Neighborhoods and Health

Abstract

Neighborhoods and Health: Ichiro Kawachi and Lisa F. Berkman. Oxford University Press, New York, 2003. ISBN: 0195138384. Price: $58.95; 368 pp. For epidemiologists, “place” was once a central concern. In his essay “On Airs, Waters and Places,” Hippocrates advised that “when one comes into a city to which he is a stranger, he should consider its situation...and the mode in which the inhabitants live, and what are their pursuits.” Brian MacMahon saw fit to situate this quotation at the beginning of his pioneering 1960 textbook Epidemiologic Methods,1 and this same book contains an entire chapter devoted to “Place.” The epidemiologic triangle, connecting its nodes of host, agent, and environment, emphasizes each apex equally, with the environment being “All that which is external to the individual human host.”2 Developments in molecular genetics and statistical methodology during the latter decades of the 20th century, however, led epidemiologists to emphasize matters of host and agent, often at the expense of context. This paradigm reaches its most absurd pinnacle in the surfeit of articles purporting to describe “gene–environment interaction” in which the “environment” is assessed solely at the level of the host, for example by defining it as nothing more than an individual's smoking behavior. In the most recent decade, many epidemiologists have turned their attention back to the environment in a vigorous fashion, including both social and physical dimensions. An exemplary representation of this resurgent interest in the role of context, enriched by methods and concepts from the social sciences, is the new volume by Kawachi and Berkman, Neighborhoods and Health. Although “neighborhoods” are for the most part defined literally (as geographic entities), a superb chapter by Berkman and Clark extends the discussion to include social connections that defy physical boundaries. The volume is enthusiastically interdisciplinary, with chapters by sociologists, statisticians, geographers, and others. Its only obvious parochialism lies in the choice of authors, for although the editors claim to have “gathered together contributions from leading international investigators,” almost half are from a single university. The claim of an “international” span rests largely on the presence of several British contributors, but the only substantive material from outside of the United States is in the fine contribution by David Gordon on area-based deprivation measures in the United Kingdom. The editors propose, rather ambitiously, that the purpose of their text is to “inspire researchers and policy-makers to move the field ahead by refining the methodology, adding new evidence, or devising novel policy approaches to improve the health of populations.” Drawing appropriately on many of the seminal articles and texts in the field, this book does accomplish a varied and accessible overview of the modern literature. In that regard, it is perhaps best suited as an introductory text, because several contributions offer some particularly helpful summaries. The chapters by Krieger et al. and by Acevedo-Garcia and Lochner, for example, provide systematic compendia of explicit variable definitions and equations. This condensation of tools and concepts can serve as an invaluable aid to those approaching the field for the first time, and thus will surely help facilitate informed interest and more rigorous approaches to future research endeavors. In approaching such a broad and complex topic area, however, it is impossible to introduce all issues thoroughly. Potential readers should be warned that there are no spatial methods contained in this text, nor indeed any substantial geographic perspective at all. Likewise, physical neighborhood conditions that are crucial determinants of health such as housing stock—availability, price, age and quality, and inherent health risks (mold, vermin) or supports (green space)—receive little discussion. Moreover, the chosen examples are unapologetically urban in focus. Furthermore, some more challenging issues are presented without sufficient criticism of current scholarly practice. There has been considerable excitement about multilevel regression models, for example, especially the promise of separating contextual from compositional effects. Yet, the typical application of this variance decomposition methodology in the epidemiologic literature rests on a logical fallacy.3 Authors generally adjust for a few (poorly) measured individual-level (ie, compositional) predictors, and declare that any remaining between-clusters variance is contextual (eg, Subramanian et al.4). However, unexplained variability is exactly that—unexplained. It can arise from uncontrolled causes at the individual or the aggregate level. To label unexplained variability as contextual after controlling for a few compositional measures is akin to the police detective who reasons, “The victim was not murdered by either of two men I have detained and questioned, so therefore the murderer must have been a woman.” The text asserts that the time has come to use our research results to inform policy decisions, but then largely avoids the significant challenge of discerning the specific public policy implications of neighborhood-level estimated effects from observational data. When estimated effects do not correspond to specifically defined actions, it is by no means clear what they tell us about optimizing decisions for the health of communities. For example, an estimate of the effect of having a medical center located in a neighborhood is a useful piece of information in helping us to weigh the overall costs and benefits of constructing such a facility. It therefore has a direct connection to policy decisions. On the other hand, consider the exposure “social capital” as assessed by voting behavior. In this case, it is more difficult to interpret a causal effect estimate for several reasons. First, the effect corresponds to hypothetical intervention on the indicator, not the latent construct. Even if voting behavior is truly a good indicator of social capital, it does not logically follow that inducing changes in the indicator will in turn affect the latent quantity.5 Furthermore, it is not at all made clear by such a model how voting behavior would be modified. Certainly neighborhood-level interventions are possible and can be enormously beneficial for health, whether at the compositional level or the contextual level. If we truly want our research to play a role in policy decisions, we need to be providing the necessary evidence for exactly which of these specific potential interventions are most worthy of implementation. The pendulum is swinging back, and epidemiologists are now pursuing “place” with a fresh methodologic armamentarium and theoretical perspective. The book Neighborhoods and Health helps us locate this renascent field on the epidemiologic landscape, and provides for the first time an organized and readable overview of the basic issues and problems facing this research paradigm. The book authors’ enthusiasm for this research is evident throughout, and helps to make this work as enjoyable as it is instructive. Equipped with this text, public health researchers can now look forward to further progress in reconceptualizing the relationships between neighborhoods and the people who reside in them.