Informative study designs to identify true parameter-covariate relationships.

This study explored how study design influences the probability of selecting a 'true' covariate from two competing covariate models. The probability of selecting the 'True Model' (lean body weight on clearance) over the 'False Model' (total body weight (WT) on clearance) was compared for designs where WT was either lognormally distributed (i.e. non-stratified), or stratified into 3 equal strata. The probability of selecting the 'True Model' increased as the WT inclusion criterion widened, and was always greater under the stratified design. Incorporating stratification into study designs, in combination with a wide covariate range, can aid identification of true parameter-covariate relationships. This has particular importance if the model is to be extrapolated beyond the studied population (e.g. dosing in obesity).