Alternative monotonicity assumptions for improving bounds on natural direct effects

Area of Science:

• Epidemiology
• Biostatistics
• Clinical Research

Background:

• Estimating the direct effect of treatments with multiple pathways to outcomes is crucial in research.
• Unmeasured confounding can prevent identification of direct effects, necessitating bounds.
• Previous methods relied on linear programming and stricter assumptions.

Purpose of the Study:

• To propose novel, simpler bounds for the natural direct effect of a treatment.
• To relax restrictive assumptions common in prior direct effect estimation methods.
• To provide a more generally applicable method for bounding direct effects.

Main Methods:

• Developed bounds for natural direct effects without using linear programming.
• Introduced weaker monotonicity assumptions than previously used.
• Removed the assumption of a binary outcome variable.

Main Results:

• The proposed bounds are simpler and derived without linear programming.
• New, weaker monotonicity assumptions lead to narrower bounds.
• The method is applicable to non-binary outcome variables.

Conclusions:

• The novel bounding method offers a simpler and more flexible approach to estimating natural direct effects.
• These bounds are particularly useful when unmeasured confounding is present.
• The approach enhances the analysis of treatment effects in epidemiological and clinical studies, as demonstrated in a coronary heart disease trial.