Identifiability of pharmacological models from data

Sammanfattning: In order to automate anaesthesia in patients using propofol, closed-loop control systems with models that describe the time course of the drug effects on the body are required, usually being represented by pharmacokinetics and pharmacodynamics(PKPD). This thesis focused on evaluating parameter identifiability for the PK part of the model, using a model proposed by Eleveld et al. that has six parameters. Different sets of data were simulated with said model and Gaussian noise was added. To identify the parameters in the simulated data, Markov Chain Monte Carlo with the Metropolis-Hastings algorithm was applied for a set of different test cases. The results show that the estimations are dependent on the choice of priors and that the system is not uniquely identifiable. Although the estimated values differed from the parameters which were used for simulating data, the estimated parameters were able to fit the observed data very well in all trials. The conclusion of this work is that a PKPD model structure using six parameters is not practically identifiable and suggestions for future work would be to investigate whether a structure with fewer parameters could be more suitable for closed-loop control systems in anaesthesia.