S109. ADMIXTURE ANALYSIS TO MODEL CLOZAPINE PHARMACOKINETICS: COMPARISON OF TWO HOSPITAL-BASED SAMPLES

Abstract Background Clozapine and norclozapine plasma levels are routinely used for monitoring clozapine compliance, side-effects, and treatment response. As well as treatment resistant schizophrenia, clozapine is also prescribed in suicidal patients with schizophrenia to reduce suicide ideation. However, it is unknown what the ideal clozapine levels are for targeting specific treatment response, i.e., positive, negative, cognitive symptoms, as well as suicidal ideation. In this study, we propose to use the admixture analysis to generate a model to define groups of clozapine plasma levels comparing two hospital-based samples. Methods For the first sample, we have recruited 91 schizophrenia participants at CAMH treated with clozapine and monitored for clozapine plasma levels at baseline with two optional follow up visits at 2–4 months and 4–8 months. Participants were included in the study if participants had a diagnosis of schizophrenia or schizoaffective disorder confirmed by medical charts, prescribed clozapine monotherapy for 3 months, and kept on a stable dose for at least 1 week. Participants were excluded if currently on a depot antipsychotic or receiving electroconvulsive therapy in the past 3 months. Scales utilized for the admixture analysis included the Columbia-Suicide Severity Rating Scale (CSSRS), Brief Psychiatric Rating Scale (BPRS) factor scores (i.e., reality distortion, disorganization, negative symptoms, and anxiety/depression) and a cognitive assessment in the form of the Brief Neurocognitive Assessment (BNA). All were performed at each visit. For the second sample, we included 83 subjects whose clozapine levels were extracted from electronic medical records (EMR). The clozapine plasma levels were analyzed using the admixture analysis to determine participants who were on high and low levels of clozapine. Gender, age and ethnicity were also included in the model, to assess the influence of different demographics across the two different samples. The admixture analysis through the MCLUST R package determined whether subjects fell into two (low versus high) or three (low versus intermediate versus high) normal distributions with regard to clozapine levels, and the appropriate model with the lowest Bayesian information criterion was selected. Results Based on the lowest Bayesian Information Criterion, for the first sample, the admixture analysis generated one model of ideal clozapine level distributions with two groups and optimal cut-off