Oscar Della Pasqua and Alan Maloney
Worldwide Clinical Pharmacology & Clinical Pharmacology Data Sciences GlaxoWellcome Research & Development, UK
Assessment of treatment effect and its correlation to drug concentration and dosing regimen are major endpoints in clinical drug developm ent. The use of non-linear models in population PK/PD modelling, such as the sigmoid Emax model, to characterise concentration-ef fect relationships is now widespread, but its application is often poor. It is conceivable that models and parameters show considerable unce rtainty. However, the need to appropriately correct for a pre-dose (baseline) value is frequently overlooked. Data transformation is applied , the assumption of homogeneity of variance across the effect range ignored, and indeed perhaps most importantly the underlying physiologica l changes in the parameter of interest are rarely considered or investigated. The consequences of these oversights can be varied in nature, from the researcher who simply gets a somewhat imperfect fit, but nevertheless arrives at the appropriate conclusion, to one who fails to id entify the true relationship, but instead obtains biased estimates.
This work looked to investigate three underlying physiological models: a baseline that is stable across time, a baseline which is linearl y increasing across time, and a baseline which is non-linear across time. In addition, the variability of the data around these functions w as considered. These two factors determined the necessity or otherwise of modelling a physiological model [f(E0)] in parallel with the pharmacodynamic model.
We show that a sound analysis of the baseline effect yields more than just accurate final parameter estimates. It also maximises the info rmation contained within the data, effectively enhancing the statistical power. Furthermore, it can be particularly relevant for simulation purposes when inter-trial variability and other aspects of trial performance are to be estimated.
In conclusion, failure to recognise fluctuations in baseline will lead to bias in parameter estimates, whilst not employing sound analysi s techniques will lead to model parameters that are estimated with poor precision. A major gain in accuracy will still be dependent on one’s ability to prospectively define a study design based on population PK/PD modelling requirements.
Reference: PAGE 9 (2000) Abstr 96 [www.page-meeting.org/?abstract=96]
Poster: poster