Application of pharmacometric methods to clinical toxicology studies

Abstract

Risk assessment is a fundamental part of clinical toxicology. It is complicated due to a variable time course of clinical effects of drugs in clinical toxicology. Defining the time course of clinical effects of drugs in overdose will assist in accurate risk assessment and thus minimise the risk to benefit ratio for each individual patient. However assessing the time course of clinical effects of drugs is complex in overdose studies. The complexity arises from absence of accurate knowledge of the dose, the time at which the overdose was ingested and observations in the initial phase of the study after the overdose. The study designs in overdose studies are highly unbalanced which adds to the complexity. The purpose of this thesis is to apply pharmacometric methods to define the time course of clinical effects of drugs in overdose. In this thesis pharmacometric methods are applied to: 1) Quantify the effects of various decontamination procedures on pharmacokinetics of venlafaxine in overdose. 2) Quantify the effects of various decontamination procedures on pharmacodynamics of venlafaxine in overdose. 3) Develop a robust optimality criterion for designing a study to assess whether paracetamol in overdose has linear or nonlinear pharmacokinetics. In the pharmacokinetic analysis (Chapter 2), data obtained from a venlafaxine overdose study were modelled using Bayesian methodology in WinBUGS 1.4.3.The results of the analysis showed that a one-compartment model with first-order input and first-order elimination provided an adequate description of the data. Single dose activated charcoal increased the clearance of venlafaxine by 35% and a combination of single dose activated charcoal and whole bowel irrigation reduced the fraction absorbed by 29%, however the latter produced a greater reduction in maximum plasma concentration for a similar drop in area under the curve compared to single dose activate charcoal alone. In the pharmacodynamic analysis (Chapter 4), a linear logistic regression model was used to describe the influence of dose and decontamination on the probability of seizures. Simulations from the model showed that the probability of seizure increased with dose. Single dose activated charcoal and combination of single dose activated charcoal and whole bowel irrigation decreased the probability of seizure. The decrease in probability of seizure was greater with the combination when compared with single dose activated charcoal alone. A modified Gompertz model was used to define the time to first seizure using Bayesian methodology in WinBUGS 1.4.3. Simulations from the model showed that the time to 90% of first seizure was not affected by dose or decontamination procedures. The results also showed that the pharmacokinetics of venlafaxine drives the pharmacodynamics. A pharmacokinetic study of paracetamol in overdose was prospectively designed to optimally discriminate between two candidate models (Chapter 6). In this study a robust T-optimal design was developed to distinguish between two candidate models, a two compartment model with linear elimination (M1) and a two compartment model with Michaelis-Menten elimination (M2), of paracetamol without assuming the true model a priori. Optimal designs were obtained using robust T-optimal design and standard T-optimality. Standard T-optimality assumes that one of the two competing models is true. The power of a design to distinguish between the two models was calculated. The results of the analysis showed that designs obtained assuming standard T-optimality depended on the model assumed to be true. The results also revealed that the design obtained assuming M1 is correct had a low power to distinguish between the models when M2 was the correct model in the simulation (13%). Similarly the design obtained assuming that M2 is correct had a low power to distinguish between the models when M1 was the correct model in the simulation (40%). The robust design, where either M1 or M2 could be correct had a higher power to distinguish between the models in the above situations 50 and 87% respectively. Methodologies developed for pharmacokinetic analysis included 1) A method to account for missing dose history (Chapter 3). Methodologies developed during pharmacodynamic analysis included diagnostics for logistic regression models which were 1) random binning and 2) simplified Bayes marginal model plots (Chapter 5). The methods developed and applied in this thesis are not limited to the venlafaxine and paracetamol and could be applied to other drugs in overdose. Some of the methods, especially the methods to account for missing dose history, Bayesian posterior predictive checks, criteria for covariate analysis and diagnostics for logistic regression can be applied not only in clinical toxicology but also clinical pharmacology.