The effect of study design on findings regarding diabetes risk associated with antipsychotics was studied.
This study was a retrospective analysis of data from more than 100,000 Medicaid patients. Diabetes odds ratios (ORs) for patients treated with clozapine, olanzapine, quetiapine, risperidone, ziprasidone, or conventional antipsychotics versus untreated patients were estimated with and without the following design enhancements: screening for preexisting diabetes, selecting for antipsychotic monotherapy, and identifying diabetes with prescription claims only. Logistic regression controlled for patient sex, race and ethnicity, type of psychosis, length of observation and treatment, antipsychotic dosage, pre-existing excess weight or dyslipidemia, and use of other drugs with potential diabetogenic effects.
Under the weakest study design (none of the above enhancements), all antipsychotics were associated with significantly higher odds of diabetes relative to no treatment (p < 0.05). Estimated ORs were as follows: clozapine, 1.468; olanzapine, 1.108; quetiapine, 1.270; ziprasidone, 1.226; risperidone, 1.232; and conventional antipsychotics, 1.159. Under the strongest design (all of the above enhancements), ORs relative to no treatment were significant for clozapine (1.484) and olanzapine (1.149) and nonsignificant for quetiapine (0.998), risperidone (1.124), ziprasidone (0.717), and conventional antipsychotics (1.025). The data also strongly suggest selection bias by clinicians (i.e., selecting antipsychotics based on preexisting diabetes or risk factors for diabetes), disfavoring risperidone and favoring olanzapine. Although the evidence is weaker, quetiapine may also have been affected by unfavorable selection bias.
In large database studies, estimated risks of diabetes among patients treated with antipsychotics appeared to be influenced by study design. When a more rigorous design was used, only clozapine and olanzapine were associated with diabetes risk significantly greater than that in untreated patients.