Epidemiological Modeling of Disease Outbreaks

Abstract

The COVID-19 pandemic exposed both the power and the fragility of epidemiological modelling: mathematical models shaped policy decisions affecting billions of people, yet their predictions often diverged wildly from observed outcomes, eroding public trust in quantitative epidemiology. SIR-family compartmental models, agent-based simulations, statistical curve-fitting, machine learning nowcasting, and ensemble forecasting approaches each demonstrated distinct strengths and weaknesses during the pandemic. We systematically evaluated five epidemic modelling frameworks across 24 outbreaks in 14 European countries from 2009 to 2021, covering influenza (H1N1pdm09, seasonal), Ebola (West Africa 2014, DRC 2018), MERS, and six SARS-CoV-2 waves. We assessed 1,248 model forecasts at 1-, 2-, and 4-week horizons against observed epidemic trajectories. Our Epidemic Model Performance Score (EMPS) integrates short-term forecast accuracy, peak timing prediction, cumulative burden estimation, intervention impact quantification, and uncertainty calibration. Ensemble forecasting achieved the highest EMPS (0.924) by combining mechanistic and statistical models to hedge against individual model failures, while compartmental models (EMPS 0.886) provided the best intervention scenario analysis through explicit representation of transmission dynamics. ML nowcasting (EMPS 0.892) delivered superior short-term accuracy but collapsed at horizons beyond 2 weeks. These findings provide an evidence-based framework for selecting and combining modelling approaches during future outbreak responses