Non-parametric methods in population pharmacokinetics and/or pharmacodynamics.

Summary The thesis studies non-parametric methods (NP) for estimating the distribution of random effects in a non-linear mixed effects model. The objective is to evaluate the interest of these methods for population Pharmacokinetic (PK) and/or Pharmacodynamic (PD) analyses in the Pharmaceutical industry. First, the thesis reviews the statistical properties of four important NP methods. In addition, it evaluates their practical performance through simulation studies inspired by population PK analyses. The value of NP methods is established, both in theory and in practice. The NP methods are then evaluated for population PK/PD analysis of an antidiabetic drug. The objective is to evaluate the ability of the methods to detect a subpopulation of non-responders to the treatment. Simulation studies show that two NP methods seem to be better able to detect this subpopulation. The last part of the thesis is devoted to the research of stochastic algorithms to improve the computation of NP methods. A perturbed stochastic gradient algorithm is proposed.