Fat chance that obesity would turn out to be simple! But, then, what is it?

That obesity rates are high in poor neighborhoods because these areas are "food deserts", home not to grocery stores that provide fresh produce but to fast food restaurants and liquor stores, is an idea that has gained traction in recent years.  Is it true?  Gina Kolata of the New York Times reported yesterday on two new studies that say no. 

Kolata writes:

...two new studies have found something unexpected. Such neighborhoods not only have more fast food restaurants and convenience stores than more affluent ones, but more grocery stores, supermarkets and full-service restaurants, too. And there is no relationship between the type of food being sold in a neighborhood and obesity among its children and adolescents.

Within a couple of miles of almost any urban neighborhood, “you can get basically any type of food,” said Roland Sturm of the RAND Corporation, lead author of one of the studies. “Maybe we should call it a food swamp rather than a desert,” he said.

The solution to obesity, some have said, is to provide fresh produce.  Families will buy it and children will eat strawberries and lettuce rather than hamburgers and fries, and obesity rates will fall.  But Helen Lee, in her study published in Social Science & Medicine in April, finds that

...children who live in residentially poor and minority neighborhoods are indeed more likely to have greater access to fast-food outlets and convenience stores. However, these neighborhoods also have greater access to other food establishments that have not been linked to increased obesity risk, including large-scale grocery stores. When examined in a multi-level modeling framework, differential exposure to food outlets does not independently explain weight gain over time in this sample of elementary school-aged children. Variation in residential food outlet availability also does not explain socioeconomic and racial/ethnic differences. It may thus be important to reconsider whether food access is, in all settings, a salient factor in understanding obesity risk among young children.

She used data from the Early Childhood Longitudinal Study kindergarten cohort (ECLS-K), 1999-2000, as her source of data on body mass index (BMI) and residence in a random sample of children in kindergarten through fifth grade.  She used longitudinal data on all businesses in the nation from 1992 to 2006 for data on what kinds of businesses are in which neighborhoods, and she categorized food availability by type of food store from full supermarket to corner store.  From these data she concludes that "food outlet exposure holds no independent relationship to child weight gain." 

The second study, published in the American Journal of Preventive Medicine in February, reports no "robust relationship between food environment and consumption".  This study is based on dietary and BMI data from the 2005 and 2007 California Health Interview Survey (CHIS), and "food environment" data measured as "counts and density of businesses" categorized by type and distance from a respondent's home or school.  The authors found no correlation between food availability and BMI.

What do the findings of these two studies mean about childhood obesity?  And, do they prove that the idea of food deserts is an urban myth, unrelated to obesity rates?

To us, these studies suggest several things.  Both base their information about food availability on population-level data, and even though they have data on children's individual weights and heights, this tells us nothing whatsoever about where their parents were shopping for food nor what they bought nor what the children actually were in the habit of eating.  Assuming that it means anything at all about this is an example of the 'ecolological fallacy', the imputation of population-level data to individuals.  Even more problematic is the assumption that because fresh produce is available, children are eating it.  Yes, the CHIS data included some information on what children reported eating, but dietary information collected in this way is notoriously unreliable.  Indeed, dietary information collected in any way is notoriously unreliable.

So, although these studies may indeed show that poor neighborhoods are not the 'food deserts' they've been assumed to be,  as they set out to do, this tells us little to nothing about the causes of obesity in poor neighborhoods.  Or indeed anywhere -- there are plenty of overweight and obese children in middle and upper class neighborhoods as well.

And, even if produce is available, it can be more expensive than processed foods, and often requires preparation time, and, anyway, must actually be consumed to have any health effects. Reducing the very real obesity epidemic to an issue of available fresh foods ignores the complexity of the cultural issues overlaying food, eating, how children spend their free time, whether active or sedentary, and so on.

The dream of simplicity redux
The obesity epidemic is an example of another complex trait that many have wished to reduce to simple causes, be they single genes or single environmental contributors. So, while we readily criticize genetics and its many 'omic' children for its very expensive, low-payoff, often self-serving nature, it's only appropriate to extend the same critique to epidemiology that is perhaps even more costly and often--and often for the most important problems--delivers even less cogent payoff.  The epidemiology empire is every bit as large and self-serving as anything in genomics.

But then, with all of its resources, why are questions such as the origin and nature of obesity, and even to some extent its actual relationship to health outcomes, proving to be so challenging?   How can we not, after decades of 'practice' at large-scale megavariate studies, know the answers to even some of the most fundamental questions?  A large biostatistics industry has grown up around this work, with spillover into genomics.

Is chronic disease epidemiology yet another case of technology driving rather poorly framed questions?  Is it another clear indicator that our reductionistic, enumeration-driven approach to science somehow an inappropriate epistemology--that there are better ways to ask the question, or better ways to understand the probabilistic, complex world, than what we have developed so far?