Risk mapping novel respiratory pathogens with large-scale dynamic contact networks

TL;DR

A large-scale actor-based model capturing early epidemic dynamics of a novel respiratory pathogen on dynamic contact networks is developed, highlighting the necessity of integrating fine-scale human-to-human contact realism and population scale in epidemic forecasting and control.

Abstract

Background: Human-to-human transmission of pathogens fundamentally depends on interactions among infectious and susceptible individuals, yet traditional population-scale models often overlook the stochastic, behaviour-driven, and highly heterogeneous nature of these interactions. Methods: Here, we develop a large-scale actor-based model capturing early epidemic dynamics of a novel respiratory pathogen on dynamic contact networks. We build these networks upon explicitly integrating detailed demographic and residential registry data from the Netherlands. The model simulates the Dutch population characterised by age, residency and mobility patterns, with actors interacting stochastically across households, workplaces and schools. Results: We show how the geographic and demographic profiles of initial cases impact transmission trajectories, with densely populated municipalities in the country's western core acting as key hubs driving epidemic spread. The framework enables rigorous assessment of intervention strategies incorporating behavioural adaptations. As case studies, we quantify the effects of symptomatic self-isolation and travel restrictions to and from major urban centres, highlighting their potential to modulate epidemic outcomes. Conclusions: Our findings underscore the necessity of integrating fine-scale human-to-human contact realism and population scale in epidemic forecasting and control.

Figures (5)

• Figure 1: Model flowchart describing the construction of the dynamic contact network amongst actors. Step 1: actors are distributed across municipalities reflecting the demographic compositions at both the national and the municipality level. Steps 2 and 3: actors move from municipality to municipality based on a travel schedule informed by real-life data at hourly resolution. Step 2 shows, in darkness scale, how many actors from which municipality are present in a given municipality (in this case municipality of Amsterdam) at a given hour, while step 3 shows stochastically generated movement patterns for individual actors. Step 4: Once the actor composition is known in a given municipality at any hour, then they interact with amongst each other in four situations (home, school, work or other). After all this is implemented, a full actor spatiotemporal trajectory is generated [panel (a), right box] over a period of time. The spatiotemporal trajectory is encoded in the computer in terms of events that describe actor interactions [panel (b), right box]. Events, tagged E$_1$-E$_4$ represent the specific spatiotemporal anchor points for actor interactions. The event-based dynamic spatiotemporal representation yields a significant speed-up in comparison to a corresponding agent-based simulation. Time (starting arbitrarily at zero) is measured in units of an hour, with municipalities as the lowest units of space. The left box has been adapted from Ref. dekker2023.
• Figure 2: The "infection intensity maps" of the Netherlands, i.e., cumulative number of infections over 17 days, following pathogen introduction in five working adults of Delfzijl (a), Venlo (b), and Leiden (c) on day zero. For all three seed municipalities we also show the total cumulative infectious actors in panel (d) over the full 17 days. The cumulative number of infections and geographic spread are significantly greater for Leiden as the seed municipality than for Delfzijl and Venlo, due to its proximity to Amsterdam, the Hague, and Rotterdam, three large cities in the Netherlands. Results are averaged over 75 simulation runs. Stochastic variability across simulation runs are showcased for Leiden as the seed municipality, for early (e) and somewhat later (f) times (see text for details).
• Figure 3: Seed risk characterised at municipality-resolution for a novel respiratory pathogen in the Netherlands. (a) Shown in colour coding is the seed risk posed by a municipality, defined as the cumulative number of national infectious actors after 17 days following the introduction of the pathogen into five infectious actors in that municipality on day zero (see main text). The average across municipalities is displayed in panel (b). In both panels results are shown for pathogen introduction by primary school children (left), students (middle), and working adults (right). Results are averaged over 75 simulation runs per seed municipality and demographic group of introduction.
• Figure 4: Municipality-resolved transmission maps, and transmission risk score maps, stratified by the demographic group into which the pathogen is introduced. (a) Transmission risk map defined by the cumulative number of transmissions taking place within each municipality on day 17, upon randomly introducing the pathogen into five actors of a demographic group in a municipality on day zero, calculated as follows. The number of cumulative transmissions taking place in a municipality is weighted by the probability $P$ of the epidemic originating in one municipality, calculated as the fraction of actors from the seeding demographic group living in that municipality; i.e., $P(\mathrm{municipality\ } m\ \mathrm{ being\ the\ seed}) = \frac{n(g,m)}{N(g)}$, where $n(g,m)$ is the number of actors of group $g$ in municipality $m$, and $N(g)$ is the national number of actors in group $g$ (see further the Transmission section in Methods). (b) Correspondingly, the risk score map (transmissions taking place per capita in a municipality). See definition of risk score in the main text. Results are averaged over 75 runs per municipality and demographic group.
• Figure 5: Impact of interventions on pathogen transmission. (a) and (b): symptoms-based self-isolation; (c) and (d): movement restrictions to and from municipalities with more than $100,000$ inhabitants). See main text for details. (a) and (c): mean cumulative national number of transmissions; (b) and (d): reduction of transmissions relative to no intervention for four different adherence rates. Actors are randomly selected to be adherent. The seed actors are working adults. Results are averaged over 25 simulations per municipality per adherence rate. The 90% range of the transmissions for various adherence rates can be found in SI D (they have been left out of the plots in order to avoid cluttering them).