COMPARISON OF THE BAYESIAN-APPROACH AND A SIMPLE ALGORITHM FOR ASSESSMENT OF ADVERSE DRUG EVENTS

The differential diagnosis of severe adverse drug events can be based on clinical judgment, algorithms, or the Bayesian approach. The Bayesi an Adverse Reactions Diagnostic Instrument (BARDI) calculates the post erior probability (PsP) in favor of a specific drug cause based on bac kground (e.g., epidemiologic) and case information (e.g., time of onse t). Although BARDI discriminates between drug- and nondrug-induced adv erse events, its apparent complexity may limit its use, BARDI results were compared with those from an algorithm for rating the probability that an adverse drug event is drug-induced (Adverse Drug Reaction Prob ability Scale, or APS) that is still commonly used, APS scores were ob tained by two independent raters for 106 challenging cases that had be en analyzed from 1 to 5 years ago with BARDI (91 cases of hypersensiti vity, 12 cases of hematologic toxicity, and three cases of pulmonary t oxicity); 130 ratings were generated because of the use of multiple dr ugs. APS scores for the two raters were highly correlated (r = 0.79; p < 0.0001). Probabilities of drug causation with use of BARDI versus a verage APS scores were significantly correlated (r(s) = 0.45; p < 0.00 01). However, BARDI better distinguished cases that were highly probab le (n = 83; PsP greater than or equal to 0.75) or highly improbable (n = 30; PsP less than or equal to 0.25), whereas the BPS rated the majo rity of these cases in the midrange (n = 128; range of APS, 1 to 8.9). These results suggest that APS and BARDI evaluations are concordant. Thus the APS may be an effective screening tool, although BARDI can be tter discriminate drug from nondrug-induced cases and may be more appr opriate for serious cases of adverse drug reactions.