
Background and Aim
Over 3.6 billion people live in areas highly susceptible to climate change, with direct health costs projected to reach US$2-4 billion annually by 2030 and greater indirect costs. Despite over 800 petabytes of environmental observation data, growing by over 100 petabytes annually, this evidence remains largely untapped for public health action. Communities with the weakest health infrastructure face the greatest climate-related health burdens yet are least equipped to respond. This systematic review aimed to map current AI and machine learning applications to climate-health big data, identify gaps, and develop a framework guiding equitable, cross-sectoral use to support data-driven policy and efficient resource allocation to those most at risk.
Methods and Analysis
A systematic review was conducted of peer-reviewed literature intersecting climate change, AI/big data/machine learning, and health or health inequity. Thematic analysis identified patterns of application, collaboration, and challenge across research and practice settings.
Outcomes
Analysis revealed a fragmented landscape: AI applications are growing but unevenly distributed, with high-burden, low-resource settings underrepresented in both data generation and tool development. From these findings, a framework was developed articulating conditions for beneficial AI application, centring equity, data sharing and governance, alongside structural barriers limiting impact.
Conclusion and Future Actions
Closing the gap between available data and public health action requires deliberate investment in equitable AI infrastructure, workforce capability, and cross-sector collaboration. Embedding AI-generated insights into planning and policy cycles with transparency and accountability will be critical to ensuring resources reach the populations and geographies where the health burden of climate change is greatest. The framework developed through this review offers a practical foundation for policymakers, researchers and practitioners to prioritise communities at risk and ensure AI becomes a tool for health equity, not a mechanism that deepens existing disparities.
Over 3.6 billion people live in areas highly susceptible to climate change, with direct health costs projected to reach US$2-4 billion annually by 2030 and greater indirect costs. Despite over 800 petabytes of environmental observation data, growing by over 100 petabytes annually, this evidence remains largely untapped for public health action. Communities with the weakest health infrastructure face the greatest climate-related health burdens yet are least equipped to respond. This systematic review aimed to map current AI and machine learning applications to climate-health big data, identify gaps, and develop a framework guiding equitable, cross-sectoral use to support data-driven policy and efficient resource allocation to those most at risk.
Methods and Analysis
A systematic review was conducted of peer-reviewed literature intersecting climate change, AI/big data/machine learning, and health or health inequity. Thematic analysis identified patterns of application, collaboration, and challenge across research and practice settings.
Outcomes
Analysis revealed a fragmented landscape: AI applications are growing but unevenly distributed, with high-burden, low-resource settings underrepresented in both data generation and tool development. From these findings, a framework was developed articulating conditions for beneficial AI application, centring equity, data sharing and governance, alongside structural barriers limiting impact.
Conclusion and Future Actions
Closing the gap between available data and public health action requires deliberate investment in equitable AI infrastructure, workforce capability, and cross-sector collaboration. Embedding AI-generated insights into planning and policy cycles with transparency and accountability will be critical to ensuring resources reach the populations and geographies where the health burden of climate change is greatest. The framework developed through this review offers a practical foundation for policymakers, researchers and practitioners to prioritise communities at risk and ensure AI becomes a tool for health equity, not a mechanism that deepens existing disparities.
Biography
Bianca Lau-Goodchild is an Early Career Researcher in climate change and public health at the Sydney School of Public Health and Melbourne Climate Futures. Her research explores how AI can advance climate change and global health outcomes, alongside climate-sensitive infectious disease and climate-resilient health policy and governance.
Outside of her research work, Bianca is the Partnerships Coordinator and Youth Lead of the Climate and Health Alliance, supporting implementation and policy integration of climate and health research, and coordinating youth engagement across Australia and the Pacific through the CAHA Youth Leadership Council. Bianca recently represented Australia as a youth delegate at COP30 Brazil and will continue her role as a strong advocate for intergenerational equity and health-focused climate policy regionally and internationally at COP31 in Türkiye.
Dr Rachel Rowe
Lecturer
UNSW
What do public health actors in Australia think about AI?
Abstract
Background and Aim:
It has been suggested that limited governance models, underinvestment in infrastructure, skills gaps, and a lack of strategic partnerships are preventing public health actors from realising the promise of big data and algorithmic technologies (Fisher & Rosella, 2022). However, there is much missing from this picture. Looking beyond technical examples of applications, this study explores how public health actors in Australia are interpreting the social, economic, and political implications of big data and complex algorithms within the objectives and values of public health.
Methods:
Twenty-three semi-structured interviews were conducted with Australian public health researchers, practitioners, infrastructure managers, and policymakers between 2019 and 2023. Interview transcripts were thematically analysed.
Outcomes:
Three primary concerns were identified among public health actors. These related to (1) the divergent ways colleagues are responding to “AI hype”, including the adoption of a strategy of non-exceptionalism with respect to AI; (2) perceived power asymmetries between public health institutions and commercial technology producers and consultancy firms; and (3) concern that the dominant individualised framing of predictive analytics and risk assessment algorithms may jeopardise attention to the social, environmental and commercial determinants of population health.
Conclusions:
Expectations surrounding AI may function to legitimise and stabilise innovation trajectories, reorganise and coordinate intersectoral relationships, and reorient public health knowledge practices and interventions. This study identifies concerns that should be addressed in appropriate regulation, governance and use of AI in public health programs. Ensuring that public health actors retain meaningful capacity to question, refuse, or set boundaries around data practices and algorithmic tools is essential for maintaining the credibility and trustworthiness of public health institutions.
It has been suggested that limited governance models, underinvestment in infrastructure, skills gaps, and a lack of strategic partnerships are preventing public health actors from realising the promise of big data and algorithmic technologies (Fisher & Rosella, 2022). However, there is much missing from this picture. Looking beyond technical examples of applications, this study explores how public health actors in Australia are interpreting the social, economic, and political implications of big data and complex algorithms within the objectives and values of public health.
Methods:
Twenty-three semi-structured interviews were conducted with Australian public health researchers, practitioners, infrastructure managers, and policymakers between 2019 and 2023. Interview transcripts were thematically analysed.
Outcomes:
Three primary concerns were identified among public health actors. These related to (1) the divergent ways colleagues are responding to “AI hype”, including the adoption of a strategy of non-exceptionalism with respect to AI; (2) perceived power asymmetries between public health institutions and commercial technology producers and consultancy firms; and (3) concern that the dominant individualised framing of predictive analytics and risk assessment algorithms may jeopardise attention to the social, environmental and commercial determinants of population health.
Conclusions:
Expectations surrounding AI may function to legitimise and stabilise innovation trajectories, reorganise and coordinate intersectoral relationships, and reorient public health knowledge practices and interventions. This study identifies concerns that should be addressed in appropriate regulation, governance and use of AI in public health programs. Ensuring that public health actors retain meaningful capacity to question, refuse, or set boundaries around data practices and algorithmic tools is essential for maintaining the credibility and trustworthiness of public health institutions.
Biography
Dr Rachel Rowe is a multidisciplinary researcher with expertise in public health, political economy, sociology, social policy, and science and technology studies. Her current research examines the power dynamics, discourses and logics shaping algorithmic technologies in public health. A core focus of this work is on how systems of knowledge-production support or curtail attention to inequality as a core driver of population health.
Adjunct Associate Professor Leonie M. Short
Dental Practitioner
Seniors Dental Care Australia
Smilo.ai, an Australian AI tool for oral health: from inception to commercialisation
Abstract
Background and Aim: Oral health neglect is prevalent in residents in aged care facilities with cost effective and in-house solutions needed. There is a significant gap in oral health care with many residents not receiving care aligned with the Aged Care Quality Standard 5.5.7 and facing limited access to affordable oral health care. This project aimed to develop, adapt and evaluate an AI enabled oral health screening tool to improve access to oral health care in residential aged care.
Methods and Analysis: Smilo.ai, a web-based digital health application, was conceived in Brisbane during the COVID-19 lockdown (2021). An Aged Care Research & Industry Innovation Australia (ARIIA) grant (2023-25) supported a mixed methods evaluation and adaption and customisation of Smilo.ai for aged care. Resident demographic data was collected, and quantitative analysis was conducted on resident oral health surveys and screening results. Pre and post staff education and multidisciplinary team surveys were analysed and resident journeys documented. Additional components included dental aids, staff training and manager interviews.
Outcomes: AI enabled oral health screening, Smilo.ai is a game changer in residential care. Just 3% of residents completed an oral health assessment in the 12 months preceding Smilo implementation; 35% brushed their teeth twice daily, 25% reported pain and 25% had denture discomfort. Smilo.ai facilitated baseline data using a traffic light system on oral health screening of residents and the establishment of preventative oral health screening practices. 22.5% rated in the green category; 77.5% rated either amber or red. Customisation of the AI enabled program assured residents of inclusive, subsequent care planning with an integrated decision-making tool for registered nurses in aged care compliant with the Aged Care Quality Standard 5.5.7. Cost was identified as the major barrier to dental access, Smilo.ai enabled access and affordability. Education identified that basic practices like denture care, were misunderstood. Mean scores for oral health knowledge from the pre-and-post survey identified an increase in staff knowledge from 5.3 (SD=1.93) before education to 16.61 (SD=1.94) post education.
Conclusion and Future Actions: Smilo.ai is now customised to an ageing population and has demonstrated improvement in a resident oral health. Integrating Smilo.ai enabled individualised reports that empowered staff and residents through easy access to knowledge, aids and in-house oral health assessment. Dissemination of Smilo.ai in vulnerable populations includes oral health for persons with a disability (MRFF) and for people with diabetes (NHMRC); Smilo.ai is now readily available as an app, no internet connectivity is required and with availability in 35 languages.
Commercialisation in aged care for twice-yearly screening of over 1000 residents has occurred. AI can offer benefits for public oral health practice if it is clinician led, research driven, and policy compliant. Ai generated in-house screening provides and affordable oral health screening and care planning strategy that addresses dental neglect and aligns with Australian aged care reporting standards.
Methods and Analysis: Smilo.ai, a web-based digital health application, was conceived in Brisbane during the COVID-19 lockdown (2021). An Aged Care Research & Industry Innovation Australia (ARIIA) grant (2023-25) supported a mixed methods evaluation and adaption and customisation of Smilo.ai for aged care. Resident demographic data was collected, and quantitative analysis was conducted on resident oral health surveys and screening results. Pre and post staff education and multidisciplinary team surveys were analysed and resident journeys documented. Additional components included dental aids, staff training and manager interviews.
Outcomes: AI enabled oral health screening, Smilo.ai is a game changer in residential care. Just 3% of residents completed an oral health assessment in the 12 months preceding Smilo implementation; 35% brushed their teeth twice daily, 25% reported pain and 25% had denture discomfort. Smilo.ai facilitated baseline data using a traffic light system on oral health screening of residents and the establishment of preventative oral health screening practices. 22.5% rated in the green category; 77.5% rated either amber or red. Customisation of the AI enabled program assured residents of inclusive, subsequent care planning with an integrated decision-making tool for registered nurses in aged care compliant with the Aged Care Quality Standard 5.5.7. Cost was identified as the major barrier to dental access, Smilo.ai enabled access and affordability. Education identified that basic practices like denture care, were misunderstood. Mean scores for oral health knowledge from the pre-and-post survey identified an increase in staff knowledge from 5.3 (SD=1.93) before education to 16.61 (SD=1.94) post education.
Conclusion and Future Actions: Smilo.ai is now customised to an ageing population and has demonstrated improvement in a resident oral health. Integrating Smilo.ai enabled individualised reports that empowered staff and residents through easy access to knowledge, aids and in-house oral health assessment. Dissemination of Smilo.ai in vulnerable populations includes oral health for persons with a disability (MRFF) and for people with diabetes (NHMRC); Smilo.ai is now readily available as an app, no internet connectivity is required and with availability in 35 languages.
Commercialisation in aged care for twice-yearly screening of over 1000 residents has occurred. AI can offer benefits for public oral health practice if it is clinician led, research driven, and policy compliant. Ai generated in-house screening provides and affordable oral health screening and care planning strategy that addresses dental neglect and aligns with Australian aged care reporting standards.
Biography
Associate Professor Melissa Taylor is a registered nurse and researcher who brings deep expertise in rural workforce development, aged care, and end-of-life care to her practice. She has a strong track record in qualitative and mixed-methods research, with a focus on co-design and co-created solutions that address real-world healthcare challenges. As an early adopter of artificial intelligence in healthcare, Melissa champions the use of AI to identify areas of unmet need and neglect, enabling timely, targeted interventions and evidence-based, point-in-time care. Her work integrates innovation with practical application, ensuring technologies are meaningful and accessible in rural settings. Melissa is committed to delivering sustainable, person-centred solutions that strengthen health systems and improve outcomes for individuals and communities across rural and regional Australia.
Mrs. Manoja Gamage
Phd Candidate
Queensland University Of Technology
Food literacy is contextual and complex: Exploring AI-assisted advice for diabetes care
Abstract
Background and aim
Sri Lanka is a hotspot within the South Asian type 2 diabetes mellitus (T2DM) epidemic. Our co-design research found that generic dietary advice was impractical for people living with T2DM in Sri Lanka and identified gaps in the capabilities required to follow dietary advice in daily life. Food literacy may provide an approach to developing these capabilities. Healthcare professionals (HCPs) prioritised knowledge of dietary recommendations within routine diabetes care because of limited consultation time, high patient volumes, and the absence of efficient mechanisms to deliver personalised advice. These findings highlighted the need for an efficient approach to delivering personalised, context-specific food literacy support. This study aimed to explore perspectives of people living with T2DM and HCPs on the potential role of artificial intelligence (AI) in addressing these challenges.
Methods and analysis
A series of qualitative co-design workshops was conducted with people living with T2DM (n = 10) and HCPs involved in T2DM management (n = 10) in Sri Lanka. Participants discussed the potential role of AI in delivering food literacy advice. Content analysis was undertaken.
Outcomes
Participants agreed opportunities for AI in food literacy support in diabetes care. HCPs agreed that AI-assisted tools may efficiently identify individual circumstances and food literacy support needs, enabling personalised advice within time-constrained consultations. AI-assisted tools were also viewed as assisting HCPs to deliver advice regardless of the complexity of individual food literacy needs, while maintaining consistency across HCPs and reducing the need for extensive expertise in individual food cultures. People living with T2DM supported HCPs’ use of AI-assisted tools and valued ongoing support beyond clinical settings.
Conclusion and future actions
Participants supported AI-assisted tools for personalised food literacy support in diabetes care. Further research is underway to develop an AI-assisted food literacy tool for context-specific diabetes care.
Sri Lanka is a hotspot within the South Asian type 2 diabetes mellitus (T2DM) epidemic. Our co-design research found that generic dietary advice was impractical for people living with T2DM in Sri Lanka and identified gaps in the capabilities required to follow dietary advice in daily life. Food literacy may provide an approach to developing these capabilities. Healthcare professionals (HCPs) prioritised knowledge of dietary recommendations within routine diabetes care because of limited consultation time, high patient volumes, and the absence of efficient mechanisms to deliver personalised advice. These findings highlighted the need for an efficient approach to delivering personalised, context-specific food literacy support. This study aimed to explore perspectives of people living with T2DM and HCPs on the potential role of artificial intelligence (AI) in addressing these challenges.
Methods and analysis
A series of qualitative co-design workshops was conducted with people living with T2DM (n = 10) and HCPs involved in T2DM management (n = 10) in Sri Lanka. Participants discussed the potential role of AI in delivering food literacy advice. Content analysis was undertaken.
Outcomes
Participants agreed opportunities for AI in food literacy support in diabetes care. HCPs agreed that AI-assisted tools may efficiently identify individual circumstances and food literacy support needs, enabling personalised advice within time-constrained consultations. AI-assisted tools were also viewed as assisting HCPs to deliver advice regardless of the complexity of individual food literacy needs, while maintaining consistency across HCPs and reducing the need for extensive expertise in individual food cultures. People living with T2DM supported HCPs’ use of AI-assisted tools and valued ongoing support beyond clinical settings.
Conclusion and future actions
Participants supported AI-assisted tools for personalised food literacy support in diabetes care. Further research is underway to develop an AI-assisted food literacy tool for context-specific diabetes care.
Biography
I am a PhD candidate at Queensland University of Technology. My research focuses on utilising food literacy in the dietary management of people living with type 2 diabetes mellitus. Using co-design approaches, I work with people living with diabetes and healthcare professionals to develop a culturally appropriate food literacy tool that supports personalised, context-specific diabetes care in Sri Lanka. My current research explores how artificial intelligence and other digital approaches may enhance the delivery of food literacy support, with the broader aim of improving dietary management and supporting more effective care for people living with non-communicable diseases.
Dr Victor Gallegos-Rejas
Postdoctoral Research Fellow
The University Of Queensland
Evaluating AI-Generated Arabic Vaccine Information: Quality and Accuracy Across Major Platforms
Abstract
Background and Aim:
AI platforms are increasingly used for health information, yet most are developed in Western contexts. As vaccine hesitancy remains a global public health challenge, it is critical to evaluate how effectively AI communicates vaccine information to culturally and linguistically diverse populations. This study assessed the quality and accuracy of Arabic-language vaccine responses generated by major AI platforms.
Methods and Analysis:
Thirteen literature-informed standardised vaccine-related questions were entered in Arabic across four AI platforms: ChatGPT Free, ChatGPT Plus, Gemini, and Claude, yielding 39 evaluated responses. Responses were independently assessed by three Arabic-speaking reviewers using the Quality Analysis of Medical Artificial Intelligence (QAMAI) and CLEAR tools. Cross-platform differences were analysed using Friedman tests with Dunn’s post hoc pairwise comparisons and Bonferroni correction.
Outcomes:
All platforms generated accurate, relevant and understandable Arabic-language vaccine information and actively corrected common vaccine misconceptions, including false claims. Factual accuracy was consistent across platforms on both the QAMAI (p=0.062) and CLEAR (p=0.118) scales. However, statistically significant differences were observed across quality domains including clarity, relevance, completeness, evidence, and usefulness (all p<0.05). QAMAI total scores ranged from 23.02 (ChatGPT Plus) to 25.90 (Claude), and CLEAR total scores from 21.06 (ChatGPT Plus) to 23.45 (Gemini). Claude achieved the highest scores for evidence attribution and structural clarity, while Gemini excelled in completeness, relevance, and usefulness. Evidence-based content was the weakest domain across all domains, with ChatGPT Free scoring particularly low (mean=1.51).
Conclusion and Future Actions:
AI platforms can deliver accurate Arabic-language vaccine information, but significant gaps remain in evidence transparency and communication quality. Future efforts should focus on developing multilingual benchmarking standards for AI-generated health information, requiring source citation as a minimum platform standard, and expanding evaluation to additional Arabic dialects and languages to ensure equitable and trustworthy health information globally.
AI platforms are increasingly used for health information, yet most are developed in Western contexts. As vaccine hesitancy remains a global public health challenge, it is critical to evaluate how effectively AI communicates vaccine information to culturally and linguistically diverse populations. This study assessed the quality and accuracy of Arabic-language vaccine responses generated by major AI platforms.
Methods and Analysis:
Thirteen literature-informed standardised vaccine-related questions were entered in Arabic across four AI platforms: ChatGPT Free, ChatGPT Plus, Gemini, and Claude, yielding 39 evaluated responses. Responses were independently assessed by three Arabic-speaking reviewers using the Quality Analysis of Medical Artificial Intelligence (QAMAI) and CLEAR tools. Cross-platform differences were analysed using Friedman tests with Dunn’s post hoc pairwise comparisons and Bonferroni correction.
Outcomes:
All platforms generated accurate, relevant and understandable Arabic-language vaccine information and actively corrected common vaccine misconceptions, including false claims. Factual accuracy was consistent across platforms on both the QAMAI (p=0.062) and CLEAR (p=0.118) scales. However, statistically significant differences were observed across quality domains including clarity, relevance, completeness, evidence, and usefulness (all p<0.05). QAMAI total scores ranged from 23.02 (ChatGPT Plus) to 25.90 (Claude), and CLEAR total scores from 21.06 (ChatGPT Plus) to 23.45 (Gemini). Claude achieved the highest scores for evidence attribution and structural clarity, while Gemini excelled in completeness, relevance, and usefulness. Evidence-based content was the weakest domain across all domains, with ChatGPT Free scoring particularly low (mean=1.51).
Conclusion and Future Actions:
AI platforms can deliver accurate Arabic-language vaccine information, but significant gaps remain in evidence transparency and communication quality. Future efforts should focus on developing multilingual benchmarking standards for AI-generated health information, requiring source citation as a minimum platform standard, and expanding evaluation to additional Arabic dialects and languages to ensure equitable and trustworthy health information globally.
Biography
Dr Amalie Dyda is an infectious disease epidemiologist working as a teaching and research academic in the School of Public Health. In 2009 she completed a Master of Applied Epidemiology at the Australian National University, followed by a PhD investigating vaccine preventable diseases in adults at the University of New South Wales in 2017. She has experience working as a field epidemiologist in numerous health departments throughout Australia and has research experience in infectious diseases, data linkage and public health informatics. She is currently working on projects investigating the use of technology and machine learning methods to assist the public health response to infectious diseases, and links between social media use and health. Additionally, Amalie does a lot of work to improve gender equity in health and medical research, including working as part of the peer advisory committee for Franklin Women.
Mrs Tammy Sooveere
Clinical Facilitator
Metro South Health
The Artificial Intelligence Reasoning Cycle (AiRC): Navigating the Healthcare Frontier
Abstract
Background and Aim
Healthcare practitioners are increasingly required to integrate AI into clinical workflows, yet existing global AI ethical frameworks do not provide a practical clinical reasoning tool to guide safe, accountable decision‑making. This gap risks inappropriate reliance on AI, erosion of clinical judgement, and unintended harm to patients. This study aimed to develop a structured tool to support healthcare professionals in evaluating the ethical adoption—or withdrawal—of AI across diverse healthcare contexts, from point‑of‑care decision‑making to digital health leadership and policy environments.
Methods and Analysis
A literature synthesis identified the ethical foundations relevant to AI in healthcare. This analysis integrated five established AI ethical pillars—transparency, justice and fairness, non‑maleficence, responsibility, and privacy—with the core healthcare ethics principles of autonomy, beneficence, non‑maleficence, and justice. These elements were mapped to determine the requirements of a practical ethical reasoning framework suitable for clinical, organisational, and policy use.
Outcomes
A robust tool—the Artificial Intelligence Reasoning Cycle (AiRC)—was developed, comprising six steps: Assess, Identify, Review, Action, Verify and Validate, and Reflect. In the Review step, AI applications are scrutinised against the four crucial ethical healthcare pillars to evaluate appropriateness, risks, and implications. This presentation will introduce the AiRC framework, demonstrate its application through a comparative risk profile analysis, and empower clinicians to shape AI implementation in ways that strengthen clinical judgement, safeguard care quality, and uphold the profession’s ethical foundations.
Conclusion and Future Actions
AiRC provides a practical blueprint to guide ethical AI adoption in healthcare, supporting safe, equitable, and person‑centred care. As a conceptual model, it requires empirical validation through comparative risk profiling and real‑world testing to determine its impact on decision quality and patient outcomes. Future work should refine the tool for use across education, governance, and policy contexts. As healthcare enters rapid technological acceleration, clinicians must lead in defining how AI supports safe, equitable, and person‑centred care.
Healthcare practitioners are increasingly required to integrate AI into clinical workflows, yet existing global AI ethical frameworks do not provide a practical clinical reasoning tool to guide safe, accountable decision‑making. This gap risks inappropriate reliance on AI, erosion of clinical judgement, and unintended harm to patients. This study aimed to develop a structured tool to support healthcare professionals in evaluating the ethical adoption—or withdrawal—of AI across diverse healthcare contexts, from point‑of‑care decision‑making to digital health leadership and policy environments.
Methods and Analysis
A literature synthesis identified the ethical foundations relevant to AI in healthcare. This analysis integrated five established AI ethical pillars—transparency, justice and fairness, non‑maleficence, responsibility, and privacy—with the core healthcare ethics principles of autonomy, beneficence, non‑maleficence, and justice. These elements were mapped to determine the requirements of a practical ethical reasoning framework suitable for clinical, organisational, and policy use.
Outcomes
A robust tool—the Artificial Intelligence Reasoning Cycle (AiRC)—was developed, comprising six steps: Assess, Identify, Review, Action, Verify and Validate, and Reflect. In the Review step, AI applications are scrutinised against the four crucial ethical healthcare pillars to evaluate appropriateness, risks, and implications. This presentation will introduce the AiRC framework, demonstrate its application through a comparative risk profile analysis, and empower clinicians to shape AI implementation in ways that strengthen clinical judgement, safeguard care quality, and uphold the profession’s ethical foundations.
Conclusion and Future Actions
AiRC provides a practical blueprint to guide ethical AI adoption in healthcare, supporting safe, equitable, and person‑centred care. As a conceptual model, it requires empirical validation through comparative risk profiling and real‑world testing to determine its impact on decision quality and patient outcomes. Future work should refine the tool for use across education, governance, and policy contexts. As healthcare enters rapid technological acceleration, clinicians must lead in defining how AI supports safe, equitable, and person‑centred care.
Biography
Tammy Sooveere MACN is a registered nurse, educator, and early-career researcher based in Queensland, specialising in infectious diseases and early-career nurse development. She has completed a Graduate Certificate in Clinical Nursing Education and leads initiatives that strengthen clinical capability, evidence‑based practice, and the safe integration of emerging technologies in healthcare. Tammy’s work focuses on supporting early‑career nurses, advancing ethical decision‑making frameworks, and improving patient outcomes through education and system‑level innovation.
Mitchell Bannah MACN is a clinical nurse consultant with experience across acute, community palliative, and regional health services, currently working within Darling Downs Health. With qualifications from the Queensland University of Technology, he brings strengths in adaptable clinical practice, workforce development, and patient‑centred care. He has also contributed to specialist palliative care services, supporting patients and families through compassionate, person‑centred end‑of‑life care. Mitchell is committed to supporting nurses through mentorship, leadership, and advocacy for safe, high‑quality nursing practice.
Dr Louise van Herwerden
Community And Public Health Domain Lead. Master Of Nutrition And Dietetic Practice Program
Bond University
Generative Artificial Intelligence in Dietary Analysis: A Cross-Sectional Study
Abstract
Background and Aim
Generative artificial intelligence (Gen AI) tools are increasingly being explored for dietary analysis; however, their agreement and accuracy compared to established dietary analysis methods remains unclear. This study aimed to evaluate the level of agreement and efficiency of a Gen AI tool (Copilot) compared to traditional dietary analysis software (Foodworks) for estimating energy, macronutrients, sodium, and dietary fibre.
Methods and Analysis
An analytical cross-sectional study was conducted using three-day food diaries from 88 adults. Dietary intake of energy, macronutrients, sodium, and dietary fibre was analysed using both Foodworks (by an Accredited Practising Dietitian) and Copilot, guided by a standardised prompt. Agreement between methods was assessed using Bland–Altman analysis and proportional bias through linear regression. Time required for dietary analysis was compared using paired t-tests.
Outcomes
Mean differences between methods were small across all nutrients, indicating minimal to moderate systematic bias. However, wide limits of agreement demonstrated substantial variability at the individual level, with more than 50% of nutrient estimates falling outside ±10% of Foodworks values. Significant proportional bias was observed for fat and dietary fibre (p<0.001), indicating increasing underestimation at higher intake levels. Copilot was significantly faster than Foodworks (p<0.001), with a large reduction in analysis time.
Conclusion and Future actions
While Gen AI offers substantial efficiency benefits, its variability and inconsistent agreement with established methods limit its suitability for individual-level dietary assessment. Future research should focus on improving accuracy through model refinement, database integration, and prompt optimisation. In practice, cautious and evidence-informed use of AI is recommended to maximise efficiency gains while minimising risks to clinical quality and patient safety.
Generative artificial intelligence (Gen AI) tools are increasingly being explored for dietary analysis; however, their agreement and accuracy compared to established dietary analysis methods remains unclear. This study aimed to evaluate the level of agreement and efficiency of a Gen AI tool (Copilot) compared to traditional dietary analysis software (Foodworks) for estimating energy, macronutrients, sodium, and dietary fibre.
Methods and Analysis
An analytical cross-sectional study was conducted using three-day food diaries from 88 adults. Dietary intake of energy, macronutrients, sodium, and dietary fibre was analysed using both Foodworks (by an Accredited Practising Dietitian) and Copilot, guided by a standardised prompt. Agreement between methods was assessed using Bland–Altman analysis and proportional bias through linear regression. Time required for dietary analysis was compared using paired t-tests.
Outcomes
Mean differences between methods were small across all nutrients, indicating minimal to moderate systematic bias. However, wide limits of agreement demonstrated substantial variability at the individual level, with more than 50% of nutrient estimates falling outside ±10% of Foodworks values. Significant proportional bias was observed for fat and dietary fibre (p<0.001), indicating increasing underestimation at higher intake levels. Copilot was significantly faster than Foodworks (p<0.001), with a large reduction in analysis time.
Conclusion and Future actions
While Gen AI offers substantial efficiency benefits, its variability and inconsistent agreement with established methods limit its suitability for individual-level dietary assessment. Future research should focus on improving accuracy through model refinement, database integration, and prompt optimisation. In practice, cautious and evidence-informed use of AI is recommended to maximise efficiency gains while minimising risks to clinical quality and patient safety.
Biography
Dr Isabella Maugeri is an Accredited Practising Dietitian with Health and Wellbeing Queensland and a collaborator with Bond University’s Nutrition and Dietetics program. Working at the intersection of clinical care, public health, and emerging technology, Bella is particularly focused on how artificial intelligence is reshaping the future of nutrition practice.
Her research examines the accuracy, opportunities, and risks of generative AI in dietary assessment, with a strong emphasis on ensuring that innovation enhances—rather than compromises—clinical quality and patient safety. She brings a critical, evidence-informed lens to the integration of AI in healthcare, exploring how dietitians can harness its efficiencies while maintaining professional judgement and accountability. Bella is passionate about supporting the responsible adoption of AI in nutrition, aligning with the broader public health goal of maximising benefits while minimising harms.
Prof Simone Pettigrew
Director, Food Policy
The George Institute for Global Health
Keeping the b******* honest: an AI perspective (literally)
Abstract
Background and Aim:
Harmful products industries—including those supplying unhealthy food, alcohol, and nicotine products—continually adapt their strategies to influence policy, public opinion, and consumer behaviour. Public health surveillance has struggled to keep pace with the scale, speed, and sophistication of these activities. This work explores how emerging AI tools could transform the way researchers detect, monitor, and respond to harmful industry practices.
Methods and Analysis:
Ways in which AI can be applied across the harmful products research cycle include: AI agents that continuously monitor parliamentary proceedings, regulatory consultations, corporate communications, and media coverage; large language models that compare policy submissions to identify coordinated messaging across industries; automated mapping of relationships between industry actors, front groups, and decision-makers; monitoring of the composition and evolution of food and alcohol markets through analysis of product databases, retail environments, reformulation, and new product launches; computer vision systems that detect and quantify marketing in digital and physical environments; analyses of corporate reports and scientific publications to identify conflicts of interest; and conversational AI interfaces that enable researchers, advocates, and policymakers to interrogate complex surveillance datasets using natural language. Associated challenges relating to transparency, bias, validation, privacy, and governance will be discussed.
Outcomes:
AI has the potential to shift surveillance from periodic, labour-intensive reviews to continuous, near real-time public health intelligence. AI could provide early warning of changes in product portfolios, market concentration, marketing strategies, political activity, and scientific influence, which could inform policy development and strengthen advocacy.
Conclusion and Future Actions:
AI offers an opportunity to transform surveillance of harmful products industries from retrospective evidence gathering to proactive intelligence generation. Realising this potential will require investment in AI capability, interdisciplinary collaboration, shared methods, and governance frameworks that ensure these technologies strengthen accountability while maintaining scientific rigour, transparency, and public trust.
Harmful products industries—including those supplying unhealthy food, alcohol, and nicotine products—continually adapt their strategies to influence policy, public opinion, and consumer behaviour. Public health surveillance has struggled to keep pace with the scale, speed, and sophistication of these activities. This work explores how emerging AI tools could transform the way researchers detect, monitor, and respond to harmful industry practices.
Methods and Analysis:
Ways in which AI can be applied across the harmful products research cycle include: AI agents that continuously monitor parliamentary proceedings, regulatory consultations, corporate communications, and media coverage; large language models that compare policy submissions to identify coordinated messaging across industries; automated mapping of relationships between industry actors, front groups, and decision-makers; monitoring of the composition and evolution of food and alcohol markets through analysis of product databases, retail environments, reformulation, and new product launches; computer vision systems that detect and quantify marketing in digital and physical environments; analyses of corporate reports and scientific publications to identify conflicts of interest; and conversational AI interfaces that enable researchers, advocates, and policymakers to interrogate complex surveillance datasets using natural language. Associated challenges relating to transparency, bias, validation, privacy, and governance will be discussed.
Outcomes:
AI has the potential to shift surveillance from periodic, labour-intensive reviews to continuous, near real-time public health intelligence. AI could provide early warning of changes in product portfolios, market concentration, marketing strategies, political activity, and scientific influence, which could inform policy development and strengthen advocacy.
Conclusion and Future Actions:
AI offers an opportunity to transform surveillance of harmful products industries from retrospective evidence gathering to proactive intelligence generation. Realising this potential will require investment in AI capability, interdisciplinary collaboration, shared methods, and governance frameworks that ensure these technologies strengthen accountability while maintaining scientific rigour, transparency, and public trust.
Biography
Professor Simone Pettigrew is the Program Director of Health Promotion at The George Institute for Global Health. She has qualifications in Economics, Commerce, and Consumer Research. Her broad areas of expertise include behavioural psychology, health promotion, health policy, communications, social marketing, and intervention research. Her substantive areas of research include nutrition, alcohol consumption, smoking/vaping, physical activity, and active transport.
