The subtype-free average causal effect for heterogeneous disease etiology.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Disease effects can differ across subtypes.
• Current methods often ignore causality when assessing exposure-disease links.
• Colorectal cancer (CRC) presents subtypes based on microsatellite instability (MSI).

Purpose of the Study:

• To propose and evaluate a novel causal estimand, the Subtype-Free Average Causal Effect (SF-ACE).
• To investigate the causal effect of smoking on colorectal cancer, considering MSI subtypes.
• To address limitations in existing causal inference methods for disease subtypes.

Main Methods:

• Utilized principal stratification to define the SF-ACE.
• Explored non-parametric identification and nuanced monotonicity assumptions.
• Developed sensitivity analysis methods to relax untestable assumptions.
• Proposed three estimators for SF-ACE, including a doubly robust option.

Main Results:

• The SF-ACE provides a causal effect estimate for individuals unaffected by other subtypes.
• Sensitivity analyses were developed to assess the robustness of findings.
• Methodology was applied to two large cohorts examining smoking's heterogeneous effect on CRC by MSI status.

Conclusions:

• The SF-ACE offers a causally-informed approach to studying exposure effects on disease subtypes.
• The developed methods and sensitivity analyses enhance the reliability of causal effect estimation.
• Findings contribute to understanding smoking's varied impact on colorectal cancer subtypes.