Finding Causal Mechanistic Drug-Drug Interactions from Observational Data

Area of Science:

• Pharmacovigilance
• Computational Epidemiology
• Drug Safety

Background:

• Adverse drug reactions (ADRs) frequently result from drug-drug interactions (DDIs) during concurrent medication use.
• Existing methods for DDI detection from observational data face challenges including the vast number of potential interactions, confounding factors, and distinguishing true causal interactions from mere co-occurrence.

Purpose of the Study:

• To develop and apply a rigorous causal inference framework for identifying true drug-drug interactions (DDIs) from observational health data.
• To address limitations in existing DDI detection methods by accounting for confounding and defining causal interaction.

Main Methods:

• Utilized data mining algorithms to pre-filter a large set of potential drug-drug interactions (DDIs).
• Employed causal interaction models to adjust for observed confounders and assess the impact of unobserved confounding via sensitivity analyses.
• Ranked candidate DDIs based on their robustness to unobserved confounding.

Main Results:

• The proposed causal approach significantly reduces false positives compared to methods that do not account for confounding and causal interaction.
• Candidate DDIs were identified and ranked, prioritizing those less likely to be explained by confounding factors.
• The methodology provides a more reliable set of potential drug-drug interactions (DDIs) associated with adverse drug reactions (ADRs).

Conclusions:

• Causal inference methods are crucial for accurate detection of drug-drug interactions (DDIs) from observational data.
• This study presents a robust framework for identifying true DDIs, improving drug safety surveillance and reducing the risk of adverse drug reactions (ADRs).