Sensitivity analysis for causality in observational studies for regulatory science.

Area of Science:

• Regulatory Science
• Data Science
• Biostatistics

Background:

• The 21st Century Cures Act mandates FDA guidance on using real-world evidence (RWE).
• Stakeholders convened to establish best practices for real-world data (RWD) in regulatory science.
• A causal roadmap for RWD study designs was recommended.

Purpose of the Study:

• To detail the specification of sensitivity analyses for RWD studies.
• To test the robustness of causal models against assumption violations.
• To provide practical considerations for regulatory applications.

Main Methods:

• An example sensitivity analysis from a Nifurtimox effectiveness study for Chagas disease.
• Overview of various sensitivity analysis methods.
• Emphasis on practical considerations for regulatory use.

Main Results:

• Sensitivity analyses require careful design and prespecification to prevent bias.
• Methods need auxiliary information and should rely on verifiable assumptions.
• Auxiliary information must be learnable from existing scientific knowledge.

Conclusions:

• Prespecified and assumption-lean sensitivity analyses are crucial for regulatory science.
• These methods strengthen the validity and trustworthiness of effectiveness conclusions.
• Robustness testing is key for reliable RWD utilization in regulatory decision-making.