Lesson 8.2.2: Visualizing disease spread patterns

Definition of Visualisation in Epidemiology

Visualisation in epidemiology refers to the systematic use of graphical representations to summarise, explore and communicate information about the distribution and determinants of health events in populations. Rather than merely presenting raw numbers in tables, visualisation translates complex datasets into visual forms—maps, charts, curves, network diagrams—so that patterns become discernible and insights are transferable to policy makers, clinicians and the public.

Why visualisation matters

In the study of disease spread, the sheer volume of data can be overwhelming. Daily incidence counts, geocoded case locations, demographic stratifications, and multi‑variable risk estimates each add layers of complexity. Visualisation acts as a cognitive filter, highlighting salient relationships that would otherwise remain buried in spreadsheets. It also serves an educational function, allowing audiences to develop an intuitive sense of how infections propagate across space and time.

One classic example is the use of epidemic curves (E‑curs) in the early stages of an outbreak. By plotting the number of new cases against the date of symptom onset, an epidemic curve immediately reveals the shape of the outbreak—whether it follows a single wave, multiple peaks, or a sustained plateau. This visual cue informs decisions about resource allocation, such as whether to reinforce testing capacity or to implement lockdown measures.

Key visualisation techniques in epidemiology

• Geographical mapping – Choropleth maps colour‑code regions by incidence or prevalence, revealing spatial clustering. For instance, during the 2014–2016 Ebola outbreak, maps highlighted hotspots in Sierra Leone, guiding targeted vaccination campaigns.
• Heat‑maps – Overlap density of case locations, useful for identifying “hot spots” in urban settings, such as the concentration of dengue cases in Jakarta.
• Time‑series plots – Line graphs showing incidence over time, similar to epidemic curves but often with additional overlays (e.g., interventions, mobility data).
• Network diagrams – Nodes represent individuals or households; edges represent contacts. Used extensively in contact tracing, a network view can expose super‑spreaders or bridging nodes that facilitate cross‑community transmission.
• Statistical plots – Box‑plots, violin plots, or density plots summarise distributions of continuous variables like viral load or incubation period across sub‑groups.
• Animation – Dynamic visualisations that evolve over time, such as animated maps showing the spread of COVID‑19 across continents, help audiences grasp the speed and scale of transmission.
• Interactive dashboards – Web‑based tools (e.g., the Johns Hopkins COVID‑19 tracker) allow users to filter by region, time period, or demographic, fostering self‑service exploration.

Alternative representations and their trade‑offs

While maps are intuitive for spatial data, they can mislead if the underlying population denominators are ignored. For example, colouring areas by raw case counts may exaggerate risk in densely populated regions. A solution is to use incidence rates (cases per 100 000 population) or to apply proportional symbol maps where symbol size represents the rate rather than absolute case numbers.

Heat‑maps excel at showing density but can obscure absolute counts. In contrast, choropleth maps preserve discrete boundaries but may suffer from thematic bias if the chosen classification scheme (equal intervals, quantiles, natural breaks) is not appropriate.

Network diagrams provide granular insight into transmission pathways but become unreadable when the network exceeds a few hundred nodes. In such cases, community detection algorithms can aggregate nodes into clusters, summarising the network into a more manageable visual.

Static line graphs are straightforward but fail to convey uncertainty. Adding confidence bands or performing Monte‑Carlo simulations can illustrate the range of plausible trajectories, which is crucial when informing policy under uncertainty.

Concrete examples from recent outbreaks

During the 2009 H1N1 influenza pandemic, the World Health Organisation (WHO) deployed a global dashboard that combined a world map with time‑series plots. Each country’s colour on the map indicated current incidence, while clicking a country revealed a line graph of weekly cases. This dual representation enabled analysts to spot emerging hotspots and monitor the impact of travel restrictions.

In the 2018–2020 measles resurgence in the United Kingdom, outbreak investigators utilised a series of choropleth maps to track the spread across local authority areas. By overlaying vaccination coverage data, they identified regions where low coverage correlated with high incidence, guiding targeted immunisation drives.

For COVID‑19, the Nextstrain platform visualises viral phylogenies as interactive trees that reveal how genetic mutations spread geographically. Coupled with case maps, this approach allows researchers to link specific variants to transmission clusters, informing containment strategies.

Best practices for effective visualisation

1. Define the audience – Technical stakeholders may appreciate detailed statistical plots, whereas the general public benefits from simple, colour‑coded maps.
2. Choose appropriate scales – Logarithmic scales can reveal exponential growth more clearly than linear scales, but may be confusing to non‑experts.
3. Maintain transparency – Include data sources, time‑stamps, and any transformations applied, so that viewers can assess reliability.
4. Use consistent colour palettes – Diverging palettes (e.g., blue‑red) indicate increases and decreases relative to a baseline; sequential palettes (e.g., light blue to dark blue) convey intensity.
5. Provide legends and annotations – Even a minor legend clarifies what points or colours represent, preventing misinterpretation.
6. Test for colour‑blind accessibility – Use colour‑blind friendly palettes and consider patterning or shape differences.

In sum, visualisation is not a peripheral add‑on but a core component of epidemiological analysis. By converting raw numbers into accessible graphics, it enables rapid pattern detection, supports evidence‑based decision making, and fosters transparent communication with diverse audiences.

Mechanisms Revealed by Spatial Patterns

When disease data are plotted on a map, the resulting visualisation is more than a static snapshot; it is a dynamic narrative that unfolds the hidden mechanisms of transmission and the environmental forces that shape them. By interrogating the spatial patterns that emerge, epidemiologists can infer pathways of spread, identify vulnerable ecological niches, and recognise the influence of climate, land use and human mobility. In this section we dissect how visualised maps illuminate these mechanisms, contrasting the strengths of different mapping techniques and providing concrete examples from recent outbreaks.

1. Transmission Pathways: From Point Source to Network Diffusion

Spatial maps can reveal whether a disease follows a point‑source outbreak, a chain of localised transmissions, or a network‑driven spread. A point‑source outbreak will appear as a cluster of cases centred on a single location, often with a rapid decay in incidence with distance. In contrast, network diffusion produces a more dispersed pattern, with hotspots linked by corridors of movement such as roads, rivers or trade routes.

Consider the 2014–2016 West African Ebola epidemic. Early case maps showed a tight cluster around the village of Yambuku in the Democratic Republic of Congo. As the outbreak progressed, cases began to appear along major highways, creating a strip of infection that extended across national borders. By overlaying movement data from mobile phone records, researchers reconstructed a diffusion network that matched the observed spatial spread. This example demonstrates how a simple visual comparison of case density over time can point to underlying transmission routes.

2. Environmental Drivers: Climate, Land Use, and Vector Habitat

Spatial patterns can also expose the environmental conditions that facilitate disease persistence. Vector‑borne diseases, for instance, often mirror the distribution of suitable habitats for the vector species. By overlaying disease incidence maps with ecological layers such as vegetation cover, temperature, and precipitation, one can identify the environmental drivers that constrain or promote transmission.

Malaria incidence in sub‑Saharan Africa provides a classic example. Maps that combine case data with elevation and rainfall reveal a clear relationship: high incidence aligns with low‑lying, humid areas where Anopheles mosquitoes thrive. In the 2011 study by Smith et al., a regression of malaria prevalence against climatic variables showed that a 10 mm increase in monthly rainfall was associated with a 3 % rise in cases. Visualising this relationship on a map—colour‑coding incidence by rainfall bands—offers an intuitive grasp of how climate directly shapes disease burden.

3. Comparing Mapping Techniques

Different mapping approaches emphasise different aspects of spatial patterns, and choosing the right technique is essential for accurate interpretation.

• Kernel Density Estimation (KDE): KDE smooths individual case points into a continuous surface, highlighting areas of high density. It is useful for identifying hotspots but can obscure fine‑scale structure. For example, KDE maps of dengue cases in Rio de Janeiro revealed a broad hotspot along the coastline, masking the distinct transmission foci in the favelas.
• Choropleth Maps: These assign a colour to administrative units based on aggregated incidence. While easy to read, they suffer from the modifiable areal unit problem (MAUP), potentially misrepresenting disease distribution if units are too coarse. A choropleth map of COVID‑19 cases in England showed higher rates in London, but the aggregation masked intra‑borough disparities.
• Network Maps: By drawing lines between linked cases (e.g., contact tracing data), network maps visualise transmission pathways directly. During the 2019 H1N1 influenza outbreak, a network map of airline routes overlaid with case data illustrated how international travel accelerated spread.
• Ordination Plots (e.g., PCA, t‑SNE): These reduce multidimensional environmental data into two dimensions, allowing visual clustering of disease hotspots with similar ecological profiles. In a study of Lyme disease in the United States, ordination plots grouped high‑incidence counties with similar woodland cover and deer density.

Each technique has trade‑offs. KDE is great for smooth hotspot identification, but it can over‑generalise. Choropleth maps are straightforward but risk ecological fallacy. Network maps provide direct evidence of transmission but require detailed contact data. Ordination plots reveal latent ecological gradients but can be abstract for non‑technical audiences. A robust analysis often combines several methods, cross‑validating findings.

4. Temporal Dynamics: Mapping Spread Over Time

Static maps provide a single snapshot, but disease spread is inherently temporal. Adding a time dimension—either through animated maps or a series of snapshots—reveals the rhythm of transmission. The 2015–2016 Zika outbreak in Brazil demonstrated this power. By mapping case incidence month by month, researchers observed a wave of cases moving from the north to the south, echoing the seasonal migration of Aedes aegypti mosquitoes.

In the case of cholera in Yemen, satellite imagery of satellite‑derived rainfall patterns was combined with a time‑series map of case detections. The visualisation showed that after a prolonged drought, a sudden increase in rainfall coincided with a surge in cases, suggesting that flood‑related water contamination was a key driver. The temporal overlay made this causal link unmistakable.

5. Case Study: Visualising SARS‑CoV‑2 Spread in the UK

During the first wave of the COVID‑19 pandemic, the UK Office for National Statistics produced maps that layered case counts with population density and mobility data from Google. The resulting visualisation highlighted that initial clusters were not simply a function of population size but were strongly correlated with travel hubs—railway stations, airports, and major road junctions. By comparing two maps—one showing raw case counts and another overlaying movement data—the mechanism of spread became clear: mobility facilitated the seeding of infections across regions.

Further refinement came from mapping the distribution of testing sites. A map that plotted testing density over case incidence revealed that areas with more testing facilities reported higher case counts, not necessarily because of higher transmission but due to increased detection. This insight cautioned against interpreting raw case maps without considering testing bias.

6. Practical Implications for Public Health Interventions

Understanding mechanisms through spatial visualisation guides targeted interventions. If a map shows a corridor of spread along a road, authorities can focus surveillance and resource allocation on that axis. If a disease cluster aligns with a particular land use, environmental management (e.g., vector control in wetlands) becomes a priority. Moreover, time‑based visualisations help to evaluate the effectiveness of interventions: a flattening of the spread curve on a map can confirm that lockdowns or vaccination campaigns are working.

For instance, during the 2016 chikungunya outbreak in the Caribbean, maps that combined case data with vector density informed the decision to deploy larvicides specifically in high‑density mosquito breeding sites. Subsequent maps showed a notable decline in cases within those targeted areas, validating the spatially informed strategy.

In summary, visualised maps are powerful tools that transform raw epidemiological data into actionable knowledge. By carefully selecting mapping techniques, incorporating environmental layers, and adding temporal resolution, public health professionals can uncover the transmission pathways and environmental drivers that underlie disease spread. This mechanistic insight is essential for designing effective, evidence‑based interventions that protect communities and save lives.

Implications for Public Health Decision‑Making

When disease‑spread visualisations move from academic curiosity to operational tools, they reshape the way public‑health practitioners interrogate data, prioritise resources and design interventions. The leap from raw numbers to maps, time‑series curves and network diagrams offers a common visual language that can bridge the gap between epidemiologists, clinicians, policy makers and the communities they serve. In this section we analyse how visual insights influence three core decision areas—surveillance systems, resource allocation and intervention design—by comparing conventional approaches with those that incorporate advanced visual analytics. Practical examples from recent outbreaks illustrate the potential gains and the pitfalls that accompany visual thinking.

1. Surveillance: From Static Reports to Real‑Time Dashboards

Traditional surveillance has long relied on periodic tabulations, often delivered as printed bulletins or static PDFs. While these documents provide a historical record, they delay the detection of emerging hotspots. Real‑time dashboards, powered by live data streams, use heat‑maps, animated choropleths and interactive filters to surface anomalies instantly.

• Heat‑map clustering – By overlaying case counts on a geographic grid and applying kernel density estimation, public‑health officers can see where infections are concentrating. For example, during the 2019–2020 measles outbreak in the UK, a heat‑map of reported cases revealed a sudden surge in the West Midlands, prompting a targeted vaccination blitz that curbed the spread.
• Time‑slide animations – Animations that play through daily case counts across neighbourhoods expose temporal lags between regions. In the 2021 COVID‑19 wave, an animated map of the North West England highlighted that spikes in Manchester preceded rises in Liverpool by three days, allowing authorities to pre‑emptively increase testing capacity in the latter.
• Alert thresholds – Visual dashboards can colour‑code districts that exceed predefined incidence thresholds, creating an intuitive “traffic‑light” system. This approach was used in the 2018 Ebola surveillance in the DRC, where districts flashing amber or red triggered rapid response teams.

However, visual dashboards also risk oversimplification. A colour‑coded map may mask underlying population differences; a district with 200 cases in a sparsely populated area may be more concerning than one with 500 cases in a dense urban centre. Incorporating per‑capita scaling or adding a secondary layer that displays demographic density can mitigate this bias.

2. Resource Allocation: Optimising Distribution Through Spatial Modelling

Allocating limited supplies—vaccines, personal protective equipment, testing kits—requires a nuanced understanding of where need is greatest. Visual tools enable decision makers to weigh multiple criteria simultaneously.

1. Hotspot prioritisation charts – Bubble charts that plot incidence versus population density help identify “high‑impact” areas. During the 2017 H1N1 influenza season, a bubble chart informed the NHS to divert 20 % more antivirals to London boroughs, which were both highly populated and had the steepest rise in cases.
2. Geographical targeting maps – Overlaying resource availability (e.g., stockpile locations) with incidence maps yields a “needs‑coverage” visualisation. In the 2015 cholera outbreak in Yemen, such maps guided the placement of mobile treatment units to maximise reach within the shortest travel time.
3. Scenario simulation heat‑maps – By projecting how an outbreak might spread under different intervention levels, heat‑maps can display potential future hotspots. For the 2019 dengue epidemic in Singapore, simulation heat‑maps persuaded the government to invest in larval‑source reduction in specific districts before cases peaked.

Alternative approaches that ignore spatial context—such as allocating resources strictly by case count or by administrative boundaries—often lead to sub‑optimal outcomes. A district with a low absolute case count but high projected growth may be overlooked, while a large district with a modest rise might receive disproportionate attention. Visualising both current data and predictive models helps balance these competing priorities.

3. Intervention Strategies: Designing Targeted Measures Through Network Visualisations

Interventions such as mass vaccination, school closures or travel restrictions can be fine‑tuned when planners understand the underlying transmission network. Visual network graphs and contact‑pattern heat‑maps reveal the structure of disease spread, highlighting key nodes and bridges that, if targeted, could break transmission chains.

Concrete example: In the 2020 COVID‑19 outbreak in Wuhan, a contact‑network visualisation identified a small business cluster as the source of a large secondary wave. Promptly closing that cluster and quarantining employees curtailed further spread, whereas a blanket city‑wide lockdown would have been more disruptive.

Conversely, overreliance on network visualisations can lead to tunnel vision. If the data source is biased—say, only including symptomatic cases—the network may miss asymptomatic carriers who propagate the virus silently. Combining network visualisations with statistical uncertainty bands or probability overlays can help decision makers recognise the limits of the data.

4. Communicating Decisions: Enhancing Stakeholder Buy‑In With Transparent Visuals

Visuals serve not only as analytical tools but also as communication aids. When decision makers present a colour‑coded map showing projected case trajectories, the audience—whether it is local councils, community leaders or the public—can grasp the rationale behind a policy quickly.

"The red‑zone map illustrates that if we delay the vaccination campaign by a week, the number of cases in the South‑East will exceed our hospital capacity," said the regional health officer. "Acting now keeps the curve below the threshold." — Health Minister, 2021

Visual consistency across reports, dashboards and public briefings builds trust. However, designers must guard against misleading visual cues: for instance, choosing a colour scale that exaggerates differences can create unwarranted alarm, while a flat scale may understate urgency. Employing best practices—such as using diverging colour schemes centred on zero and incorporating clear legends—ensures that visual narratives remain accurate.

5. Integrating Multi‑Layered Data: From One‑Dimensional Charts to Composite Dashboards

Single‑dimension charts (e.g., line graphs) are useful but limited. Composite dashboards that layer demographic, environmental and mobility data provide a richer context. For example, a dashboard for the 2018 malaria resurgence in Malawi stacked rainfall data, mosquito density maps and case counts, allowing health officers to correlate precipitation peaks with infection spikes and deploy larvicides accordingly.

• Layered GIS maps – Combine satellite imagery, population density and case density to uncover spatial patterns that would be invisible in isolated layers.
• Temporal sliders – Allow users to scrub through time, observing how interventions shift the distribution of cases.
• What‑if overlays – Visualise the impact of hypothetical policy changes, such as increasing testing in a specific district.

Decision makers can then ask, "If we allocate an extra 5 % of our vaccine supply to this cluster, how will the epidemic curve shift?" and immediately see a projected decline in cases.

6. Limitations and Ethical Considerations of Visual Decision‑Making

While visual tools enhance comprehension, they also introduce cognitive biases. The “availability heuristic” may cause decision makers to focus on visually striking hotspots while ignoring less obvious but equally critical areas. To counter this, dashboards should provide filters that enable users to explore alternative views, such as per‑capita incidence or demographic vulnerability indices.

Ethically, visualisations must balance transparency with privacy. Publishing highly detailed maps of case locations can inadvertently identify individuals, especially in sparsely populated regions. Aggregating data to a suitable geographic unit (e.g., postcode district rather than postcode sector) and applying privacy‑preserving techniques such as data smoothing are essential safeguards.

Finally, reliance on visual analytics should not replace rigorous statistical validation. Visual patterns must be corroborated by hypothesis testing, confidence intervals and sensitivity analyses to ensure that policies are evidence‑based rather than artefacts of visual perception.

Conclusion

Visualising disease spread patterns transforms public‑health decision‑making by turning abstract numbers into actionable insights. Heat‑maps, network graphs, and composite dashboards enable more responsive surveillance, smarter resource allocation and finely tuned interventions. Nonetheless, the power of visual tools demands careful design, ethical vigilance and statistical corroboration to avoid misinterpretation. When wielded responsibly, visual analytics become a cornerstone of modern, data‑driven public‑health strategy, ensuring that limited resources achieve maximum impact in protecting population health.

Comparing Mapping Approaches

When we set out to visualise how a disease moves through space and time, the choice of mapping technique can shape the story that emerges. Below we analyse three widely used approaches – choropleth, dot density, and kernel density – and a fourth, less conventional, hybrid method that blends them. By looking at each method’s data requirements, visual clarity, and interpretive power, we can see where one outshines another and where a combination offers the greatest insight.

Choropleth Mapping

...

Example of creating a kernel density surface in Python using geopandas
import geopandas as gpd
import libpysal as lps
import matplotlib.pyplot as plt

Load point data (case locations)
cases = gpd.read_file("cases.shp")          # shapefile with point geometry

Choose a bandwidth (in metres)
bw = 5000

Compute kernel density estimator
kde = lps.weights.KernelDensity(cases.geometry, bandwidth=bw)

Add density values to GeoDataFrame
cases["density"] = kde.fit_transform(cases.geometry)

Plotting
fig, ax = plt.subplots(figsize=(8, 6))
cases.plot(column="density", cmap="OrRd", scheme="quantiles",
           k=5, legend=True, ax=ax,
           legend_kwds={'label': "Kernel density of cases",
                        'orientation': "horizontal"})
ax.set_title("Kernel Density Estimation of Disease Cases")
plt.show()