Network epidemics 2

Wastewater-based epidemiology (WWBE) emerged as a method for monitoring infectious diseases SARS-COV-2 during the COVID-19 pandemic [1]. One strength of WWBE is that diseases can be monitored passively over entire populations, removing demographic biases due to, for example, reporting behaviour, but introducing challenges due to multiple sources of variation associated with signal detection. A key issue is the intersection between two complex networks – mobility networks spreading the disease, and the network of sewage pipes accumulating virus (Figure 1). Here, we develop an algorithm for choosing where to place a limited number of monitoring sites, given the overall structure of the wastewater network and the distribution of residents on that network, and test this method against a set of real wastewater systems, where the aim is to either quickly detect a novel disease outbreak or target containment measures, or both. We simulate disease surveillance in sewer catchment areas containing up to half a million residents and compare it to (i) COVID-19 case data distributions and (ii) simulations where infection is driven by recorded mobility networks and spatial proximity. We quantify the tradeoffs in spatial and temporal resolution associated with having only a limited sampling capacity (e.g. is it better to test 2x per week in a single location, or 1x per week at two?). We show that our placement of monitoring sites further "upstream" substantially outperforms simple monitoring at the "sink" as well as random placement, by more quickly and precisely detecting infected individuals which may otherwise fall under the threshold for detection at the "sink" node. We then use a previous approach [2] to quantify the sensitivity and specificity that our approach has to detect incidence patterns as COVID-19 burdens rise and fall. [1] Fitzgerald et al. https://doi.org/10.1021/acs.est.1c05029[2] Colman & Kao, ttps://doi.org/10.1101/2023.03.07.23286904

Herd immunity depends heavily on the structure of contact networks, and also on the mechanism through which individuals have gained immunity. Extending the work of Hiraoka et al. (2024), here we demonstrate that the average degree in geometric networks plays a crucial role in the effectiveness of vaccine- and disease-induced immunization strategies. Our results show that vaccine-induced immunity is more effective in sparse, moderately degree-heterogeneous networks. However, as the edge density increases, disease-induced immunity becomes more advantageous - a switchover phenomenon absent in non-geometric network models. In addition to the epidemiological implications, we also discuss the fundamental properties of disease-induced immunization as a spatially-correlated node percolation process, including the susceptibility, the degree distribution and the component statistics, from a statistical physics perspective.

The spread of epidemics when coupled with human behavior dynamics presents a challenge for accurate modeling and prediction.Leveraging on previous studies, we construct an epidemiological model that simultaneously accounts for the age structure of individuals, the heterogeneity of mobility, and the interplay between the time scales of infection and mobility processes.Starting from a metapopulation SIR model with reaction diffusion epidemic dynamic, we introduce an alternative formulation based on the force of infection framework, showing how both methods produce qualitatively comparable results and how the presence of heterogeneity in mobility reduces the final fraction of infected subpopulations.Additionally, the force of infection framework, capable of preserving the initial spatial distribution of population, is used to forecast the number of seasonal influenza infections in Italy.Overall, this work demonstrates how the impact of mobility varies depending on the modeling of the spreading process and provides an independent estimate for the Italian seasonalinfluenza forecasting service.

In countries with growing elderly populations, multimorbidity poses a significant healthcare challenge.The trajectories along which diseases accumulate as patients age and how they can be targeted by prevention efforts are still not fully understood. We propose a compartmental model, traditionally used in infectious diseases, describing chronic disease trajectories across 132 distinct multimorbidity patterns (compartments). Leveraging a comprehensive dataset from approximately 45 million hospital stays spanning 17 years in Austria, our compartmental disease trajectory model (CDTM) forecasts changes in the incidence of 131 diagnostic groups and their combinations until 2030, highlighting patterns involving hypertensive diseases with cardiovascular diseases and metabolic disorders. We pinpoint specific diagnoses with the greatest potential for preventive interventions to promote healthy aging. According to our model, a reduction of new onsets by 5% of hypertensive diseases (I10--I15) leads to a reduction in all-cause mortality over a period of 15 years by 0.57 (0.06)% and for malignant neoplasms (C00--C97) mortality is reduced by 0.57 (0.07)%.Furthermore, we use the model to assess the long-term consequences of the Covid-19 pandemic on hospitalizations, revealing earlier and more frequent hospitalizations across multiple diagnoses. Our fully data-driven approach identifies leverage points for proactive preparation by physicians and policymakers to reduce the overall disease burden in the population, emphasizing a shift towards patient-centered care.

Using contact tracing data provided by the Cyprus Ministry of Health, infection networks are constructed for the first four waves of the COVID-19 pandemic. In these networks, nodes represent infected individuals, while links indicate the direction of transmission between them. For each transmission chain, the distribution of the hopcount of the infection path from the root node to all other nodes is calculated, aggregating the data of all the trees of the same size N in each wave. The COVID-infection process, which generates the observed infection trees, is modeled as a homogeneous non-Markovian SI process on the complete graph KN with N nodes, where the infection times T are Weibull distributed with shape parameter α. The shape parameter α gives an indication of the deviation of the underlying spreading process from a Markovian (α = 1) epidemic. The results comparing the empirical hopcount distribution of infection trees computed from data, non-Markovian SI simulations with Weibull infection times, and the approximated distribution Pr[hN = k] suggest that the non-Markovian properties of the epidemic process can be quantified using real-world infection trees. Nearly all α values are larger than one, suggesting that real-world epidemics are very likely characterized by non-exponential infection processes and therefore by non-Markovian dynamics. The infection trees extracted from the data also allow to estimate the average number of secondary infections caused by a primary case (basic reproduction number R0). The distribution of the hopcount ratio r[k]/= /Pr[HN/=/k]/Pr[HN/=/k/−/1] gives a more complete indication of the number of secondary infections compared to the R0.The study shows that analyzing the infection trees from contact tracing data, allows us to approximate the disease dynamics (via α) and quantify the strength of the epidemic process (approximations of R0). This knowledge refines epidemic spreading models and improves preparedness for future pandemics.

In this paper, we develop a network model for disease transmission risk between livestock and wild animals in Scotland, focusing on bovine tuberculosis and wildlife reservoirs. The network of land parcels connecting livestock to wild animals is crucial for inter-species infection problems and is influenced by land use change.Governments provide financial incentives to encourage woodland plantation to meet climate change and biodiversity targets. While these effects are largely positive, woodland expansion could increase wildlife presence near farms, enhancing the risk of disease transmission between wildlife and livestock.The network includes economic, ecological, and epidemiological considerations. First, an economic model predicts land use changes due to subsidies for woodland planting. Then the impact of these changes on wild deer populations in new woodland areas is estimated. The disease risk is then modelled by the proximity network of deer populations and cattle holdings.With increased subsidies, deer populations become more numerous and better connected to cattle holdings, increasing the risk of cross-species infection. In South-West Scotland, with diverse farms and existing woodland, we consider varying subsidy levels and new woodland scenarios. Although deer populations increase incrementally, contact between deer and cattle areas rises by 26\% to 35\% compared to no subsidy.The model shows that increasing woodland subsidy, despite its benefits, also increases inter-species connectivity and disease risk. It provides a foundation for examining potential risk mitigation strategies, such as targeting subsidies in low-risk areas or creating buffer zones between woodland and agricultural holdings.