State the population selected by a landmark analysis

Name the landmark question

A landmark analysis evaluates outcomes after a fixed time among people who meet specified conditions at that time. It can help analyse information acquired during follow-up without treating that information as if it were known at baseline. However, it changes the population and the question.

The survival package time-dependent-covariate vignette explains the importance of using information available at the relevant time and avoiding future information in risk comparisons. Before using a landmark, state why the selected time is scientifically meaningful and which events or exclusions determine eligibility there.

Freeze the classification window

Specify the landmark relative to the common origin. Define which measurements or exposure records are available by that point and how missing information is handled. Start outcome follow-up at the landmark for included participants, using a consistent time scale.

Do not classify exposure using events after the landmark. Document how many participants were excluded because they died, had the outcome, left observation or lacked required information beforehand. These exclusions are part of the design's interpretation, not merely data cleaning.

Hypothetical example

A fictional study classifies response during the first 90 days and compares subsequent outcomes among people alive and under observation at day 90. The resulting comparison applies to that selected day-90 population. It does not estimate an effect among everyone present at the original baseline, including those who died earlier.

Moving the landmark to day 180 would create another selected population and another classification window. A difference between the analyses may reflect selection as well as additional information about response.

Evaluate the remaining bias risks

A landmark can prevent one form of immortal-time misclassification, but it does not automatically remove confounding or selection bias. Response or treatment status at the landmark may be related to prognosis, prior care and survival to that point. Record the causal or associational interpretation and the adjustment rationale explicitly.

Prespecify the main landmark and label alternative times as sensitivity analyses when justified. Report the original cohort, landmark-eligible cohort and subsequent follow-up separately. Use the time-zero alignment worksheet to draw both clocks and the collider-selection worksheet to consider conditioning on survival or observation. The conclusion should name the landmark population rather than allowing readers to assume that the result describes the full baseline cohort.

Source verification: 6 November 2026. Examples are hypothetical.