Article Text

Download PDFPDF

On the percent of excess risk explained
  1. Ashley H Schempf1,
  2. Jay S Kaufman2
  1. 1Office of Epidemiology, Policy and Evaluation, Maternal and Child Health Bureau, Rockville, Maryland, USA
  2. 2Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Quebec, Canada
  1. Correspondence to Dr Ashley H Schempf, Office of Data and Program Development, Maternal and Child Health Bureau, 5600 Fishers Lane Rm 18-46, Rockville, MD 20857, USA; aschempf{at}

Statistics from

Bleich et al1 provide a novel examination of black–white disparities in women's health by comparing only women who live in two Baltimore census tracts that are integrated and low income, thereby controlling for social context and contemporaneous income. Many national and regional datasets include black and white women who tend to live in very different neighbourhoods due to systematic residential segregation, and unless the neighbourhood environment is controlled with fixed effects,2 3 disparities may be overestimated. The authors report that the obesity disparity is no longer statistically significant within the integrated low-income communities (OR 1.25, 95% CI 0.90 to 1.75) in contrast to that found in a national sample (OR 1.99, 95% CI 1.71 to 2.32), suggesting an important role of social context in generating racial disparities.

Despite the clever design and impressive findings, the authors report that the adjusted OR in the integrated community is only 37% lower than the national OR. The formula they used is (OR1−OR2)/OR1*100. Without subtracting one from the denominator, however, a 100% reduction could only be achieved with an OR of 0 even though it is a value of 1 that indicates equality. The correct expression should be (OR1−OR2)/(OR1−1)*100.4 No similar subtraction is necessary in the numerator because the two terms cancel each other. When the correct formula is applied, the OR is reduced by 75%—a large portion of the overall effect, and a reduction that is much more consistent with the paper's conclusion. This algebraic error and consequent underestimation appear in other articles.5 6

The cited formula in Szklo and Nieto4 also uses the RR rather than OR as the measure of effect. For common outcomes (>10%), such as obesity, odds do not provide a good approximation of risk and should be avoided. Using the unadjusted percentages of obesity in table 2 of Bleich et al, the black–white OR in the national sample is 2.24 but the RR is 1.83. In the integrated communities, the OR is 1.21, whereas the RR is 1.13. There are several simple methods of obtaining adjusted risk differences and RR from logistic regression and other models.7–9

We applaud the design and overall conclusions of the study and wished only to mention some statistical issues that probably served to underestimate the contribution of social context. The real challenge will be to affect policies that reduce segregation and/or to modify the integral neighbourhood features that contribute to obesity in disadvantaged communities.



  • Linked articles 119578.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; not externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles

  • PostScript
    Sara N Bleich Roland J Thorpe Hamidah Sharif-Harris Ruth Fesahazion Thomas A LaVeist