Abstract

The reproducibility and validity of an algorithm for diagnosis of adverse drug reactions (ADRs) were tested in a clinical spectrum of 30 suspect cases. Using a questionnaire derived from the algorithm the three algorithm developers (nonexperts) agreed on the probability of ADR in 67% of cases, with pair-wise agreement varying from 73% to 87%. The pair-wise agreement of two clinical pharmacologic experts rose from 47% without the algorithm to 63% with the algorithm, with Kw, a chance-corrected index of weighted agreement, increasing from 0.26 to 0.57. The algorithmic assessments of the three nonexperts agreed with expert consensus in 80% to 83% of cases. The ADR algorithm appears to provide a reproducible and valid method of evaluating the likelihood of ADRs in individual patients. Its use can help improve the diagnostic and epidemiologic approach to these important, complex clinical phenomena.