Immortal time bias in pharmaco-epidemiology

Area of Science:

• Epidemiology
• Pharmaco-epidemiology
• Biostatistics

Background:

• Immortal time bias, identified in the 1970s, affects cohort studies by including follow-up time where outcomes are impossible.
• This bias has resurfaced in pharmaco-epidemiology, potentially overestimating medication benefits.
• Observational studies using computerized databases are particularly susceptible.

Purpose of the Study:

• To describe cohort study designs that lead to immortal time bias.
• To quantify the magnitude of this bias under various survival distributions.
• To illustrate the bias using lung cancer patient data.

Main Methods:

• Analysis of time-based, event-based, and exposure-based cohort definitions.
• Quantification of bias magnitude using survival distributions (Weibull, exponential).
• Illustration with a real-world cohort of lung cancer patients.

Main Results:

• Bias in rate ratio increases proportionally with the duration of immortal time.
• Bias is more pronounced with decreasing hazard functions (e.g., Weibull distribution) compared to constant hazards (exponential distribution).
• Misclassification or exclusion of immortal time inflates observed effects.

Conclusions:

• Observational studies on drug benefits using computerized databases require careful design and analysis to prevent immortal time bias.
• Accurate handling of follow-up time is essential for valid pharmaco-epidemiological research.
• Understanding and mitigating immortal time bias is critical for reliable evidence synthesis.