Abstract
Thunderstorm asthma events can overwhelm emergency departments within hours, making rapid detection critical for public health preparedness and patient safety. We examined whether thunderstorm asthma presentations show distinctive patterns compared to routine asthma cases in emergency department data, with an aim of enhancing the detection of thunderstorm asthma incidence. Using Victoria’s SynSurv surveillance system, we analyzed triage notes from 3 thunderstorm asthma events occurring between 2022 and 2024 in Victoria, Australia, comparing 687 thunderstorm-related presentations with 687 presentations from 8 control days. The study population consisted of emergency department attendees with respiratory presentations. Triage notes were processed using natural language processing and manual review, with demographic, symptom-onset, and medical-history data compared using chi-square tests and temporal analysis. Thunderstorm cases showed significantly higher rates of acute symptom onset within 24 hours (63.0% vs 47.2%), with presentations peaking 3-8 hours after storms and returning to baseline within 24 hours. The affected population skewed markedly toward young adults aged 21-40 years (35.9% vs 15.6% in controls). Immunological and allergic comorbidities were 3 times more prevalent in thunderstorm cases (16.6% vs 5.6%). These distinct demographic, temporal, and allergic profiles could enhance real-time surveillance and emergency preparedness through refined detection models.
Keywords: thunderstorm asthma, syndromic surveillance, emergency preparedness, natural language processing, population health
Introduction
Thunderstorm asthma is a potentially life-threatening phenomenon where meteorological conditions during thunderstorms interact with high pollen concentrations to trigger widespread asthma across affected populations.1 These events can significantly stress emergency departments, causing substantial surges in patient presentations that strain healthcare resources, challenge clinical workflows, and potentially compromise patient care.2 The scale of this threat was demonstrated in 2016 when Melbourne, in the south-eastern Australian state of Victoria, suffered the biggest thunderstorm asthma event ever recorded, affecting thousands of people and overwhelming medical systems with severe asthma cases, including ten deaths.3
The ability to rapidly detect thunderstorm asthma events has important clinical and public health implications. Emergency departments face major challenges during these unpredictable surges, which can overwhelm systems within hours.4,5 Effective surveillance also requires timely identification of respiratory threats to enable prompt public health responses such as community alerts and preventive messaging.6
Most existing research to characterize patients affected by thunderstorms has relied on hospital and emergency department records,7,8 and patient questionnaires.9,10 Although these approaches have successfully documented clinical presentations and outcomes, they predominantly use structured administrative data or processed clinical documentation. These data sources may not be available until hours or days after initial patient contact and cannot be used to assist with detecting a developing thunderstorm asthma event.
In response to the 2016 Melbourne event, the Victorian Department of Health developed “SynSurv,” a syndromic surveillance system that has operated a thunderstorm asthma module in production mode since November 2022. SynSurv collects triage notes in near real time from most public hospital emergency departments across Victoria. Triage notes offer a unique opportunity for near-real-time surveillance as they contain clinically relevant information available within minutes of patient presentation, before formal diagnostic processes are completed.
The SynSurv thunderstorm asthma module uses natural language processing (NLP) with transformer-based models to analyze triage text in real time. The current model successfully detects increases in asthma presentations at individual hospitals. However, the ability to distinguish thunderstorm-associated asthma presentations from routine asthma presentations during non-thunderstorm periods could enable even shorter detection times and more targeted responses.
This study aimed to identify the clinical, demographic, and linguistic markers within emergency department triage notes that distinguish thunderstorm asthma presentations from routine asthma cases, with the explicit purpose of providing actionable features to refine NLP detection algorithms for more precise and timely event identification.
Methods
Data
The data for this study come from the SynSurv near-real-time syndromic surveillance system, which captures clinical information from most emergency departments of public hospitals across Victoria, Australia, serving a population of 7.2 million. SynSurv collects triage notes written by emergency department nurses during their initial assessment of each patient at the point of first clinical contact. These notes possess a distinctive structure characterized by clinical abbreviations (such as “SOB” for shortness of breath) and concise phrases. Text length varied from comprehensive narratives to brief statements. Patient age and gender data, which are routinely collected alongside triage notes, were incorporated into the analysis as they provide important contextual information for characterizing respiratory presentations during thunderstorm asthma events.
All data analyzed in this study come from emergency department triage notes that were either flagged as asthma-related by a RoBERTa-based transformer classifier model developed by our team11 or manually identified as respiratory-related.
The NLP model was already deployed within the SynSurv operational system prior to this study and was developed by our team using an active learning approach with uncertainty sampling on SynSurv triage notes.11 The model was trained on SynSurv data and developed specifically to detect asthma-related presentations for monitoring sudden increases in acute asthma reporting that might indicate unfolding health events, particularly thunderstorm asthma. The model identifies asthma presentations based on symptoms with relevant history, asthma medication use, or likely asthma diagnosis, but is trained to exclude patients under 2 or over 70 years of age unless asthma is explicitly documented. On a held-out test dataset of 250 records (125 asthma and 125 non-asthma), the classifier achieved an F1-score of 0.91, precision of 0.90, and recall of 0.93.11 The study data were used in inference mode only and was not used in model training.
Manual pattern matching additionally identified patients who had no history of asthma or chronic lung disease, but who presented with a sudden onset of shortness of breath and/or wheeze. The model had not identified these non-asthmatic presentations, but temporal analysis indicated that they were a strong additional signal during thunderstorm asthma events. This identical classifier and manual augmentation process was applied to both the thunderstorm event data and the control-day data, ensuring consistent case ascertainment across both groups. The dataset comprises 2 components: records from 25 hours following 3 prospectively identified thunderstorm asthma events (see “Thunderstorm Asthma Event Identification” section) from affected hospitals and a comparison cohort from non-thunderstorm control days.
Patient demographics (age and gender) were obtained from the triage records, and key variables were extracted from the triage notes, including clinical presentation data (primary complaint and symptom duration) and medical history variables (previous respiratory conditions, documented asthma, wheeze, allergies, and other respiratory comorbidities). A clinician reviewed the analysis to ensure accuracy.
Ethical considerations
The study used routinely collected de-identified syndromic surveillance data from public hospital emergency departments across Victoria, Australia. Ethical approval was granted by Department of Health, Human Research Ethics Committee in Victoria, Australia (project ID: HREC/83486/DOH-2022-298485), with a waiver of individual consent as the data were collected for public health surveillance purposes and contained no identifiable information.
Thunderstorm asthma event identification
Thunderstorm asthma events were identified using SynSurv’s operational surveillance algorithm, which detects abnormal clustering of asthma presentations at individual emergency departments based on inter-arrival times between presentations. When consecutive presentations arrive closer than typical baseline patterns, the system generates automated alerts.
The 3 events analyzed in this study (November 6-7, 2024; November 7-8, 2023; and November 8-9, 2022) were flagged by this surveillance system during routine operations. All 3 events subsequently met established meteorological criteria for thunderstorm asthma, including documented thunderstorm activity with convergence lines and at least moderate pollen forecasts, providing independent validation of the algorithmic detection. A 25-hour window following each thunderstorm event was selected to capture the complete clinical surge associated with thunderstorm asthma, ensuring no truncation of event-related cases.
Control days were selected using the same combination of asthma model identification and manual pattern matching, and from the same November period when: (1) no surveillance alerts were generated at any ED, (2) meteorological records confirmed absence of thunderstorm and convergence line activity, and (3) days were matched for weekday to control for healthcare-seeking patterns. For 2024, 7 control days were selected (November 1, 4, 5, 12, 13, 15, and 19); for 2022, 1 control day (November 1) was selected from the same November period. Control days maintained sufficient temporal distance from thunderstorm events to avoid residual effects while remaining within the same asthma season to control for seasonal factors affecting asthma prevalence. The imbalanced ratio of control days to event days (8:3) reflects the need to ensure sufficient non-event data while avoiding days too proximate to actual events that might reflect residual effects or altered care-seeking behavior. To create balanced comparison groups for statistical testing, control-day presentations were randomly sampled from the larger pool identified across the 8 non-thunderstorm control days to match the number of thunderstorm presentations (n = 687). Unsampled control presentations were not included in the analysis.
Analysis
This retrospective comparative study analyzed respiratory presentations during thunderstorm events with the objective of characterizing the clinical and linguistic features of thunderstorm asthma. A retrospective comparative design with matched control days was employed rather than a case-crossover approach. The case-crossover design is intended to estimate the causal effect of transient exposures on acute outcomes, with each individual serving as their own control.12,13 Our objective was instead to characterize the clinical and linguistic features of thunderstorm asthma presentations for surveillance algorithm refinement, rather than to estimate the causal effect of thunderstorm exposure on individual health-seeking behavior. The latter would also require individual-level time-varying meteorological exposure data linked to exact presentation times, which are not available within the SynSurv dataset. Descriptive statistics were calculated for all variables, with categorical variables presented as frequencies and percentages. Age was grouped into 10-year categories to examine age-specific patterns.
Comparisons between thunderstorm and control periods were conducted using chi-square tests for categorical variables including gender distribution, age groups, past medical history categories, and symptom-onset duration. The chi-square test was specifically used to assess differences in immunological and allergic conditions between thunderstorm and non-thunderstorm cohorts, with statistical significance set at P < .01.
Temporal analysis of presentation patterns was conducted by calculating the percentage of total presentations occurring in each hour over the 25-hour study period following thunderstorm events. Peak presentation periods were identified by examining the distribution of cases across time intervals, with particular focus on the first 8 hours post-thunderstorm.
Daily presentation rates were calculated by dividing total presentations by the number of study days for each period (storm vs non-thunderstorm) to enable direct comparison of presentation intensity. Symptom onset was categorized as acute (≤24 hours) vs subacute/chronic (>24 hours) to examine differences in presentation acuity between thunderstorm and control periods. Gender-age interactions were examined through stratified analysis, calculating male-to-female ratios within each age group for both thunderstorm and non-thunderstorm periods.
Results
Symptom onset
Over the study period, thunderstorm days were associated with a higher proportion of acute presentations (symptoms ≤24 hours): 433/687 (63.0%) compared to 324/687 (47.2%) on non-thunderstorm days. Conversely, subacute and chronic presentations (symptoms >24 hours) were less frequent during thunderstorm days: 112/687 (16.3%) vs 241/687 (35.1%) on non-thunderstorm days (Table 1).
Table 1.
Comparison of symptom duration between thunderstorm and non-thunderstorm days.
| Thunderstorm days | Non-thunderstorm days | Difference | |||
|---|---|---|---|---|---|
| Symptom duration category | (n = 687) | (n = 687) | |||
| n | % | n | % | (%) | |
| Acute (≤24 hours) | 433 | 63 | 324 | 47.2 | 15.8 |
| Subacute/chronic (>24 hours) | 112 | 16.3 | 241 | 35.1 | −18.8 |
| Unknown/unclear | 142 | 20.7 | 122 | 17.8 | 2.9 |
| Chi-square test | χ2 = 64.3, df = 2, P < .001 | ||||
| Effect size (Cramér’s V) | 0.216 (moderate effect) | ||||
Acute category combines presentations with symptom duration ≤24 hours (sudden onset and recent); subacute/chronic category combines presentations with symptom duration >24 hours (short-term and chronic/ongoing).
Demographics
Presentation patterns showed distinct temporal characteristics across the 3 years as depicted in Figure 1, which shows the percentage of patients presenting within 25 hours following thunderstorm passage over an automatic weather station in the years 2022 to 2024.
Figure 1.
Yearly comparison of patient presentations within 25 hours following thunderstorm passage (2022-2024). Line chart showing the percentage of patients presenting within 25 hours after a thunderstorm event. Separate lines represent years 2022, 2023, and 2024. The 2024 event shows a sharper early spike (secondary Y-axis, 0%-30%) due to a single-hospital cluster, while 2022-2023 show wider regional patterns (primary Y-axis, 0-9%).
In 2023, 420 presentations were recorded from 13 hospitals, with 1 additional case each from 7 other hospitals. In 2022, there were 222 presentations from 8 hospitals. The 2024 event consisted of 31 presentations from a single regional hospital. All years demonstrated similar patterns: minimal presentations in hour 1, rapid increase peaking between hours 3 and 8, and then gradual decline toward baseline by hour 25. Peak activity concentrated in the first 8 hours: 42% of presentations in 2023, 50% in 2022, and 80% in 2024. The 2024 event is indicative of the potential suddenness of the onset of thunderstorm asthma events, compared to the other years studied where the thunderstorms covered a wider area and where more hospitals were able to absorb the intake of affected people.
Gender was recorded for 605/687 (88.1%) in the thunderstorm cohort and 670/687 (97.5%) in the non-thunderstorm cohort—shown in Table 2. The overall male-to-female ratio was 1.22:1 across both groups (699 male and 576 female).
Table 2.
Age and gender distribution of presentations.
| Thunderstorm cohort | Non-thunderstorm | |||||
|---|---|---|---|---|---|---|
| Age group | Male (% gender per age group) | Female (% gender per age group) | Total (% age group) | Male (% gender per age group) | Female (% gender per age group) | Total (% age group) |
| 0-10 | 109 (70.8%) | 45 (29.2%) | 154 (25.5%) | 201 (64.2%) | 112 (35.8%) | 313 (46.7%) |
| 11-20 | 49 (57.6%) | 36 (42.4%) | 85 (14.0%) | 37 (47.4%) | 41 (52.6%) | 78 (11.6%) |
| 21-30a | 45 (48.9%) | 47 (51.1%) | 92 (15.2%) | 19 (41.3%) | 27 (58.7%) | 46 (6.9%) |
| 31-40a | 51 (40.8%) | 74 (59.2%) | 125 (20.7%) | 23 (39.7%) | 35 (60.3%) | 58 (8.7%) |
| 41-50 | 32 (53.3%) | 28 (46.7%) | 60 (9.9%) | 26 (46.4%) | 30 (53.6%) | 56 (8.4%) |
| 51-60 | 15 (45.5%) | 18 (54.5%) | 33 (5.5%) | 19 (50%) | 19 (50%) | 38 (5.7%) |
| 61-70 | 21 (72.4%) | 8 (27.6%) | 29 (4.8%) | 16 (42.1%) | 22 (57.9%) | 38 (5.7%) |
| 71-80 | 6 (60%) | 4 (40%) | 10 (1.7%) | 12 (54.5%) | 10 (45.5%) | 22 (3.3%) |
| 81-90 | 7 (63.6%) | 4 (36.4%) | 11 (1.8%) | 8 (47.1%) | 9 (52.9%) | 17 (2.5%) |
| Over 90 | 2 (33.3%) | 4 (66.7%) | 6 (1.0%) | 1 (25%) | 3 (75%) | 4 (0.6%) |
| Total | 337 (55.7%) | 268 (44.3%) | 605 | 362 (54%) | 308 (46%) | 670 |
| Overall age distribution | χ2 = 151.34, df = 9, P < .001 | |||||
| Overall gender distribution | χ2 = 4.85, df = 1, P = .028 | |||||
Combined 21-40 years significantly higher in storm period (35.9% vs 15.6%, P < .001).
Age distribution showed a highly significant association with presentation period (thunderstorm vs non-thunderstorm) (χ2 = 151.34, df = 9, P < .001), with a moderate effect size (Cramér’s V = 0.345). Gender distribution also differed between periods (χ2 = 4.85, df = 1, P = .028), though with a small effect size (Cramér’s V = 0.062). Age-gender interactions within each period were statistically significant for both thunderstorm events (χ2 = 34.72, df = 9, P < .001) and non-thunderstorm presentations (χ2 = 28.95, df = 9, P = .001).
Age distribution patterns differed between thunderstorm and non-thunderstorm periods (P < .001). Thunderstorm presentations demonstrated a more even distribution across age groups, with the 0-10 years age bracket representing 25.5% (n = 154) of cases, followed by the 31-40 years group at 20.7% (n = 125) and the 21-30 years group at 15.2% (n = 92). The 11-20 years age group comprised 14.0% (n = 85) of thunderstorm presentations. Notably, patients aged 21-40 years collectively represented 35.9% (n = 217) of all thunderstorm-related presentations.
In contrasts, non-thunderstorm presentations were heavily concentrated in the youngest demographic, with 46.7% (n = 313) occurring in the 0-10 years age bracket. The remaining age groups showed substantially lower representation: 11-20 years (11.6%, n = 78), 31-40 years (8.7%, n = 58), 41-50 years (8.4%, n = 56), and 21-30 years (6.9%, n = 46). The combined 21-40 years age groups represented only 15.6% (n = 104) of non-thunderstorm presentations, a significantly lower proportion compared to thunderstorm events (35.9% vs 15.6%, P < .001, OR = 3.02, 95% CI, 2.28-4.00).
Patients over 40 years of age comprised 24.8% (n = 150) of thunderstorm presentations compared to 26.9% (n = 180) of non-thunderstorm presentations, indicating no significant difference in older age group representation between the 2 periods. These demographic patterns are illustrated in Figure 2.
Figure 2.
Age and gender distribution of presentations during thunderstorm and non-thunderstorm periods. Grouped bar chart comparing age and gender categories (0-10, 11-20, …, ≥90 years) between thunderstorm and non-thunderstorm cohorts. Data show higher proportions of children (0-10 years) in non-thunderstorm periods, and a more even age distribution, including greater representation of 21-40 years, in thunderstorm periods.
Gender distribution
Overall, gender distribution showed male predominance in both periods, with thunderstorm events demonstrating a slightly higher male-to-female ratio of 1.26:1 (55.7% male, 44.3% female) compared to non-thunderstorm periods (1.18:1 ratio; 54.0% male, 46.0% female). This difference in overall gender distribution was statistically significant, but with a small effect size. Male predominance was pronounced in the 0-10 year age group in both cohorts, while female predominance was observed in the 31-40 year age group during both, but was more markedly increased during thunderstorm events. These age-gender patterns were consistent across individual thunderstorm events, as shown in Figure 3.
Figure 3.
Percentage distribution of age and gender per thunderstorm event. Stacked bar chart illustrating proportions of males and females within each age group for individual thunderstorm events (2022, 2023, and 2024). The figure highlights recurring male predominance in the 0-10 year group and female predominance in the 31-40 year group, especially in 2023-2024 events.
Past medical history
Respiratory conditions: the dominant category
Respiratory conditions represented the most prevalent medical history in both cohorts. The thunderstorm cohort recorded 291 respiratory cases (61.9% of documented histories), while the non-thunderstorm cohort had 340 cases (68.4%). While respiratory conditions remained the leading category in both groups, the non-thunderstorm cohort demonstrated a 6.5 percentage point higher prevalence, suggesting a relatively lower proportion of respiratory histories among thunderstorm-related presentations.
The most notable finding emerged in immunological and allergic conditions (including hay fever, eczema, and atopic disorders), which demonstrated a 3-fold disparity between cohorts. The thunderstorm cohort recorded 78 cases (16.6% of documented histories, 95% CI, 13.4%-20.2%), compared to only 28 cases (5.6%, 95% CI, 3.8%-8.0%) in the non-thunderstorm cohort. This difference of 11.0 percentage points (95% CI, 7.8%-14.2%) highlights a substantial variation between groups.
A chi-square test of independence confirmed that the distribution of medical categories differed significantly between cohorts (χ2 = 40.0, df = 8, P < .001, Cramér’s V = 0.204), indicating a medium effect size for the association between thunderstorm exposure and medical history patterns. Table 3 depicts the relative differences between past medical histories in the storm and non-storm cohorts, highlighting respiratory and immunological conditions as the categories with the greatest differential patterns.
Table 3.
Prevalence of patient medical histories during thunderstorm and non-thunderstorm periods.
| Medical category | Thunderstorm cohort n (%) | Non-thunderstorm cohort n (%) | Total | Difference (thunderstorm − non-thunderstorm) |
|---|---|---|---|---|
| Respiratory | 291 (61.9%) | 340 (68.4%) | 631 | −6.5% |
| Immunological allergic | 78 (16.6%) | 28 (5.6%) | 106 | 11.0% |
| Neurological psychiatric | 18 (3.8%) | 21 (4.2%) | 39 | −0.4% |
| Cardiovascular | 12 (2.6%) | 22 (4.4%) | 34 | −1.9% |
| Renal urological | 9 (1.9%) | 5 (1.0%) | 14 | 0.9% |
| Endocrine metabolic | 8 (1.7%) | 15 (3.0%) | 23 | −1.3% |
| Pediatrics/neonatal | 4 (0.9%) | 14 (2.8%) | 18 | −2.0% |
| Oncology | 4 (0.9%) | 8 (1.6%) | 12 | −0.8% |
| Other | 46 (9.8%) | 44 (8.9%) | 90 | 0.9% |
| Total | 470 | 497 | 967 | |
| Chi-square test | χ2 = 40.0, df = 8, P < .001 | |||
| Effect size (Cramér’s V) | 0.204 (moderate effect) | |||
Discussion
This study identified clear clinical, demographic, and temporal patterns that distinguish thunderstorm asthma from routine asthma presentations. Key findings include a marked increase in acute-onset symptoms (≤24 hours), a bimodal age distribution with a pronounced peak in young adults (21-40 years), and a 3-fold higher prevalence of immunological and allergic conditions. These patterns provide strong evidence that thunderstorm asthma represents a distinct clinical phenomenon rather than a simple exacerbation of baseline asthma.
The notable shift toward acute presentations during thunderstorm days provides important insights into the pathophysiology of thunderstorm-related respiratory symptoms. The 15.8 percentage point increase in acute presentations (symptoms ≤24 hours) suggests that thunderstorms predominantly trigger rapid-onset respiratory symptoms, which is consistent with the established clinical characteristics of epidemic thunderstorm asthma.14,15 The concurrent 18.8 percentage point decrease in subacute and chronic presentations (symptoms >24 hours) during thunderstorm days indicates that patients with pre-existing, ongoing respiratory conditions are not disproportionately seeking care during these events.
The temporal specificity of thunderstorm-related presentations is reflected in distinctive language patterns in triage notes. Clinical notes during thunderstorm events frequently contain precise temporal markers indicating rapid symptom onset, such as “coughing fit @ 2230” and “30/60 ago onset SOB.” This linguistic signature distinguishes thunderstorm-related presentations from typical respiratory complaints with less precisely documented onset timing.
The distinctive patterns identified in this study provide actionable features for enhancing real-time thunderstorm asthma detection algorithms. Building upon existing NLP classifier frameworks that identify asthma presentations through symptom-history combinations and clinical assessments, additional high-value distinguishing features emerge: acute-onset timing with precise temporal markers in triage notes, young adult age groups, and documented allergic conditions showing higher prevalence during thunderstorm events.
A refined detection approach could flag standard asthma presentations as potential thunderstorm cases when they include documented hay fever/eczema/atopic conditions and acute-onset language patterns, while recognizing the relative absence of established respiratory histories during thunderstorm events. This enhancement could enable surveillance systems to distinguish thunderstorm-characteristic presentations as an adjunct to general asthma increases, potentially reducing detection time and enabling targeted alerts for at-risk atopic populations before widespread emergency department surges develop. The predictable temporal pattern (peak presentations 3-8 hours post-thunderstorm) provides a critical 2-3 hour preparation window for healthcare systems, though the rapid escalation requires flexible surge capacity planning.
Given that climate change is expected to increase the frequency of thunderstorm asthma events through prolonged pollen seasons and more frequent extreme weather events,16 our findings demonstrate the potential for triage note analysis to provide essential near-real-time surveillance capabilities. The combination of temporal precision (specific times, recent-onset markers) and acute symptom descriptors (sudden awakening, rapid deterioration, and precise time stamps) creates a textual fingerprint that can enhance automated case detection algorithms. This linguistic approach to case identification could complement traditional diagnostic coding methods, particularly important given that thunderstorm asthma may not be immediately recognized or appropriately coded during acute presentations.
Thunderstorm asthma events demonstrate a characteristic demographic profile that differs fundamentally from typical asthma presentations. The bimodal age distribution, with peaks in both pediatric (0-10 years) and young adult (21-30 years) populations, contrasts sharply with the heavy pediatric concentration seen during non-thunderstorm periods. The predominance of young adults aged 21-40 years, representing over one-third of thunderstorm cases vs only 15.6% of non-thunderstorm presentations, may reflect greater outdoor exposure and occupational activities in this demographic during adverse weather events.
Our findings align with emerging evidence that thunderstorm asthma reflects a unique exposure rather than a simple exacerbation across all demographic groups.17 Studies have consistently demonstrated that thunderstorm asthma disproportionately affects adults while showing minimal impact on pediatric populations.17-19 This pattern directly contradicts baseline asthma demographics, where children typically experience the highest emergency department visit rates and represent the most vulnerable population for asthma exacerbations.20 The low representation of patients over 40 years during thunderstorm events, despite our study showing some presentations in this age group, indicates either protective factors in older populations or fundamental differences in susceptibility mechanisms. The concentration of risk in specific adult age groups suggests that behavioral factors, occupational exposures, or physiological changes across the life course may influence thunderstorm asthma susceptibility.15
The age-specific gender patterns reveal intriguing variations that may provide insights into underlying mechanisms. The strong male predominance in the youngest age group aligns with typical childhood asthma patterns, where males show higher incidence rates,21 but becomes more pronounced during thunderstorm events. However, the shift to female predominance specifically in the 31-40 years age group during thunderstorm events contrasts with the balanced gender distribution in adjacent age groups and differs from studies identifying men aged 18-44 as the highest risk group.17 These variations may reflect regional differences in exposure patterns, occupational factors, or the well-established hormonal influences on asthma pathogenesis.22
The 3-fold increase in presentations among patients with immunological/allergic conditions (16.6% vs 5.6%, P < .001) provides compelling evidence that thunderstorm asthma affects atopic individuals.23 This finding has critical clinical implications, as many affected individuals may not recognize their respiratory risk or be adequately prepared for severe symptoms, potentially leading to delayed self-management and higher emergency department acuity. The demographic profile concentrating in young adults with allergic histories rather than established respiratory patients necessitates targeted public health messaging for atopic individuals who may not recognize their risk. Healthcare systems should prepare resource allocation and staffing strategies specifically for the 21-40 age demographic, who represent the highest relative increase during events.
This study has some limitations. First, the study analyzed only 3 thunderstorm events from consecutive years, with the 2024 event comprising presentations from a single regional hospital (n = 31), limiting the robustness of observed patterns and requiring validation across a larger number of independent events. Second, the imbalanced control day selection (8 non-thunderstorm days vs 3 event days) may introduce temporal bias, though days were selected to avoid residual effects while maintaining seasonal comparability. Third, the 25-hour observation window, while capturing the acute surge, does not account for delayed presentations or longer-term health impacts of thunderstorm exposure. Fourth, this analysis did not examine patient-level geographic indicators and therefore did not account for local socio-economic or environmental factors that may influence thunderstorm asthma susceptibility and health-seeking behavior. While participating hospitals were located within affected regions for each event, Victoria encompasses diverse populations and meteorological zones, and the results should be interpreted considering this geographic heterogeneity. Despite these limitations, the consistent patterns observed across 3 independent events and the large sample size provide valuable insights for enhancing real-time detection systems.
Conclusions
The identification of key differentiating features, acute symptom-onset markers, documented hay fever/eczema/atopic conditions, and specific demographic patterns offers actionable inputs for improving automated surveillance systems. These findings support the development of targeted detection criteria that could reduce surveillance response times and enable more precise public health alerts for at-risk atopic populations. The substantial effect sizes observed across demographic and clinical variables establish thunderstorm asthma as a predictable public health phenomenon requiring specialized monitoring approaches and population-specific emergency preparedness protocols.
Acknowledgments
The authors acknowledge the emergency department nurses who created the triage documentation and the SynSurv development team for their contributions.
Contributor Information
Sedigh Khademi, Health Informatics Group, Centre for Health Analytics, Melbourne Children's Campus, Melbourne, Australia; Department of Pediatrics, University of Melbourne, Melbourne, Australia.
Gerardo Luis Dimaguila, Health Informatics Group, Centre for Health Analytics, Melbourne Children's Campus, Melbourne, Australia; Department of Pediatrics, University of Melbourne, Melbourne, Australia.
Christopher Palmer, Health Informatics Group, Centre for Health Analytics, Melbourne Children's Campus, Melbourne, Australia.
Muhammad Javed, Health Informatics Group, Centre for Health Analytics, Melbourne Children's Campus, Melbourne, Australia.
Jim Buttery, Health Informatics Group, Centre for Health Analytics, Melbourne Children's Campus, Melbourne, Australia; Department of Pediatrics, University of Melbourne, Melbourne, Australia; Infectious Diseases, Royal Children’s Hospital, Melbourne, Australia.
Jim Black, Melbourne School of Population and Global Health, University of Melbourne, Melbourne, Australia; Department of Health, State Government of Victoria, Melbourne, Australia.
Conflicts of interest
None declared.
Funding
None declared.
References
- 1. D’Amato G, Liccardi G, Frenguelli G. Thunderstorm-asthma and pollen allergy. Allergy. 2007;62(1):11-16. 10.1111/j.1398-9995.2006.01271.x [DOI] [PubMed] [Google Scholar]
- 2. D’Amato G, Annesi-Maesano I, Urrutia-Pereira M, et al. Thunderstorm allergy and asthma: state of the art. Multidiscip Respir Med. 2021;16(1):806. 10.4081/mrm.2021.806 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Thien F, Beggs PJ, Csutoros D, et al. The Melbourne epidemic thunderstorm asthma event 2016: an investigation of environmental triggers, effect on health services, and patient risk factors. Lancet Planet Health. 2018;2(6):e255-e263. 10.1016/S2542-5196(18)30120-7 [DOI] [PubMed] [Google Scholar]
- 4. Hew M, Sutherland M, Thien F, et al. The Melbourne thunderstorm asthma event: can we avert another strike? Intern Med J. 2017;47(5):485-487. 10.1111/imj.13413 [DOI] [PubMed] [Google Scholar]
- 5. Elliot AJ, Hughes HE, Hughes TC, et al. The impact of thunderstorm asthma on emergency department attendances across London during July 2013. Emerg Med J. 2014;31(8):675-678. 10.1136/emermed-2013-203122 [DOI] [PubMed] [Google Scholar]
- 6. Andrew E, Nehme Z, Bernard S, et al. Stormy weather: a retrospective analysis of demand for emergency medical services during epidemic thunderstorm asthma. BMJ. 2017;359:j5636. 10.1136/bmj.j5636 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Lee J, Kronborg C, O’Hehir RE, et al. Who’s at risk of thunderstorm asthma? The ryegrass pollen trifecta and lessons learnt from the Melbourne thunderstorm epidemic. Respir Med. 2017;132:146-148. 10.1016/j.rmed.2017.10.012 [DOI] [PubMed] [Google Scholar]
- 8. Stewart CG, Mahesh A, Mulvenna C. Thunderstorm asthma: a paediatric emergency department experience in London. BMJ Paediatr Open. 2024;8(1):e002572. 10.1136/bmjpo-2024-002572 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Douglass JA, Lodge C, Chan S, et al. Thunderstorm asthma in seasonal allergic rhinitis: the TAISAR study. J Allergy Clin Immunol. 2022;149(5):1607-1616. 10.1016/j.jaci.2021.10.028 [DOI] [PubMed] [Google Scholar]
- 10. Girgis ST, Marks GB, Downs SH, et al. Thunderstorm-associated asthma in an inland town in south-eastern Australia: who is at risk? Eur Respir J. 2000;16(1):3-8. 10.1034/j.1399-3003.2000.16a02.x [DOI] [PubMed] [Google Scholar]
- 11. Khademi S, Palmer C, Javed M, et al. Detecting asthma presentations from emergency department notes: an active learning approach. In: Proceedings of the 21st Australasian Data Mining Conference (AusDM 2023). Vol. 1943. Springer Nature; 2023. [Google Scholar]
- 12. Lewer D, Petersen I, Maclure M. The case-crossover design for studying sudden events. BMJ Med. 2022;1(1):e000214. 10.1136/bmjmed-2022-000214 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol. 1991;133(2):144-153. 10.1093/oxfordjournals.aje.a115853 [DOI] [PubMed] [Google Scholar]
- 14. Chatelier J, Chan S, Tan JA, et al. Managing exacerbations in thunderstorm asthma: current insights. J Inflamm Res. 2021;14:4537-4550. 10.2147/JIR.S324282 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Harun NS, Lachapelle P, Douglass J. Thunderstorm-triggered asthma: what we know so far. J Asthma Allergy. 2019;12:101-108. 10.2147/JAA.S175155 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Thien F, Davies JM, Douglass JA, et al. Thunderstorm asthma: current perspectives and emerging trends. J Allergy Clin Immunol Pract. 2025;13(6):1273-1280. 10.1016/j.jaip.2025.04.001 [DOI] [PubMed] [Google Scholar]
- 17. Smith ML, MacLehose RF, Wendt CH, et al. Sex and age characteristics of thunderstorm asthma emergency department visits. Hyg Environ Heal Adv. 2024;11:100099. 10.1016/j.heha.2024.100099 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Forouzan A, Masoumi K, Haddadzadeh Shoushtari M, et al. An overview of thunderstorm-associated asthma outbreak in southwest of Iran. J Environ Public Health. 2014;2014(1):504017. 10.1155/2014/504017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Hew M, Lee J, Susanto NH, et al. The 2016 Melbourne thunderstorm asthma epidemic: risk factors for severe attacks requiring hospital admission. Allergy. 2019;74(1):122-130. 10.1111/all.13609 [DOI] [PubMed] [Google Scholar]
- 20. (CDC) Centers for Disease Control and Prevention . Asthma emergency department (ED) visits 2010-2018. Published 2021. Accessed July 19, 2025. https://archive.cdc.gov/#/details?url=https://www.cdc.gov/asthma/asthma_stats/asthma-ed-visits_2010-2018.html
- 21. Johnson CC, Chandran A, Havstad S, et al. US childhood asthma incidence rate patterns from the ECHO consortium to identify high-risk groups for primary prevention. JAMA Pediatr. 2021;175(9):919-927. 10.1001/jamapediatrics.2021.0667 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Chowdhury NU, Guntur VP, Newcomb DC, et al. Sex and gender in asthma. Eur Respir Rev. 2021;30(162):210067. 10.1183/16000617.0067-2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Rangamuwa KB, Young AC, Thien F. An epidemic of thunderstorm asthma in Melbourne 2016: asthma, rhinitis, and other previous allergies. Asia Pac Allergy. 2017;7(4):193-198. 10.5415/apallergy.2017.7.4.193 [DOI] [PMC free article] [PubMed] [Google Scholar]