Comparison of the Bayesian approach and a simple algorithm for assessment of adverse drug events

Abstract

The differential diagnosis of severe adverse drug events can be based on clinical judgment, algorithms, or the Bayesian approach. The Bayesian Adverse Reactions Diagnostic Instrument (BARDI) calculates the posterior probability (PsP) in favor of a specific drug cause based on background (e.g., epidemiologic) and case information (e.g., time of onset). Although BARDI discriminates between drug- and nondrug-induced adverse 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 Probability Scale, or APS) that is still commonly used. APS scores were obtained by two independent raters for 106 challenging cases that had been analyzed from 1 to 5 years ago with BARDI (91 cases of hypersensitivity, 12 cases of hematologic toxicity, and three cases of pulmonary toxicity); 130 ratings were generated because of the use of multiple drugs. APS scores for the two raters were highly correlated (r = 0.79; p < 0.0001). Probabilities of drug causation with use of BARDI versus average APS scores were significantly correlated (rs = 0.45; p < 0.0001). However, BARDI better distinguished cases that were highly probable (n = 83; PsP > or = 0.75) or highly improbable (n = 30; PsP < or = 0.25), whereas the APS rated the majority 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 better discriminate drug from nondrug-induced cases and may be more appropriate for serious cases of adverse drug reactions.