Protective behaviors and social contacts in epidemiology: from data collection to data-driven models

ABSTRACT
The COVID-19 pandemic has intensified long-standing epidemiological needs: timely behavioral data, its integration into mechanistic models, and auditing which data components most influence outcomes. This presentation addresses these interconnected areas, moving from data collection to data-driven modeling and systematic model auditing.
First, I present two data collections using original surveys that investigated public awareness for dengue and behavioral relaxation post-COVID-19 vaccination. This was complemented by analyzing Italian social contact data (CoMix) to quantify the impact of interventions on contact patterns. These studies highlighted important predictors of behavioral changes such as risk perception and constituted a rich empirical foundation for the second part of the work. 
Second, these insights were translated into mathematical models. I built epidemiological frameworks to study the impact of heterogeneities in behaviors due to risk perception on the disease dynamics and to audit the impact of gender-stratified data (behaviors and susceptibility) on COVID-19 mortality.
Key findings reveal that behavioral dynamics, like the relaxation of measures by those with low risk perception, can impact epidemic trajectories. The audit demonstrated the importance of the contact structure to obtain realistic results and how gender-stratification of contacts and infectious fatality ratios increase the performances of models. 
This work concludes that a symbiotic relationship exists between empirical data and modeling: data grounds models, while models test the importance of observed behaviors. It underscores that heterogeneity is a core driver of epidemic spread and that models should be developed “fit for purpose.” This work argues for an integrated epidemiological approach that centers the measurement and modeling of human behavior to build effective and equitable public health tools.