A new probabilistic rule for drug-dug interaction prediction

Area of Science:

• Pharmacokinetics
• Drug Interactions
• Bayesian Modeling

Background:

• Drug-drug interactions (DDIs) significantly impact drug efficacy and safety.
• Predicting the clinical significance of DDIs is crucial for drug development and patient care.
• Existing deterministic methods for DDI prediction are sensitive to sampling variations.

Purpose of the Study:

• To develop an innovative probabilistic rule for predicting the clinical significance of DDIs.
• To integrate a hierarchical Bayesian model for summarizing substrate/inhibitor pharmacokinetic (PK) models.
• To incorporate between-subject and between-study variances into DDI predictions for both population-average and subject-specific outcomes.

Main Methods:

• A hierarchical Bayesian model was used to synthesize PK models from multiple sources.
• A probabilistic rule was developed to classify DDIs based on predicted AUC ratio (AUCR) intervals.
• The model incorporated between-subject and between-study variances to predict population-average and subject-specific AUCRs.

Main Results:

• The probabilistic rule classifies DDIs into clinically significant, weak, or insignificant based on AUCR probabilities.
• Predictions incorporating between-subject variability showed greater variance than population-average predictions.
• The study highlighted the dependence of predicted AUCRs on interaction constants and dose combinations, using ketoconazole and midazolam as an example.

Conclusions:

• The proposed probabilistic rule offers a robust method for DDI clinical significance prediction, outperforming deterministic rules by accounting for sample variability.
• This approach enables early decisions on DDI significance, potentially reducing the need for extensive in vivo studies.
• Subject-specific predictions are essential, as population-average insignificance does not guarantee individual safety due to inherent variability.