Spreading potential in disease relevant networks: Predicting centralities in rural Northeast Madagascar

Abstract

Heterogeneity in contact patterns can have marked effects on disease transmission, including through superspreading where few individuals drive most infections. Networks based on different types of human-human contacts quantify individuals' centrality, which can be used to identify individuals or sub-populations who are at increased risk of spreading disease. By understanding the predictors of centrality, high-risk individuals and sub-populations can be targeted to improve public health intervention strategies, even when detailed network data are unavailable. This study inferred transmission potential networks representing different pathogen transmission pathways among people living in rural villages of northeast Madagascar. We constructed four network types: social, close contact, household proximity, and environmental overlap using survey data and global positioning system (GPS) trackers. We then investigated how sociodemographic and anthropometric variables predicted different types of network centralities using multiple mixed effects linear models. Gender and wealth based on household material quality tended to be the most important sociodemographic predictors of centrality, but centrality outcomes varied by network type and had wide confidence intervals. Men tended to be more central to their environmental overlap network than women. Further, wealth based on household materials was an important, positive predictor of close contact network centrality. Gender and wealth were associated with centrality in transmission-potential networks but varied in their importance across different network types. Our study results suggest that targeted intervention efforts focused on diseases that are transmitted through shared environments (i.e., parasites shared through soil or water) or direct contact (i.e., respiratory infections) in similar agricultural settings should consider gender- and wealth-associated differences in contact patterns.

Fig 1. Network visualizations for each village and network type.

Nodes represent individuals and edges represent their connections. Substructures that appear disconnected in these networks were based on applied thresholds that helped visualize connections; there were no completely disconnected network fragments in our study.

Fig 2. Coefficient plots of the relationships between sociodemographic variables and centrality for each network type (a. Social Network, b. Close Contact Network, c. Household Network, and d. Environmental Network).

Each panel represents a different model. Points represent estimated effects; thick bars represent 90% confidence intervals and thin bars represent 95% confidence intervals; color represents variable importance. Models control for village, season, and centrality metric type (S7 Fig).