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

Area of Science:

• Pharmacovigilance and Drug Safety
• Causal Inference and Data Science
• Biostatistics and Epidemiology

Background:

• Disproportionality analysis is crucial for detecting adverse drug reaction safety signals in pharmacovigilance.
• Existing methods often suffer from biases, leading to discrepancies between detected associations and true causation.
• A comprehensive framework to address these inherent biases is currently lacking.

Purpose of the Study:

• To demonstrate how directed acyclic graphs (DAGs) can enhance disproportionality analysis inferences.
• To better qualify the limitations of disproportionality analysis.
• To facilitate the integration of disproportionality analysis into the broader evidence landscape.

Main Methods:

• Introduction of a DAG-based causal framework to systematically identify and mitigate biases (e.g., confounding, colliders, measurement, reporting).
• Application of the framework to case studies using the FDA Adverse Event Reporting System.
• Utilizing the Information Component as a disproportionality metric and restriction as a conditioning method.

Main Results:

• DAGs enable formalization of causal assumptions and existing knowledge.
• Optimization of disproportionality analysis design to enhance sensitivity, specificity, and transparency.
• Improved ability to critique findings, highlight limitations, and guide future research for evidence synthesis.

Conclusions:

• DAGs aid in mapping and mitigating biases in disproportionality analysis, though caution is needed.
• The approach yields more reliable, knowledge-based safety signals, reducing the association-causation gap.
• Further research is recommended to tailor DAGs for pharmacovigilance, map data generation mechanisms, and integrate findings into evidence synthesis.