Under-Compliance And Population Pharmacokinetics
Pascal Girard
Service De Pharmacologie Clinique, 162 Av. Lacassagne, 69003 Lyon, France
One of the key considerations in designing experiments for population PK-PD determinations, is knowing as precisely as possible the inputs, among which the time and amount of each of the doses taken by the patient are the most important, in order to fit a model to the outputs, concentrations and effects. This is particularly important in data arising from clinical practice, where oral drug intake is under the control of ambulatory patients. Until recently it was impossible to know dosing histories precisely. Most often, the history consisted of the time and amount of the last dose(s) taken, as reported by the patient, and the nominal prescribed dose regimen for all previous doses. The development of micro-electronic devices now provides access to much more reliable information: such devices can record the time and date of openings and closures of a drug container, which, along with the assumption that at each such opening one dose is removed and ingested, yields a complete dose history [1]. Very few attempts have been made to use this extensive compliance information for population pharmacokinetic studies[2]. A recent study of diltiazem showed that using full MEMS data records improved the parameter estimates[3]. This study was an experimental one, with many samples per patient. Use of full MEMS data was justified by the fact that samples were taken on many different days over two weeks. The ratio between number of pharmacokinetic samples and number of recorded doses was sizable[3]. The situation examined here is different [4].
We address the case of sparse PK data collected in ambulatory patients over a long period of time. In such a case the amount of information given by the electronic monitor becomes overwhelming for each patient, while the pharmacokinetic information per patient remains small as it is in most population pharmacokinetic studies. To investigate different strategies for using and summarizing this new abundant information, a Markov chain process model was developed, that simulates compliance data from real data from electronically monitored patients, and data simulations and analyses were conducted.
Results indicate that traditional population pharmacokinetic analysis methods that ignore actual dosing information tend to estimate biased clearance and volume and markedly over-estimate random interindividual variability. The best dosing information summarization strategies consist of initially estimating population pharmacokinetic parameters, using no covariates and only a limited number of dose records, the latter chosen based on an a priori estimate of the half-life of the drug in the compartment of interest; then resummarizing the dose records using either population or individual posterior Bayes parameter estimates from the first population fit; and finally reestimating the population parameters using the newly summarized dose records. Such summarization strategies yield the same parameter estimates as using full dosing information records while reducing by at least 75% the cpu time needed for a population pharmacokinetic analysis.
1. Urquhart. Role of patient compliance in clinical pharmacokinetics. Clin. Pharmacokinet. 27: 202 - 215 (1994).
2. Levy. A pharmacokinetic perspective on medicament noncompliance. Clin. Pharmacol. Ther. 54: 242 - 243 (1993).
3. Rubio, C. Cox, and M. Weintraub. Prediction of diltiazem plasma concentration curves from limited measurements using compliance data. Clin. Pharmacokinet. 22: 238 - 246 (1992).
4. Girard P, Sheiner LB, Kastrissios H, Blaschke T. Do we need full compliance data for population pharmacokinetic analysis? J Pharmacokin Biopharm.. 24: 265 - 282 (1996).