Network Analysis for Infectious Disease Spread in Animal Populations

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-11

Key Takeaways

• Network analysis models infectious disease spread by representing epidemiological units (animals, farms) as nodes and contact events (movements, shared resources) as edges, enabling quantification of transmission pathways beyond simple incidence.
• Node-level metrics like degree (number of contacts), betweenness (bridging capacity), and eigenvector centrality identify critical units for targeted surveillance or intervention, such as high-degree farms acting as hubs for rapid dissemination.
• Network topology, including density (overall connectedness) and assortativity (tendency to connect to similar nodes), significantly influences epidemic trajectory, with dense or assortative networks potentially sustaining endemic transmission.
• Dynamic and multiplex network approaches are crucial for accurately representing time-varying contact patterns and multiple transmission routes (e.g., movement vs. spatial proximity), providing a more complete risk assessment than static, single-layer networks.
• Incomplete or biased data necessitate statistical inference methods, such as hurdle models for generating synthetic networks, to accurately assess connectivity and inform control strategies, while acknowledging data limitations and performing sensitivity analyses.
• Practical application involves defining the relevant contact system, constructing and validating the network, computing targeted metrics, and simulating interventions to compare strategies like movement restrictions or vaccination based on network position.

Network analysis provides a formal framework for describing how individuals, farms, or other epidemiological units are connected through contact events that can transmit infectious agents. In veterinary epidemiology, the approach treats each animal or holding as a node and each potentially infectious contact as an edge, allowing researchers to quantify pathways of spread that conventional incidence measures cannot capture. This article explains the conceptual foundations of network analysis, the metrics used to characterize contact structure, and the practical applications of these methods for surveillance design, outbreak response, and policy evaluation. It is written for veterinary researchers and graduate students who need a working understanding of network methods and their evidentiary basis, and it addresses the question of how contact topology, instead of simple contact frequency, determines the trajectory of an epidemic through an animal population.

The relevance of this material extends across species and production systems. Livestock movement records, wildlife telemetry data, and within-herd behavioral observations all generate network data, and the analytical principles apply regardless of whether the nodes are individual animals, barns, markets, or administrative regions. The methods described here complement, but do not replace, compartmental modeling approaches, the focus is on the relational structure of contacts and the inferences that structure supports.

At a Glance

Graph Theory Foundations for Epidemiological Networks

A network is a mathematical graph composed of nodes and the edges that connect them. In veterinary applications, the graph is typically directed when the contact has a clear origin and destination, as with livestock movements from a seller to a buyer, and undirected when the contact is reciprocal, as with shared water sources or fence-line contact between herds. Edges may carry weights representing the number of animals moved, the duration of contact, or the probability of transmission per contact event. The choice of edge definition is the most consequential decision in any network study, because it determines which epidemiological processes the network can represent. A network built from cattle movement records, for example, captures only the movements that were reported, and it omits indirect transmission through contaminated vehicles, personnel, or wildlife vectors unless those pathways are added as separate edge types.

The structure of a network is described at three levels. At the node level, degree measures the number of contacts an individual unit has. In directed networks, in-degree counts incoming contacts and out-degree counts outgoing contacts, and these can differ substantially for a single node. A livestock market, for instance, typically has high in-degree from farms supplying animals and high out-degree to farms purchasing them. At the network level, density, path length, and clustering coefficients describe how readily a pathogen can traverse the population. At the component level, the network partitions into connected subgroups, and a pathogen introduced into one component cannot reach nodes outside it unless the network is dynamic and new edges form over time.

Contact Heterogeneity and Its Epidemiological Consequences

Observed livestock movement networks consistently show marked heterogeneity in node degree. A small number of holdings account for a large proportion of movements, while the majority of holdings have few contacts. This scale-free-like property has direct epidemiological consequences: a pathogen introduced at a high-degree node spreads rapidly and widely, whereas introduction at a peripheral node may produce only a small, contained outbreak. The practical implication is that surveillance and control resources should be allocated according to network position, not simply by herd size or geographic location. Reviews of network analysis applied to foot-and-mouth disease modeling have emphasized that this heterogeneity is a consistent feature across countries and production systems, and that ignoring it in simulation models produces misleading predictions of outbreak size and duration.

Clustering, the tendency of nodes to form densely interconnected groups, also shapes transmission dynamics. Highly clustered networks sustain transmission within groups but slow spread between groups, whereas networks with low clustering and long-range connections allow rapid geographic dissemination. The distinction matters for control: in clustered networks, targeted vaccination or movement bans within affected clusters can be highly effective, while in networks with many long-range edges, such measures must be applied more broadly to prevent reintroduction.

Dynamic and Multiplex Networks

Contact networks in animal populations are rarely static. Livestock movements follow seasonal patterns, market cycles, and trade restrictions, and the network at the time of pathogen introduction may differ substantially from the network during the peak transmission period. Dynamic network analysis treats edges as time-stamped events, allowing researchers to ask whether a node that was central in the aggregated network was actually central at the moment transmission was occurring. This temporal resolution can change intervention priorities, because a node that is briefly but intensely connected during a high-risk period may matter more than a node with moderate, constant connectivity.

Multiplex networks extend this idea by layering multiple types of contact onto the same set of nodes. A cattle population, for example, may be connected simultaneously by direct movements, shared grazing land, and common attendance at livestock markets. Each layer has different transmission characteriztics, and pathogens with different routes of spread will exploit different layers. Analyzes of livestock movement data from northern Tanzania have demonstrated that combining movement networks with spatial contact layers produces a more complete picture of transmission risk than either layer alone, and that targeted interventions on such multiplex networks substantially outperform random interventions for both fast- and slow-spreading pathogens.

Network Metrics Used in Veterinary Epidemiology

Several node-level metrics have proven useful in veterinary applications. Degree centrality identifies nodes with many contacts, which are obvious surveillance targets. Betweenness centrality identifies nodes that lie on many shortest paths between other nodes, making them potential bridges for pathogen spread between otherwise separated network regions. A farm with moderate degree but high betweenness may be more important for disease control than a farm with high degree but redundant connections. Eigenvector centrality weights a node's importance by the importance of its neighbours, which can identify nodes that are connected to other well-connected nodes, a pattern associated with efficient transmission.

Network-level metrics describe the population as a whole. The size of the largest connected component indicates the maximum potential extent of an outbreak under static network assumptions. Assortativity measures whether nodes tend to connect to similar nodes, and assortative mixing by infection status has been documented in live poultry trade networks, where traders tend to source birds from communes with similar outbreak histories. This pattern can sustain endemic infection by repeatedly reintroducing the pathogen into susceptible populations within the same network neighbourhood.

Network Generation and Inference from Incomplete Data

Complete network data are rarely available. Movement records may be missing, illegal movements go unrecorded, and wildlife contact networks are inherently difficult to observe. Statistical approaches can partially compensate for these gaps. Hurdle models, which separately predict the probability of a movement occurring and the number of animals moved, have been used to generate synthetic complete networks from partial movement permit data, and these synthetic networks capture a significant proportion of the observed variation in real movement patterns. The resulting networks can then be used to simulate disease spread and evaluate control strategies under realistic contact assumptions.

Network generation algorithms also serve a second purpose: testing the effect of structural interventions. By constraining the algorithm to prevent high in-degree nodes from selling to high out-degree nodes, researchers can construct counterfactual networks that preserve overall movement volumes while altering the topology most conducive to transmission. Simulation studies using this approach on British cattle movement data have predicted substantially lower endemic disease prevalence under such trade-restricted networks compared with baseline networks, suggesting that targeted restrictions on specific classes of connections may be more effective than blanket movement bans.

Applied Network Analysis Workflow

Network analysis in veterinary epidemiology follows a structured sequence that begins with defining the epidemiological question and ends with actionable recommendations for surveillance or intervention. The workflow differs from conventional outbreak investigation because the unit of analysis is the relationship, not the individual animal or farm.

Step 1: Define the Contact System and Data Source

The first decision is which contact layer matters for the pathogen of interest. Directly transmitted infections such as foot-and-mouth disease spread through livestock movements, whereas pathogens with environmental persistence may require spatial proximity layers in addition to movement data. The choice of contact definition determines the network's nodes and edges, and this choice must be made explicit before any metric is calculated.

Data sources fall into three categories. Movement records from mandatory reporting systems provide the most complete picture where they exist, and many countries now require farmers to report livestock movements to authorities, creating large datasets suitable for analysis. Permit-based systems in emerging economies can be exploited even when incomplete, using statistical approaches to infer missing movements. The third category, observational contact data collected specifically for research, is resource-intensive but allows inclusion of indirect contacts such as shared equipment, personnel, or wildlife interfaces.

The completeness of the data determines which analytical methods are valid. A network constructed from incomplete movement records will underestimate connectivity, and metrics derived from it will be biased toward the observed portion of the network. Where data are fragmented, hurdle model approaches that predict both the probability of movement and the number of animals moved can generate synthetic complete networks for simulation purposes.

Step 2: Construct and Validate the Network

Network construction requires decisions about temporal aggregation. A static network that aggregates all movements over a year will overstate connectivity for pathogens with short infectious periods, because it treats a farm that received animals in January as connected to a farm that shipped animals in December. Dynamic networks that respect the timing of contacts are more realistic for fast-moving pathogens, but they require more data and more complex analysis.

Validation should compare the constructed network against known epidemiological events. If the network correctly identifies farms that experienced outbreaks during a historical epidemic, confidence in its structure increases. Where possible, sensitivity analysis should test whether conclusions change when the network is reconstructed with different temporal windows or different contact thresholds.

Step 3: Compute Metrics and Identify Targets

The metrics selected depend on the intervention question. For targeted surveillance, node-level metrics such as degree, betweenness, and closeness identify farms that are disproportionately likely to acquire or disseminate infection. Highly connected livestock holdings can be identified through social network analysis, which supports surveillance and disease prevention activities. For policy design, network-level metrics such as density, clustering coefficient, and assortativity describe the system's overall vulnerability.

The interpretation of metrics must account for the production system. In a cattle movement network, farms with high in-degree are at risk of acquiring infection, while farms with high out-degree are potential disseminators. The combination of both, high in-degree and high out-degree, identifies farms that could amplify an epidemic by receiving infection and passing it on. Targeted manipulation of contact network structure, for example preventing farms with high in-degree from selling to farms with high out-degree, has been predicted to reduce endemic disease prevalence in simulation models of the British cattle industry.

Step 4: Simulate Interventions and Compare Strategies

Network models allow comparison of intervention strategies before field implementation. Random vaccination or movement bans can be contrasted with targeted approaches that prioritize nodes identified through network metrics. In a simulated cattle movement network for northern Tanzania, targeted interventions substantially outperformed random interventions for both fast and slow pathogens, with the advantage most pronounced for pathogens with lower basic reproduction numbers.

The choice of intervention depends on the pathogen's transmission dynamics and the network's structure. Movement bans are disruptive to production systems and may be economically unsustainable for prolonged periods. Vaccination targeted at high-betweenness nodes can interrupt transmission with less disruption, but requires that an effective vaccine exists and that the target population is accessible. The simulation should test multiple intervention combinations and report the trade-offs between epidemiological benefit and operational cost.

Case Study: Live Poultry Trade Network in Northern Vietnam

A published network analysis of the live poultry trade in northern Vietnam during the 2003 to 2006 HPAI H5N1 epidemic illustrates the applied workflow. The study documented the flow of live poultry from production flocks through live poultry traders to retail and wholesale markets, constructing a network in which communes were nodes and trade relationships were edges.

The analysis produced two findings of direct epidemiological relevance. First, traders who had been operating for less than one year and who sold at retail markets were more likely to source poultry from communes with a past history of HPAI H5N1 outbreaks than traders operating longer and selling at wholesale markets. This identifies a specific, actionable surveillance target: new entrants to the trade network operating at retail level. Second, the network analysis showed that traders tended to link communes of similar infection status, a pattern of assortative mixing that can sustain endemic infection within clusters while limiting spread to naive populations.

The study's utility lies in its translation to policy. Surveillance resources can be directed toward retail markets supplied by recently established traders, and biosecurity interventions can be designed to break the assortative links that maintain transmission within infected clusters. The case demonstrates that network analysis does also describe contact patterns, it identifies specific nodes and edges where intervention has the highest expected value.

Metrics Summary for Applied Use

The choice of metric must match the decision being made. A surveillance program for an exotic pathogen incursion should prioritize betweenness and closeness, because these identify nodes that would spread infection quickly if infected. A control program for an endemic pathogen should prioritize degree and assortativity, because these identify the structures that maintain transmission. The same network can support both objectives, but the metrics reported will differ.

Species and Production System Considerations

The correct analytical approach varies with the production system. Cattle movement networks in countries with mandatory reporting are typically well characterized and support detailed dynamic analysis. Poultry networks, as in the Vietnam case, are often informal and require primary data collection, with the resulting network reflecting the sampling frame instead of the true contact structure. Wildlife networks require observational contact data and are limited by the feasibility of tracking individuals, the resulting networks are usually sparser and less complete than livestock movement networks.

The epidemiological question also changes the analysis. For a fast-moving pathogen with a basic reproduction number of 3, the network's overall connectivity dominates the outcome, and interventions must be rapid and broad. For a slow pathogen with a basic reproduction number of 1.5, the network's local structure matters more, and targeted interventions at specific nodes can be effective. The same network data can support both analyzes, but the metrics reported and the interventions simulated will differ.

Where movement data are not routinely collected, the argument for establishing routine data collection is strong, because network analysis of well-recorded movements has been used successfully to identify system properties, highlight vulnerabilities to transmission, and inform targeted surveillance and control. The investment in data infrastructure is justified by the improved targeting of interventions that network analysis enables.

Limitations of Network Inference from Movement Data

Livestock movement records are the most common data source for veterinary network analysis, yet they carry systematic biases that affect interpretation. Movement permits and interstate health certificates capture regulatory compliance, not actual contact behavior. Farms that under-report movements, use informal channels, or move animals without documentation create networks that underestimate true connectivity. Analyzing livestock network data for infectious disease control highlights this problem directly, noting that where movement data are poorly recorded, network tools must infer structure from biased or fragmented datasets.

Temporal resolution poses a second constraint. A network built from annual movement summaries treats a farm pair as connected if any movement occurred during the year, regardless of whether the movement happened in January or December. For fast-spreading pathogens with short infectious periods, this aggregation overstates the risk of transmission along stale edges. Dynamic network methods that timestamp edges and allow them to expire reduce this error, but they require movement data with reliable dates, which many recording systems do not provide.

The direction of movement also carries meaning that static metrics can obscure. A farm that sells many animals is epidemiologically different from a farm that buys many animals, even when their degree centrality is identical. Network analysis terminology applied to foot-and-mouth disease modeling emphasizes that directed networks reveal the role each holding plays in acquiring versus spreading infection, information that undirected summaries discard.

Common Errors in Applied Network Analysis

Less experienced analysts frequently confuse node-level metrics with system-level properties. High degree centrality identifies a farm with many contacts, but it does not identify the farm most likely to introduce a novel pathogen. That distinction requires considering the network position of the farm's contacts, which is what eigenvector centrality or k-core decomposition captures. The corrective action is to state the epidemiological question before selecting metrics, then choose the metric that answers that specific question.

A second recurring error is treating network structure as static when the underlying contact process is seasonal. Calving seasons, market cycles, and weather-driven movement patterns all change edge density over the year. Analyzing a single aggregated network hides these dynamics and can mislead intervention timing. The remedy is to construct time-windowed networks and compare metric stability across windows before committing to a single representation.

A third error is overinterpreting centrality rankings when the network is incomplete. If movement data are missing for a subset of farms, the farms with complete records will appear more central simply because their edges are observed. Social network analysis in preventive veterinary medicine notes that the technique has been applied only recently in this field, and the characteriztics of the method remain unfamiliar to many practitioners. Sensitivity analysis that removes nodes or edges and re-computes metrics helps distinguish robust findings from artefacts of missing data.

Failure Modes and Detection

Evidence Gaps and Divergent Expert Opinion

The evidence base for network-informed veterinary interventions remains uneven. Most published work concentrates on livestock movement networks in high-income countries with mandatory reporting, particularly cattle systems in Europe and North America. Controlling infectious disease through targeted manipulation of contact network structure demonstrates that restricting specific connection types can reduce endemic prevalence in simulation, but the translation of these findings to field conditions depends on compliance with movement restrictions, which varies by production system and jurisdiction.

Expert opinion diverges on how much network detail is necessary for practical decision-making. Some analysts argue that simple degree-based targeting captures most of the benefit, while others maintain that dynamic, multiplex representations are required to avoid systematically misidentifying risk. The evidence does not yet settle this question. Introduction to network analysis for animal disease modeling argues that observed networks show substantial contact heterogeneity and clustering, and that understanding this architecture improves infection spread predictions. The practical question is whether the improvement justifies the additional data collection and analytical complexity.

For emerging economies, the evidence base is thinner still. Routine movement data are often absent, and the argument for routine data collection in emerging economies demonstrates that statistical inference can generate synthetic networks from partial permit data, but the approach depends on government records that many countries do not yet maintain. Until such systems exist, network analysis in these settings will rely on cross-sectional surveys and expert elicitation, methods with their own biases.

Referral, Consultation, and Regulatory Reporting

Network analysis findings should trigger consultation when they inform disease control decisions that affect multiple farms or jurisdictions. A veterinary practitioner who identifies a highly connected node in a local movement network should discuss the finding with the regional veterinary authority before acting, because targeted interventions may have unintended consequences for animal welfare or trade. WOAH terrestrial animal health standards and WOAH surveillance standards define reporting obligations for notifiable diseases, and these obligations take precedence over any network-derived prioritization scheme.

Laboratory involvement becomes necessary when network analysis identifies a suspected transmission pathway that cannot be confirmed by movement records alone. Pathogen sequencing can distinguish between direct transmission along network edges and introduction from unsampled sources, and the combination of network analysis with molecular epidemiology is increasingly the standard for outbreak investigations. Regulatory reporting is warranted when analysis reveals systematic under-reporting of movements, because this compromises both disease surveillance and traceability systems that support trade.

Frequently Asked Questions

How Much Does Network Analysis Cost, and Is It Feasible for a Small Practice or Regional Authority?

Costs scale with data availability and analytical depth. If movement records, herd registries, or traceability data already exist, the marginal cost is mainly analyst time and software. Free or low-cost network packages in common statistical environments handle most descriptive metrics and simulation tasks. Regional authorities with limited budgets can prioritize descriptive metrics such as degree distribution and connected components before investing in dynamic simulation. For practices without in-house expertise, collaboration with a veterinary school or national reference laboratory is often the most efficient route. The investment is justified when it directly informs surveillance targeting or movement restrictions, as demonstrated in livestock systems where targeted interventions outperformed random ones network analysis for infectious disease control in emerging economies.

What Should We Do When Movement Data Are Incomplete or Biased?

Treat missing data as a modeling problem, not a reason to abandon the analysis. Permit-based movement records often undercount smallholder movements or informal trade. Statistical approaches such as hurdle models can predict both the probability of a movement and the number of animals moved, generating synthetic networks that preserve observed structure analyzing livestock network data for infectious disease control. Validate inferred networks against independent data such as abattoir throughput, market surveys, or veterinary inspection records. Report uncertainty explicitly and run sensitivity analyzes over plausible network reconstructions. When data quality is poor, restrict conclusions to robust properties such as the presence of highly connected hubs, which tend to persist across inference methods, instead of precise path lengths or centrality rankings.

How Do Network Findings Translate into Practical Surveillance or Control Actions?

Identify nodes with high in-degree or out-degree and prioritize them for targeted surveillance or movement restrictions. Farms with high in-degree are acquisition risks, while those with high out-degree are dissemination risks. Connections from high in-degree to high out-degree nodes are disproportionately influential in disease spread, and restricting these specific contacts can reduce endemic prevalence in simulation models targeted manipulation of contact network structure. Use network metrics to stratify sampling frames for prevalence surveys and to target tracing efforts during outbreaks. For reportable diseases, align these priorities with international surveillance and notification standards WOAH animal health surveillance standards.

Does Network Analysis Work Differently for Wildlife, Companion Animals, or Aquatic Species?

The framework transfers, but data sources and network dynamics differ substantially. Wildlife networks rely on GPS collars, proximity loggers, or genetic relatedness instead of movement permits, producing smaller, shorter-duration networks with more missing data. Companion animal networks are built from veterinary clinic visits, registries, or shelter intake records, and they change rapidly with adoption and relocation. Aquatic systems involve farm-to-farm movements, shared water sources, and fomite transmission via equipment, requiring multiplex networks that combine multiple contact types. In all cases, the epidemiological questions are the same: who contacts whom, how often, and which nodes bridge otherwise separate groups. The limiting factor is data collection, not methodology social network analysis in preventive veterinary medicine.

What Records Should We Keep to Make Future Network Analysis Possible?

Maintain structured, machine-readable records of every between-unit contact: unique identifiers for source and destination, date, number of animals, species, and production type. For livestock, this means consistent use of ear tags or electronic identification and digital movement documentation. For companion animals, record owner and location changes at each visit. For wildlife, archive GPS and proximity logger data with metadata on deployment and retrieval. Standardize identifiers across databases so records can be linked. Centralized, routine collection of movement data enables retrospective network reconstruction and prospective surveillance analyzing livestock network data for infectious disease control. Even simple spreadsheets are useful if identifiers are consistent and dates are accurate.

How Should I Present Network Findings to a Farm Owner, Practice Manager, or Regulatory Body?

Frame findings in terms of risk and action, not graph theory. Explain that a small number of farms or animals account for most transmission potential and that targeted measures outperform blanket restrictions. Use visualizations such as node-link diagrams or heat maps of contact frequency, but pair them with plain-language interpretation. For regulators, reference the relevant international standards for surveillance and trade-related disease control WOAH terrestrial animal health code. For farm owners, focus on specific biosecurity actions: which suppliers or buyers pose the greatest risk, which movements to avoid during an outbreak, and how record keeping improves their ability to respond. Emphasize that network analysis supports decisions, it does not replace clinical judgment or statutory obligations.

Related Clinical & Scientific Guides

• Evaluating Veterinary Surveillance System Attributes
• Randomized Controlled Trials in Veterinary Field Settings
• Regression Analysis in Veterinary Epidemiology: Logistic and Poisson Models

References and Further Reading

• Social network analysis. Review of general concepts and use in preventive veterinary medicine.. 2009.
• A review of network analysis terminology and its application to foot-and-mouth disease modeling and policy development.. 2009.
• Controlling infectious disease through the targeted manipulation of contact network structure.. 2015.
• Introduction to network analysis and its implications for animal disease modeling.. 2011.
• Analyzing livestock network data for infectious disease control: an argument for routine data collection in emerging economies.. 2019.
• Associations between attributes of live poultry trade and HPAI H5N1 outbreaks: a descriptive and network analysis study in northern Vietnam.. 2010.
• WOAH Animal Health Surveillance Standards. WOAH.
• CDC Principles of Epidemiology in Public Health Practice. CDC.
• MSD Veterinary Manual, Professional Edition. MSD Veterinary Manual.

Related Articles

• Risk Factor Analysis for Disease in Animal Populations
• Sensitivity Analysis in Veterinary Disease Models
• Time Series Analysis for Veterinary Disease Surveillance
• Model Validation and Calibration for Animal Disease Spread
• Measuring Disease Frequency: Incidence and Prevalence in Animal Populations

This article is educational professional reference material for veterinary audiences. It is not a substitute for veterinary diagnosis, individual clinical judgment, current product labeling, or applicable regulatory requirements.