K.C. Carlsson (1), N.O. Hoem (2), T. Glauser (3), A. A. Vinks (4).
(1)School of Pharmacy, Faculty of Mathematics and Natural Sciences
Objectives: Population pharmacokinetics (PPK) models for implementation in Bayesian adaptive control strategies have been developed for several drugs and patient populations. A prerequisite for a good structural PK model is informative data. Therapeutic drug monitoring (TDM) data are available for routinely monitored drugs. This analysis explores the utility of sparse carbamazepine TDM data for the development of clinically useful population PK models.
Methods: Patient and TDM data were identified from our neurology outpatient clinic database. For carbamazepine 556 entries with one or more serum concentration measurements were found. Each entry contained information on patient demographics, date of birth, gender, ethnicity, drug used and concomitant meds, dosing regimen, time and amount of last dose and a laboratory result from a draw made at a stated time. Exclusion criteria were; combination therapy with other anticonvulsants, concomitant medication known to interact with carbamazepine, carbamazepine therapy of shorter duration than one month, suspected non-compliance and missing or conflicting information in the database vs. medical chart. Data were analyzed with an iterative-two stage Bayesian (IT2B) and a nonparametric EM algorithm (NPEM).
Results: 57 carbamazepine mono therapy patient data sets were identified from the database and used in the analysis. A one-compartment model with first-order absorption based on mean population data gave poor predictions. The individual Bayesian posterior models gave better predictions for all subjects, but showed large inter patient variability in the estimated parameters. To explore the potential benefits of splitting the data set into groups depending on drug formulation used several other models were tested.
Conclusion: Routinely collected TDM data can be used for PPK modelling. However, it will be necessary to combine such data with additional richer data. This will allow for the development of more informative and clinically useful PPK models that can then be used as part of Bayesian individualization strategies.
Reference: PAGE 13 () Abstr 510 [www.page-meeting.org/?abstract=510]
Poster: poster