Estimating compliance to study medication from serum drug levels: application to an AIDS clinical trial of zidovudine.

This paper examines the use of serum drug levels to assess compliance to study medication in a clinical trial. We discuss problems of false-positivity, false-negativity, and bias that arise because of experimental errors in the drug assays, pharmacokinetic variations of the drug, and differential dosing levels. Basic concepts in probability are applied to derive a simple model that quantifies these problems. This model is used to obtain an estimate of compliance rate that corrects for these problems. However, derivation of this estimate requires additional information about false-positive and false-negative rates of the assay as well as some knowledge of the pharmacokinetic properties of the drug. We illustrate the evaluation of such a compliance estimate in the setting of an AIDS clinical trial of zidovudine (ZDV), in which some accessory data are available on the properties of ZDV serum assays and on the pharmacokinetic behavior of ZDV. We also describe a method that uses the accessory data to provide the additional information needed for computing the compliance estimate.