Abstract:
|
Longitudinal drug concentration or pharmacokinetic data are often analyzed using nonlinear mixed models, which provide interpretable parameter estimates and concentration predictions. This compartmental approach relies on a biological framework for the concentration trajectory and allows for differing assessment times between subjects and/or sparse data. The disadvantage of this method is reliance on special software, difficulty in developing and validating the model and optimization issues. A simpler traditional approach is a two-stage, non-compartmental model. An outcome summary variable, such as area under the curve, is calculated which is subsequently analyzed. This approach can utilize standard software, but assumes similar sampling times and may not be appropriate for sparse data. We consider a different analysis approach when non-sparse, data are collected, but there is between-subject variability in measurement times. A semi-parametric model utilizing natural b-splines, which may be interacted with drug category, provides a smooth concentration trajectory and allows prediction within the data range utilizing a linear mixed model and standard software.
|