A comparative study of PK/PD and neural network modeling.

摘要

This work has three parts of which the first two parts use simplifying assumptions and a compartmental approach to describe drug movement in complex biological system. The first part deals with the development of a Pharmacokinetic and Pharmacodynamic (PK/PD) model for propofol, which is an intravenous anesthetic agent. The model developed could be used to simulate data sets that would be employed in developing a strong fuzzy logic algorithm for propofol infusion using EEG power distribution variables as surrogate measures for assessing depth of anesthesia. In the second part circadian production of saliva melatonin was described by model based on the biochemical production of melatonin. This was extended into a PK/PD model by establishing temporal and quantitative relationship between saliva melatonin and core body temperature. With this model we can use wake time saliva melatonin levels as a circadian phase marker instead of core body temperature and the model was further extended for the quantification of activity. In the third part a model independent approach was applied by utilizing an Artificial Neural Network models (NN) for both propofol and melatonin. The advantage of this method is that no assumptions are made about the complex biological system and drug movement in the system. This method involves training the network with known inputs and outputs and testing the trained network with unknown data. A comparison was made for the predicting ability of the PK/PD model and neural network model. PK/PD models described very well the observed data for both propofol and melatonin and predictions were found to be better with the PK/PD model than the NN model.