03 · What You Need to Know
How Staggered Entry Affects Cohort Membership, Follow-Up, and Analysis
What Is Staggered Entry in a Cohort Study?
Staggered entry occurs when participants begin cohort follow-up at different times rather than sharing one calendar starting date.
For example, a university may include all students entering a doctoral program between 2026 and 2030. Each student's enrollment date could serve as the individual start of follow-up.
The resulting cohort contains participants who begin contributing observation time during different calendar periods.
Staggered entry is common in studies using administrative records, institutional enrollment databases, employment registries, and other systems in which eligible individuals enter continuously or periodically.
It is not inherently a methodological weakness. Problems arise when differences in entry timing are ignored or when participants are incorrectly included in risk sets before they become eligible.
Is Staggered Entry the Same as an Open Cohort?
Not exactly. The concepts are related but describe different features of a study.
An open cohort permits new eligible individuals to join during the observation period. Staggered entry describes the fact that participants begin observation at different times.
Open cohorts commonly have staggered entry, but closed cohorts may also contain participants with different starting dates.
For example, a closed cohort consisting of everyone who enrolled during the 2026 academic year may include students who began in January and August. No students from subsequent academic years are admitted, but individual entry dates still differ.
The distinction between open and closed cohort membership should therefore not be confused with the analytical treatment of individual follow-up dates.
Why Does Staggered Entry Create Unequal Follow-Up?
When a study has a common administrative closing date, participants entering later generally have less potential observation time.
Suppose a cohort database closes on December 31, 2030. A student entering in January 2026 could contribute almost five years of follow-up. A student entering in January 2029 could contribute approximately two years.
Even if neither student experiences the outcome, their observed follow-up durations differ.
This matters because the probability of experiencing an outcome often increases with the amount of time under observation.
A student observed for five years has more opportunity to experience doctoral withdrawal than a student observed for six months, although the actual event process may be more complex.
Researchers should not interpret differences in crude event proportions as though all participants had equal opportunity to experience the outcome.
How Should Time Zero Be Defined?
Time zero is the beginning of follow-up for the relevant participant or comparison.
In a study of doctoral attrition after enrollment, each student's initial enrollment date may provide a defensible time origin.
In a study of outcomes following eligibility for an intervention, the relevant origin may instead be the date the student first becomes eligible for the specified exposure strategies.
The starting point should be determined by the research question, not merely by the first date appearing in the dataset.
For causal comparisons, eligibility, exposure strategies, and follow-up must be appropriately aligned to avoid time-related selection and exposure-classification bias.
The principles of defining the start of cohort follow-up remain essential even when every participant has a different calendar entry date.
What Is the Difference Between Calendar Time and Time Since Entry?
Cohort studies can organize follow-up using different time scales.
Calendar time refers to actual dates or historical periods, such as January 2026 or the 2028 academic year.
Time since entry refers to elapsed time from each participant's individual starting point, such as six months or two years after enrollment.
These time scales answer different questions.
Calendar time
Useful for investigating changes associated with historical periods, institutional policies, technology availability, or other conditions occurring at particular dates.
Time since cohort entry
Useful for examining how outcome risk develops as participants progress through a defined period after becoming eligible, such as months since doctoral enrollment.
For example, doctoral withdrawal risk may be highest during particular stages of a program. That pattern concerns time since enrollment.
However, a university-wide policy introduced in 2028 may influence students regardless of when they enrolled. That pattern concerns calendar time.
Some studies need to account for both time scales. The appropriate analytical approach depends on the outcome process, data structure, and research question.
How Is Person-Time Calculated With Staggered Entry?
Participants contribute person-time only during periods when they are eligible, under observation, and at risk of the specified outcome.
For a simple first-event analysis, observed time commonly begins at the participant's valid time zero and ends at the event, censoring, or administrative closing date.
The calculation assumes that the specified intervals qualify as time at risk. Competing events, prior outcomes, and changes in eligibility may require different handling.
Person-time methods can accommodate unequal observation, but they do not automatically correct selection bias or confounding.
Can You Compare Crude Event Proportions Across Entry Groups?
Such comparisons can be misleading when entry groups have different follow-up durations.
Suppose 20% of students entering in 2026 have withdrawn by the end of 2030, while 8% of students entering in 2029 have withdrawn by the same date.
The earlier group had substantially more time to experience withdrawal.
The difference may therefore reflect observation duration rather than a genuinely lower withdrawal risk among later entrants.
A more meaningful comparison might estimate withdrawal risk within a common period after enrollment, such as 12 or 24 months, provided the available data support that horizon.
Alternatively, time-to-event analyses may use unequal follow-up appropriately, with relevant assumptions about censoring and the event process.
The intended estimand should determine whether the analysis compares fixed-horizon risks, incidence rates, hazards, or another quantity.
Can Kaplan–Meier Analysis Handle Staggered Entry?
Yes, when participants are followed from their own valid time zero and the analysis uses time since entry as its time scale, conventional Kaplan–Meier estimation can accommodate participants who enter on different calendar dates.
Each participant contributes to the risk set after their individual entry, expressed as elapsed follow-up time. Administrative censoring may produce unequal observation durations.
However, the validity of Kaplan–Meier estimates depends on assumptions including appropriate handling of censoring.
In particular, censoring should not convey unaccounted information about future event risk under the assumptions of the analysis.
Kaplan–Meier methods also require care when competing events occur. Treating competing events as ordinary censoring does not generally estimate the real-world cumulative incidence of the event of interest.
Researchers should not assume that every staggered-entry problem can be resolved through a standard Kaplan–Meier procedure without considering the underlying time scale and observation mechanism.
What Is Delayed Entry or Left Truncation?
Delayed entry, often associated with left truncation, occurs when participants become observable only after the origin of the analytical time scale.
For example, suppose researchers study the time until doctoral withdrawal measured from initial enrollment, but their database includes only students who were still enrolled when an electronic record system began operating two years later.
Students who withdrew before the electronic system began may be absent from the observed cohort.
The included participants necessarily survived in the relevant state long enough to become observable. This creates a selection process that must be considered.
Left-truncated survival methods can construct risk sets so that participants contribute observation only after their observed entry time.
However, valid estimation generally requires assumptions about the entry mechanism and its relationship with the outcome process.
Left truncation is not identical to simply enrolling different participants on different calendar dates. The distinction depends on the analytical time origin and when participants become observable relative to it.
How Do Risk Sets Work With Delayed Entry?
A risk set contains the participants eligible to experience the outcome and under observation at a particular analytical time.
With delayed entry, a participant should not contribute to the observed risk set before their entry time.
For example, suppose age is used as the time scale in a cohort study. A participant who becomes observable at age 40 should not contribute observed person-time to risk sets at ages 25 or 30.
The participant enters the risk set at the appropriate observed entry age, assuming the relevant eligibility and event-free conditions are satisfied.
In survival analysis, this can be represented using start–stop intervals or other methods supporting left-truncated observations.
The precise implementation depends on the statistical model and software.
Correct risk-set construction is necessary, but it does not by itself establish that the observed participants represent everyone who would have belonged to the target population at earlier times.
What Is the Difference Between Staggered Entry and Immortal Time Bias?
Staggered entry is a feature of the observation structure. Immortal time bias is a potential analytical bias.
Staggered entry does not automatically produce immortal time bias.
However, bias may arise if researchers incorrectly assign exposure or observation time before participants meet eligibility conditions, or classify participants according to future events that require remaining outcome-free.
For example, students who eventually complete an AI literacy program should not automatically be classified as trained from enrollment if training completion occurs months later.
The problem concerns misalignment of eligibility, exposure classification, and follow-up rather than the mere existence of different calendar entry dates.
Researchers should examine the implications of exposures changing during follow-up separately from the mechanics of staggered cohort entry.
Can Entry Timing Be Associated With the Outcome?
Yes. Participants entering at different calendar periods may experience different environments, policies, resources, or external conditions.
For example, students entering a doctoral program after a university introduces an AI-supported advising system may receive different academic support from earlier entrants.
Differences in withdrawal outcomes could reflect these institutional changes rather than entry timing itself.
Researchers may need to account for entry cohort, calendar period, or other relevant contextual variables.
However, simply adjusting for calendar year does not automatically eliminate confounding, especially when entry period is closely related to exposure availability or other institutional changes.
The appropriate comparison depends on the causal or descriptive question and the available variation in the data.
What If Participants Enter at Different Stages of the Outcome Process?
Not all entrants are necessarily at the same stage of risk.
Suppose researchers recruit doctoral students currently enrolled in a university rather than students entering their programs for the first time.
Some participants may be in their first semester, while others may already be completing dissertations.
Using recruitment date as time zero would describe outcomes after recruitment, but it would not automatically estimate withdrawal risk from initial doctoral enrollment.
If the intended question concerns duration since enrollment, researchers may need historical enrollment dates and an appropriate method for handling students who entered observation after surviving earlier program stages.
Alternatively, the research question could be framed explicitly around outcomes among students enrolled at the recruitment date.
Both approaches can be legitimate, but they target different populations and estimands.
How Does Administrative Censoring Affect Later Entrants?
Administrative censoring occurs when follow-up ends at a predefined study closing date.
Later entrants often have shorter potential follow-up before this date.
For example, a student entering in 2029 cannot contribute a complete three-year follow-up period to a database closing in 2030.
The student should not be classified as having remained free of withdrawal for three years simply because no withdrawal was recorded before the database closed.
Appropriate time-to-event methods can account for the shorter observed period under relevant censoring assumptions.
However, estimates at longer horizons may depend heavily on earlier entrants if later participants have not yet accumulated sufficient observation time.
The implications for choosing and reporting cohort follow-up duration should therefore be considered before defining the study's closing date.
Can Entry Timing Affect Generalizability?
Yes. A cohort assembled over several years may contain participants exposed to different recruitment practices, eligibility criteria, or institutional conditions.
For example, changes in university admissions requirements may alter the characteristics of successive doctoral student intakes.
The combined cohort may represent a broader historical population, but its overall estimates may not describe any one entering class accurately.
Researchers should clarify whether their intended population consists of all eligible entrants during the observation period, one particular entry cohort, or a standardized population.
Subgroup analyses or standardization may sometimes be appropriate, but they should follow the research question and be supported by adequate data.
How Should Staggered Entry Be Reported?
Researchers should report the cohort entry period, individual time-zero definition, administrative closing date, and the distribution of follow-up durations.
They should explain whether the analysis uses calendar time, time since entry, age, or another time scale.
Where delayed entry or left truncation is relevant, the methods should describe how participants enter the risk sets and the assumptions underlying that approach.
Researchers should also report the numbers at risk across meaningful follow-up times, censoring patterns, and relevant differences between entry groups.
The STROBE reporting recommendations emphasize clear descriptions of participant selection, relevant dates, follow-up, and statistical methods. These details are particularly important when cohort membership and observation time vary across participants.
