Efficient identification of the optimal dosing regimen with a nonparametric method
Thomas Bouillon
Dept. of Pharmaceutical Sciences, KU Leuven, Belgium
Objectives: To propose a method for efficient identification of a covariate adjusted dosing regimen in "open loop" situations using a nonparametric algorithm.
Methods: Data from 70 adult patients (CYP3A5.0: n=55; CYP3A5.1: n=15) receiving tacrolimus at steady state was used for demonstration. All patients underwent therapeutic drug monitoring (TDM). 1. Standard PK parameter estimation was done with Pmetrics (NPAG)[1]. 2. Probability weighted support points from this analysis were implemented as covariates in a new dataset, “actual” covariates (age, weight, CYP3A5, day after transplantation, hct), a unit dose and a PK target (Cmin,ss 8mcg/L (median in the study population)) were added for every patient. 3. Using bioavailability as "surrogate parameter", the best dose for target attainment (TA) was estimated. 4. Bayesian estimates (EBE) of the individual doses were used for covariate identification. 5. Selected covariates were implemented into the model and the difference in -2log likelihood evaluated. 6. Candidate dosing regimens and their TAs were displayed as empirical cumulative distribution functions (ecdf’s).
Results: A 2 compartment model with first order input adequately described the PK of tacrolimus. Although this type of model can be solved analytically for the optimal dose, the "dose estimation" approach was favored in order to demonstrate applicability to more complex systems. As expected, the ecdf of the EBEs of dose was virtually identical to that of the support points, justifying the use of nonparametric EBEs vs. covariate plots. CYP3A5 expression was the only covariate identified and decreased -2LL by 50. Weight based dosing was deliberately added to demonstrate the (lack of) effect of a naively complex dosing regimen. Rounded Dose(s) were 2-13 mg/d (Q5-Q95) for TDM based, 5 and 9 mg/d (95% CI: 4.3-6.1; 8.0-11.2 mg/d) for CYP3A5 based, 3.5-11 mg/d (Q5-Q95) for CYP3A5+weight based and 5 mg/d (95% CI: 4.7-6.8 mg/d) for unique dose. TDM based dosing achieved Cmin,ss concentrations between 5-11 mcg/L in 95% of the investigated patient population. The respective values for CYP3A5 based, CYP3A5+weight based and unique dose are 65%, 65% and 50%. CYP3A5 is relevant for TA of tacrolimus, but not sufficient to replace TDM adjusted dosing.
Conclusions: The proposed method correctly represents dose as a parameter to be optimized. Covariates are implemented directly on the relevant dose and their effect(s) regarding target attainment assessed.
References:
[1] Neely MN, van Guilder MG, Yamada WM, Schumitzky A, Jelliffe RW. Accurate detection of outliers and subpopulations with Pmetrics, a nonparametric and parametric pharmacometric modeling and simulation package for R. Therapeutic Drug Monitoring. 2012; 34(4): 467-476.
