Computerized Aids for Probabilistic Assessment of Drug Safety II: An Expert System *

Abstract

We describe a probabilistic expert system for assessment of cases of pseudomembranous colitis suspected to be due to antibiotics. The system uses a prospective model for the occurrence and timing of the disease given the exposure to a single antibiotic. This model takes into account the mechanism of colitis and expert opinion on the probability of occurrence of the component events leading to colitis. The model is used in two ways: 1. To provide inputs to a spreadsheet program that uses Bayse theorem to sort out the relative likelihood of causation by competing antibiotics. This approach assumes that competing causes are mutually exclusive; and 2. As a building block of a more complex model that allows for simultaneous causation by two drugs

We demonstrate both approaches on a case of pseudomembranous colitis exposed to similar antibiotics at different times. They produce similar results except for the possibility of overlap which is not allowed for in the first approach. Both approaches provide a mechanism to encode expertise for repeated and coherent assessment of similar cases of suspected adverse drug reactions. We believe that application of this approach can improve the validity of adverse drug reaction (ADR) causality assessment and the primary use to which this assessment should be put — predication of the rate of adverse drug reactions in future patients to be treated with the drug