Introduction
In both observational and Randomised Clinical Trial studies, the validity of causal inferences depends on the proper classification of exposure/treatment and follow-up time. Immortal time bias is a form of time-related misclassification that can seriously distort associations between exposure/treatment and outcome. It occurs when a period of follow-up during which the outcome could not have occurred — because of the way exposure is defined — is incorrectly attributed to the exposed/treated group. This “immortal” time gives exposed/treated individuals an artificial survival advantage, leading to biased estimates of effect, most often in favour of the exposure/treated.
The term immortal refers not to biological immortality but to the fact that, during this time window, participants must survive (i.e., cannot experience the event of interest) to be classified as exposed. The bias is particularly problematic in observational pharmacoepidemiology, clinical cohort studies, and health-service evaluations that involve time-varying exposures.
Conceptual Illustration
Consider a cohort study examining whether statin use after hospital discharge reduces mortality among patients with myocardial infarction. Researchers may classify patients who ever receive a statin prescription after discharge as “exposed.” However, those individuals must first survive long enough to fill their prescription. The time between discharge and the first statin prescription is immortal time because the patient had to remain alive (and event-free) to become exposed. If this immortal time is erroneously counted as exposed person-time, it will inflate survival in the statin group and underestimate mortality risk, creating a spurious protective effect.
Mathematically, immortal time bias leads to differential misclassification of person-time. The denominator for the exposed group includes follow-up time when no exposure actually occurred but during which the event could not have happened. Consequently, the incidence rate for the exposed group is artificially reduced compared with that of the unexposed.
Sources of Immortal Time Bias
Immortal time bias can arise in several research designs:
- Post-exposure definition bias
When exposure is defined based on an event that happens after cohort entry (e.g., initiation of treatment, receipt of transplant), any survival time between cohort entry and exposure initiation becomes immortal if attributed to the exposed group. - Eligibility criteria requiring survival
Some studies include only participants who survive a minimum time after diagnosis or enrollment to meet eligibility (e.g., “survivor cohorts”). This conditioning on future survival introduces bias analogous to immortal time. - Time-dependent covariate misclassification
When exposure is treated as a fixed variable instead of a time-dependent one, follow-up prior to actual exposure may be misclassified as exposed person-time.
Consequences of Immortal Time Bias
The main consequence is biased estimation of effect size, typically an overestimation of benefit or underestimation of harm. Because immortal time artificially improves survival or prolongs event-free duration in the exposed group, hazard ratios or risk ratios become biased toward protective values.
This bias can alter clinical interpretation, policy decisions, and cost-effectiveness estimates. In extreme cases, it can reverse the direction of association — showing apparent benefit where none exists. Moreover, immortal time bias is insidious: standard regression adjustment for confounders does not eliminate it, because it arises from incorrect exposure classification rather than confounding by covariates.
Analytical Strategies to Avoid or Correct Immortal Time Bias
Proper study design and analytical methods are essential to prevent immortal time bias. Common approaches include:
- Time-dependent exposure modeling:
The gold-standard approach is to treat exposure as a time-varying variable in survival analysis. In Cox proportional hazards models, the exposure indicator changes from 0 (unexposed) to 1 (exposed) at the time of initiation. This ensures that person-time before exposure is correctly attributed to the unexposed group. - New-user (incident-user) designs:
Restricting the cohort to new initiators of the treatment eliminates pre-exposure immortal time. Follow-up begins at the time of first exposure rather than at cohort entry. This design is standard in pharmacoepidemiology for assessing drug effectiveness and safety. - Matching or weighting on time of exposure:
When using retrospective data, matching exposed individuals to unexposed comparators with the same time-since-entry or using inverse probability weighting based on exposure timing can mitigate bias.
Reporting and Transparency
Because immortal time bias is easily overlooked, explicit reporting of how exposure and follow-up are defined over time is critical. Best-practice guidelines, such as STROBE and RECORD, recommend describing:
- The timing of cohort entry, exposure initiation, and outcome ascertainment;
- Methods used to handle time-varying exposures; and
- Any sensitivity analyses addressing immortal or misclassified periods.
Clear timelines and flow diagrams help readers verify whether exposure assignment preceded outcome risk.
Conclusion
Immortal time bias is a pervasive but avoidable threat to validity in epidemiological research. It occurs when survival time that is guaranteed — by design or definition — is wrongly credited to an exposure group, thereby exaggerating protective effects. Recognizing and addressing immortal time bias requires aligning exposure definition, eligibility, and follow-up in time; employing appropriate statistical models that handle time-varying exposures; and reporting these methods transparently.
For epidemiologists and the like, awareness of immortal time bias is not merely technical — it is fundamental to causal reasoning. Only by accurately representing when exposure and risk truly begin can we draw reliable conclusions about interventions, treatments, and public health policies.