To the Editor: We welcome the chance to respond to the questions raised by Drs. Ertugrul and Meltzer because it provides us with an additional opportunity to explain why we feel our assessment that clozapine is not associated with a lower than usual risk of suicide is based on a more sound evaluation method than in previous studies.
Dr. Ertugrul’s letter confirms our primary claim that previous studies of this issue have employed mostly weak designs. The most informative control groups had characteristics demonstrably similar to those of the specified treatment groups in every respect except treatment assignment. As summarized by Dr. Ertugrul, previous studies have employed historical control groups (Meltzer and Okayli, 1995), control groups about which so little is known that comparability could not be assessed (Reid et al., 1998), and control groups that have included no patients who were not previously exposed to clozapine (Walker et al., 1997). Thus, we feel that Dr. Ertugrul’s comments provide substantial support for our evaluation of the literature.
Dr. Meltzer asserts that we did not match for the “four most important” variables for the matching for suicide risk in patients with schizophrenia. While we agree that this is a limitation, as discussed at length in our article, it applies to all of the studies on this subject. Dr. Meltzer draws special attention to the fact that when the patients never exposed to clozapine were compared with the patients who had been receiving clozapine for more than 12 months (groups that exclude many who showed no improvement while taking clozapine), there was a trend (at the p=0.10 level) for those receiving clozapine to have a lower rate of suicide. We do not think this was a relevant comparison for evaluating the effect of clozapine on suicidal risk.
The comparison that best reflects clinical practice and is most informative compares all patients who received clozapine with a carefully matched group of patients who were never exposed to clozapine, yielding a nonsignificant (p=0.76) difference in the rates of suicide. We presented the comparison between those who completed 1 year of treatment (excluding a large number of treatment nonresponders) and the entire control group (from which nonresponders had not been excluded) to illustrate the bias inherent in previous studies comparing patients who dropped out of treatment and those who continued. The fact that this highly biased comparison still failed to yield a significantly lower rate of suicide in our study adds further support to the nonsignificant findings of our main analysis.
We appreciate the thoughtful comments of our colleagues, which allowed us to emphasize and further clarify some of the more informative points of our analysis.
