LEVEILLE ET AL. RESPOND

Abstract

We thank Flegal and her colleagues for their suggestions. As noted in the articles they cite, a number of methods have been proposed for estimating relative risk. There is ongoing debate in the literature about the choice of models and methods to address potential bias in estimates of attributable fractions.1 Measures of attributable risk using the standard formulas must be considered within the limitations of the data and potential biases, as we discussed in our article. In our analyses, the adjusted odd ratios differed from the unadjusted odds ratios modestly in the earlier waves of the National Health and Nutrition Examination Survey (NHANES) and very little in the recent waves, indicating more confounding from demographic factors in the earlier waves. Using the unadjusted odds ratios to calculate the population attributable risk percent (PAR%) for the influence of obesity on arthritis prevalence yielded the following results: NHANES I, 7.4%; NHANES II, 10.7%, NHANES III, 19.9%, NHANES 1999–2002, 27.5%. Although these estimates are higher than those presented in Table 3 of our article, the observable trend across waves of the NHANES is the same and our conclusions are unchanged. In our initial analyses, we used the odds ratio approximation to the relative risk from our models in calculating the PAR%.2 On the basis of reviewers’ concerns about the potential “common outcome” bias, we applied the relative risk conversion,3 which yielded more conservative estimates of the PAR%. We chose to present these more conservative estimates. Regardless of the analytic approach, our conclusions remain the same. As is true of all observational research, our results should be considered within the full context of the existing research and confirmed or refuted by future studies.