An Examination of Heterogeneity in Treatment Response to Antipsychotic Medications

Objective: To examine treatment heterogeneity in the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE); a large RCT comparing effectiveness of several oral second-generation antipsychotics (SGAs) and a first-generation antipsychotic. We tested whether subgroup characteristics modified treatment effects on time to all-cause treatment discontinuation and change in neurocognition measured by the MATRICS Consensus Cognitive Battery (MCCB). Methods: Data is from the intent-to-treat cohort without tardive dyskinesia (N=1206). Participants aged 18-65 years, with a DSM-IV diagnosis of schizophrenia were randomly assigned to SGA olanzapine (OLAN), quetiapine (QUET), risperidone (RISP), or first-generation perphenazine (PERP) and followed for 18 months. For the primary analyses of overall time to all-cause treatment discontinuation and change in least squares mean (LSMean) MCCB composite Z-scores from baseline to month 2 between the four treatments (3 degrees of freedom, significance of p<0.05), we tested the modification of assigned treatment by subgroup characteristics (i.e., age, sex, race, substance use disorder, baseline symptom severity (PANSS) and baseline adherence) using Kaplan-Meier curve estimation and cox proportional hazards regression models, and analysis of covariance, adjusted for treatment site and exacerbation of schizophrenia symptoms in the previous three months. Second, we tested the modification of bivariate pairs of SGAs versus PERP by subgroup characteristics on both outcomes. Results: The primary analysis of overall time to all-cause treatment discontinuation found no interactions, but bivariate analyses found longer time to all-cause treatment discontinuation in females assigned PERP versus QUET (χ2=5.93, p=0.02), and no difference in males. For the primary analysis of overall change in MCCB Z-score between the treatments, persons with PANSS ≥75 had the greatest LSMean increases when assigned PERP and RISP (increase=0.33 and 0.25 respectively) and the smallest increases when assigned QUET and OLAN (increase=0.10 and 0.14 respectively). Persons with PANSS <75 at baseline had Z-score increases when assigned any treatment, but not significantly different between treatments (increases from 0.16-0.19). Bivariate analyses found persons with PANSS ≥75 had a greater increase when taking PERP versus QUET but the increase was not significantly different between QUET and PERP among persons with PANSS <75 at baseline. Conclusions: In some comparisons, sex modified assigned treatments’ effects on time to all-cause treatment discontinuation and symptom severity modified assigned treatments’ effects on change in neurocognition. Although these analyses do not reach statistical significance after multiple comparisons adjustment, they are of potential clinical importance; further study is warranted. Examination of heterogeneity of treatment effects has the potential to lead to more personalized treatment choices.