Application of population pharmacokinetic/pharmacodynamic methodologies to current clinical issues facing drug development

Pharmaceutical drug development is a lengthy process and has been associated with rising development costs. Modelling and simulation (M&S) approaches employed across preclinical development phases to phase IV provide an evolving quantitative framework and thus may speed up this process. In this thesis, some population PK/PD M&S techniques to aid in dose optimisation and trial simulations applied to some current clinical issues in drug development are explored and presented. In the area of pharmacogenetics and personalised medicine, warfarin was used as an example, where ethnic differences were initially observed in maintenance dose requirements. It was later found that this might be attributed to the differences in VKORC1 haplotype frequency distribution. Empirical PK/PD modelling showed that whilst CYP2C9 genotype status was predictive of S-warfarin clearance, VKORC1 haplotype status was more predictive of warfarin response. The models presented illustrate a starting point to characterise the pharmacogenetics effects on drug PK and PD. Together with the inclusion of more patient relevant characteristics, these may be used to assess the need for dose optimisation and personalised medicine. The integration of PK and PD data in a drug-disease model was illustrated through a meta-analysis of 3 phase III trials investigating the use of duloxetine in the treatment of diabetic peripheral neuropathy. Two population PK/PD models were developed: a continuous descriptive model for pain score reduction on the 11-point Likert scale and a proportional odds model describing the probability of achieving pain relief. The results from these models were similar to those presented in the literature for each trial. In addition, the agreement and definition of clinically meaningful pain relief between several efficacy scales against the patient global improvement (PGI) scale was explored. The results showed that whilst the mean difference between the scales was close to zero, precision was poor. By defining clinically meaningful pain relief as at least ‘much better’ on the PGI scale, the equivalent on the Likert scale was a minimum of 2.90 points, in agreement with previous literature findings. These analyses identifying correlations between scales provide opportunities for clinical validation and enable more accurate extrapolation of clinical data across studies. Missing data is present in most clinical trials. Since PK/PD models are frequently developed using non-missing longitudinal data, the influence of missing data on accuracy of parameter estimations should be investigated. This can be achieved through comparing different missing data imputation methods on the estimated parameters. The analysis results showed that regardless of imputation method, PK/PD parameter estimates were largely similar to those estimated with non-missing data, suggesting that missing data had little impact at an overall completion rate of approximately 75%. Selection models incorporating dropout were also developed to enable more accurate trial simulations. The results showed that the dropout mechanism was likely to be missing at random, and non-inclusion of dropout may lead to underestimation of variability and therefore the estimation of effect size. This thesis presented only a small subset of areas in which PK/PD M&S can play a role in drug development by focusing on some of the issues and challenges that currently face this process. Since M&S methodologies bridge across various disciplines, it should be well-placed to help handle any future issues that might face the development of pharmaceuticals.