Background

Human behavior is a key factor in the spread of infectious diseases, with age as a particularly important predictor of social mixing behavior. Existing data on contact mixing patterns typically involve healthy individuals or those whose health status is unknown; thus, the use of contact matrices derived from data on healthy individuals to evaluate public health interventions may paint an inaccurate picture of their effectiveness. This study evaluated how social contact patterns over the course of acute infection may affect the estimated impact of interventions.

Methods 

SickMix is a longitudinal study of social mixing patterns among individuals in Portland, OR, experiencing acute respiratory illness or acute gastroenteritis. Using contact matrices from SickMix, we developed two age-structured susceptible-infectious-recovered (SIR)-type models with acute and late infectious compartments: a behavior-dynamic model accounts for reductions in social contacts during acute and late infectious phases, while a behavior-naive model does not. Simulating the spread of a generic respiratory pathogen among the U.S. population, we compared cumulative cases, differences in the basic reproductive number (R0), and outcomes of public health interventions for the dynamic and naive models. 

Results 

The behavior-naive model yielded 37% more cases compared to the dynamic model, with an R0 ratio of 1.35 (2.37/1.76). Age-specific and total attack rates were lower for the dynamic model (Figure). Failing to account for reduced social mixing associated with illness resulted in an overestimation of the reduction in total cases for daycare closures (5.6% vs. 4.9%) and nursing home isolation (2% vs. 0.8%) and similar reductions for K-12 school closures (99.9% vs. 99.9%). 

Conclusion 

Our results demonstrate the importance of accounting for changes in social behavior during illness when modeling infectious disease transmission and evaluating the effectiveness of public health interventions.