Enhancing Vaccine Safety Through AI-Powered Social Media Surveillance
Purpose and Statement of Need

This award entry seeks support for the continued development of an innovative AI-powered social media surveillance system that detects adverse events following immunisation (AEFIs) in near real-time. While vaccines undergo rigorous pre-licensure testing, traditional spontaneous reporting systems like SAEFVIC face limitations including underreporting and delayed signal detection. Our system addresses these challenges by leveraging social media data—where vaccine experiences are shared in real-time—enhanced by advanced artificial intelligence technologies to identify safety signals earlier and more comprehensively.

Work to Date
Our research has demonstrated results using social media for vaccine safety monitoring:
• Development of topic modelling with custom scoring techniques that isolated posts containing vaccine adverse event mentions with 99% accuracy.
• Creation of a Transformer-based classifier to detect vaccine-related personal health mentions across social media platforms.
• Successful detection of the AstraZeneca clotting signal weeks before official recognition by regulatory authorities.
• Exploration of large language models (LLMs) to enhance classifier development and optimize data labelling processes.

Implementation Plan
Our 12-month implementation strategy follows three phases:
1. Expanding Data Collection (Months 1-4): Broadening our dataset beyond text to incorporate multimodal data from multiple platforms including Reddit, YouTube, and BlueSky.
2. Developing Classification Algorithms (Months 5-8): Training and refining AI models to process multimodal data efficiently with enhanced precision and recall.
3. Extracting & Normalizing Vaccine Reactions (Months 9-12): Transforming detected AEFI mentions into structured medical reports compatible with official surveillance systems.

Background
Vaccination is one of public health’s greatest achievements, second only to clean water in preventing death and disease globally. While intensified scrutiny obviously accompanies new vaccines, variants, or use in new populations, sustained monitoring of established vaccines is equally critical to ensure detection of rare adverse events, identify risks in sub-populations and rapidly recognise anomalies by brand, batch, provider-group, recipient characteristics, or even spurious findings. Because safety signals may emerge from multiple pathways, effective vigilance demands diverse surveillance systems and analytic methods to ensure that no meaningful signal goes unnoticed.

Aim
To describe Victoria’s routine weekly vaccine safety vigilance supporting safe immunisation program delivery.

Methods & Analysis
Victoria has established rigorous multi-modal routine vaccine safety surveillance. Each week, spontaneous reports to SAEFVIC are systematically reviewed for disproportionality in vaccine-adverse-event pair ratios and as rates per 100,000 doses administered, stratified by sex, age-group and vaccine target disease; influenza, COVID and RSV are additionally analysed by brand. Serious adverse-event reports are line-list reviewed for epidemiological clustering. Telephone helpline and emergency department presentations are monitored using modified CUSUM-based statistics to detect deviation from historic patterns in post-vaccination contacts or within-28-day presentations, with ability to drill into event-specific topics when increases are observed. Monthly VaxPulse social-listening monitors online personal adverse-event discussions and misinformation topic-trends for prioritised vaccines.

Signal hypotheses are assessed for cross-method signalling, biological plausibility, known associations, and clinical significance to determine whether they require escalation to enhanced monitoring or investigation, or return to routine surveillance.

Outcome
Robust, multi-method vaccine safety monitoring sustains provider and consumer confidence in immunisation programs.

Conclusion
Victoria’s integrated, multi-source, sex- and age-disaggregated routine vaccine safety surveillance is a global exemplar for contemporary immunisation safety and provides an aspirational model for ongoing assurance that vaccine programs remain safe while enabling rapid detection and response to emerging concerns.

Background and Aim
Maternal vaccination against respiratory syncytial virus (RSV) (Abysvo) was introduced in Australia on 01-February-2025 under the National Immunisation Program (NIP) to protect infants through passive immunity. Timely safety monitoring is essential to support public confidence. This analysis aimed to describe the shorterm safety profile of maternal RSV vaccination using active surveillance data.

Methods and Analysis
We assessed the short-term safety of Abysvo, including co-administration with other vaccine(s), among pregnant women aged 15–49 years who received Abrysvo from 28 weeks’ gestation between 1-February-2025 and 31-January-2026 at participating general practices in AusVaxSafety. Online surveys solicting adverse events following immunisation (AEFI) were sent three days post-vaccination. Demographic and vaccination data were retrieved from general practice records.

Outcomes
A total of 1,715 pregnant women (median age 33 years [IQR 30–36]) responded to the survey (response rate 29%), with 873 (51%) receiving Abrysvo alone and 842 (49%) receiving concomitant vaccines. Pertussis was the most common additional vaccine (618, 73%). Most respondents (96%) were vaccinated between 28 and 36 weeks’ gestation. Overall, 510 (29.7%) reported at least one AEFI within three days post-vaccination (26.8% Abrysvo alone and 32.8% concomitant Abrysvo). Medical review was uncommon (overall: 1.6%, Abrysvo alone: 1.7%; concomitant Abrysvo: 1.4%), including 0.3% emergency department visits. The most frequently reported symptoms were local reactions (24.7%), fatigue (18.3%), myalgia (12.9%), and headache (12.3%), reported slightly more often in the concomitant group. Cardiorespiratory symptoms (palpitations, chest pain/discomfort, dyspnoea) were rare (0.35%). Daily activities were affected in 6.1% of respondents (Abrysvo alone: 4.9%; concomitant Abrysvo: 7.4%), with most impacts (67%) lasting one day or less.

Conclusion and Future Actions
Active surveillance identified predominantly mild and short-lived AEFI following maternal RSV vaccination. Medical review rate was low in both groups, suggesting no need to separate RSV from other maternal vaccines. Continued monitoring will support public confidence and future work should focus on longer-term follow-up.

Background and Aim: The benefits of influenza vaccination extend beyond influenza prevention, to reductions in secondary complications and associated healthcare utilisation. We recently demonstrated 19.6% relative effectiveness of cell-based compared to egg-based influenza vaccines against test-confirmed influenza among individuals aged 6 months to 17 years during the United States 2023-2024 influenza season. Modelling translated this to 605,278 additional influenza outpatient visits prevented if all paediatric vaccinated individuals received cell-based instead of egg-based vaccines. Here, we estimate secondary complications attributable to influenza to inform the broader public health benefit of cell-based influenza vaccines.

Methods and Analysis: Within the vaccinated paediatric population included in the United States 2023-2024 effectiveness study, the risk of selected secondary events within 14 days following test-confirmed influenza cases was compared to the risk within the same period in up to 10 controls without influenza, matched on sex, age, and region. We estimated relative risk and risk difference, with the latter interpreted as excess risk attributable to influenza. The burden of secondary events that could be additionally prevented by use of cell-based vaccine was calculated by applying the estimated risk difference to the previously estimated 605,278 additional influenza outpatient visits prevented.

Outcomes: Influenza was associated with 6- to 20-fold increases in selected secondary complications and a 7-fold increase in antibiotic use. Use of cell-based instead of egg-based vaccine was estimated to additionally prevent 23,364 episodes of acute otitis media, 12,408 asthma exacerbations, 5,932 episodes of acute bronchitis and 1,937 bacterial pneumonia cases, as well as 92,486 antibiotic prescriptions.

Conclusion and Future actions: In the paediatric population, use of a more effective cell-based influenza vaccine could confer substantial public health benefits beyond influenza prevention, through clinically meaningful reductions in secondary complications and antibiotic use. These findings support consideration of downstream outcomes, including antibiotic prescribing, in influenza vaccination strategy evaluation.

Understanding the short-term safety of concomitant vaccination in older adults is essential for public confidence and vaccine uptake, as co-administration reduces missed vaccination opportunities and improves protection in a population at higher risk of severe disease.

We assessed the short-term safety of influenza, COVID-19, recombinant zoster vaccine (Shingrix®) and 13vPCV (13-valent pneumococcal conjugate vaccine) vaccines when administered alone or concomitantly to adults aged ≥60 years, between 1 January 2024 and 31 December 2025 at participating AusVaxSafety sentinel sites. We analysed adverse event following immunisation (AEFI) reports collected three days post-vaccination and compared the proportions of AEFI reports, medical attendance (MA; self-reported ED or GP visit), and the most common AEFI between vaccines administered alone and those given concomitantly.

Among our vaccine recipients (N=517,901, 58.2% female, median [IQR] age 73 [68,78]), 86.7% received a single vaccine, while 13.3% received two or more vaccines on the same day. Most received either influenza (37.2%) or Shingrix (28.2%) alone; the most common concomitant vaccination combination was influenza+COVID-19 (7.8%). Overall, 139,966 respondents (27.0%) reported an AEFI within three days post-vaccination. AEFI reporting increased with concomitant administration compared to standalone administration for influenza (12.0% alone; 27.2% concomitant)), COVID-19 (23.7% alone; 27.3% concomitant), and 13vPCV (13.3% alone; 26.7% concomitant). In contrast, AEFI reporting was higher for Shingrix when given alone (47.8%) compared to concomitant administration (42.2%). Across all vaccine combinations, the most frequently reported symptoms were local pain (8.3-43.3%), fatigue (6.1-30.9%) and myalgia (4.1-21.6%). MA for adverse events was overall similar between vaccines given alone compared to those given concomitantly (0.2%, all vaccine combinations).

Overall, while there was a higher proportion of AEFI reporting with concomitant vaccination, they were mostly expected side effects, and MA rates remained low and comparable to standalone vaccination. This supports the practice of concomitant vaccination with some communication around expected side effects.

Background and Aim
Influenza vaccination coverage among Australian children under 5 years declined from 44.6% in 2020 to 25.7% in 2025. Parental concerns about vaccine reactogenicity may be one factor influencing vaccine uptake. This study quantifies adverse events (AE) following influenza vaccination in young children and time loss from carers’ daily activities after vaccination.

Methods and Analysis
Online surveys were sent between 2022 to 2025 to carers of influenza vaccinees in this age group three days post-administration, soliciting reports of AE as measures of short-term reactogenicity and impact (time loss) on daily activities. Bayesian hierarchical modelling was used to estimate the probability of experiencing an AE, seeking medical advice for an AE, and any loss of time for normal daily activities after vaccination, adjusted for age, sex, year, and co-administered vaccines.

Outcomes
Of the 144,329 surveys sent, 40,940 (28.4%) responses were received within seven days post-vaccination. Unadjusted proportions of responders with self-report of any AE were similar across 2022-2025, sex and age. The aggregated proportions were 20.3% for Fluarix Tetra, 21.8% for Flucelvax Quad, 23.2% for FluQuadri, and 22.2% for Vaxigrip Tetra. Across these vaccines, the model-adjusted probability of any AE and of any loss of time for daily activities was <25% and <1.5%, respectively, and the proportion of respondents who sought medical advice was only 1.5-1.7%. The most frequently reported AEs were 12.3-15.1% for injection site reaction, 10.9-13.2% for fatigue and 8.0-10.8% for fever.

Conclusion and Future actions
Reactions are reported in ~20% of children under 5 years after influenza vaccination, although their causal attribution to the vaccine is uncertain due to absence of a comparator. The reactions are typically mild, rarely require medical review, and have minimal disruption of daily activities. These findings support vaccine providers in addressing parental reactogenicity concerns and may contribute to improved uptake.

Background
The study aimed to evaluate the safety and impact of maternal influenza, pertussis, COVID-19 vaccine on perinatal outcomes and identify determinants of uptake among pregnant women in South Australia.

Methods
This retrospective cohort study included 19,098 pregnant women who delivered at the Women’s and Children’s Hospital in South Australia between 2020 to 2023. Cox proportional-hazards models estimated hazard ratios (HRs) for time-sensitive perinatal outcomes, and log-binomial models were used for time-independent outcomes. Poisson regression models were used to estimate adjusted prevalence ratios (aPRs) to identify factors influencing maternal vaccine uptake.

Results
Overall, 64.1% received influenza, 74.1% pertussis, and 29.5% received ≥1 COVID-19 dose during pregnancy. Maternal influenza, pertussis, and COVID-19 vaccination were not associated with increased risks of hypertensive disorders, chorioamnionitis, premature rupture of membranes, antepartum or postpartum haemorrhage, small-for gestational-age, or low birth-weight. Following maternal influenza vaccination, a 13% protective effect against preterm birth was observed, with similar reductions after pertussis (17% and nearly 50% fewer stillbirths) and COVID-19 vaccination (aHR 0.79, 95% CI 0.68–0.92). Both maternal influenza (aRR 0.94, 95% CI 0.91–0.98) and pertussis vaccination (aRR 0.89, 95% CI 0.86–0.92) were associated with a lower likelihood of experiencing at least one obstetric complication. A small increase in diagnosed gestational diabetes was observed among women vaccinated against influenza or pertussis. Neonatal outcomes were favourable, with lower rates of resuscitation, early-onset sepsis, and low Apgar scores among infants of vaccinated mothers. Younger, multiparous, unmarried, and First Nations women were less likely to be vaccinated, with lower uptake also observed among mothers whose infants were discharged to non-parental care.

Conclusions
Findings reaffirm the safety and added benefits of maternal influenza, pertussis, and COVID-19 immunisation, and highlight persistent inequities in uptake that require targeted, culturally safe strategies to ensure equitable protection for all pregnant women and their infants.

Background
Australia is on track to eliminate cervical cancer, driven by sustained high coverage of the national HPV vaccination program. Victoria’s continuous, population-based vaccine safety surveillance since program inception generates one of the longest-running HPV safety datasets globally—from early post-licensure monitoring of quadrivalent vaccines in girls to current surveillance of nonavalent vaccines across all adolescents.

Aim: To describe Victoria’s long-term HPV vaccine safety surveillance findings.

Methods & Analysis
Adverse events following immunisation (AEFI) reported from 2007–2025 were analysed by age/sex, seriousness and temporal trends in relation to programmatic and vaccine changes. Findings were contextualised using sex-disaggregated background rates and targeted signal investigations addressing clinical and consumer concerns.

Of 2,258 AEFI reported (rate 65.0 per 100,000 doses), reactions were predominantly mild and self-limiting (18% syncope, 12% rash, 10% headache). Serious AEFI were uncommon (5.0%) with no new safety signals. AEFI reporting rate by sex was similar overall (male:female (M:F) risk ratio (RR) 1.13) and for serious AEFI (p=0.781) but with notable sex disparities with fever (M:F RR 1.82, p<0.001), syncope (1.50, p<0.001) higher in males and allergic rash higher in females (0.57, p=0.003). Four reports (<0.2%) related to postural intolerance, an oft-mentioned consumer concern. Male preponderance of Guillain Barre syndrome (RR 2.17, p=0.596) was consistent with background rate expectations. Reporting patterns aligned appropriately with population and 3,2,1 dose scheduling shifts, while remaining stable across vaccine formulation and delivery setting evolutions.

Outcome
Victoria’s long-standing HPV vaccine safety surveillance underscores the role of transparent safety systems in sustaining public confidence, high coverage and program continuity across generations, integral to Australia’s cervical cancer elimination trajectory.

Conclusion
With national coverage declining since 2020, maintaining elimination gains requires ongoing, credible safety monitoring—paired with proactive communication to address real and misinformation-driven concerns—to support program resilience and continued regional scale-up of HPV prevention.