How a Single Concert Can Pull Forward a City’s Epidemic Peak by Weeks

A single mass gathering — a dance festival in Amsterdam — can advance the arrival of an epidemic across an entire city by up to 20 days, even if the event itself lasts only hours. In Madrid, simulations show that such an event reduced the time to reach a critical outbreak threshold by nearly half, from 130 to 70 days on average, and increased the probability of a full-scale epidemic from 33% to over 85%. These aren’t projections based on guesswork. They come from a new hybrid model that fuses fine-grained human contact data from real-world events with citywide mobility networks, revealing how fleeting moments of dense human interaction can reshape the trajectory of disease across metropolitan populations.

The finding flips a common assumption: that mass gatherings mainly matter because they generate more cases. They do — but their deeper impact lies in when and where those cases emerge. By injecting a concentrated burst of infection into a mobility network, these events don’t just amplify outbreaks — they accelerate them, synchronize them, and compress the window for public health response. And crucially, the structure of the gathering — how people move, cluster, and interact — determines how powerful that ripple effect becomes.

The Science

The challenge of modeling epidemics across scales has long plagued public health. At the city level, researchers rely on metapopulation models: simplified frameworks that divide urban areas into districts and simulate disease spread using aggregate flows of people and average transmission rates. These models are computationally efficient and excellent for capturing large-scale patterns, such as how commuters carry infections between neighborhoods. But they fail to resolve what happens inside a packed concert hall or sports stadium, where transmission depends on dynamic, person-to-person proximity — not population averages.

At the other extreme, individual-based models (IBMs) simulate every person, their movements, and their interactions. These can capture the chaotic swirl of a crowd, but simulating millions of agents across a city for weeks or months demands immense computing power. As the authors note, “applying an individual-based representation indiscriminately across both the gathering and the surrounding metropolitan population is often impractical.”

So Kwak, Lees, Ooi, and Cai built a bridge. Their hybrid framework splits the problem in space and time. First, an individual-based model simulates transmission during a mass gathering using empirically derived contact networks — real data on how people actually interact at events. Then, the infections generated during that event are injected into a stochastic SEIR (susceptible-exposed-infected-recovered) metapopulation model of Madrid, which simulates the subsequent spread through 21 districts connected by real commuting patterns.

The model uses data from five distinct events: a football match and a dance festival in Amsterdam, and three indoor concerts in Leipzig with progressively stricter crowd control. Each event’s contact network — who was near whom, and for how long — was derived from prior empirical studies using wearable sensors or video tracking. These networks vary dramatically: the Amsterdam dance event involved 1,048 people with an average of 162 contacts each, while the most controlled Leipzig concert had only 8.7 contacts per person.

Once the event simulation ends, the new infections become an “event-induced epidemic seed” — a localized outbreak that enters the city’s mobility network. From there, the metapopulation model takes over, tracking how infections spread via daily commutes, with transmission governed by a baseline reproduction number $R_0$ (set at 1.5 in the main simulations, comparable to seasonal influenza).

This coupling is the innovation: it preserves the realism of individual-level interactions where they matter most — during the event — while maintaining computational feasibility for citywide simulation. It’s like using a microscope for the spark, then a telescope for the fire.

What They Found

The results are striking. Every mass gathering accelerated the epidemic, but not equally. The Amsterdam dance event — dense, dynamic, and prolonged — had the largest effect. It generated a median of 120 new infections (

Median number of new infections generated during each mass gathering event

Label	Value
Baseline	5
AMS_football	25
Leipzig_1	35
Leipzig_2	45
Leipzig_3	15
AMS_dance	120

), more than ten times the baseline seeding from imported cases. This larger seed translated directly into faster citywide spread.

Median number of new infections generated during each mass gathering event

Label	Value
Baseline	5
AMS_football	25
Leipzig_1	35
Leipzig_2	45
Leipzig_3	15
AMS_dance	120

The dance event also pulled forward the epidemic peak by nearly three weeks and more than doubled the peak incidence, from about 1.8 to 4.2 active cases per 1,000 people (

Median time to exceed 1 cumulative case per 1000 residents

Label	Value
Baseline	130
AMS_football	90
Leipzig_1	100
Leipzig_2	110
Leipzig_3	120
AMS_dance	70

). Even more consequential, it slashed the time to cross the epidemic threshold — defined as one cumulative case per 1,000 residents — from a median of 130 days to just 70. In practical terms, that’s the difference between having months to prepare and being blindsided by rapid spread.

Median time to exceed 1 cumulative case per 1000 residents

Label	Value
Baseline	130
AMS_football	90
Leipzig_1	100
Leipzig_2	110
Leipzig_3	120
AMS_dance	70

But the impact wasn’t just temporal — it was spatial. Without a mass gathering, the epidemic spread gradually from central districts outward, like ripples from a stone dropped in water. With a major event, the wavefront became jagged and fast-moving.

shows how the dance event caused early arrival in peripheral districts that would otherwise have been spared for weeks. The synchronization of invasion — how closely in time different districts were hit — increased dramatically. In the baseline, the interval between first and last district invasion spanned over 100 days. With the dance event, it shrank to under 60 days (

Synchronization of epidemic invasion across districts

Label	Value
Baseline	100
AMS_football	80
Leipzig_1	90
Leipzig_2	95
Leipzig_3	105
AMS_dance	55

). The city didn’t just get infected faster — it got infected together.

Synchronization of epidemic invasion across districts

Label	Value
Baseline	100
AMS_football	80
Leipzig_1	90
Leipzig_2	95
Leipzig_3	105
AMS_dance	55

Not all events were equally disruptive. The Leipzig_3 scenario, which simulated a concert with strict crowd control (reduced density, limited movement), had effects much closer to the baseline. It generated only about 15 new infections and delayed the threshold crossing by 20 days compared to the uncontrolled dance event. This suggests that interventions like spacing, capacity limits, or timed entry aren’t just about reducing infections at the event — they preserve the city’s natural buffering against synchronized outbreaks.

Perhaps the most counterintuitive finding came from survival analysis of district-level arrival times. The researchers used Cox proportional hazards models to ask: what predicts how soon a district gets hit? Unsurprisingly, holding the event constant, districts with stronger commuter links to the seed location were invaded earlier. But when they adjusted for the size of the epidemic seed, the effect of the event itself on arrival time dropped by more than half. For example, in one model, the hazard ratio for earlier arrival dropped from 2.90 to 1.45 after accounting for seed size (

Hazard ratios for earlier district-level epidemic arrival

Label	Value
District 0	2.9
District 7	1.99
District 18	2.2

). This means that while the event matters, much of its accelerating effect is mediated by the number of infections it generates.

Hazard ratios for earlier district-level epidemic arrival

Label	Value
District 0	2.9
District 7	1.99
District 18	2.2

Yet seed size wasn’t the whole story. The researchers also tested whether post-event transmission conditions — the level of ongoing transmission in the days after the gathering — could predict invasion timing. They found that a 14-day average of transmission pressure provided the best predictive signal beyond seed size. In other words, it’s not just the spark, and not just the kindling — it’s how easily the fire catches in the immediate aftermath.

Why This Changes Things

Public health has long treated mass gatherings as either black boxes (“ban them”) or afterthoughts (“they’ll happen, so test people”). This study reframes them as amplifiers of spatial-temporal risk — not just sources of cases, but accelerators of urban vulnerability.

Consider the implications for pandemic planning. If a city knows a major event is coming, this model suggests it should treat the post-event period as a high-risk window for secondary transmission. Enhanced surveillance, targeted testing, or temporary mobility advisories could buy critical time. Conversely, if an event is unavoidable, the study shows that crowd-control measures aren’t just about reducing on-site infections — they’re about preserving the city’s epidemic resilience. Slowing the initial spark can delay the cascade.

The findings also challenge how we define “superspreading.” Traditionally, the term refers to an individual who infects many others. But here, the event itself acts as a superspreader — not because of one person, but because of the collective contact structure. The Amsterdam dance event didn’t just produce more cases; it produced a different kind of outbreak — one that is faster, broader, and harder to contain.

This has equity implications too.

reveals that the acceleration effect isn’t uniform: districts with weaker healthcare access or higher population density may be hit earlier and harder when spread is synchronized. A staggered invasion allows resources to be rotated; a synchronized wave overwhelms them all at once.

More broadly, the hybrid framework offers a new tool for evaluating non-pharmaceutical interventions. What if masks were worn? What if ventilation improved? The model could test these scenarios by adjusting the event-specific transmission rate ${\beta }_{\text{event}}$ and measuring the downstream impact on citywide dynamics. It turns abstract “risk reduction” into concrete, spatially explicit outcomes.

And it’s not just for respiratory diseases. The same logic could apply to foodborne outbreaks at festivals, or even the spread of misinformation at political rallies — any process where transient, dense interactions seed broader network diffusion.

What’s Next

The study has limitations. It’s a simulation, not real-world data. The results depend on the quality of the contact networks and the accuracy of Madrid’s commuting flows. The model assumes perfect mixing within districts post-event, which may smooth over local heterogeneities. And it focuses on a single event; in reality, cities host many gatherings, sometimes overlapping.

Future work could integrate real-time mobility data — from phones or transit cards — to make the metapopulation layer more dynamic. It could also model behavioral feedback: what if people change their movement after hearing about an outbreak at an event? And while the current model treats the event as exogenous, a next step would be to simulate how public health messaging or policy changes attendance or behavior during the event.

Perhaps most importantly, the framework needs validation. Can it predict actual outbreak patterns following real events? If a city hosts a large festival and then sees a spike in cases, can the model retroactively estimate how much of the acceleration was due to the event’s contact structure versus background transmission?

But even as a proof of concept, this work shifts the conversation. It shows that the cost of a mass gathering isn’t just measured in immediate infections — it’s measured in lost time. Every day by which an epidemic is delayed is a day to vaccinate, to prepare hospitals, to communicate with the public. And those days can hinge on decisions made in the design of a single event: where the exits are, how dense the crowd is, whether people are moving or static.

In an age of recurring pandemics and resurgent mass events, that’s a powerful insight. We can’t stop people from gathering. But we can learn to gather in ways that don’t hand the virus a head start.