Development and Validation of Estimate Equations for Adverse Drug Reactions Using Risk Factors and Subjective Symptoms

The purpose of this study was to develop and validate estimate equations for preventing adverse drug reactions (ADRs). We conducted five case-control studies to identify individual risk factors and subjective symptoms associated with the following five ADRs: drug-induced ischemic heart disease; renal damage; muscle disorder; interstitial pneumonia; and leucopenia. We performed logistic regression analysis and obtained eight regression equations for each ADR. We converted these to ADR estimate equations for predicting the likelihood of ADRs. We randomly selected 50 cases with non-individual ADRs from the Case Reports of Adverse Drug Reactions and Poisoning Information System (CARPIS) database of over 65000 case reports of ADRs, and assigned these cases to a validation case group. We then calculated the predictive probability for 50 cases using the eight estimate equations for each ADR. The highest probability for each ADR was set as the probability of each ADR. If the probability was over 50%, the case was interpreted as ADR-positive. We calculated and evaluated the sensitivity, specificity, and positive likelihood ratio of this system. Sensitivity of the estimate equations for muscle disorder and interstitial pneumonia were ≥90%. Specificity and positive likelihood ratios of estimate equations for renal damage, interstitial pneumonia and leucopenia were ≥80% and ≥5, respectively. Our estimate equations thus showed high validity, and are therefore helpful for the prevention or early detection of ADRs.