Target trial emulation under nonmutually exclusive assignment: structural pitfalls and methodological remedies

Area of Science:

• Epidemiology
• Causal Inference
• Observational Data Analysis

Background:

• Traditional epidemiologic studies assume mutually exclusive exposure groups for causal inference.
• Target Trial Emulation (TTE) is a framework for causal effect estimation from observational data.
• Non-mutually exclusive treatment assignment in real-world settings presents challenges for TTE validity.

Purpose of the Study:

• To evaluate TTE implementation strategies under non-mutually exclusive treatment assignment.
• To assess the impact of treatment overlap and covariate alignment on causal estimation bias.
• To understand how TTE performs when treatment groups are not mutually exclusive.

Main Methods:

• A simulation study was conducted to evaluate TTE strategies.
• Systematic variation of treatment overlap and covariate alignment.
• Analysis of propensity score estimation and outcome modeling under non-mutually exclusive assignment.

Main Results:

• Non-mutually exclusive assignment can introduce bias if treatment overlap and positivity are not addressed.
• Sufficient covariate overlap allows TTE to recover marginal effects accurately.
• Poor covariate overlap hinders accurate marginal effect recovery, even with advanced TTE strategies.

Conclusions:

• Explicitly addressing treatment overlap and positivity is essential for valid TTE under non-mutually exclusive assignment.
• TTE can yield valid causal estimates comparable to mutually exclusive designs when overlap is adequate.
• Aligning study design, estimand, and treatment structure is critical for successful TTE in real-world applications.