To the Editor: We welcome the opportunity to respond to the questions raised by Drs. Geller, MacFadden, and Gianfrancesco about our recent article reporting an association between prescription of atypical neuroleptics and diagnosis of diabetes mellitus in a group of nearly 40,000 patients. Although data on matched comparison subjects were not obtained, a multivariate adjustment procedure was employed. This statistical procedure is more appropriate in situations in which multiple agents are compared.

While the number of patients did vary from group to group, we do not think that this invalidates our analysis. To take the example offered by Drs. Geller and MacFadden, we suggest that, given the small numbers of patients receiving quetiapine (probably because of the short amount of time that quetiapine had been available at the time of the study), our findings of an association of the diagnosis of diabetes mellitus with quetiapine appear all the more striking. Indeed, the larger problem in studies of this kind is that a large number of patients increases the probability of finding statistically significant findings that are not clinically significant.

We agree with Drs. Geller and MacFadden that observational nonexperimental designs do not support causal conclusions, although their strength is that they allow evaluation of the outcomes of large numbers of patients treated under real-world conditions. However, we point out that a recent study designed to investigate just this causal connection between atypical neuroleptics and diabetes mellitus (1) produced findings consistent with many of ours.

Dr. Gianfrancesco reiterates several limitations that we stated in our article. Still, although we know very little about Dr. Gianfrancesco’s data, they do appear to validate our supposition that some neuroleptics were much more likely to be switched to after the development of diabetes. However, the reliable determination of preexisting diabetes was beyond our capability at the time. Our group is working on another data set in an attempt to address this necessary compromise. The suggestion of controlling for treatment duration is also well taken. However, it seems that this would tend to bias the results in favor of the newest available antipsychotic—at the time, quetiapine—having the lowest odds ratio for the diagnosis of diabetes mellitus, which did not appear to occur.

While changing the case definition of diabetes would certainly change the number of patients so classified, how this would introduce biases against any particular antipsychotics is not clear, and, in fact, an analysis of prescriptions for hypoglycemic agents yielded substantially the same results as reported in the article.

The authors of both letters raise legitimate concerns about some of the limitations of our study that should, of course, be noted. However, we do not believe that these either represent a lack of care or substantively challenge the conclusions reached in our article.