Quantitative Evaluation of Initial Symptoms as Predictors to Detect Adverse Drug Reactions Using Bayes’ Theory: Expansion and Evaluation of Drug-Adverse Drug Reaction–Initial Symptom Combinations Using Adverse Event Reporting System Database

Abstract

In prescription dispensing in Japan, to avoid adverse drug reactions (ADR) pharmacists provide patients with information concerning the initial symptoms (IS) of any ADR that might be caused by the drugs they have been prescribed. However, the usefulness of such information for preventing ADR has not been quantitatively evaluated. We previously performed a trial calculation of the usefulness of rash as a predictor of drug-induced liver disorders by applying Bayes' theorem and showed that the predictive utility of IS can be quantitatively evaluated using likelihood ratios. However, for other drug-ADR-IS combinations it was difficult to obtain the information required for the calculations from Japanese data alone. In this study, using the Adverse Event Reporting System (AERS) database of the U.S. Food and Drug Administration (FDA), we evaluated 132 drug-ADR-IS combinations that were considered to be potentially clinical significant. Regarding bezafibrate-associated rhabdomyolysis and cibenzoline-associated hypoglycemia, these ADR were not detected in cases involving monotherapy. For 58 combinations, no events that were considered to be IS of the target ADR developed. Fever, nausea, and decreased appetite were the IS of many ADR, making them very useful predictors. In contrast, pruritus and rash were not very useful. Fever might be a predictor of thiamazole-induced agranulocytosis or levofloxacin- or terbinafine-induced liver disorder, tremors might be useful for predicting paroxetine-induced serotonin syndrome, and decreased appetite might be a useful indicator of terbinafine-induced liver dysfunction.