Discontinuing and switching antipsychotic medications: understanding the CATIE schizophrenia trial.

A new standard in effectiveness research on schizophrenia medications has been established by the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study. The study used an innovative approach to determining relative effectiveness of medications by using time until medication discontinuation or switch as the primary outcome criterion. The study is perhaps best known for the overall high proportion of subjects (74%) meeting all-cause discontinuation (ACD) criteria within the 18-month time frame of being assigned to their phase 1 antipsychotic. However, some of the drawbacks of the ACD approach are not well understood, in part because of unfamiliarity with the way ACD was assessed and problems with the use of hierarchical criteria to establish the primary reason for medication discontinuation. Using the time until ACD as an endpoint cannot by itself capture the complexity of the trajectory of a patient's response to a new medication. In particular, it is quite plausible that switching medications upon entering CATIE phase 1 would reduce some symptoms, which then would lead to a greater desire to make another medication switch. Using ACD criteria, a comparison with CATIE subjects who coincidentally remained on treatment with their preswitch medication would make it seem that switching was detrimental when in fact it could have been helpful. Another major limitation of the ACD was omitting the recording of the reason for stopping study medication whenever the ACD was considered to be a "patient-decision" discontinuation. This means that patient-initiated discontinuations could never be classified as a tolerability discontinuation. Since the ACD was done by the patients' clinicians, this approach may have underestimated the proportion of side effect discontinuations whenever the patient disagreed with his or her clinician. Moreover, retaining the "patient-decision" discontinuation subgroup in the attributable risk estimates of tolerability discontinuations further minimizes the attributable risk estimate of the role of side effects relative to other causes of discontinuation. For these assumptions to be valid would require the very optimistic assumption that CATIE clinicians never underestimated tolerability concerns in their patients. Otherwise, this mutually exclusive approach will lead to significant underestimation of the proportion of CATIE discontinuations caused by tolerability problems. It can be argued that excluding the "patient decision" subgroup from the attributable risk estimate of role of tolerability in medication discontinuation is a better approach to mitigate against these biases. A reanalysis using an adjusted N of 1061 evaluable subjects changes the attributable portion of tolerability discontinuations from 14.9% to 38.5%. Regarding specific side effect-medication pairs of interest, the attributable risk of extrapyramidal symptoms as a reason for discontinuing perphenazine increases from 8% to 21%, and weight-related discontinuations from olanzapine from 9% to 28%. Therefore, the clinical implications of the CATIE phase 1 findings may depend, in part, on the underlying assumptions of the ACD outcome measure.