Instrumental variable additive hazards models.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Instrumental variable (IV) methods are crucial for estimating causal effects in non-experimental studies, particularly when unobserved confounding is present.
• Existing IV methods have limited extensions to censored data scenarios, posing a challenge for medical intervention studies.
• The proportional hazards model, commonly used for survival data, presents difficulties when integrated with IV techniques.

Purpose of the Study:

• To address the challenges of applying instrumental variable (IV) techniques to censored data within the proportional hazards model.
• To demonstrate the effectiveness of an additive hazards formulation for instrumental variable (IV) analyses involving censored data.
• To develop and evaluate a novel two-stage estimator for causal effects in additive hazard models with censored outcomes.

Main Methods:

• Development of a closed-form, two-stage instrumental variable (IV) estimator tailored for the additive hazard model with censored data.
• Application of linear structural equation models for the hazard function to facilitate the estimation process.
• The proposed methods accommodate both continuous and discrete exposure variables.

Main Results:

• The developed additive hazards formulation provides a robust framework for instrumental variable (IV) analyses with censored data.
• The two-stage estimator enables the estimation of causal effects and causal relative survival measures.
• Simulation studies and a real-world application to colon cancer chemotherapeutic agent data confirm the performance of the proposed methods.

Conclusions:

• The additive hazards model offers a viable and effective extension of instrumental variable (IV) methods for censored data.
• The novel two-stage estimator performs well in estimating causal effects, offering valuable insights for medical intervention research.
• This approach enhances the ability to conduct causal inference in observational studies with survival data.