Validation of Population Pharmacokinetic Models for Clozapine Dosage Prediction

Massimo Berneri, Uma Jha, Seán O'Halloran, Sam Salman, Shanek Wickramasinghe, Kevin Kendrick, Jessica Nguyen, David A. Joyce

Research output: Contribution to journalArticlepeer-review

2 Citations (Scopus)

Abstract

Background:Clozapine is unique in its capacity to ameliorate severe schizophrenia but at high risk of toxicity. A relationship between blood concentration and clinical response and evidence for concentration-response relationships to some adverse effects justify therapeutic drug monitoring of clozapine. However, the relationship between drug dose and blood concentration is quite variable. This variability is, in part, due to inductive and inhibitory interactions varying the activity of cytochrome P450 1A2 (CYP1A2), the principal pathway for clozapine elimination. Several population pharmacokinetic models have been presented to facilitate dose selection and to identify poor adherence in individual patients. These models have faced little testing for validity in independent populations or even for persisting validity in the source population.Methods:Therefore, we collected a large population of clozapine-treated patients (127 patients, 1048 timed plasma concentrations) in whom dosing and covariate information could be obtained with high certainty. A population pharmacokinetic model was constructed with data collected in the first 6 weeks from study enrolment (448 plasma concentrations), to estimate covariate influences and to allow alignment with previously published models. The model was tested for its performance in predicting the concentrations observed at later time intervals up to 5 years. The predictive performances of 6 published clozapine population models were then assessed in the entire population.Results:The population pharmacokinetic model based on the first 6 weeks identified significant influences of sex, smoking, and cotreatment with fluvoxamine on clozapine clearance. The model built from the first 6 weeks had acceptable predictive performance in the same patient population up to the first 26 weeks using individual parameters, with a median predictive error (PE) of -0.1% to -15.9% and median absolute PE of 22.9%-27.1%. Predictive performance fell progressively with time after 26 weeks. Bayesian addition of plasma concentration observations within each prediction period improved individual predictions. Three additional observations extended acceptable predictive performance into the second 6 months of therapy. When the published models were tested with the entire data set, median PE ranged from -8% to +35% with a median absolute PE of >39% in all models. Thus, none of the tested models was successful in external validation. Bayesian addition of single patient observations improved individual predictions from all models but still without achieving acceptable performances.Conclusions:We conclude that the relationship between covariates and blood clozapine concentrations differs between populations and that relationships are not stable over time within a population. Current population models for clozapine are not capturing influential covariates.

Original languageEnglish
Pages (from-to)217-226
Number of pages10
JournalTherapeutic Drug Monitoring
Volume46
Issue number2
DOIs
Publication statusPublished - Apr 2024

Fingerprint

Dive into the research topics of 'Validation of Population Pharmacokinetic Models for Clozapine Dosage Prediction'. Together they form a unique fingerprint.

Cite this