Inherent correlation between dose and clearance in therapeutic drug monitoring settings: possible misinterpretation in population pharmacokinetic analyses.

During the course of therapeutic drug monitoring (TDM), doses are adjusted to attain a target concentration range and a correlation between clearance (CL) and dose is introduced. In population pharmacokinetic analyses of such TDM data, CL has frequently been modeled as a function of dose. This paper demonstrates by simulation methodology that the TDM process does indeed introduce a correlation between dose and CL which can be interpreted as a nonlinearity. Using literature values of carbamazepine pharmacokinetics, three steady-state concentrations were simulated following a standard 1000 mg total daily dose (TDD) regimen in 100 in silico subjects. A set of clinical rules was established to adjust the TDD based on these three concentrations, as might be done in the clinical setting. Another set of concentrations using these TDM-derived TDDs for each subject (600-1600 mg) was simulated. A standard population pharmacokinetic analysis of the post-TDM data was conducted using NONMEM. This process was replicated 100 times to estimate the type II error rate. When CL was modeled without TDD, plots of WRES versus PRED demonstrated a clear pattern, as did the delta plots of CL (CL minus TVCL) versus TDD, suggesting the covariate TDD should be incorporated into the model. After TDD was included in the model for CL, the objective function value decreased by an average of 75.7 (p < 0.001). In addition, the inter-individual variability in CL expressed as a coefficient of variation decreased by an average of 9.9% and the diagnostic plots improved. Although CL was simulated to be independent of TDD, it was identified as an important covariate using standard approaches in a simulated TDM setting in 100% of the replicated simulation studies. The TDM process introduces a correlation between CL and TDD that can be misinterpreted as nonlinearity in the system.