The rich get richer, even in epidemiologic studies.

In a previous post, I indicated that I would eventually post epidemiologic analyses.

This is the first such post. Please refer back to the previous post, as needed, for a brief overview of epidemiologic methods and key concepts.

**********

My doctoral thesis in epidemiology focused on the health impacts of neighborhood walkability:

“Thus, more proximate destinations (e.g., shopping, work, recreation) and higher population density are thought to promote walking over driving, while walkable streets and proximate public transportation are thought to increase physical activity generally.” (pg. 1)

Specifically, I hypothesized that women who lived in less walkable neighborhoods at baseline would be more likely to be diagnosed with incident diabetes and/or have higher levels of depressive symptoms in the future than women living in more walkable neighborhoods. I used data from the Black Women’s Health Study (BWHS) as well as other data gathered by two members of my doctoral committee. Residential neighborhood walkability (most, 2nd most, 2nd least, least walkable) was assessed for BWHS residents of the New York City, Chicago and Los Angeles metropolitan areas in 1995. In the diabetes study (n=13,519), follow-up lasted 16 years, while in the depressive symptoms study (n=17,886), Center for Epidemiologic Studies-Depression (CES-D) scores were obtained in 1999 and 2005.

When I looked at the unadjusted association between neighborhood walkability and each outcome, however, I was surprised to find associations in the opposite direction than hypothesized. That is, women living in least walkable neighborhoods had 14% LOWER rates of incident diabetes over 16 years of follow-up than living in most walkable neighborhoods (which increased to 22% LOWER after adjusting for age[1]). The findings were similar (albeit even closer to the null) for depressive symptoms.

I was baffled…and more than a little nervous.

And then I started adjusting for other possible confounders (identified from the literature, confirmed using directed acyclic graphs).

The most obvious confounder was the socioeconomic status (SES)[2] of the woman’s neighborhood. First, it was strongly inversely associated with the exposure: the Pearson correlation[3] between continuous measures of neighborhood walkability and neighborhood SES was -0.31, and mean scaled neighborhood SES (mean=0, standard deviation [SD]=1) was 1.06 SD higher in least walkable neighborhoods than in most walkable neighborhoods.

In other words, less walkable neighborhoods tended to be much “better off” than more walkable neighborhoods.

Second, neighborhood SES was a risk factor for diabetes and for higher depressive symptoms in most walkable neighborhoods (my designated “unexposed” category). For example, every 1 SD decrease in neighborhood SES increased the risk of incident diabetes 29%. Results were similar for depressive symptoms[4]

Finally, neighborhood SES was not on the causal pathway between neighborhood walkability and either outcome.

And…huzzah!

After adjusting for SES (and city of residence), women living in any less walkable neighborhood had a 6% higher incidence of diabetes over 16 years of follow-up than women living a most walkable neighborhood.[5]

And…after adjusting for SES (and marital status), women living in a least walkable neighborhood had an 18% higher risk of having CES-D≥25[6] in 1999 and/or 2005 than women living in a most walkable neighborhood.[7]

In other words, adjusting for neighborhood SES turned living in a less walkable neighborhood from a protective factor to a risk factor.

Why would that be?

Because neighborhood walkability and neighborhood SES are inextricably tangled (at least within this primarily urban/suburban population of better-educated black women). The least walkable neighborhoods—hypothesized to yield more poor health outcomes—are much better off than the most walkable neighborhoods—hypothesized to yield better health outcomes.

Just bear with me while I put on my rarely-worn economic-populist hat to think through the policy implications of these findings.

The primary recommendation, of course, would be to find a way to increase the walkability of the less walkable neighborhoods, with the goal of improving health outcomes.

But those neighborhoods ALREADY have better health outcomes because of their higher SES.

And the most walkable neighborhoods?

Well, there is little you can do to improve their health outcomes via walkability, which is a pity because of the lower SES of these neighborhoods (and attendant less good health outcomes).

In other words: if you follow the policy recommendations suggested by my doctoral research, the rich (or, at least, the relatively better off) would get richer (improved neighborhood walkability), while the poor (relatively worse off) would get nothing.

Hmm…maybe it is a good thing I have not yet published my doctoral research.

Until next time…

[1] It has become routine practice (foolish, in my opinion) always to adjust for age, gender and race/ethnicity in any epidemiologic study, regardless of whether they meet the three criteria for confounding or not. Given that this study only assessed black women, the only variable of these three left to adjust was age.

[2] SES combines income and education level into a general “status” measure. For my doctorate, we concatenated “median household income; median household value, percentage of houses receiving interest, dividends or net rental income; percentage of persons aged≥25 with college degrees; persons of employed persons aged≥16 in white collar occupations; percentage of families not headed by a single female.” (pg. 8)

[3] A measure of the linear association between two variables, ranging from -1.00 (as one increases, the other always decreases) to 1.00 (as one increases, the other always increases), with 0.00 a purely random association.

[4] I can’t seem to locate the actual analyses at this time.

[5] IRR=1.06, 95% confidence interval (CI)= 0.90 to 1.24.

[6] The scale runs from 0 to 60, with a higher score indicating higher depressive symptom levels. CES-D≥16 is the typical cut-point for an indication of clinical depression (though the CES-D is not an official diagnostic tool). The association between neighborhood walkability and CES-D score above/below 16 was essentially null. CES-D≥25 has also been cited as an indicator of clinical depression, so I tested both in my doctoral research.

[7] RR=1.18, 95% CI=1.02-1.37.

2 thoughts on “The rich get richer, even in epidemiologic studies.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s