K. E. Fattinger and D. Verotta
In a population pharmacokinetic analysis we generally use parametric functions (e.g. exponentials) to describe drug disposition and input rate functions. By selecting one particular parametric function we make stringent assumptions about the shape of the disposition and input rate function. These assumptions might possibly be misleading. An alternative approach is to use more flexible non-parametric functions, e.g. longitudinal splines, to describe drug disposition and input rate functions. A longitudinal spline (D. Verotta, Longitudinal Splines, Technical Report 36, Department of Biostatistics, University of California, San Francisco, 1993) is composed of a template spline which is in common to all subjects and a distortion spline representing the subjects’ differences from the template spline. We consider different estimation methods including (1) parametric nonlinear mixed effect , (2) least squares and (3) two-stage.
The possibilities of nonparametric population deconvolution will be illustrated using data from a pharmacokinetic study in which cocaine was given intravenously, orally and nasally to 17 healthy volunteers. The cocaine serum concentration time course was thereafter observed over 6 hours. The input rate functions for oral cocaine administration are estimated to be nearly zero at times immediately after the dose, increase thereafter, peak at about 45 minutes, then decrease and become negligible at about 150 minutes. For some subjects the peak occurs delayed and some subjects show a small second peak at about 75 minutes. The nasal input rate functions start well above zero, peak at about 10 minutes, then decrease and show a second peak at about 45 minutes for most subjects. The second peak of the nasal input rate functions resembles the peak of the oral input rate function and occurs at the same time. Therefore the hypothesis is suggested that after nasal administration a part of the cocaine administered is swallowed and thereafter absorbed gastrointestinally. We therefore reanalyze the data from the nasal occasion under the assumption that the input rate function for cocaine swallowed after nasal administration is the same as after oral administration and estimate that 7 % of dose is absorbed in the nose and that 27 % of the total drug exposure is due to nasally absorbed cocaine.
The different estimation methods are also compared using simulated data. In a first set of simulations the data were obtained using the same models as in the analysis, i.e. longitudinal splines were used as drug input rate and disposition functions. In this case all estimation methods performed similarly well. In a second set of simulations parametric functions were used as drug input rate and disposition functions. In this case method 2 and 3 again performed well, whereas the results obtained with methods 1 were partly biased. The reason for this bias is probably the misspecification of the variance model.
In summary, nonparametric subject-specific deconvolution using longitudinal splines is a tool which allows to estimate template and subjects input rate (and disposition) functions without stringent assumptions about function shape. This can be especially helpful in case of data, where frequent samples were taken during drug absorption and where standard parametric models might fail to adequately describe the data.
Reference: PAGE 6 () Abstr 603 [www.page-meeting.org/?abstract=603]
Poster: oral presentation