Abstract
Background: The CATIE trial compared several antipsychotics and concluded that all antipsychotics compared in Phase 1 are equal in their efficacy. It is unclear however whether there are separable patterns of antipsychotic response that may suggest a drug-by-patient effect. Latent variable modeling allows researchers to test for individual differences in developmental trajectories that may underlie psychiatric data. The goal of the study is to use latent variable modeling to investigate differential patterns of antipsychotic response in the CATIE trial.
Methods: Participants were 1460 people with schizophrenia drawn from the CATIE trial. Participants completed the Positive and Negative Syndrome Scale (PANSS), neurocognitive measures, and the Quality of Life Interview at baseline and several follow-up periods. First, we submitted participant scores to Latent Growth Curve Analysis (LGCA) to examine changes in outcomes during the study from baseline to follow-up periods. Next, we used Latent Class Growth Modeling (LCGM) and Growth Mixture Modeling (GMM) to investigate the presence of subpopulations that may differ in response trajectory. We tested models with predictors including demographics, illness duration, CGI, and medication type. Fit indices including likelihood-based statistics and information criteria evaluated the fit of models to data.
Results: A linear function best explained change in PANSS, cognition, and function scores when fit indices are examined. Age, race, years of education, and years of present ill/treatment were significant predictors of change in symptoms, whereas education level and ethnicity were significant predictors of change in cognition and functional status during antipsychotic treatment. Both LCGM and GMM supported a 3-class model of change in psychotic symptoms when fit indices are examined. The information criteria indices were lower for the 3-class GMM than the 3-class LCGM, suggesting that a 3-subgroup pattern of change in which there exists variability within subgroups best explains the pattern of antipsychotic response in PANSS scores. For cognition and function, the LCGM tended to favor the 4-class model when the information criteria and likelihood-based indices are examined. In contrast, the GMM favored a 2-class model. The 2-class GMM better fits the cognition and psychosocial function data compared to the 4-class LCGM. A group with better antipsychotic response was apparent in all outcomes.
Conclusion: Latent trajectories are apparent in the pattern of response to antipsychotic medications in Phase 1 of the CATIE trial when PANSS, cognition, and psychosocial function scores are considered. Age, race, education, and illness duration predicted treatment response within unveiled subgroups. Current dogma about CATIE results are challenged. Latent variable models are adaptable to studying repeated measures data when collected longitudinally at known time points.