Confounding by Indication and the Risk of Hyperlipidemia in Observational Studies of Antipsychotics

To the Editor: Using an elegant case-control design, Mark Olfson, M.D., M.P.H., et al. have reported a somewhat weak but significant association between treatment with clozapine, risperidone, olanzapine, quetiapine, or ziprasidone, and the risk of developing hyperlipidemia (1) . The authors stated that the strength of the association for olanzapine was smaller than previously reported. We agree with them that a possible explanation for this discrepancy is that physicians might be aware of the metabolic risks of olanzapine and thus be reluctant to prescribe it to those patients at high risk of developing hyperlipidemia. However, we think that this effect, known as confounding by indication, also could have biased the results against risperidone and especially ziprasidone. Since these two drugs appear to be associated with a lower risk of hyperlipidemia (2) , it is possible that physicians might be prone to prescribe them to patients with a higher risk for hyperlipidemia.

Confounding by indication is probably the most important confounder when evaluating treatment effects in observational studies (3) . Although there are several methods for dealing with confounding by indication (4) , randomization is the best way to avoid it (5) . Randomized controlled trials appear to indicate that neither risperidone (6 , 7) nor ziprasidone (7 , 8) is associated with an increased risk of hyperlipidemia. While two of these studies were short-term (6 , 8) , the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study (7) was long-term and therefore more suitable for evaluating drug effects on plasma lipids. In the CATIE study, after adjusting for the duration of treatment, olanzapine was associated with greater increases in total cholesterol and triglycerides, while risperidone produced almost no changes, and ziprasidone was associated with improvement in these metabolic variables (7) .