How unmeasured confounding in a competing risks setting can affect treatment effect estimates in observational...

Area of Science:

• Biostatistics
• Epidemiology
• Health Research Methods

Background:

• Competing risks analysis is vital in observational studies.
• Standard methods like Cox and Fine & Gray models are susceptible to bias from unmeasured confounding.
• Limited research exists on bias in competing risks frameworks.

Purpose of the Study:

• To investigate the impact of unmeasured confounding on treatment effect estimates.
• To evaluate bias in Cox and Fine & Gray models under varying confounding strengths.

Main Methods:

• Simulations were designed to assess bias in competing risks models.
• Scenarios varied the strength of unmeasured confounding on treatment and outcomes.

Main Results:

• Unmeasured confounding biased treatment effect estimates in both Cox and Fine & Gray models.
• Bias direction depended on the confounder's effect on the event of interest versus competing events.
• Uneven treatment arms amplified these biases.

Conclusions:

• Unmeasured confounding significantly biases treatment effect estimates in competing risks analysis.
• Subdistribution models (Fine & Gray) are more sensitive to bias than cause-specific models (Cox).
• Sensitivity analyses are recommended to ensure the robustness of findings from observational studies.