Pandemic waves as the outcome of coupled behaviour and disease dynamics: a mathematical modelling study

BackgroundThe COVID-19 pandemic was strongly shaped by the interaction between population behaviour and transmission dynamics. Standard mathematical models do not account for this interaction, however. Objectivewe tested whether adding a mechanistic representation of population behavioural dynamics improves the ability of a mathematical model to explain and predict COVID-19 pandemic waves. MethodsWe compared a standard Susceptible-Infected-Recovered (SIR) model to a variant (SIRx) with a mechanistic representation of behavioural processes, including two-way coupling between behaviour and transmission dynamics. We used approximate Bayesian computation to parameterise the models with SARS-CoV-2 case incidence and the Oxford stringency index from 13 European countries. Models were fitted to the Spring 2020 wave, and their out-of-sample prediction for the Summer/Fall 2020 wave was tested. Outcome measures included the Akaike Information Criterion (AICc), the area between empirical and model epidemic curves, and predicted timing/magnitude of the second wave. ResultsThe average AICc for the SIRx model across all 13 countries was lower (-2638{+/-}345 versus - 2295{+/-}212 for SIR), meaning that the SIRx model explains the data more parsimoniously. The average area-between-curves was also lower (0.072{+/-}0.071 versus 0.16{+/-}0.11). The predicted peak magnitude for the SIRx model (0.0015{+/-}0.0014) was closer to the data (0.0006{+/-}0.0005) than the SIR prediction (0.0083{+/-}0.0090). The average day-of-peak for the SIRx model (283{+/-}19 days from first data point) was also closer to the data (278{+/-}25), than the SIR prediction (253{+/-}31), although the 95% credible intervals for individual countries were very large. ConclusionCoupling behavioural and disease dynamics improves the ability of mathematical models to explain and predict crucial features of pandemic waves. Research in contextO_ST_ABSEvidence before this studyC_ST_ABSMost mathematical models of infectious disease transmission do not explicitly account for behaviour, but the COVID-19 pandemic clarified the role of behavioural processes in determining the trajectory of infectious diseases in populations. On the other hand, many theoretical models of coupled behaviour-disease processes exist, although relatively few attempt to validate these models against data. We searched Google Scholar using the terms COVID-19 model, and behavio*-disease or behavio* epidem* from March 1, 2020 to October 8, 2025. We did not find any papers that compared retrospective out-of-sample model predictions of COVID-19 pandemic waves of a non-behavioural transmission model to the predictions of a coupled behaviour-disease model, in multiple populations. Added value of this studyWe carried out such a comparison for 13 European countries, by fitting models to the first COVID-19 wave in Spring 2020 and testing how well they would have predicted the second wave. We found that the coupled behaviour-disease model predicted the second wave better than the non-behavioural model, and was also more parsimonious, despite having more parameters. This study shows that feedback between disease dynamics and behavioural dynamics is a significant factor for determining the timing and magnitude of pandemic waves caused by an acute respiratory infection. It also shows that integrating population behaviour dynamics into transmission models is feasible, and can better explain observed temporal patterns in case incidence. Implications of all the available evidenceMathematical models that endogenously include the feedback between infectious disease dynamics and behavioural dynamics can add a unique and complementary tool to the public health modelling toolbox during a pandemic. Such models could help design public health interventions by improving our ability to anticipate population responses to both the interventions themselves, and a rapidly evolving epidemiological landscape.