Agent-based simulation of epidemic incorporating real data

Resumo

The spread of infectious diseases in large urban populations is strongly influenced by individual-level contact heterogeneity and network structure. Agent-based models combined with network representations can capture this complexity beyond the homogeneous mixing assumptions of classical compartmental models. We developed an agent-based simulator using a Watts-Strogatz small-world network to represent interaction, and variable contact probabilities to reflect patterns observed in data. The model was programmed to generate execution logs summarizing agent histories. The software was tested with actual data from a covid outburst in Curitiba, Brazil, a city of approximately 1.77 million inhabitants, covering a period of 240 days. Contact probabilities were adjusted in eight intervals optimised against actual data using the nomad optimization tool. Three counterfactual scenarios were evaluated: public transport restrictions, maintained public transport capacity, and delayed school closure.  The model achieved a good fit to actual data (RMSE = 9125; R² = 0.767) and sensitivity analysis showed that contact probability perturbations during the first 150 days produced the largest effects. The contact probabilities exhibited a temporal pattern consistent with changes in social behaviour and policy interventions. Counterfactual experiments showed strong nonlinear responses. The model demonstrated that the integration of demographic realism, small-world network topology, stochastic simulation, and data-driven calibration produced a computationally feasible framework that reproduces empirical epidemic dynamics of a large urban setting. The simulation outputs (contagion trees, secondary-case distributions, serial intervals) provide epidemiological information not accessible from compartmental approaches, supporting decision making for public health planning and epidemic preparedness.