2D - Innovation
Tracks
Track 4
| Monday, June 15, 2026 |
| 1:30 PM - 3:00 PM |
Speaker
Dr Richard Creswell
Research Fellow
University of Melbourne
Designing infectious disease surveillance systems for learning infection dynamics
Abstract
Background and Aim
Infection prevalence and antibody seroprevalence surveys, which assess randomly sampled people for being infected or for having antibodies indicating a history of infection (or vaccination), provide essential information about the proportion of the population which is susceptible to infection. Information about the true number of infections, rather than reported cases (which typically underestimate infections), is valuable for accurately predicting the future course of the epidemic and planning effective interventions.
Disease surveillance based on random surveys is expensive to conduct, motivating survey designs which do not place excessive demands on limited public health resources. However, the relationship between survey design parameters and the usefulness of the resultant observations is highly complex, necessitating new strategies to determine survey design parameters which are economically feasible while still providing the most useful information for downstream public health tasks.
Methods and Analysis
We developed a simulation platform allowing us to generate simulations of respiratory virus epidemics and accompanying simulations of disease surveillance systems such as population-wide infection prevalence and seroprevalence surveys. We simulate various choices of survey design parameters, and quantitatively evaluate the corresponding quality of the surveillance data, focusing on the ability to accurately and precisely infer the true number of infections over time.
Outcomes
Our results demonstrate the importance of frequent sampling to achieve precise inference results, and the presence of diminishing returns in inference precision with increasing sample sizes. We also demonstrate the potential for inaccurate assumptions about seroreversion to substantially decrease the ability to accurately learn infections.
Conclusion and Future actions
These analyses illustrate the ability of epidemiological modelling to make helpful contributions to the design of efficient infectious disease surveillance systems. The work aims to support improved predictions of infectious disease burdens and a more efficient use of limited public health resources.
Infection prevalence and antibody seroprevalence surveys, which assess randomly sampled people for being infected or for having antibodies indicating a history of infection (or vaccination), provide essential information about the proportion of the population which is susceptible to infection. Information about the true number of infections, rather than reported cases (which typically underestimate infections), is valuable for accurately predicting the future course of the epidemic and planning effective interventions.
Disease surveillance based on random surveys is expensive to conduct, motivating survey designs which do not place excessive demands on limited public health resources. However, the relationship between survey design parameters and the usefulness of the resultant observations is highly complex, necessitating new strategies to determine survey design parameters which are economically feasible while still providing the most useful information for downstream public health tasks.
Methods and Analysis
We developed a simulation platform allowing us to generate simulations of respiratory virus epidemics and accompanying simulations of disease surveillance systems such as population-wide infection prevalence and seroprevalence surveys. We simulate various choices of survey design parameters, and quantitatively evaluate the corresponding quality of the surveillance data, focusing on the ability to accurately and precisely infer the true number of infections over time.
Outcomes
Our results demonstrate the importance of frequent sampling to achieve precise inference results, and the presence of diminishing returns in inference precision with increasing sample sizes. We also demonstrate the potential for inaccurate assumptions about seroreversion to substantially decrease the ability to accurately learn infections.
Conclusion and Future actions
These analyses illustrate the ability of epidemiological modelling to make helpful contributions to the design of efficient infectious disease surveillance systems. The work aims to support improved predictions of infectious disease burdens and a more efficient use of limited public health resources.
Mrs Zahra Ahsani
Biostatistician/data Analyst
ASPREN
Forecasting Influenza-Like Illness Using Machine Learning in Australian Primary Care Influenza Surveillance
Abstract
Background:
Influenza-like illness (ILI) surveillance from primary care sentinel sites is critical for monitoring influenza activity in Australia. However, current surveillance is largely retrospective, limiting timely preparedness and response. This study aimed to improve short-term forecasting of weekly ILI rates using data from the Australian Sentinel Practices Research Network (ASPREN).
Methods:
We used ASPREN influenza surveillance data from 2011 to 2019 to compare two forecasting methods: a traditional SARIMAX model and a machine learning approach using XGBoost. Both models incorporated weekly ILI rates and laboratory results to predict one-week-ahead ILI activity. Performance was evaluated in two stages: (1) expanding window cross-validation across six influenza seasons (2014–2019), forecasting each season using only previous years’ data, and (2) on an independent 2019 holdout set excluded from training. Accuracy was assessed using RMSE, MAE, and R².
Outcomes:
Across expanding-window evaluations (2014–2019), XGBoost model achieved an average predictive accuracy of RMSE = 1.21 cases per 1,000 consultations with best accuracy in 2018 (RMSE 0.53; R² 0.93). SARIMAX was more variable (RMSE 1.31-2.68), performing best in 2017 (RMSE 1.31; R² 0.97) but exhibits reduced performance in seasons with high variability. On the 2019 holdout set, XGBoost further outperformed SARIMAX, achieving a lower RMSE (1.27 vs. 2.62) and higher R² (0.93 vs. 0.71).
Conclusion:
XGBoost substantially improves short-term ILI forecasting, achieving a 47% improvement in accuracy over the traditional SARIMAX method on the 2019 holdout set. While SARIMAX may perform better in atypical seasons, such as 2017, XGBoost provides consistent forecasts under typical conditions and handles missing data and complex patterns without manual adjustments. Its ability to be fully automated monthly makes it a practical and scalable tool for routine influenza surveillance and health service planning.
Influenza-like illness (ILI) surveillance from primary care sentinel sites is critical for monitoring influenza activity in Australia. However, current surveillance is largely retrospective, limiting timely preparedness and response. This study aimed to improve short-term forecasting of weekly ILI rates using data from the Australian Sentinel Practices Research Network (ASPREN).
Methods:
We used ASPREN influenza surveillance data from 2011 to 2019 to compare two forecasting methods: a traditional SARIMAX model and a machine learning approach using XGBoost. Both models incorporated weekly ILI rates and laboratory results to predict one-week-ahead ILI activity. Performance was evaluated in two stages: (1) expanding window cross-validation across six influenza seasons (2014–2019), forecasting each season using only previous years’ data, and (2) on an independent 2019 holdout set excluded from training. Accuracy was assessed using RMSE, MAE, and R².
Outcomes:
Across expanding-window evaluations (2014–2019), XGBoost model achieved an average predictive accuracy of RMSE = 1.21 cases per 1,000 consultations with best accuracy in 2018 (RMSE 0.53; R² 0.93). SARIMAX was more variable (RMSE 1.31-2.68), performing best in 2017 (RMSE 1.31; R² 0.97) but exhibits reduced performance in seasons with high variability. On the 2019 holdout set, XGBoost further outperformed SARIMAX, achieving a lower RMSE (1.27 vs. 2.62) and higher R² (0.93 vs. 0.71).
Conclusion:
XGBoost substantially improves short-term ILI forecasting, achieving a 47% improvement in accuracy over the traditional SARIMAX method on the 2019 holdout set. While SARIMAX may perform better in atypical seasons, such as 2017, XGBoost provides consistent forecasts under typical conditions and handles missing data and complex patterns without manual adjustments. Its ability to be fully automated monthly makes it a practical and scalable tool for routine influenza surveillance and health service planning.
Dr Katharine Senior
Senior Research Officer/Research Fellow
The Kids Research Institute Australia/The University of Melbourne
Optimising early epidemic phase infection prevalence surveys to learn infection severity metrics
Abstract
Background and Aim
Public health decision making during epidemics relies on gaining a rapid understanding of key disease parameters, including severity measures such as infection-hospitalisation rate (IHR). IHR is often measured in early pandemic phases through First Few X (FFX) studies, which track households of infected individuals to learn transmission patterns. However, historically, FFX studies have been slow and costly to set up. A potential alternative is to use infection prevalence surveys to gain an understanding of infection burden in the community, and pair this with routine surveillance data to inform our estimates of IHR. However, significant uncertainty remains about how best to implement infection prevalence surveys to accurately measure infection severity, particularly in the early epidemic phase when prevalence is low. This study explores how infection prevalence surveys can be optimised to learn infection severity.
Methods and Analysis
We simulate a series of epidemics across different disease settings to create infection prevalence, case and hospitalisation timeseries. We then use a multi-stream infection incidence model that incorporates multiple datasets to model the underlying infection timeseries, from which we calculate IHR. Finally, we vary the frequency and number of infection prevalence surveys across different disease settings and compare our estimates of infections and IHR to the simulated ‘truth’ using continuous ranked probability scores (CRPS).
Outcomes
Using simulation modelling, we can gain an understanding of when and where resources should be allocated to infection prevalence surveys during the early epidemic phase, to learn key metrics such as infection severity. We provide guidelines for how these survey programs should be structured and the disease scenarios under which they will be most useful.
Conclusion and Future actions
Future disease outbreaks could implement a structured infection prevalence survey program, using our guidelines, to learn disease parameters such as IHR in the early epidemic phase.
Public health decision making during epidemics relies on gaining a rapid understanding of key disease parameters, including severity measures such as infection-hospitalisation rate (IHR). IHR is often measured in early pandemic phases through First Few X (FFX) studies, which track households of infected individuals to learn transmission patterns. However, historically, FFX studies have been slow and costly to set up. A potential alternative is to use infection prevalence surveys to gain an understanding of infection burden in the community, and pair this with routine surveillance data to inform our estimates of IHR. However, significant uncertainty remains about how best to implement infection prevalence surveys to accurately measure infection severity, particularly in the early epidemic phase when prevalence is low. This study explores how infection prevalence surveys can be optimised to learn infection severity.
Methods and Analysis
We simulate a series of epidemics across different disease settings to create infection prevalence, case and hospitalisation timeseries. We then use a multi-stream infection incidence model that incorporates multiple datasets to model the underlying infection timeseries, from which we calculate IHR. Finally, we vary the frequency and number of infection prevalence surveys across different disease settings and compare our estimates of infections and IHR to the simulated ‘truth’ using continuous ranked probability scores (CRPS).
Outcomes
Using simulation modelling, we can gain an understanding of when and where resources should be allocated to infection prevalence surveys during the early epidemic phase, to learn key metrics such as infection severity. We provide guidelines for how these survey programs should be structured and the disease scenarios under which they will be most useful.
Conclusion and Future actions
Future disease outbreaks could implement a structured infection prevalence survey program, using our guidelines, to learn disease parameters such as IHR in the early epidemic phase.
Dr Surendra Karki
Research Fellow
Australian Red Cross Lifeblood
The Australian Blood Donor Study and Biobank- contributing to Australian Health Research
Abstract
Background and Aim
Recruiting participants into long-term health research involving biological sampling and epidemiological data requires substantial resources and infrastructure. National blood collection agencies such as Australian Red Cross Lifeblood (Lifeblood) can be leveraged to enrol donors into such studies and collect samples efficiently. This provides a practical pathway to recruit large numbers of participants. To pilot this approach, we recruited whole blood and apheresis donors into the Australian Blood Donor Study (ABDS).
Methods and Analysis
Donors were invited to participate via email two-weeks before their donation appointment. Consenting donors provided demographic, lifestyle, and health information through online surveys, and blood samples were collected at the time of donation. Recruitment commenced in November 2022 and was completed in August 2024. Samples were transported to the ABDS Biobank in Sydney using existing Lifeblood logistics. Serum, plasma, and buffy coat were processed and stored at -80 °C for future analyses and DNA extraction. DNA extraction was performed in-house and by an external provider. Yearly follow-up surveys are currently underway.
Outcomes
A total of 37,766 donors were approached, and 9,082 (24.0%) consented. The mean age was 50.5 years (SD 14.9); 55% were female, 97.1% consented to external data linkage and 71.1% provided blood samples. Participants were slightly older and predominantly of European ancestry. Consistent with other population cohorts, participants appeared healthier than the general Australian population, reflected by a lower prevalence of current smoking and a higher proportion reporting excellent or very-good health. Several Lifeblood studies are using the samples and data for donor health research. Governance frameworks are being finalised to enable access for external researchers.
Conclusion and Future actions
We demonstrated that recruitment of donors into a longitudinal cohort study and biobank for health research is feasible within Lifeblood operations. Appropriate funding mechanisms to expand the biobank are being explored.
Recruiting participants into long-term health research involving biological sampling and epidemiological data requires substantial resources and infrastructure. National blood collection agencies such as Australian Red Cross Lifeblood (Lifeblood) can be leveraged to enrol donors into such studies and collect samples efficiently. This provides a practical pathway to recruit large numbers of participants. To pilot this approach, we recruited whole blood and apheresis donors into the Australian Blood Donor Study (ABDS).
Methods and Analysis
Donors were invited to participate via email two-weeks before their donation appointment. Consenting donors provided demographic, lifestyle, and health information through online surveys, and blood samples were collected at the time of donation. Recruitment commenced in November 2022 and was completed in August 2024. Samples were transported to the ABDS Biobank in Sydney using existing Lifeblood logistics. Serum, plasma, and buffy coat were processed and stored at -80 °C for future analyses and DNA extraction. DNA extraction was performed in-house and by an external provider. Yearly follow-up surveys are currently underway.
Outcomes
A total of 37,766 donors were approached, and 9,082 (24.0%) consented. The mean age was 50.5 years (SD 14.9); 55% were female, 97.1% consented to external data linkage and 71.1% provided blood samples. Participants were slightly older and predominantly of European ancestry. Consistent with other population cohorts, participants appeared healthier than the general Australian population, reflected by a lower prevalence of current smoking and a higher proportion reporting excellent or very-good health. Several Lifeblood studies are using the samples and data for donor health research. Governance frameworks are being finalised to enable access for external researchers.
Conclusion and Future actions
We demonstrated that recruitment of donors into a longitudinal cohort study and biobank for health research is feasible within Lifeblood operations. Appropriate funding mechanisms to expand the biobank are being explored.
Mr Michael Dymock
Biostatistician
The Kids Research Institute Australia
Vaccine trials with adaptive designs: Promise or peril?
Abstract
Background and Aim
Adaptive designs allow for pre-specified modifications to the trial design based on accumulated data. Examples of trial adaptation include sample size re-estimation, early stopping for efficacy or futility, and arm dropping. Over the past decade, adaptive trial designs have become increasingly popular due to their perceived flexibility, operational efficiency, attractiveness to funders, and potential for reduced sample sizes. However, compared to fixed designs, adaptive trials typically have higher setup overheads and ongoing resource use (e.g., database costs, statistical analyses), and are not always beneficial (e.g., when recruitment is fast relative to outcome ascertainment).
Alongside their growing use in other clinical areas, adaptive designs have recently been used for vaccine trials. Therefore, clinicians planning future vaccine trials may ask themselves: should an adaptive design be considered?
Methods and Analysis
PICOBOO and BOOST-IC are two Australian clinical trials with adaptive designs that evaluated the immunogenicity and reactogenicity of COVID-19 booster vaccines in immunocompetent and immunocompromised populations, respectively. The trials both evaluated short-term immune responses for the primary outcome and both allowed for early stopping and arm dropping based on accumulated data, and a novel decision rule based on the precision of the estimate of the average immune response.
Outcomes
Both trials were conducted during a period of shifting policy recommendations and vaccine supply. Their adaptive designs allowed the trials to flexibly adapt with the evolving landscape, but they were challenged by a lack of enabling infrastructure (e.g., laboratory and statistical resource constraints).
Conclusion and Future actions
In this talk we will describe our experience implementing adaptive designs for vaccine trials in Australia. We will provide guidance for clinicians considering adaptive designs for their future vaccine trials and respond to the question: does the inclusion of adaptive design features in vaccine trials create promise or peril?
Adaptive designs allow for pre-specified modifications to the trial design based on accumulated data. Examples of trial adaptation include sample size re-estimation, early stopping for efficacy or futility, and arm dropping. Over the past decade, adaptive trial designs have become increasingly popular due to their perceived flexibility, operational efficiency, attractiveness to funders, and potential for reduced sample sizes. However, compared to fixed designs, adaptive trials typically have higher setup overheads and ongoing resource use (e.g., database costs, statistical analyses), and are not always beneficial (e.g., when recruitment is fast relative to outcome ascertainment).
Alongside their growing use in other clinical areas, adaptive designs have recently been used for vaccine trials. Therefore, clinicians planning future vaccine trials may ask themselves: should an adaptive design be considered?
Methods and Analysis
PICOBOO and BOOST-IC are two Australian clinical trials with adaptive designs that evaluated the immunogenicity and reactogenicity of COVID-19 booster vaccines in immunocompetent and immunocompromised populations, respectively. The trials both evaluated short-term immune responses for the primary outcome and both allowed for early stopping and arm dropping based on accumulated data, and a novel decision rule based on the precision of the estimate of the average immune response.
Outcomes
Both trials were conducted during a period of shifting policy recommendations and vaccine supply. Their adaptive designs allowed the trials to flexibly adapt with the evolving landscape, but they were challenged by a lack of enabling infrastructure (e.g., laboratory and statistical resource constraints).
Conclusion and Future actions
In this talk we will describe our experience implementing adaptive designs for vaccine trials in Australia. We will provide guidance for clinicians considering adaptive designs for their future vaccine trials and respond to the question: does the inclusion of adaptive design features in vaccine trials create promise or peril?
Dr Cristyn Davies
Senior Research Fellow
University of Sydney
Usability and Acceptability of HD-MAP Administration Among Healthcare Professionals and Lay Adults
Abstract
Background and Aim
High-density microarray patch (HD-MAP) vaccines may improve uptake through needle-free delivery and potential self-administration. Successful use requires end users to follow the Instructions for Use (IFU) correctly. This study assessed the usability and acceptability of HD-MAP administration using prototype IFUs among healthcare professionals (HCPs) and lay adults.
Methods and Analysis
This mixed-methods study used a single-group intervention approach involving two simulated administrations of a prototype HD-MAP applicator to the deltoid. HCPs administered to an adult volunteer using an HCP IFU, while lay adults self-administered to their non-dominant upper arm using a lay IFU. Administrations were video-recorded and assessed against predefined essential criteria. Participants completed questionnaires, interviews, and hand strength testing. Quantitative data were summarised descriptively and compared between groups, and qualitative data were analysed thematically.
Outcomes
Sixty-seven participants were recruited (25 HCPs; 42 lay adults). The proportion meeting essential criteria improved between attempts (HCPs: 40% to 60%; lay adults: 56% to 66%). The most frequently missed criterion was maintaining the 10-second hold following activation, although mean hold times exceeded 10 seconds (HCPs: 10.9 [SD 1.2] to 11.8 [SD 0.9] seconds; lay adults: 16.8 [SD 2.8] to 12.8 [SD 1.1] seconds). Eight participants were unable to activate the applicator; all demonstrated reduced index finger and thumb strength. During the interview, participants reported reading the IFU in full (HCPs: 88%; lay adults: 95%) and preferred the HD-MAP over needle-and-syringe vaccination (HCPs: 72%; lay adults: 83%). Qualitative findings indicated that the IFUs were easy to follow, and confidence improved with practice, although uncertainty regarding activation force and delivery confirmation remained.
Conclusion and Future Actions
HD-MAP applicator administration demonstrated high usability and acceptability among HCPs and lay adults when guided by IFUs. Refinements to IFUs, clearer activation guidance, and confirmation of delivery may enhance usability and support future implementation, including self-administration.
High-density microarray patch (HD-MAP) vaccines may improve uptake through needle-free delivery and potential self-administration. Successful use requires end users to follow the Instructions for Use (IFU) correctly. This study assessed the usability and acceptability of HD-MAP administration using prototype IFUs among healthcare professionals (HCPs) and lay adults.
Methods and Analysis
This mixed-methods study used a single-group intervention approach involving two simulated administrations of a prototype HD-MAP applicator to the deltoid. HCPs administered to an adult volunteer using an HCP IFU, while lay adults self-administered to their non-dominant upper arm using a lay IFU. Administrations were video-recorded and assessed against predefined essential criteria. Participants completed questionnaires, interviews, and hand strength testing. Quantitative data were summarised descriptively and compared between groups, and qualitative data were analysed thematically.
Outcomes
Sixty-seven participants were recruited (25 HCPs; 42 lay adults). The proportion meeting essential criteria improved between attempts (HCPs: 40% to 60%; lay adults: 56% to 66%). The most frequently missed criterion was maintaining the 10-second hold following activation, although mean hold times exceeded 10 seconds (HCPs: 10.9 [SD 1.2] to 11.8 [SD 0.9] seconds; lay adults: 16.8 [SD 2.8] to 12.8 [SD 1.1] seconds). Eight participants were unable to activate the applicator; all demonstrated reduced index finger and thumb strength. During the interview, participants reported reading the IFU in full (HCPs: 88%; lay adults: 95%) and preferred the HD-MAP over needle-and-syringe vaccination (HCPs: 72%; lay adults: 83%). Qualitative findings indicated that the IFUs were easy to follow, and confidence improved with practice, although uncertainty regarding activation force and delivery confirmation remained.
Conclusion and Future Actions
HD-MAP applicator administration demonstrated high usability and acceptability among HCPs and lay adults when guided by IFUs. Refinements to IFUs, clearer activation guidance, and confirmation of delivery may enhance usability and support future implementation, including self-administration.
Dr Sashika Harasgama
Public Health Registrar
Western Public Health Unit
Use of Generative AI to assist Surveillance in Local Public Health Units
Abstract
Background and Aim:
Timely epidemiological surveillance data from international and national sources is essential for effective local health protection responses. Online resources are increasingly behind paywalls and provide a level of detail unnecessary for day-to-day local communicable disease control. Manually collating communicable disease reports is both time- and resource-intensive. Generative artificial intelligence (AI) provides the opportunity for timely and automated information gathering.
Methods and Analysis:
We developed an AI agent using ChatGPT to generate weekly surveillance reports from publicly available national and international sources. Between January-February 2026, two users with different levels of health protection experience independently validated four reports against ProMED. Outputs were assessed for sensitivity (proportion of ProMED-reported events correctly identified), citation accuracy (correct hyperlinked sources) and rate of hallucinations.
Outcomes:
Across four validation attempts, the tool achieved 67.4% overall sensitivity (31/46 ProMED events correctly identified) and 69.6% citation accuracy (32/46 sources correctly linked). Sensitivity varied across geographies: 55.0% for Australian data versus 76.9% for international data. No hallucinations were identified by either user. The majority of events (89.1%) were deemed relevant to the local public health unit context. Citation and formatting errors were more common for the user not involved with tool development.
Conclusion and Future Actions:
The AI agent serves as a quick and very low-cost adjunct to regular surveillance reporting methods that can be easily contextualised to the health protection needs of a local public health unit. The non-deterministic nature of generative AI and user behaviour results in inconsistent outputs across users. However, the surveillance content reported remained relatively accurate. Continuing to train and develop AI agents to perform effective searches of public health intelligence is a safe and appropriate use case in local communicable disease control. Careful validation and quality assurance must be employed when using AI.
Timely epidemiological surveillance data from international and national sources is essential for effective local health protection responses. Online resources are increasingly behind paywalls and provide a level of detail unnecessary for day-to-day local communicable disease control. Manually collating communicable disease reports is both time- and resource-intensive. Generative artificial intelligence (AI) provides the opportunity for timely and automated information gathering.
Methods and Analysis:
We developed an AI agent using ChatGPT to generate weekly surveillance reports from publicly available national and international sources. Between January-February 2026, two users with different levels of health protection experience independently validated four reports against ProMED. Outputs were assessed for sensitivity (proportion of ProMED-reported events correctly identified), citation accuracy (correct hyperlinked sources) and rate of hallucinations.
Outcomes:
Across four validation attempts, the tool achieved 67.4% overall sensitivity (31/46 ProMED events correctly identified) and 69.6% citation accuracy (32/46 sources correctly linked). Sensitivity varied across geographies: 55.0% for Australian data versus 76.9% for international data. No hallucinations were identified by either user. The majority of events (89.1%) were deemed relevant to the local public health unit context. Citation and formatting errors were more common for the user not involved with tool development.
Conclusion and Future Actions:
The AI agent serves as a quick and very low-cost adjunct to regular surveillance reporting methods that can be easily contextualised to the health protection needs of a local public health unit. The non-deterministic nature of generative AI and user behaviour results in inconsistent outputs across users. However, the surveillance content reported remained relatively accurate. Continuing to train and develop AI agents to perform effective searches of public health intelligence is a safe and appropriate use case in local communicable disease control. Careful validation and quality assurance must be employed when using AI.
Mr Anthony Renehan
Digital Health Implementation Specialist
Vitavo Health
Digital Transformation of Local Government Immunisation: A Case Study in Operational Efficiency
Abstract
Background: Local government immunisation programs face mounting pressure from rising operational costs, limited state funding, and growing community demand. Traditional paper-based processes and fragmented technology systems create administrative bottlenecks that constrain service capacity and threaten program viability.
Aim: To evaluate the impact of implementing an integrated digital immunisation management platform on service efficiency, capacity, and financial sustainability within a metropolitan local government setting.
Methods: Knox City Council (population 160,000+) implemented Vitavo, a digital platform automating 90% of administrative and clinical workflows, across community clinic and School Immunisation Program (SIP) settings in July 2023. A comparison was undertaken to evaluate pre- and post-implementation data across key metrics including appointment duration, staff hours, operating costs, and paid vaccine revenue over the first financial year.
Results:
• Community sessions achieved a 50% reduction in appointment times (10 to 5 minutes), enabling increased client throughput without additional resources. Administration time per encounter dropped from 10 minutes to 30 seconds.
• Community session operating costs reduced by $71,239 annually, with cost per session lowered by $697. The net loss per encounter decreased by 90.5% (from $24.73 to $2.01), positioning community sessions toward breakeven.
• Paid vaccine revenue margin increased by $18,659.
• School programs achieved $70,000+ in annual savings through 50% reduction in administrative hours (1,482 hours saved).
• Total program savings reached $167,843, representing a 1,145% return on investment.
Conclusions: Digital transformation of immunisation service delivery can dramatically improve operational efficiency while creating pathways to financial sustainability.
This model demonstrates how local governments can maintain and expand immunisation services despite funding constraints, with the potential to reinvest savings into programs reaching underserved populations or realise new program delivery models to increase access and uptake of vaccinations.
Aim: To evaluate the impact of implementing an integrated digital immunisation management platform on service efficiency, capacity, and financial sustainability within a metropolitan local government setting.
Methods: Knox City Council (population 160,000+) implemented Vitavo, a digital platform automating 90% of administrative and clinical workflows, across community clinic and School Immunisation Program (SIP) settings in July 2023. A comparison was undertaken to evaluate pre- and post-implementation data across key metrics including appointment duration, staff hours, operating costs, and paid vaccine revenue over the first financial year.
Results:
• Community sessions achieved a 50% reduction in appointment times (10 to 5 minutes), enabling increased client throughput without additional resources. Administration time per encounter dropped from 10 minutes to 30 seconds.
• Community session operating costs reduced by $71,239 annually, with cost per session lowered by $697. The net loss per encounter decreased by 90.5% (from $24.73 to $2.01), positioning community sessions toward breakeven.
• Paid vaccine revenue margin increased by $18,659.
• School programs achieved $70,000+ in annual savings through 50% reduction in administrative hours (1,482 hours saved).
• Total program savings reached $167,843, representing a 1,145% return on investment.
Conclusions: Digital transformation of immunisation service delivery can dramatically improve operational efficiency while creating pathways to financial sustainability.
This model demonstrates how local governments can maintain and expand immunisation services despite funding constraints, with the potential to reinvest savings into programs reaching underserved populations or realise new program delivery models to increase access and uptake of vaccinations.
Dr Fiona May
Advanced Epidemiologist
Gold Coast Public Health Unit
Thinking outside the box: using REDCap for public health operations
Abstract
Background and Aim:
Public health workforce issues are common in Australia. With Public Health Units experiencing increasingly complex workloads, new technologies must be harnessed to streamline, automate and reduce manual processes. REDCap is an internationally recognised digital data collection tool, often used in public health to collect enhanced disease surveillance for notifiable conditions. We describe the varied uses of REDCap for operational tasks in a Public Health Unit.
Methods and Analysis:
All teams in the Public Health Unit were consulted to identify potential projects that would benefit from REDCap. The Health Surveillance team developed tools guided by the consultation with an aim to improve processes and data quality. Teams were consulted again to test and fine tune the REDCap tool before deployment.
Outcomes:
Tools for institutional settings to actively report gastrointestinal outbreaks (childcare centres and residential aged care homes) replaced intensive phone calls with facilities. A tool for reporting rabies vaccine and immunoglobulin use by immunisation service providers using government funded stock improved the ease and accuracy of data collection. A quiz for assessing knowledge for administration of these products helps with educating providers. During Tropical Cyclone Alfred, REDCap was used to collect information on immunisation service provider cold chain breaches. This tool provided the service provider with immediate information in an automated email, allowing the Public Health Team time to triage and prioritise response after the emergency had passed.
Conclusion and future actions:
REDCap is proving to be a useful tool in the public health unit. While not yet formally evaluated, staff report that the use of these tools is easing workload, allowing them to focus on other priority work. The list of future projects continues to grow as more staff become familiar with its utility.
Public health workforce issues are common in Australia. With Public Health Units experiencing increasingly complex workloads, new technologies must be harnessed to streamline, automate and reduce manual processes. REDCap is an internationally recognised digital data collection tool, often used in public health to collect enhanced disease surveillance for notifiable conditions. We describe the varied uses of REDCap for operational tasks in a Public Health Unit.
Methods and Analysis:
All teams in the Public Health Unit were consulted to identify potential projects that would benefit from REDCap. The Health Surveillance team developed tools guided by the consultation with an aim to improve processes and data quality. Teams were consulted again to test and fine tune the REDCap tool before deployment.
Outcomes:
Tools for institutional settings to actively report gastrointestinal outbreaks (childcare centres and residential aged care homes) replaced intensive phone calls with facilities. A tool for reporting rabies vaccine and immunoglobulin use by immunisation service providers using government funded stock improved the ease and accuracy of data collection. A quiz for assessing knowledge for administration of these products helps with educating providers. During Tropical Cyclone Alfred, REDCap was used to collect information on immunisation service provider cold chain breaches. This tool provided the service provider with immediate information in an automated email, allowing the Public Health Team time to triage and prioritise response after the emergency had passed.
Conclusion and future actions:
REDCap is proving to be a useful tool in the public health unit. While not yet formally evaluated, staff report that the use of these tools is easing workload, allowing them to focus on other priority work. The list of future projects continues to grow as more staff become familiar with its utility.
Mrs Jenny Watts
Public Health Epidemiologist
Western Sydney Public Health Unit
Digital Innovation in Communicable Disease Management: A Quality‑Improvement Approach in Western Sydney
Abstract
Background/Aim
Communicable diseases remain an ongoing threat to population health. Timely case investigation, contact management and rapid information sharing are essential to managing public health risks, particularly when notification surges increase workload, delay transmission control measures and contribute to staff fatigue. To improve efficiency and reduce stress during peak periods, the Western Sydney Public Health Unit introduced digital tools using GoShare for bulk SMS messaging, embedding MS Forms for rapid data collection and language-specific factsheets. Initially employed for measles responses, this approach was then used for other notifiable conditions and surge events.
Methods/Analysis
The impact of integrating digital tools into communicable disease case and contact follow-up was evaluated using routine public health data (2023–2025). Quantitative analysis included GoShare delivery metrics and MS Forms response data, and qualitative insights were drawn from team meeting minutes, after-action reviews and consumer feedback. Descriptive analysis explored trends in response rates, workload impacts and case management efficiency. Thematic analysis examined staff experience, tool usability and community perceptions of message clarity and trust.
Outcomes
The adoption of digital tools improved efficiency and effectiveness in case and contact management, with bulk messaging streamlining follow-up, reducing time-consuming calls and enabling rapid delivery of postexposure advice. In measles responses, efficiency gains were significant: a 2023 case managed using traditional follow-up methods required 19 staff over 2.5 days to manage 239 contacts, whereas comparable 2024 cases managed with GoShare (209 and 179 contacts) were completed by 4 staff in 2 days and 2 staff over a weekend. Contacts sought prophylaxis promptly after receiving messages. One symptomatic contact presented for care with the message, facilitating immediate isolation and testing. These outcomes demonstrated improved public responsiveness, increased staff capacity and reduced workload stress.
Conclusion
Integrating digital tools into routine communicable disease management improved timeliness, scalability and staff agility during surge events.
Communicable diseases remain an ongoing threat to population health. Timely case investigation, contact management and rapid information sharing are essential to managing public health risks, particularly when notification surges increase workload, delay transmission control measures and contribute to staff fatigue. To improve efficiency and reduce stress during peak periods, the Western Sydney Public Health Unit introduced digital tools using GoShare for bulk SMS messaging, embedding MS Forms for rapid data collection and language-specific factsheets. Initially employed for measles responses, this approach was then used for other notifiable conditions and surge events.
Methods/Analysis
The impact of integrating digital tools into communicable disease case and contact follow-up was evaluated using routine public health data (2023–2025). Quantitative analysis included GoShare delivery metrics and MS Forms response data, and qualitative insights were drawn from team meeting minutes, after-action reviews and consumer feedback. Descriptive analysis explored trends in response rates, workload impacts and case management efficiency. Thematic analysis examined staff experience, tool usability and community perceptions of message clarity and trust.
Outcomes
The adoption of digital tools improved efficiency and effectiveness in case and contact management, with bulk messaging streamlining follow-up, reducing time-consuming calls and enabling rapid delivery of postexposure advice. In measles responses, efficiency gains were significant: a 2023 case managed using traditional follow-up methods required 19 staff over 2.5 days to manage 239 contacts, whereas comparable 2024 cases managed with GoShare (209 and 179 contacts) were completed by 4 staff in 2 days and 2 staff over a weekend. Contacts sought prophylaxis promptly after receiving messages. One symptomatic contact presented for care with the message, facilitating immediate isolation and testing. These outcomes demonstrated improved public responsiveness, increased staff capacity and reduced workload stress.
Conclusion
Integrating digital tools into routine communicable disease management improved timeliness, scalability and staff agility during surge events.
Miss Maggie Miller
Senior Epidemiologist
Metro North Public Health Unit
User-centric approach to design an outbreak management tool for aged care facilities
Abstract
Background and aim: In the Metro North Hospital and Health Service region, the surveillance and management of outbreaks of acute respiratory infections in residential aged care facilities has traditionally been resource intensive and induced a large workload upon public health unit staff. This study involved a qualitative design to investigate the factors affecting outbreak management in aged care facilities and to incorporate the preferences and needs of end-users in a revised system. Methods and Analysis: Maximum variation purposive sampling identified aged care facility and public health unit staff to participate in a semi-structured interview. Interviews included questions relating to factors affecting outbreak management as well as the requirements of a prospective digital tool. Interviews were conducted by a single researcher and recordings were transcribed verbatim. Thematic analysis followed the Braun and Clarke method, using an inductive approach. Outcomes: Ten interviews were conducted with intended users of the digital tool. Transcript analysis identified several key factors influencing outbreak management, including the involvement of families as stakeholders, workforce capacity and confidence, access to resources and support, and the physical and policy environment. Interviewees provided suggestions to incorporate in the design of digital tool, specifically relating to its simplicity, security, timeliness, and educational components. Feedback informed the following inclusions in the revised tool: clear outbreak definition guidance, embedded links to educational resources, a basic dashboard to summarise outbreak metrics, validated data entry fields, and secure password-protect access. Conclusion and Future actions: This qualitative study highlighted the value of incorporating a user-centric approach to explore influential factors of outbreak management in aged care settings and to gauge stakeholder perspectives on a prospective digital tool for outbreak response. Findings from stakeholder interviews were used to inform the design of the digital tool prior to its trial within the Metro North Hospital and Health Service region.
