Instrumental variable based estimation under the semiparametric accelerated failure time model.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Randomized controlled trials (RCTs) are ideal but often infeasible for medical research.
• Observational studies are alternatives but susceptible to unmeasured confounding, biasing treatment effect estimates.
• Existing instrumental variable (IV) methods are limited in survival analysis, especially with semiparametric models and censoring.

Purpose of the Study:

• To introduce a novel instrumental variable method for estimating causal treatment effects in survival analysis.
• To relax restrictive assumptions of prior IV techniques in the context of censored survival data.
• To provide a robust approach for analyzing observational data when RCTs are not possible.

Main Methods:

• Developed a semiparametric instrumental variable approach tailored for survival data.
• The method accounts for unmeasured confounding and censoring inherent in observational studies.
• Validated through extensive simulations and applied to real-world Medicare data.

Main Results:

• The proposed IV method yields consistent estimates of causal effects, outperforming traditional methods under unmeasured confounding.
• Demonstrated robust performance across various simulated scenarios, confirming theoretical properties.
• Successfully applied to Medicare data, comparing surgical procedures for abdominal aortic aneurysm, providing reliable causal insights.

Conclusions:

• The novel instrumental variable method offers a powerful tool for causal inference in survival analysis using observational data.
• This approach addresses critical limitations of existing methods, particularly in the presence of unmeasured confounding and censoring.
• Enables more reliable estimation of treatment effects in medical research where RCTs are impractical.