by Thomas H Lee
Christakis and Fowler found that many of the friends and relatives named were also members of a Framingham study, and so there were data on them as well. The researchers were able to map network links among the participants at each examination and from one examination to the next. They were able to plot network dynamics over time: changes that were due to birth and death, the formation of new links through friendship and marriage, and links dissolved by relocation and divorce.
They found that having a friend who became obese increased one’s chances of becoming obese by 57 percent. The effect was even greater for mutual friends: pairs of participants who named each other as a friend. It was greatest in friends of the same sex. In fact, a close friend might be more likely to influence one’s weight than was a spouse. If a woman’s husband’s body mass index (BMI) topped 30, the chance that hers might follow suit was 37 percent. It was 40 percent if a sibling became obese. The authors published the study in the New England Journal of Medicine in 2007.8
The analysis seemed to rule out a host of factors that might have been responsible, including one of those most often blamed for the obesity epidemic: a change in the eating habits or availability of food in the community at large, such as a fantastic doughnut shop opening in the neighborhood. People didn’t gain weight at the same time as their neighbors. In fact, social ties were much stronger influences than was geographic proximity. Friends who lived hundreds of miles away had the same effect on their friends’ weight as did those who lived next door, whereas neighbors who weren’t friends had no effect.
Christakis and Fowler postulated that the influence wasn’t due to homophily, or “the birds of a feather flock together” effect. Obese people weren’t likely to be linked as friends initially. In other words, they didn’t seem to be attracted to one another because they felt comfortable or shared a love of food. One or both of a pair of friends became obese over time, and not necessarily at the same time.
Their evidence indicates that obesity became normalized in networks of friends. Having a friend of a friend become obese raised one’s chance of becoming obese by 20 percent. Christakis suggests that this effect is due to subtle changes in attitude. Here’s how it might work: You and Amy have been friends for a long time. You have lunch every couple of weeks, and Amy’s disapproving glance keeps you from ordering dessert. Amy also lunches with Sheila, a friend from childhood whom you don’t know. Sheila has been gaining weight gradually, and now she’s obese. Amy has become accustomed to seeing Sheila as a larger woman who thoroughly enjoys her food, and she appreciates Sheila’s healthy self-image. During one of your lunches, you order dessert and Amy doesn’t react. She’s become accustomed to eating with Sheila and has started to interpret overeating as having a healthy appetite. Soon you’re ordering dessert every time you eat out.
In that 2007 study, the effect declined with more distant relationships but was still present at a third degree of separation. Having a friend of a friend of a friend become obese is associated with a 10 percent increase in a person’s risk of obesity. Christakis refers to this as the Three Degrees of Influence Rule. Those authors have since shown, using a variety of ingenious experiments involving thousands and even millions of people, both online and offline, that findings similar to those in their observational studies can be documented with respect to behaviors as diverse as cooperation and vitamin use.
Christakis and Fowler’s study was controversial not just because they concluded that obesity is a contagious disease but also because of some of the analytic methods they employed. They responded to their critics in a paper published by Statistics in Medicine in February 2013.9 They also used alternative methods suggested by their critics in a December 2013 analysis tracing divorce through the Framingham cohort and came up with similar findings: Having a divorcing friend increases the likelihood that you and your spouse will split up. The likelihood was 75 percent if your own friends divorced and 33 percent if your friends’ friends split up. There was no association farther down the network.10
Rules of Life in the Network
From their study of many social networks and by observing the spread of norms and social values, Christakis and his coauthor have described “Rules of Life in the Network.” They begin by pointing out two fundamental aspects of the network, the first of which is the connections themselves: who has ties to whom and how strong and durable those ties are. The patterns are important as well. That is, if you have three friends, that is great. But if those three friends also have relationships among them, you have a stronger social network around you, and a new norm such as an epidemic of empathy may spread more quickly.
The second aspect is contagion, which they define as “what, if anything, flows across the ties. … [It] could be germs, money, violence, fashions, kidneys, happiness, or obesity. Each of these flows might behave according to its own rules.”11 They then define several rules that characterized the ways networks actually work.
Rule 1: We Shape Our Network
We choose whom we want to have in our network, how many close relationships we will have, and how they will be organized. In a survey of 3,000 randomly sampled Americans, Christakis and his colleagues found that people have just four close social contacts on average, with most having between two and six. In that survey, 12 percent of respondents listed no one with whom they could discuss important matters or spend free time and 5 percent had eight such people. Other researchers have found that the “core discussion networks” of Americans tend to decrease with age, that there is no difference between the sexes, and that college-educated Americans have core discussion networks nearly twice as large as those who do not graduate from high school.
In Christakis’s work, his team next asks people how interconnected their social contacts are with one another. If their contacts also have close relationships, there is greater transitivity. Some people live in a tightly woven network in which everyone knows everyone, whereas people who live in low-transitivity networks may have multiple contacts but those contacts come from different subgroups of the world around them.
In healthcare, the issues raised by this rule include how connected the groups of personnel taking care of patients are. Do they actually know one another? Refer to Figure 2.1 in Chapter 2, which shows how important the coordination of care is to patients. If patients feel that they have a real team rather than a loosely connected group that Michael Porter likes to call a pickup team, those patients have greater peace of mind, and a reasonable assumption is that they are more likely to recommend the providers to others.
Real teams are in many ways the antidote to the chaos of modern medicine. Real teams meet to improve the care they provide. They meet formally to discuss their performance data and what they can and should do to improve. They meet informally, bumping into one another in the hallways or chatting at social gatherings. When they meet informally, they often discuss the care of individual patients, and the care of those patients is better because of those interactions. On real teams, there are clear roles and everyone knows what the others are doing and what he or she is going to do.
Rule 2: Our Networks Shape Us
Individuals turn out differently as a result of their environments. Firstborn children are different from their younger siblings in statistically significant ways. They don’t all fit the stereotypes of eldest children, but there is a reason the stereotypes exist.
The transitivity of individuals’ social networks can change outcomes. For example, one of my colleagues and his wife met in high school. They dated for a while, but she broke it off. By that point, however, he had become good friends with her brothers and parents, and they would routinely invite him over for Sunday dinner. Once he even brought his new girlfriend with him. After high school graduation, they went to the same college and resumed dating. They married when she was a senior, and three decades and three children later, things seem to have worked out just fine. Would there have been this outcome if she hadn’t had a big, happy welcoming family t
hat took him in and didn’t let him go even after she broke things off? Probably not.
For healthcare providers, this rule has implications for real teamwork. If information can easily hop from person to person to person (“Mrs. Chan is worried about her knee surgery, but she is really worried about how her husband with Alzheimer’s is going to do while she is laid up”), one can imagine how much more likely it is for care to be both coordinated and empathetic.
Rule 3: Our Friends Affect Us
What flows around the network can be negative (e.g., racist ideology) or positive (e.g., studious habits, aversion to litter, even happiness). Christakis points out that social networks are agnostic: they can spread both good things and bad things. In healthcare, the implication is that we should highlight positive examples in an effort to make them the norm.
For example, my colleagues at Brigham and Women’s Hospital and I started a series in 2013 in which we put a physician-patient pair with a wonderful relationship on stage and interviewed them about what made the relationship so valuable to each of them. The format was modeled after the interviews of couples that are interspersed through the movie When Harry Met Sally. My Press Ganey colleagues and I have now done the same type of session elsewhere, and we have called this series “Love Stories.” The questions we ask the pair are the kinds of questions you might ask two spouses: “How did you meet?” “What was your first impression?” and “How did you know it was going to be good?”
From these sessions, I have been reminded of just how rewarding a good doctor-patient relationship can be for both parties. I also have learned little tips, such as specific questions and comments that clinicians can use to convey that they understand patients’ issues and are going to stick with them all the way and see them through.
Rule 4: Our Friends’ Friends’ Friends Affect Us
Christakis and his colleagues have shown that norms and behaviors spread in more complex ways than one individual infecting another or passing information on to another. Reinforcement of values such as the fundamental importance of empathic care must come from multiple contacts.
Social network scientists have done studies to determine how large a “stimulus crowd” is needed to get people walking down the street to look up at a high window on a building. When one person was staring up at the window, just 4 percent of passersby would stop and look up, but when 15 were staring up, 40 percent stopped to look. A stimulus crowd of just five produced almost as much effect as did much bigger groups.
The implication for healthcare organizations seeking to drive an epidemic of empathy is that highlighting individual role models may not be enough. Creating a critical mass of role models, getting clinicians together, and surrounding nonempathic clinicians with groups of empathic ones are all tactics that are likely to increase the effectiveness of spread of the desired value.
In fact, groups of clinicians often flip to better practice styles almost all at once. First, one or two clinicians start behaving in a different way, and others may or may not notice. But when three or four clinicians start behaving in the new way, most of the rest of the group will suddenly come along.
Rule 5: The Network Has a Life of Its Own
Christakis and his colleagues have shown that social networks have properties and functions that are not controlled or even perceived by the individuals within them; in other words, groups can be understood only as a whole. There is no one bird leading a flock of birds, but somehow they soar and dive and turn as a group. The researchers use the example of a cake, which has a taste that no single one of its components does. The whole is greater than the sum of its parts.
The question for healthcare organizations is this: How can we make our care seem to our patients more like a cake and less like the aisle in the grocery store with all the baking ingredients?
Mapping Social Networks in Healthcare
For social networks, function follows form. As Christakis notes, a pencil tip and a diamond are both composed of carbon, but their molecular structures determine whether they are soft and gray or hard and clear. The unique personality of a social network at any point in time is determined by the people in it and the types of connections among them. A network is built of nodes and connections.
The nodes are individuals. Although their genetics and acquired personality traits may determine how readily they form connections, their place in the network is determined by the number and strength of the connections they form.
The connections are the bonds between individuals. Reciprocal friendships form strong two-way bonds. One-way friendships form weak bonds. If you work in a large corporation with no middle management and your boss communicates by memo, he is likely to have hundreds of weak connections because few probably dare to write back to him. However, adding a layer of middle management whose feedback he values creates new stronger connections for the boss and adds more strong bonds between the managers and the workers they supervise. The workers are also likely to have a number of strong and weak bonds among one another that have sprung up for other reasons, such as working on projects together or sitting in the same pod.
Nodes with both strong and weak connections have a function and a value in a social network. People who form a lot of strong connections are likely to be near the center of a network, and they tend to form clusters with similar hubs. Those with weak connections are more likely to be at the periphery of the network. In that position they are well positioned to form bridges with other social networks, with which they may have stronger connections. Figure 5.1, which Christakis and Fowler created to illustrate the spread of obesity through social networks, represents the connections among 2,200 people in the Framingham Heart Study.
Figure 5.1 Largest connected subcomponent of the social network in the Framingham Heart Study in the year 2000. Reprinted from Nicholas A. Christakis and James H. Fowler, “The Spread of Obesity in a Large Social Network over 32 Years,” New England Journal of Medicine 357, July 26, 2007: 370–379.
Medical centers are rife with social networks that provide ideal structures for spreading values such as empathy, but those networks are not always apparent in the way one might imagine. The obvious networks of specialty divisions or service-worker unions may prove to be dead ends. Instead, the social networks with the greatest potential impact on the empathy and coordination of care are the teams that take care of patients, some of which are well defined, others of which are informal, and all of which are assembled differently for each individual patient.
How can you identify these social networks? Several software packages can be used to map all kinds of networks, but simple measures may suffice, as Christakis explained in the 2015 Lancet article referenced in note 11. In that study, Christakis and his colleagues recruited participants from three sets of nine villages in Honduras to receive nutritional supplements to hand out to their fellow villagers. In one set, the supplements were sent to random samples of villagers. In the second, they were sent to an equal number of villagers chosen because they had the strongest social ties. In the third, the researchers randomly identified the same number of villagers and asked them to name a friend, who then received the supplements to distribute.
The friends in the third village handed out 75 percent of the supplements, the randomly selected group in the first village delivered 66 percent, and the well-connected villagers in the second village delivered only 61 percent of the supplements they were given. This type of network targeting, without increasing the number of individuals targeted or the resources used, could be used to deliver all sorts of interventions, Christakis suggests.
In other words, simply ask a modest sample of clinical personnel who their friends are or who are the people they respect. See whose names come up more than once and you are on your way.
Tracking the Spread of Emotion
Just as collecting data from patients is the best way to measure whether patients’ needs are being met, the best way to track the spread of emotions in a social network
of healthcare personnel is to collect data from them. In most organizations, employee engagement data are collected only every year or two, but the clear trend is toward more frequent surveys. (It is easy to see that reporting very frequently on how things are going is likely to be part of the job for healthcare personnel in the future.)
The extreme version of data collection used in social network science is known as the experience-sampling method, which uses a series of alerts such as signals sent to a cell phone at varied (and thus unexpected) times, asking subjects to document their feelings, thoughts, and actions at that moment. For example, people may carry pocket devices and record moods four times per day over a five-day period. The data provide a detailed series of “emotional biopsies” from which a picture of the state of the social network can be defined.
From such data, researchers have been able to track the spread of emotions and values among teams of nurses and other groups of people working together, such as athletic teams. Data show that one positive person can improve the mood and performance of a group. Positive moods are associated with performance-enhancing changes, including more altruistic behavior, creativity, and efficient decision making.
