To the Editor: Dr. Sernyak et al. compared the presence of diabetes mellitus among schizophrenia patients within the Department of Veterans Affairs system who were treated with atypical neuroleptics (clozapine, risperidone, olanzapine, and quetiapine) and those who were treated with typical neuroleptics. Diagnostic data covered the period October 1998 through September 1999, and prescription data covered the 4-month period from June through September 1999. Associations between diabetes mellitus and neuroleptic treatment were made within this 4-month period, with index neuroleptics being essentially identified on the basis of the last neuroleptic prescription written during the period. While the study controlled for several confounding factors, limitations in design may have undermined the validity of its findings.

Patients were not screened for preexisting diabetes mellitus, although it appears that this could have been done with available data. Identification and exclusion of preexisting cases would have avoided the possibility of assigning to specific neuroleptics cases of diabetes mellitus due to previous causes, including prior neuroleptic use. There is reason to believe that the likelihood of preexistence was not the same for all neuroleptics. For example, in a study on this subject by me and my colleagues (unpublished study by F.D. Gianfrancesco et al.), we found that 76% of the diabetes mellitus observed during treatment with quetiapine already existed within the 4 months before treatment, versus 70% for risperidone, 67% for olanzapine, and 71% for typical neuroleptics. Also, in 49% of the quetiapine-treated patients, treatment with quetiapine was immediately preceded by treatment with another neuroleptic, often olanzapine, whereas treatments with risperidone, olanzapine, and typical neuroleptics were immediately preceded by treatment with other neuroleptics in only 16%, 18%, and 13% of patients, respectively.

Exclusive use of ICD-9-CM codes to identify diabetes mellitus cases could have further affected results, especially given that preexisting diabetes mellitus cases were not removed from study. The presence of diabetes mellitus in a patient is more definitely determined if there is treatment, as evidenced by prescriptions for antidiabetics or insulin. An ICD-9-CM code for diabetes mellitus does not necessarily mean that a patient has tested positive or tested sufficiently positive to warrant more than monitoring. In some instances, ICD-9-CM codes for diabetes mellitus may have reflected follow-up examinations for patients whose diabetes mellitus had already disappeared because of a change in neuroleptics, for example.

Last, the study did not control for differences in treatment duration among the neuroleptic categories, which may have further biased its findings. It is reasonable to assume that the likelihood of acquiring diabetes mellitus from a neuroleptic treatment increases with exposure to that neuroleptic. Treatments with index neuroleptics could have started at any time before or during the 4-month study period. Neuroleptics with longer treatment durations may have been disadvantaged by the study design.

In summary, a more careful study would have controlled for preexisting diabetes mellitus and for differences in neuroleptic exposure and would have used more definite indicators of this condition, such as prescriptions for antidiabetic medications and insulin.