Most epidemiological models operate at the macro-scale, providing national or state-level forecasts. While these models offer a broad view of a pandemic's trajectory, they often suffer from a "masking effect": by aggregating data over large geographic areas, they obscure the highly localized hotspots where transmission actually occurs.
Research across several infectious diseases—from COVID-19 and Dengue to Malaria—has converged on a shared insight: to implement effective public health interventions, we must move toward the meso-scale. This involves modeling at the level of cities, counties, or even individual neighborhoods to capture the unique social fabric and environmental drivers of a specific community.
The Problem of Geographic Aggregation
When data is aggregated at a coarse resolution, such as a province or country, the resulting "average" transmission rate may not describe any single location accurately. As Li-Qun Fang and colleagues demonstrated in their study of hemorrhagic fever in Beijing, provincial-level data can hide specific environmental risk factors, such as the presence of orchards or rice paddies, that are only visible at the township level Spatio-temporal Dynamics of Hemorrhagic Fever with Renal Syndrome, Beijing, People’s Republic of China.
This lack of granularity creates a "mismatch" between the scale of the forecast and the scale of decision-making. Bin Yu and Chandan Singh noted that national models often fail to provide the spatial detail required to direct critical resources, like personal protective equipment (PPE), to the specific hospitals that need them most Seven Principles for Rapid-Response Data Science: Lessons Learned from Covid-19 Forecasting.
The Meso-scale Sweet Spot
The "meso-scale" represents the intermediate level between individual-level micro-models and national-level macro-models. Sofia Hurtado and her team argue that city-level dynamics are the optimal scale for situational awareness because they capture uniform social and economic interactions that state-level analyses miss Pruning Digital Contact Networks for Meso-scale Epidemic Surveillance Using Foursquare Data.
By focusing on this scale, researchers can identify "social bubbles" and stable, recurring interactions rather than fleeting encounters. This approach has been shown to reduce model uncertainty by up to 95%, allowing for a much closer fit to real-world infection curves.
Digital DNA and Mobility Networks
To achieve this granularity, researchers are looking beyond traditional clinical reports to digital "proxies" of human behavior:
- Location-Based Social Networks: Xuelian Long and James Joshi used Foursquare data to identify the digital "DNA" of popular venues, showing how local ecosystems influence venue popularity and, by extension, potential transmission hubs Understanding venue popularity in Foursquare.
- Cell Phone Signaling: High-resolution mobility data allows models to move beyond static population density. Sirisha Rambhatla and Cyrus Shahabi applied Hawkes process modeling to cell phone data to generate neighborhood-level risk scores, providing a more accurate picture of how people actually move between regions Towards Accurate Spatiotemporal COVID-19 Risk Scores using High Resolution Real-World Mobility Data.
- Search Trends: Shihao Yang 0002 and Samuel C. Kou demonstrated that Internet search data can "nowcast" epidemics at the city scale, identifying hotspots before official government statistics are even released Advances in using Internet searches to track dengue.
From Blunt Lockdowns to Surgical Interventions
The ultimate goal of granular modeling is to replace blunt, socio-economically costly measures—like national lockdowns—with "surgical" interventions.
Sara Najem and her colleagues used network science to show that isolating just 20% of the most "central" localities (based on betweenness centrality) can trigger an 80% collapse in a disease's transmission network A Framework for Reconstructing COVID-19 Transmission Network to Inform Betweenness Centrality-Based Control Measures. Similarly, Jingyuan Wang found that while city-wide lockdowns stop transmission between neighborhoods, the virus often continues to "smolder" through movement within subdistricts, suggesting that local quarantine and testing are more effective than broad mobility restrictions High-Resolution Data on Human Behavior for Effective COVID-19 Policy-Making — Wuhan City, Hubei Province, China, January 1–February 29, 2020.
Operationalizing Local Insights
For these models to be useful, they must be accessible to local authorities. Johannes O. Ferstad and Joshua A. Salomon developed county-level tools that allow hospital leaders to simulate demand for ICU beds based on local census data and social distancing behaviors A model to forecast regional demand for COVID-19 related hospital beds.
By integrating high-resolution neighborhood data with regional mobility patterns through multi-scale Graph Neural Networks, as proposed by Giannis Nikolentzos and Konstantinos Skianis, the next generation of models can capture the "dialogue" between different geographical scales Predicting COVID-19 positivity and hospitalization with multi-scale graph neural networks. This ensures that a localized surge in one neighborhood is recognized as a potential regional crisis before it escalates.