Causal Inference Tools for Pharmacovigilance: Using Causal Graphs to Identify and Address Biases in...

Disproportionality analysis, finding associations in the co-reporting of drugs and events, is widely used in pharmacovigilance to detect potential safety signals of adverse drug reactions. However, inherent biases and unique data features often cause disproportionality to diverge from causation, and a comprehensive framework to address these issues is lacking.

We showcase how directed acyclic graphs (DAGs) can enhance disproportionality analysis-related inferences, better qualifying its limitations and catalysing its inclusion in the broader evidence landscape.

We introduce a DAG-based causal framework to systematically document and address biases in disproportionality analyses (e.g., confounding, colliders, measurement and reporting biases). We illustrate its application to case studies from the Food & Drug Administration (FDA) Adverse Event Reporting System-using the Information Component as a disproportionality metric and restriction as conditioning.

Directed acyclic graphs facilitate the formalisation of existing knowledge and causal assumptions, optimise the design of disproportionality analysis to mitigate biases-thereby enhancing sensitivity and specificity-improve transparency, better enable the formulation of critiques, highlight limitations of disproportionality and guide follow-up studies to address residual confounding and broader evidence synthesis.

Using DAGs to map and mitigate biases requires caution and does not allow to obtain definitive answers to causal questions. Still, it results in more reliable and knowledge-based safety signals, reducing and mapping the gap between what we find (association) and what we look for (causation). Additional research should further tailor DAGs to pharmacovigilance challenges, map the generative mechanisms of pharmacovigilance data, and better integrate disproportionality analysis results into evidence-synthesis workflows.