We thank Dr. Melzer and co-authors for their views. We recognize that our paper was just recently published and there was not much time for review, so we thought we would take this opportunity to provide a few clarifications to issues raised by Dr. Melzer et al. First, readers of our paper should be able to reproduce our results as we provide clear descriptions of the populations of interest, clinical definitions, NHANES variable names and statistical approach. We were surprised to see our paper described by Dr. Melzer et al. as “unfocussed” as we generally followed the approach used in their papers, examining the same outcomes. We did, however, believe that additional confounders not considered previously were important and we incorporated these into our analyses. Incorporating additional confounders results in additional missing values, a problem that might be addressed using imputation approaches. Removing participants with missing data or imputing values is a methodologic choice – both have been used by statisticians and readers are welcome to redo the analyses using alternative approaches. Second, in terms of participants included in our analysis, we sought to avoid any arbitrary a priori exclusion criteria. Of course, those familiar with NHANES will recognize that NHANES does not collect data on heart disease for those younger than 20 years of age and so these participants would not be included in those analyses (as was stated in the paper). Third, while it may be worth having a broad scientific discussion on the appropriateness of excluding “outliers,” at the very least, should participants be excluded based on their exposures, a sensitivity analyses that describes the effect of that removal ought to be conducted. This was not done in the papers published by Dr. Melzer et al. Finally, we can certainly argue about the merits of different methodological decisions, but it is hard to deny that our relatively minor decisions (such as our clinical definition of diabetes and inclusion of participants with higher levels of BPA, decisions which we believe are warranted and which were made a priori) change the results and conclusions as compared to the results and conclusions of Dr. Melzer and colleagues..
The key conclusion from our research is that having the correct exposure metric is essential to developing hypotheses – and certainly testing hypotheses – related to chemical exposure and health outcomes. While NHANES provides a wealth of important data that has answered an abundance of public health questions, NHANES is not an appropriate survey for drawing causal inferences between exposure to short-lived chemicals and chronic disease.