In Reply: Sernyak and associates suggest that the difference in medication compliance is a form of bias that accounted for the difference in suicide rates between a subgroup of patients treated with clozapine and a much larger comparison group of patients with schizophrenia. We believe that compliance is instead a causal factor. Compliance and treatment effects are related both to each other and to the particular drug prescribed.
This relationship was demonstrated in a recent randomized, controlled trial of clozapine versus haloperidol in which treatment-compliant clozapine patients had statistically significant reductions in symptoms and improvements in quality of life compared with treatment-compliant haloperidol patients. They also had compliance rates that favored clozapine by a ratio of 2 to 1 (1). Clozapine's positive effects are thus probably related to both superior efficacy and superior compliance rates.
The presence in the comparison group of two deceased patients who had taken clozapine months or years before their deaths may suggest several things, none of which detract from our suggestion of clozapine's association with suicide prevention. It may suggest that compliance is a variable not equally represented in the two groups, or that responsiveness to clozapine is important to protection against suicide. The overall finding of fewer suicides is notable regardless of the relative importance, if any, of such factors as medication effect, individual response to medication, increased compliance, differential monitoring, and special treatment or management as a result of being on a special medication. The fact that clozapine patients were among the most seriously and intractibly ill patients in our treatment system is important to the clinical interpretation of our findings.
In their analysis of our data, Sernyak and associates claim that the two groups are statistically equivalent because the 95 percent confidence intervals around the suicide count yield overlapping ranges of suicide rates. However, application of the Poisson probability formula found in Hays (2), referring to suicide rates in particular, shows that the probability of observing one or fewer suicides in the clozapine group is .04, assuming the same base rate in the comparison group.
As we stated in our paper, it is very difficult to extrapolate from a single suicide in six years. We find it hard to envision, in the real world, the large potential for suicide by patients on clozapine—up to 92 suicides per 100,000 patients per year—suggested by Sernyak's calculations. We believe we would have seen evidence of such numbers, in our system or reports from others, if this upper limit were reasonable.
Finally, Sernyak and associates mention seeing no evidence of differences in suicide rates in their mortality study comparing episodes of care for patients started on clozapine with episodes for a large group of similar patients. However, they point out that they are not looking at suicide data per se. Despite increased suicide risk in schizophrenia, suicide nevertheless represents a relatively small portion of all causes of mortality among persons with this devastating disorder. Important differential suicide outcomes may not be detectable in a context of a review of mortality from all causes. It is reasonable to consider whether, as they adjust for clinical status in their mortality study, they might eliminate from their analysis the very factors that appear associated with favorable suicide-risk outcomes for clozapine, such as superior efficacy and compliance effects.