Semi-quantitative simulation for reasoning about physiological models of drug kinetics and effects.

Inter- and intra-individual pharmacokinetic or pharmacodynamic variability is a major cause of adverse drug reactions or ineffective therapy. We are developing a computer-based tool for predicting the consequences of different physiological and pathological states and for reasoning about the possible causes of observed variability that may be useful both in a clinical decision support environment for drug monitoring and as a research aid in the investigation of the influence of physiological factors on drug response. It is based on a physiological approach to pharmacokinetic modeling in which actual anatomical or physiological entities, such as organs, tissues or blood flows, are represented. These models serve as the basis for semi-quantitative simulation, a method linking classical quantitative simulation (by numerical integration of differential equations) with artificial intelligence-based qualitative simulation techniques. This approach retains the mathematical power of the Systems Dynamics method for solving complex, time-varying systems containing feed-back loops, which are intractable for current qualitative knowledge representation techniques, and extends it with the causal reasoning and explanation power of symbolic inference techniques used in expert systems. It also allows problem solving in situations, so common in medicine, where initial values of variables and parameters cannot be estimated precisely. Simulation outputs are intended to be qualitatively, but not necessarily quantitatively, correct. The semi-quantitative simulation method was originally developed in MacLisp on a DEC 2060 and applied to modeling cardio-vascular physiology. We are porting the code to Common Lisp on a Macintosh and adapting the approach to pharmacology, concentrating on drug metabolism issues, with lidocaine pharmacokinetics as a test case.