Sensitivity analysis using bias functions for studies extending inferences from a randomized trial to a target...

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Generalizing causal inferences from randomized trials to target populations relies on exchangeability assumptions conditional on covariates.
• These assumptions are often based on uncertain background knowledge, necessitating sensitivity analyses.
• Existing methods may require detailed knowledge of unmeasured confounders or effect modifiers.

Purpose of the Study:

• To develop simple, direct methods for sensitivity analyses to assess the impact of assumption violations in causal inference generalization.
• To provide tools that do not require extensive background knowledge of specific unmeasured factors.
• To demonstrate applicability to both non-nested and nested trial designs.

Main Methods:

• Introduced bias functions to directly parameterize violations of exchangeability assumptions.
• Developed methods applicable to combining trial data with external nonrandomized samples (non-nested designs).
• Adapted methods for trials embedded within a target population cohort (nested designs).

Main Results:

• The proposed methods offer a straightforward way to conduct sensitivity analyses for causal inference generalization.
• These methods reduce the reliance on detailed, often unavailable, background information about unknown confounders or effect modifiers.
• Demonstrated flexibility in application across different trial and data structures.

Conclusions:

• The presented sensitivity analysis techniques enhance the robustness of generalizing causal inferences from randomized trials.
• These methods provide a practical approach for researchers to assess the potential impact of assumption violations.
• The techniques are valuable for improving the validity of evidence synthesis and real-world applicability of trial findings.