Immortal Time Bias in Longevity Studies

Who This Is Useful For

This page is useful for readers evaluating observational studies of mortality, healthy lifespan, age-related disease, medication use, or behaviors that begin after cohort entry. It is especially relevant when membership in an “exposed,” “treated,” “adherent,” or “long-lived” group depends on something that happens after follow-up has already started. [1] [3] [4]

Despite its name, immortal time bias does not mean that participants are biologically immortal. It refers to a span of observed follow-up during which the study outcome cannot occur if a person is to be placed in a group defined by a later event. If someone is classified as a medication user because they fill a prescription six months after enrolment, they necessarily survived and remained eligible for those six months. Assigning those months to the user group from day one gives that group survival time it had to accumulate before exposure was known. [1] [2]

A Simple Longevity Example

Imagine a cohort of older adults followed from 1 January. Researchers compare people who begin a therapy at some point during the year with people who never begin it, then measure mortality from 1 January. A participant who starts therapy on 1 July must survive until that date to enter the treated group. If January through June is analyzed as treated time, deaths before July can occur only in the comparison group, creating an artificial survival advantage for treatment. [1] [2] [5]

The temporal misclassification in this example is the central problem described in methodological studies of immortal time. [1] [2] [8]

Why Longevity Studies Are Vulnerable

Longevity research often uses long observational follow-up because many exposures cannot be assigned experimentally for decades and outcomes such as disability, disease, and death occur over time. During that follow-up, medication use, diagnoses, retirement, adherence, biomarker-defined status, and other characteristics can change. Treating a characteristic first observed later as though it were present at baseline can therefore mix the ordering of exposure and survival. [1] [4] [7]

The bias may also arise when researchers define a group by reaching a future milestone. For example, comparing mortality according to a retirement-age category can require members of a later-retirement group to survive until the age that defines that category. The general issue is the use of post-baseline information to assign groups while counting outcomes from an earlier time. [4] [8]

How the Bias Enters an Analysis

Not the Same as Confounding or Survivorship Bias

Confounding occurs when another factor is associated with both the exposure and outcome, such as baseline health influencing both treatment uptake and mortality. Immortal time bias instead concerns how eligibility, exposure assignment, and follow-up are positioned in time. A study can adjust for many baseline confounders and still retain immortal time bias if it misclassifies pre-exposure person-time. [1] [4] [8]

Survivorship bias is a broader selection problem in which analysis is restricted to those who survive long enough to be observed. Immortal time bias overlaps with selection in some designs, but it can also be generated by exposure misclassification. Recent causal descriptions therefore emphasize identifying the specific temporal selection or misclassification mechanism rather than relying on the label alone. [8]

Methods Used to Address It

One approach represents exposure as time-varying: each participant contributes unexposed person-time before exposure and exposed person-time afterward. Another is landmark analysis, which chooses a fixed time, includes people still under observation at that point, classifies them using information available by then, and follows outcomes forward. These approaches estimate different quantities and rely on different assumptions; a landmark analysis, for example, changes the eligible population and the start of the outcome analysis. [5] [6]

Target-trial specification addresses the problem at the design stage by making eligibility, assignment to treatment strategies, and the start of follow-up explicit and synchronized. More complex longitudinal strategies may require sequential trials, cloning, weighting, or g-formula methods when exposure rules and confounding change over time. The appropriate method depends on the causal question and available data, so merely naming a Cox model or a target trial is not evidence that timing was handled correctly. [4] [8]

Questions to Ask When Reading a Study

• When does follow-up begin? The paper should state a clear time zero for every comparison group. [4]
• When is exposure first knowable? If exposure can only be identified after time zero, its earlier person-time needs an explicit treatment in the design or analysis. [1] [2]
• Could the outcome occur before exposure? If early outcomes are assigned only to the unexposed group, the comparison is at risk of immortal time bias. [2]
• Is exposure modeled as fixed or time-varying? A baseline label based on later events is a warning sign. [5]
• Are eligibility, group assignment, and follow-up aligned? Misalignment among these elements is a central diagnostic feature. [4] [8]
• Is enough detail reported to reconstruct the timeline? Missing treatment dates, unclear index dates, or unexplained exclusions can conceal the bias. [7]

What This Does Not Mean

• It does not mean every association between a later exposure and longer survival is caused by immortal time bias; the study timeline must be examined. [7]
• It does not mean observational longevity research is inherently invalid; longitudinal data can support appropriate time-aligned analyses. [4] [8]
• It does not mean a time-varying model automatically resolves confounding, selection, measurement error, or reverse causation. [4] [8]
• It does not mean simply deleting the immortal period fixes the problem; selective removal of time can create a different mismatch between groups. [2] [8]

Practical Interpretation Examples

• If “ever users” live longer than “never users”: check whether users were counted as exposed before their first prescription. [1] [2]
• If people reaching a later-life milestone show lower subsequent mortality: check whether survival required to reach that milestone was included in the comparison. [4] [8]
• If a landmark analysis is reported: interpret the result as applying to people alive and eligible at the landmark, not automatically to everyone enrolled at baseline. [6]
• If adjustment includes many health variables: remember that covariate adjustment alone does not correct wrongly assigned person-time. [1] [5]

Related Reading

• Survival Curves and Hazard Ratios in Longevity Research
• Selection Bias and Healthy User Bias in Longevity Studies
• Reverse Causation in Ageing and Longevity Research
• How to Read the Methods and Limitations of a Longevity Study

Summary

Immortal time bias appears when group membership depends on a later event but survival is compared from an earlier starting point. The resulting guaranteed event-free period can make an exposure appear protective because participants had to survive long enough to receive or qualify for it. Reading these studies requires reconstructing the timeline: when eligibility is assessed, when follow-up begins, when exposure becomes known, and where each interval of person-time is assigned. [1] [2] [4] [8]

References

1. Suissa, S. (2008). Immortal time bias in pharmaco-epidemiology. American Journal of Epidemiology. https://pubmed.ncbi.nlm.nih.gov/18056625/
2. Lévesque, L. E., et al. (2010). Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes. BMJ. https://www.bmj.com/content/340/bmj.b5087
3. van Walraven, C., et al. (2004). Time-dependent bias was common in survival analyses published in leading clinical journals. Journal of Clinical Epidemiology. https://pubmed.ncbi.nlm.nih.gov/15358395/
4. Hernán, M. A., et al. (2016). Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. Journal of Clinical Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC5124536/
5. Shintani, A. K., et al. (2009). Immortal time bias in critical care research: application of time-varying Cox regression for observational cohort studies. Critical Care Medicine. https://pubmed.ncbi.nlm.nih.gov/19770737/
6. Giobbie-Hurder, A., et al. (2013). Challenges of guarantee-time bias. Journal of Clinical Oncology. https://pmc.ncbi.nlm.nih.gov/articles/PMC3732313/
7. Vail, E. A., et al. (2021). Attention to Immortal Time Bias in Critical Care Research. American Journal of Respiratory and Critical Care Medicine. https://pubmed.ncbi.nlm.nih.gov/33761299/
8. Hernán, M. A., et al. (2025). A structural description of biases that generate immortal time. Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC11598638/