Dissertation Title: "Social Networks and Health: From Epidemiology to Intervention"
This dissertation applies network science to three foundational problems in: epidemiology (the social gradient in mortality), health economics (the geographic variation in health care spending), and public health (the diffusion of knowledge and behavioral change). Chapter 1 investigates the relationship between social network position and fibrinogen, a biomarker of inflammation and cardiac risk. Socially isolated individuals face elevated rates of illness and death not explained by behavior alone. Conventional measures of social connectedness reflect an individual’s perceived network and are subject to bias and variation in reporting. In this study of a large social network, I find that greater indegree, a sociocentric measure of friendship and familial ties identified by the ego’s connections rather than the ego herself, predicts significantly lower ego fibrinogen, after controlling for demographics, education, medical history, and known predictors of cardiac risk. The effect size of social isolation, as measured by low indegree, is comparable to that of smoking, and greater than that of low education, a conventional measure of socioeconomic disadvantage. By contrast, outdegree, which reflects an individual’s perceived connectedness, is weakly associated with fibrinogen.
Chapter 2 turns to the networks of physicians whose behavior governs the cost and quality of health care across the country. Using data on hundreds of millions of patient-sharing relationships from 2009-2014, I construct comprehensive longitudinal networks of Medicare providers. For all providers billing for at least 100 office visits in 2012, I calculate six specialty-adjusted measures of billing intensity. After accounting for attributes of individual physicians, pairs of physicians who share patients are more similar on six dimensions of billing intensity than physicians who do not share patients but share a common colleague, who in turn are more similar than pairs of physicians separated in the network by three degrees or more. Moreover, patterns of physician clustering differ dramatically by region, with implications for efforts to reduce healthcare spending, and for the detection of fraud and abuse.
Chapter 3 describes the first real-world, large-scale randomized trial of network interventions for public health. In 32 villages of rural Honduras, we delivered two dissimilar public health interventions: chlorine for water purification, and multivitamins for micronutrient deficiencies. Using a block randomized design, we assigned villages to one of three targeting methods, introducing interventions to 5% samples composed either of: randomly selected villagers, villagers with the most social ties, or nominated friends of random villagers (the last strategy exploiting the “friendship paradox” of social networks). We compared the diffusion of the products and of related knowledge across the three methods of network targeting. Targeting the most highly connected individuals produced no greater diffusion of knowledge and behavior than random targeting. Targeting nominated friends, by contrast, increased adoption of the nutritional intervention by 12.2% compared to random targeting, and also improved villagers’ knowledge of the intervention’s usage and benefits at follow-up. This method has the additional advantage of scalability, because it can be implemented without mapping the network. Deploying certain types of health interventions via network targeting, without increasing the number of individuals targeted or the resources used, may enhance the adoption and efficiency of those interventions, and thereby improve population health.