Quantitative Prediction of Adverse Event Probability Due to Pharmacokinetic Interactions

Introduction Iatrogeny due to drug–drug interactions is insufficiently documented, due to the high number of possible combinations. Objective This study aimed to design a simple but general method to predict the variation of adverse events (AE) frequency due to a pharmacokinetic or pharmacodynamic interaction. Methods Three prediction models were designed using a logistic probability density function. Each prediction model was based on three components: the AE odds ratio of each drug in the combination, and the area under the curve ratio (Rauc) of the pharmacokinetic interaction, if any. Pharmacodynamic interaction was assumed to be additive on logit scale. Rauc was predicted using a well-validated mechanistic static model, freely available online. No combination study is required. The method was evaluated against a wide range of AEs (28 High Level Terms) and 211 drug combinations (involving 43 victim drugs and 55 perpetrators), by comparing the observed and predicted frequencies. The observed odds ratios were estimated with a disproportionality analysis from the FDA Adverse Event Reporting System, using an approach that minimizes biases. Results With the best model, the rate of prediction considered as correct (within 50–200% of the observed value) was 72%, and the bias was negligible (-5%). The AE odds ratio due to pharmacokinetic and pharmacodynamic interactions was equally well predicted. Conclusions A simple workflow to implement the method in practice is proposed. This method may help to foresee and to anticipate the harmful consequences associated with drug–drug interactions, at virtually no experimental cost, when the odds ratio of an AE is known for each drug alone and the AUC ratio is known or predicted by a suitable model.