2003
   Verona, Italy

Estimating population pharmacokinetic parameters when dose and dose-time are not known accurately

Duffull SB (1), Isbister GK (2), Dawson AH (2), Hackett LP (3), Whyte IM (2).

(1) School of Pharmacy, University of Queensland, Australia. (2) Discipline of Clinical Pharmacology, University of Newcastle, Australia. (3) Clinical Pharmacology and Toxicology, Western Australian Centre for Pathology and Medical Research, Australia.

Introduction: Understanding the PKPD relationship of therapeutic agents when used in overdose is paramount for the development of treatment guidelines. In deliberate self-poisonings involving therapeutic agents it is often difficult to gain accurate information on the dose taken and the exact timing of the dose. The attending clinician may be able to ascertain some of this information from the patient or relatives at the time of admission and grade the veracity of this information.

Aim: To develop a population pharmacokinetic model for citalopram when taken as a deliberate overdose and when dose and dose-time may not be known accurately.

Methods: Eighty nine plasma citalopram concentrations were available from 29 patients. The data were modelled by means of a Monte Carlo Markov chain method using WinBUGS (ver. 1.3). Model building was based on assessment of the posterior distribution of the log-likelihood. A one-compartment model with first-order input and first-order elimination provided a good description of the data. The prior distribution for CL and V were set to be minimally informative as multivariate normal with mean values of 30 L/h and 900 L (typical values from therapeutic use studies), respectively. The prior for Ka was set to 0.5 hours (with high precision [i.e. low variance]) based on literature reports for therapeutic use, since there were few samples before 4 to 6 hours post-dose. Between subject variability was assumed to be log normal with low information priors for all parameter values. In addition, the fractional dose taken and lag-time were also estimated. Both were assumed to be normally distributed with a mean of 1 and 0, respectively. The precision was indexed to the veracity of the knowledge of dose and dose-time. Veracity was reported on a 4 point ordinal scale.

Results: The posterior mean of CL was 29 L/h (between-subject variability = 41%), and for V was 760 L (between-subject variability = 51%). The estimated actual dose ingested ranged from 0.29 to 1.40 times the nominal dose recorded. The estimated actual time of dose ingestion varied from 1.6 hours before to a few minutes after the nominal dose-time. Inclusion of informative priors on dose and dose-time improved overall model fit and decreased between subject variability in CL by 43% and in V by 1%.

Conclusion: The use of informative priors where the informativeness was indexed to clinical findings, within the framework of a fully Bayesian analysis seemed to improve the predictive ability of a model developed from pharmacokinetic data arising from self-poisonings with citalopram.