Commentary - Contagion in Prescribing Behavior Among Networks of Doctors

A foundational study regarding the diffusion of innovation involved the adoption of tetracycline by doctors in four Midwestern communities in the 1950s (Coleman et al. 1966), and it is not a coincidence that social scientists keep returning to this particular application of social network analysis (including to reanalyze this original study by Coleman and colleagues; see Van den Bulte and Lilien 2001). Studying the use of drugs by doctors involves the perfect mix of a discernable social network (among the doctors), a distinct innovation (a drug), an important area (patient care), difficult statistics (related to causal inference), and financial stakes (by pharmaceutical companies, insurers, and others). Iyengar, Van den Bulte, and Valente, in their careful and insightful study (Iyengar et al. 2011), find that even after controlling for marketing effort and arbitrary systemwide changes, there is evidence for contagion in the prescribing patterns of doctors. The modeling is excellent, including controls for secular trends (such as the emergence of new drugs or clinical evidence or a change in prevalence of the disease), and the conceptual framework is innovative and comprehensive. And so this paper represents another important link in the chain stretching back to Coleman. Like any social network, a network of doctors provides an opportunity to explore whether people who share a social connection exhibit similar behavior. However, even if we find that connected doctors do act in the same way, there are a number of possible explanations: (1) doctors might choose to associate with other doctors that they already resemble (known as homophily, or “birds of a feather flock together”— for example, infectious disease doctors might form ties with other doctors of the same specialty, or high-prescribing doctors might be connected to other high-prescribing doctors), (2) doctors might jointly experience unobserved contemporaneous events that cause their behavior to covary (omitted variables or confounding—for example, the doctors might jointly experience a marketwide ad campaign), and (3) doctors might exert social influence on each other (induction, or social contagion). Of course, all three phenomena (homophily, confounding, and induction) are likely to be present in any social system, and the methodological challenge is to disentangle them, if possible. In our original work regarding the spread of obesity within social networks, we argue that the obesity epidemic could be understood, in part, as the adoption of the “innovation” of obesity, and, among other things, we propose a test for social influence involving the asymmetry of social ties (Christakis and Fowler 2007), an approach that we have since applied in other domains (Cacioppo et al. 2009; Christakis and Fowler 2008; Fowler and Christakis 2008; Mednick et al. 2010; Rosenquist et al. 2010, 2011) and that has been explored and clarified by other scientists (Bramoulle et al. 2008, Anagnostopoulos et al. 2008, Shalizi and Thomas 2010). This approach can shed light on the existence of contagion, subject to certain assumptions about how people go about nominating friends (Shalizi and Thomas 2010). Moreover, alternative methods can help to parse how much of the observed association in adoption is due to homophily and how