People are more likely to take aspirin regularly if a friend or family member takes aspirin. A growing body of literature shows that behavioral changes that promote cardiovascular health may spread through social networks (Christakis and Fowler, 2007
; Christakis and Fowler, 2008
; Rosenquist et al., 2010
). The above results add to this literature by providing preliminary evidence that another important cardiovascular health behavior—regular aspirin use—may also be shaped by how members of one’s network behave.
Another contribution of this analysis is the examination of alters’ cardiovascular events. This approach has both conceptual and methodological advantages compared to prior work. Conceptually, it broadens the scope of what sorts of inter-personal effects might be relevant to individual and public health. Methodologically, it is advantageous because, when considering the effect of an alter event on an ego behavior, it may reduce concern about confounding (since it is a bit harder to imagine factors that are associated with alter events and ego behaviors than it is to image factors that are associated with the same behavior in both egos and alters).
While alters’ cardiovascular events were not associated with ego’s behavior when looking at average effects among all pairs in , we did find significant associations among certain subgroups. Men were more likely to take aspirin if their brothers recently had a cardiovascular event, and women were more likely to take aspirin if a same-sex friend recent suffered an event. There has been very little research into how people may learn about their own cardiovascular risk and possibly take cues from their friends or family members’ health problems (Khwaji et al., 2006
). Models and theories of behavioral change posit that health behavior depends on how people perceive their health risk and where they fall on a continuum of “readiness to change” (Prochaska and Velicer, 1997
). Having a friend or family member go through an experience like a myocardial infarction or stroke may be an important catalyst on the path toward healthier behaviors and better cardiovascular health.
Because aspirin-use is lower among women than men, and evidence of the benefits of aspirin is somewhat more controversial for women than men (Ajani et al., 2006
; Mulrow and Pignone, 2005
), we stratified the above models according to ego’s sex and the sex-composition of the relationship. While the above analysis does not allow us to draw clear conclusions about sex differences in the spread of aspirin-use through social networks, there are suggestive findings. It is notable, for instance, that only male alters, not female alters, appeared to shape ego’s aspirin use. Men’s aspirin use was associated with their male friends’ aspirin use. Women’s aspirin use was associated with their brother’s aspirin use. Female alters’ (i.e., wives, sisters, female friends) aspirin use, on the other hand, was not significantly associated with ego’s aspirin use in any of our stratified models. This greater sensitivity to male alter’s aspirin use may result from several different factors including higher rates of aspirin use among men, different perceptions of cardiovascular risk for men and women (Frijling et al., 2004
), and gender inequalities in the interpersonal dynamics of ego-alter relationships.
Our analysis also points to possible sex differences in sensitivity to alters’ cardiovascular events. Associations between ego’s aspirin use and alters’ cardiovascular events occurred only within same-sex relationship (i.e., brother pairs for men and same-sex friendships for women). There was no association between the ego’s aspirin use and the alter’s cardiovascular events within mixed-sex pairs. Although several factors may contribute to this pattern, people probably identify more strongly with, and take more health cues from, people of the same sex.
When interpreting the results of our study, it is important consider the role that subjects’ doctors may have played in their behavior. One possible explanation for associations between ego’s and alter’s aspirin use is that they both have the same physician who encouraged aspirin use. In our study sample, we were able to identify ego-alter pairs who shared the same doctor and adjust our estimates for this potential confounder. Associations between ego’s and alter’s aspirin use were quite robust to this adjustment and it does not appear that shared doctors can account for our results (see Table S1 in the online data supplement
). Unfortunately, our data do not allow us to know whether egos who were influenced by alters’ aspirin use involved their doctors in their aspirin use decisions (e.g., when an ego learns of an alter’s aspirin use, does she then turn to her physician for advice about aspirin prophylaxis?).
In an effort to focus our study on egos who are likely candidates for aspirin prophylaxis, we analyzed women ages 55 to 70 and men ages 45–79. We believe this broad, demographically-based definition of an aspirin use risk group is appropriate given that we are examining how people make choices about aspirin use outside of clinical settings. Further, because clinical recommendations regarding regular aspirin use were being developed over the period in our study (i.e., 1971–1998), it is important that we capture a broad segment of the general population.
It should be kept in mind, however, that influences on alters’ behaviors and health may operate differently on egos with specific clinical risk factors (e.g., those based on blood pressure, cholesterol, diabetes, etc). We found consistent results when we replicated our main analysis for the subset of egos with prior cardiovascular events and/or elevated 10-year coronary heart disease risk scores (based on Wilson et al., 1998
). However, when we limited the sample further to include only egos with prior cardiovascular events, we no longer found a significant association between egos’ and alters’ aspirin-use. Interpreting this non-significance is difficult, though, because relatively few egos had prior cardiovascular events, raising concerns about sparse data and large standard errors (these results are presented in the on-line data supplement
). Further research, preferably with larger samples, is needed to know whether individuals at higher risk for cardiovascular disease are more or less sensitive to the health and behaviors of their friends and family. This study makes no suggestions related to clinical guidelines or about who should take regular aspirin doses. Rather, our results document how social networks are correlated with whether or not an individual adopts a regular aspirin regimen, regardless of specific clinical risk factors.
This work has a few notable limitations. It does not randomize individuals into social networks, thus leaving open the possibility that these results may in part reflect homophily-driven selection bias on the basis of unobserved traits (e.g., avidity for drugs) that influence the use of aspirin over time. For instance, we can imagine that individuals with similar tastes, health knowledge, or physical resilience may be more likely to form relationships. Shared leisure time activities, as well as similar sociodemographic positions (e.g., educational backgrounds or professions), may increase the chances that people with underlying tendencies toward aspirin-use meet and form relationships. While observational data can never overcome all concerns about unmeasured confounding, the Framingham Heart Study Social Networks Study, which is longitudinal and provides several control variables, gives us more leverage on causality than is usually possible with non-experimental data. Most notably, our data allow us to adjust for egos’ prior aspirin use and cardiovascular events. Also, as has been widely noted, the Framingham Heart Study sample is somewhat homogenous and does not have a significant percentage of underrepresented minorities.
Because of differences in how questions about aspirin use were asked in earlier and later waves of the Framingham Heart Study, our dependent variable captures daily aspirin use, but does not detect less frequent regular aspirin doses (e.g., taking aspirin every other day). Furthermore, with these data, we can measure egos’ aspirin use only every four years or so. Lower levels or shorter-term changes in egos’ aspirin use are, therefore, not detected in our study. This may mean that we fail to capture some variation in egos’ aspirin use, which would imply our estimates may fall on the conservative side.
Because our dataset captures a limited number of relationship ties, we were unable to stratify the data in certain ways, and there were several questions about heterogeneity across subgroups that we were not able to test. For instance, after stratifying by sex and relationship type, we did not have a sufficient number of observations to test whether associations differed depending on whether ego had cardiovascular disease and/or had taken aspirin previously. It should be kept in mind that the above results reflect average associations for a general population. It remains possible that associations between egos’ aspirin use and alters’ aspirin use/events may differ when distinguishing between further subgroups. Finally, the analyses comparing results across different types of relationships and exposures are exploratory. We did not have prior hypotheses about effect sizes for different types of relationships or exposures. These results should not be interpreted as evidence of differences in causal effects across groups/exposures. Since there is very little existing research into peer influences in drug-taking behavior, we believe that this type of exploratory analysis provides a useful first step in understanding how networks may shape pharmacotherapy. In this case of an exploratory analysis, it is not clear whether Bonferroni adjustments for multiple hypothesis testing are appropriate; nevertheless, we note that all Bonferroni-adjusted p-values in our analysis were greater than .05.
Across a broad swathe of behaviors, people are influenced by those around them. Pharmacotherapy is a behavior, and so we should not be surprised by the fact that people’s drug-taking behavior is related to the behavior of those around them, and to the events occurring in those around them. Similar to the person who might stop smoking when his friend gets lung cancer, a person whose friend, sibling, or spouse has a myocardial infarction may be more inclined to take aspirin because he/she now has a palpable demonstration of the occurrences the aspirin is intended to prevent, rather than an abstract admonition. Likewise, those whose friends are taking aspirin might follow suit for a variety of reasons, including the basic realization that taking aspirin is not hard at all. People are connected, and so their health is connected.