Daniel Goldberg examines a new study establishing the link between social isolation and mortality, and asks what these findings might reveal about the pathways leading to health inequalities.
In his 2000 book Bowling Alone, sociologist Robert Putnam famously declared that “[i]f you smoke and belong to no groups, it’s a toss-up statistically whether you should stop smoking or start joining.” This is obviously a provocative phrase, especially for the contingent of public health stakeholders that are undeniably a large part of Putnam’s current audience.
Putnam is one of the world’s foremost experts on the idea of social capital, which focuses on “social relations that have productive benefits.” The concept is both highly contested and yet is generally regarded as extremely important in clear thinking about health inequities and the social determinants of health.
Social Capital & Health
Many of the contests over social capital turn on of what exactly the concept is said to consist. Szreter and Woolcock acknowledge the difficulty, and analogize social capital to the current that travels along the wires of social networks. Nevertheless, the evidence that social capital, is correlated with health and its distribution is robust, to the point where even skeptics acknowledge that something important is likely captured in the evidence that links measures of social capital to community health.
In one sense, the connection is at least superficially plausible; humans are social animals, so dissolution of preexisting social bonds is unlikely to be salubrious. The literature on social capital suggests that, in some sense at least, the greater our bonds and connections with (at least some) others, the better our health will be.
But while there is widespread agreement on the adverse effects of social isolation on health, many of the specifics remain open to question. Just how deleterious is social isolation? And how does its effect size compare to the adverse health effects of the risk factors most commonly studied in clinical epidemiology?
The authors of a recent study entitled “Social isolation: a predictor of mortality comparable to traditional clinical risk factors” set out to examine exactly these questions. Pantell et al.’s study is unique for a number of reasons, and its findings are extremely important.
Does Social Isolation Predict Mortality?
First, the authors note that while prior studies do link inversely social isolation with mortality, until the present study (which uses data from NHANES III), there has been virtually no large-scale, nationally representative study of the subject in the U.S. The primary outcome variable was mortality, and the investigators analyzed its association with social isolation. Covariates included cigarette usage, BMI, elevated blood pressure, high cholesterol, and demographic variables such as age, race/ethnicity, educational level, income level, and self-reported health.
Second, consistent with the evidence documenting a social gradient for many risky health behaviors, the authors found that social isolation is associated with higher prevalence of smoking across gender, and with elevated blood pressure and high cholesterol for women.
Third, in terms of the overall relationship between social isolation and the primary outcome of mortality, the authors found a significant correlation (hazard ratio of 1.65; 95% CI, [CI]=1.29,2.02).
Kaplan–Meier survival estimates by Social Network Index score among (a) women and (b) men: Third National Health and Nutrition Examination Survey, United States, 1988–1994.
This correlation, the authors note, is comparable to the risk-factor-mortality relationship they found for smoking, and was stronger than that found for high blood pressure. Interestingly, neither obesity nor high cholesterol correlated with mortality (more on this below).
The authors pressed the analysis further by examining whether specific aspects of social isolation predicted mortality, and then comparing these findings to the covariate clinical risk factors. Among men, the authors found that being unmarried, infrequent participation in religious activities, and lacking club associations (huzzah Putnam!) predicted mortality. The first two also predicted mortality among women, but interacting infrequently with friends and family was the third predictor for women. The authors conclude that “social isolation factors predicted mortality at hazard ratio levels similar to or higher than those of several standard clinical risk factors.”
What should we take away from this fascinating study? First, we should embrace more emphatically Putnam’s quote. Whether it was really offered more tongue-in-check, as a provocation to engage the considerable evidence suggesting the import of social connections and/or social capital, this study suggests that if we did not otherwise, we ought to take Putnam’s idea extremely seriously.
Second, note that obesity and high cholesterol did not predict mortality among the data set. The authors note several possible explanations, including the limited time horizon for showing mortality impacts (14 years), medical treatments, and, interestingly, the obesity paradox itself (that among some groups, adiposity is protective of health). Although obviously the study does not license any global inferences regarding the effect size of obesity on mortality, the fact remains that in this robust study from an excellent data set, social isolation proved a vastly better predictor of mortality than obesity (which, again, did not predict mortality at all).
Third, what are the possible causal mechanisms that could explain the social isolation-mortality correlation? The authors note that isolation may signal a lack of access to resources that can be used to promote health, and, more importantly, can buffer the physiological response to stress. This last point matters because it ties into the allostatic load hypothesis, and links one of the most promising causal models for the social determinants of health in general to social isolation.
Fourth, the authors argue that we should begin to take seriously the idea of social isolation as a “modifiable risk factor.” This is important, and I agree whole-heartedly. But how should we aspire to do so? The authors suggest that clinicians should inquire as to social isolation during the health care encounter, because doing so could “potentially help in discerning which patients have worse health outcomes and targeting those patients for increased surveillance.”
This last point is especially important to me because I disagree with it so vehemently. Of course, there is nothing whatsoever wrong with the idea that clinicians should include questions geared at detecting social isolation in the course of taking a history. Indeed, this is all to the good. But, ameliorating the devastating impacts of social disadvantage cannot be accomplished at the level of clinical intervention. It is totally unclear to me how increased surveillance is likely to alleviate social isolation when much broader structural factors converge to create the conditions in which social isolation can take root and expand. And even if clinical attention can help manage social isolation, it, like most other clinical interventions, is directed towards managing the risk factor once it is has already developed, rather than with altering the conditions so as to prevent or at least delay the onset and development of the particular risk factor.
One of the most important lessons I take from the work of the great Barbara Starfield is the danger of medicalizing prevention. If social isolation is an important risk factor, and if it is itself socially determined – and how could it NOT be? – then preventing such isolation would seem to require intervention directed much higher up the causal chain than at levels proximal or subsequent to the onset of disease. (I have a much more detailed argument regarding levels of intervention for a different health problem here).
Regardless of my disagreement with this last point, this is a hugely important study. It paves the way for future work “[e]xamining the pathways through which social ties affect mortality.”