To the editors—

Abstract

I would like to challenge the commentary on the Wong et al, NEJM piece by Drs. Gruskin and Shulman. The summary of the paper is quite good but their interpretation of the methodologic strength of this piece of research is controversial, in my view. Acceptance of this study’s conclusions verbatim could seriously impact a clinician’s approach to the febrile child with clinical signs of bacterial enteritis. Please go to the following webpage to read my interpretation of this study’s methodology http:/www.pediatric-emergeny.com/wong.htm With specific regards to Wong et al’s use of logistic regression analysis in this study please refer to the following piece of research: Peduzzi P, et al. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996;49:1373–1379.Thank you for your letter about the lead abstract and accompanying commentary in our July issue concerning the antibiotic treatment of E coli O157:H7 and the risk of hemolytic-uremic syndrome (HUS). Your thoughtful Web site commentary stimulated discussion of these methodologic issues among our editorial team and consultants. The finer points of regression analysis are “out of my league.” Nonetheless, based on the responses of my methodologic consultants I will attempt a brief, partial response to some of the complex issues you have raised. We agree that knowing whether a drug has a particular effect is best evaluated in a high quality randomized controlled trial. However, when the only known effect of a drug on a disease is adverse, such a randomized controlled trial may not be ethical or even possible. So far as we are aware, there are no studies suggesting that antibiotics on presentation ever prevent HUS. If it could be cogently argued that antibiotic therapy may benefit a large number of children (those with other infections), such a study might be justified; however, in the US delaying antibiotic treatment for the most common causes of bloody diarrhea—until culture results are available, ie, E coli, campylobacter, salmonella, shigella—seems unlikely to offset the potential adverse effects of antibiotic treatment on E coli suggested by this study.We believe Wong et al attempted to deal head-on with the problem of “confounding by indication.” For example, the collection of information from clinicians occurred prior to the development of HUS and therefore should be free from recall bias that might occur if the data were collected after the onset of disease.The number of cases of HUS relative to the number of variables remains quite small, as you point out. I believe that Wong et al’s model used three variables (antibiotics, white count, and the day of culture) and that there were in fact ten children (not five, which was the number of exposed cases) who developed HUS.My consultants suggested that you might find of interest an article on how to assess how large an influence a hidden confounder might have: Lynn DY, Psaty BM, Kronmal RA. Assessing the sensitivity of regression results to unmeasured confounders in observational studies. Biometrics. 1998;54:948–963.In sum, on reflection my colleagues and I continue to believe that this study represents the best data to date; that antibiotics appear to strongly increase the risk for the development of HUS; and under these circumstances an RCT is not feasible.Thank you again for bringing your concerns to our attention. I learned a lot from the ensuing discussions with my colleagues. I hope you will continue to be a critical reader of AAP Grand Rounds.