Illustrating the structures of bias from immortal time using directed acyclic graphs

Area of Science:

• Epidemiology
• Biostatistics
• Pharmacoepidemiology

Background:

• Immortal time, a follow-up period where outcomes are impossible by design, introduces bias in epidemiological studies.
• The fundamental causes and structures of immortal time bias remain incompletely explained.
• Recognizing and addressing immortal time bias is crucial for accurate epidemiological research.

Purpose of the Study:

• To systematically explain the fundamental causes and structures of immortal time bias.
• To illustrate how immortal time bias arises and its impact on study findings.
• To discuss solutions for mitigating immortal time bias in various study designs.

Main Methods:

• Utilized a "Nobel Prize and lifespan" example for illustration.
• Employed directed acyclic graphs to visualize time-varying variables and bias structures.
• Analyzed bias structures with and without immortal time, considering competing risks.
• Examined shared structures across different pharmacoepidemiology study designs.

Main Results:

• Immortal time bias stems from misaligned exposure allocation and eligibility criteria.
• Bias structures include confounding by survival until exposure or selection bias based on survival until eligibility.
• Excluding immortal time does not fully resolve confounding or selection bias; competing risks can exacerbate it.
• Aligning baseline, exposure, and eligibility, and excluding prior exposures can prevent bias.

Conclusions:

• Understanding immortal time bias as confounding or selection bias is key.
• This understanding empowers researchers to identify, avoid, or ameliorate bias.
• Proactive identification and mitigation strategies are essential for robust epidemiological evidence.