Mapping febrile seizure clusters from emergency department timestamps
When a parent rushes a toddler into an emergency department during a febrile convulsion, the clock starts ticking in ways that public health teams have only recently begun to exploit. The triage screen captures the arrival time, the nursing assessment field logs the temperature spike, and the treating clinician records the diagnosis within minutes. Each of those small digital breadcrumbs is a timestamp, and when millions are aggregated across a health system they can form an early-warning signal for circulating infections, sometimes days before laboratory confirmation catches up.
Australia's paediatric emergency services treat thousands of simple febrile seizures every year, and the pattern in which they appear carries information about influenza, respiratory syncytial virus, human herpesvirus 6, and other pathogens that peak in this country's warm season. Treating the timestamp stream from emergency presentations as a continuous surveillance channel, rather than waiting for a coded discharge summary or a positive PCR, opens a window for earlier action. That is the practical premise behind using ED visit timestamps to identify clustering of febrile seizures in children.
How ED timestamps reveal early signals
Modern triage systems such as those used in the Sydney Children's Hospitals Network and the Royal Children's Hospital in Melbourne record several distinct time markers for every child: registration, initial nursing assessment, medical officer review, and disposition. Each is captured to the minute and stored in the patient administration system. The chief complaint field, usually free text or a drop-down menu, often contains the words "fitting," "febrile convulsion," or "seizure with fever" in the minutes immediately after a parent arrives.
Parents in Australia typically present to an emergency department very quickly once a seizure has occurred, partly because ambulance services encourage rapid transport for any child under five who has had a convulsion. That short lag means the cluster signal is sharp: a sudden rise in arrival times tagged with febrile seizure symptoms is unlikely to be a slow-moving phenomenon. Public health teams monitoring the data feed in near real time can see the bump forming hours, not weeks, after the underlying viral activity has begun to spread.
The vocabulary of the chief complaint is where the discipline comes in. Many hospitals use synonyms that must be harmonised before analysis: "febrile fit," "convulsion with fever," "hot and fitting." A small mapping exercise at the start of the surveillance programme pays dividends later, because it stops the algorithm from missing cases that would otherwise be obvious to a clinician reading the notes.
Why febrile seizures are a useful indicator condition
Febrile seizures are common enough to produce stable statistics, bounded enough in age to keep the denominator clean, and biologically linked to the infections that public health teams care about. They occur in roughly two to five per cent of children between six months and five years, peaking around 18 months. In Australia, the warmest months from December through February often see the highest rates because influenza and enteroviruses co-circulate during the summer holiday period, when families travel and gather in ways that accelerate transmission.
The age band matters for the math. Unlike an all-age signal such as ambulance dispatches for breathing difficulty, febrile seizures sit within a narrow paediatric window, which means the population at risk is well-defined and per-capita rates can be calculated accurately from Australian Bureau of Statistics projections. When the rate climbs above a moving baseline, it is more likely to reflect a real change in community transmission than a shift in coding practice or hospital catchment.
The clinical presentation is also dramatic and memorable. Parents rarely wait until morning to seek help, and ED staff rarely miss the diagnosis. That makes the signal robust to under-ascertainment, which is a constant worry in milder syndromes like gastroenteritis or upper respiratory complaints. That combination is what makes febrile seizures such a productive target for cluster detection.
Statistical methods for cluster detection
The workhorse of cluster detection in syndromic surveillance is the temporal scan statistic developed by Kulldorff and implemented in SaTScan. For ED timestamp data, the algorithm slides a window of variable width across the time series and asks, for each window, whether the observed count of febrile seizure presentations is higher than would be expected under a null model of random timing. When the window covers a real outbreak, the test statistic climbs; when it does not, the statistic stays close to its expected distribution.
Practical implementations in Australia often use a space-time permutation model rather than a pure temporal scan. This approach compares the pattern of ED arrivals inside a defined hospital catchment against the pattern in the surrounding state, which filters out statewide events already well known. The prospective version of the test matters most for outbreak response, because it only looks forward in time and flags clusters as they form. Retrospective analysis is reserved for end-of-season review and model refinement.
Setting the threshold is a judgement call. A high threshold reduces false alarms but risks missing small clusters in rural Western Australia or the Northern Territory, where ED volumes are lower. A low threshold catches more signals but generates alerts that public health teams must triage. The most successful programmes, including the one that links Queensland Health ED data to the Public Health Laboratory Network, run two thresholds: a soft one above the 95th percentile, and a hard one above the 99th that triggers an immediate investigation.
Operational workflow from triage to public health action
Once a cluster is flagged, the next step is to push the signal out to the people who can do something about it. In NSW, that means the Health Protection NSW team; in Victoria, the Communicable Disease Section at the Department of Health; in Queensland, the regional public health units. Each of these bodies has standing arrangements to receive automated feeds from major hospital networks, and febrile seizure clustering fits naturally alongside school absenteeism monitoring, pharmacy over-the-counter sales for paediatric antipyretics, and ambulance dispatch data.
The workflow typically starts at the hospital's ED information system, where a nightly extract pulls all presentations with the relevant chief complaint codes or free-text flags. The extract is sent through a secure channel to the state health department, where it joins a wider data warehouse. The cluster detection algorithm then runs, and any alert is routed to an on-call epidemiologist who checks the signal against laboratory reports, looks for a plausible infectious driver, and decides whether to issue a clinician alert, a public-facing message, or both.
Useful adjacencies include school health room records for cross-referencing absenteeism spikes, pharmacy sales of paediatric paracetamol and ibuprofen as a complementary demand signal, and ambulance dispatch codes for convulsion. The multi-channel system in Australia, integrating clinics, hospitals, ambulances, pharmacies, schools, and aged-care facilities, means a febrile seizure cluster is rarely interpreted in isolation. The convergence of several signals pointing in the same direction gives public health teams the confidence to act.
Australian paediatric context and local realities
Australian paediatric practice has a few wrinkles worth noting. The Royal Australian College of General Practitioners' Red Book guidelines shape how children are managed in primary care, and many simple febrile seizures are never seen in an ED at all, particularly in urban Sydney or Melbourne where after-hours GP services are common. That means the ED signal is the tip of a larger iceberg, with the submerged portion estimated from sentinel general practice networks such as ASPREN and the Paediatric Active Enhanced Disease Surveillance (PAEDS) platform. The state-level systems are calibrated against those sentinel sources, not a replacement for them.
Climate and geography also play their part. The 2019-2020 bushfire season left a long tail of poor air quality across the eastern seaboard, and clinicians in Brisbane and Adelaide reported a noticeable rise in paediatric respiratory illness afterwards. Febrile seizure presentations moved in step with that rise, showing how the timestamp stream can capture the secondary effects of environmental events. Daylight saving differences between Queensland and the southern states introduce small but real shifts in the diurnal pattern of ED arrivals, and experienced analysts account for that when reading the data.
Closing the gap between major metropolitan hospitals and smaller regional facilities is an ongoing priority. Aboriginal Community Controlled Health Organisations in the Northern Territory and Western Australia play a central role in primary care for many children, and their records, when properly integrated, add a layer that purely hospital-based surveillance misses. Partnerships built through the National Aboriginal Community Controlled Health Organisation, alongside the federal Department of Health and Aged Care, are slowly making that integration possible.
Limitations, ethical considerations, and data quality
Timestamp data are only as good as the discipline with which they are recorded. Smaller rural hospitals and Aboriginal Community Controlled Health Organisations may have less consistent time-stamping practices because of staffing patterns and intermittent connectivity. A febrile seizure cluster in remote Western Australia can be missed simply because the registration field is left blank or completed at the end of a shift. Closing that gap is a quiet but important part of the surveillance work, and it depends on local engagement, training, and trust-building that takes years rather than weeks.
Privacy is a perennial concern. Australian privacy legislation, including the Privacy Act 1988 and state-level health records acts, requires that identifiable paediatric data be handled carefully. The clustering algorithms work on counts and rates, not individual records, but the data feed must still be governed by data-sharing agreements between hospitals and health departments. Parents rarely object once they understand that no identifiable information leaves the hospital, but the conversation has to happen, and it has to be honest about the small residual risks of re-identification in a very small community.
A final wrinkle is the effect of media attention. When a high-profile febrile seizure case hits the news, ED presentations can rise for a week or two even when there is no change in community transmission. Sophisticated surveillance systems model this awareness effect explicitly, often by including a media-covariate term or by stepping back from the alert threshold during a defined news window. Without that adjustment, the system will generate false clusters that, while interesting, are not what public health teams are trying to find.
For those who want to look more closely at how a multi-channel surveillance system is built and what data feeds are involved, the data-use page on this site walks through the practical arrangements. The timestamp stream your service already produces has real potential as a quiet, continuous contribution to outbreak detection across the country.