Background: Utilisation of D-optimality together with clinical trial simulation are currently the primary methods used to inform clinical trial study designs and facilitate precise estimation of pharmacokinetic-pharmacodynamic (PK-PD) model parameters. A recent simulation study has extended the application of these methods to the qualitative identification of correct covariates . However, the authors did not investigate the effect of stratifying covariates to increase the ability to discriminate between competing covariates.
Aim: To examine the impact of study design on the probability of choosing the ‘true’ covariate model from two competing covariate models, using lean body weight (LBW)  and total body weight (WT) as an example.
Methods: Demographic datasets were generated using a multivariate lognormal covariate distribution with truncation at different WT limits under both a non-stratified and stratified design. For the non-stratified design, demographics were simulated by sampling from the distribution within specified WT ranges (e.g. 50-80kg). For the stratified design, demographics were also simulated from the same covariate distribution model, but WT was stratified into 3 discrete groups of equal range and size, i.e. one-third of subjects were in each stratum of 50-60, 60-70 and 70-80kg. This was repeated for 8 other WT ranges of 50-90, 50-100, 50-110, 50-120, 50-130, 50-140, 50-150 and 50-160kg.
PK data were simulated from a 1-compartment, first order input, first order elimination model with LBW as the covariate on clearance (CL), termed the ‘True Model’ . The ‘False Model’ had WT as the covariate on CL. Both models were fitted to the simulated data and the difference in objective function values computed for runs where both models converged successfully. Each design was evaluated under differing magnitudes of random effects, as well as under a D-optimal sparse sampling scheme. The performances of the 2 competing covariate models were assessed.
Results: When WT was simulated from the non-stratified design, the probability of LBW being preferred over WT increased as larger-sized subjects were recruited into the trial, from 74.2% at a WT range of 50-80kg to 81.7% at 50-110kg. Further extension of the WT range to an upper limit of 160kg did not produce any substantial improvement in the ability to select the ‘True Model’. In contrast, under the stratified design, the probability of LBW being preferred was always greater and increased steadily as larger-sized subjects were included, from 77.1% at 50-80kg to 92.7% at 50-160kg. In addition, the difference in probabilities between the 2 study designs started to diverge as the upper WT limit increased beyond 110kg. This trend was observed throughout all the simulation scenarios evaluated. The stratified design was also superior in identifying the inadequacies in model performance of the ‘False Model’.
Conclusion: We have shown under a simulation platform that designing clinical trials which stratify over WT can improve the probability of selecting the true covariate (LBW), especially when combined with a wide WT range (e.g. 50-160kg). These findings will guide the design of studies that aim to facilitate the identification of parameter-covariate relationships which are transportable across both the non-obese and obese populations, thus improving dosing guidelines for the obese.
- Ribbing et al. J Pharmacokinet Pharmacodyn 2008;35: 117-137
- Janmahasatian et al. Clin Pharmacokinet 2005;44:1051-1065
- Han et al. Clin Pharmacol Ther 2007;82:505-508