An expert system as a clinical aid in the administration and dosage adjustment of some commonly prescribed therapeutic drugs and antibiotics

Project investigates the design requirements of a computer based drug dosing system suitable for use by general physicians.. One area of medicine in which knowledge-based systems may improve day to day patient care is in the design of initial dosing regimens and dosage adjustment of certain drugs whose plasma levels correlate with their toxicity and/or efficacy. Since the general clinician is responsible for the appropriate administration of many of these drugs, a way must be found to enable these clinicians to rapidly detrmine the appropriate dose or dosage adjustment required to achieve the desired plasma concentration and thus the desired clinical effect. To obtain the best results, a good knowledge of pharrnacokinetic principles is required as well as the facility to apply these principles easily and safely. In this study, the intention was to construct a knowledge-based system for the design of drug dosing regimens and to investigate issues relating to the design of such a system which will affect its' utility in general medicine. A multidisciplinary approach to the problem was adopted. A combination of standard pharmacokinetic modelling and artificial intelligence techniques was used to design a system suitable for use by the general physician. The approach was informed by ethnography with the design incorporating features seen as desirable by prospective users and a knowledge base with facts and rules related to the safe and effective use of the system. In addition, a more general method of pharmacokinetic parameter estimation than that employed in most current pharmacokinetic systems was investigated. In the clinical environment, there are a number of sources of error which may invalidate pharmacokinetic calculations. The most important of these being those associated with the incorrect preparation of doses and the recording of incorrect times of dosing and specimen collection. It was observed that current pharmacokinetic systems do not address these sources of error directly. A more general Bayesian approach which might be extended to incorporate these 'external' errors would be appropriate. The method investigated in the report is a Bayesian formulation of the Kalman filter. It was applied to the one and two compartment linear models and to the one compartment nonlinear model. These models being sufficient to cover the majority of drugs of interest in the general hospital setting. Kalman filtering is a general method for handling state-space models which gives optimal estimates of the current state of a dynamic system. It is commonly encountered in the field of control engineering but is also used in the analysis of time series. The method was shown to be adaptable in principle to pharmacokinetic parameter estimation and it is theoretically extendable to incorporate the external sources of error described above. The positive initial results presented in the report form the basis for ongoing research into the possible extension of the system and a formal assessment of user acceptance.