Can bioavailability of low-variance drugs be estimated with an unpaired, sparse sampling design?

Bioavailability (F) with nonintravenous administration is traditionally estimated by comparison of the area under the plasma concentration versus time curve (AUC) after drug administration by each of the nonintravenous and intravenous routes in the same individual. This paired approach may not always be possible. We simulated whether F and the absorption rate constant (ka) could be estimated accurately for a drug with low variance using different patients for nonintravenous and intravenous routes and whether sparse sampling permitted accurate estimates.

Using pharmacokinetic parameters for cisatracurium besylate (INN, cisatracurium besilate), we simulated data sets representing 20 administrations (10 intravenous and 10 nonintravenous) with either three (sparse) or 16 (extensive) samples per administration. Simulations were performed twice, with ka values of 0.1 (slow absorption) or 0.3 (rapid absorption) min-1. With use of NONMEM, we estimated F and ka for each data set using both two-stage and mixed-effects modeling approaches and paired and unpaired designs to determine the percentage of estimates that deviated > 25% from the simulated value.

Estimates of F with extensive data were satisfactory for all approaches. With sparse sampling, two-stage analysis of unpaired data were not possible, two-stage analysis of paired data yielded erroneous estimates, and mixed-effects modeling gave satisfactory estimates. Estimates of ka were sometimes erroneous with all approaches except for paired analysis of extensive data with slow absorption; sparse data and two-stage analysis increased the likelihood of errors compared with extensive data and mixed-effects modeling.

Mixed-effects modeling facilitates estimation of F and ka for low-variance drugs in situations in which traditional paired extensive data designs are not possible.