How Do You Define the Start of Follow-Up in a Cohort Study?

When Should the Clock Start Running in a Cohort Study?

Suppose you are investigating whether an AI literacy program is associated with doctoral student retention. Some students complete the program immediately after enrollment, while others complete it several months later. A few discontinue their studies before they have an opportunity to participate.

Should follow-up begin when students enroll, when they become eligible for training, when they complete the program, or when researchers obtain their records?

These choices are not interchangeable. An incorrectly defined starting point can exclude early outcomes, misclassify exposure, create unequal opportunities to experience the outcome, or introduce systematic bias. Defining the start of follow-up is therefore a fundamental design decision rather than a technical detail to settle during data analysis.

Start Follow-Up When Participants Become Eligible for the Intended Observation or Comparison

How to Establish a Defensible Time Zero in Cohort Research

What Is Time Zero?

Time zero is the defined origin of follow-up for an individual or an analytical comparison. It establishes when observation begins for the outcome under investigation.

For example, if researchers study doctoral program completion, time zero might be the date a student first enrolls in the program. Follow-up then concerns outcomes occurring after that starting point.

In a study of outcomes after an educational intervention, the relevant origin may instead be the point at which eligible students are assigned to intervention strategies or become eligible to initiate them.

These choices correspond to different research questions. The first asks what happens after program entry. The second may ask what happens after a defined intervention decision.

A study should therefore specify its time origin before estimating incidence, survival probabilities, or exposure-related differences.

Why Is the Start of Follow-Up So Important?

Time zero determines which participants are eligible, which events count, and how much observation time each participant contributes.

Suppose two studies use the same administrative records but different starting points. One follows students from initial enrollment, while another begins only after students complete their first semester.

The second study excludes events occurring before the first semester ends and may select a group that has already demonstrated an ability to remain enrolled.

Both studies may answer legitimate questions, but their target populations and estimands differ. Problems arise when the later-starting study is interpreted as though it described outcomes from initial enrollment.

Time zero also affects the composition of risk sets, the interpretation of hazard estimates, and the classification of early events.

What Conditions Should Be Satisfied at Time Zero?

For a conventional first-event cohort analysis, participants should ordinarily satisfy the study's eligibility criteria and be at risk of the defined outcome when their follow-up begins.

For a causal comparison of intervention strategies, the requirements are more specific. The target-trial framework emphasizes aligning eligibility, assignment to the strategies being compared, and the beginning of follow-up.

This alignment helps prevent biases caused by selecting participants or classifying exposure groups using information that becomes available only after observation has already begun.

Researchers should ask whether:

• Participants satisfy the specified eligibility criteria at the proposed starting point.
• The outcome has not already occurred when studying first-event incidence.
• The relevant exposure or intervention strategies are defined consistently with the intended comparison.
• Participants can be observed for outcomes from the proposed starting point.
• Information occurring after time zero is being used appropriately rather than incorrectly determining baseline eligibility or exposure status.

Not every cohort analysis involves treatment strategies. Descriptive incidence studies may use an eligibility event, such as enrollment or employment commencement, as their time origin without defining an intervention comparison.

Is Time Zero the Same as the Baseline Measurement Date?

Often, but not necessarily.

Baseline usually refers to the reference point for initial measurements or characteristics. In many studies, baseline assessment occurs at or near time zero.

However, some baseline information may have been measured before cohort entry. For example, a university might use an admissions assessment completed several weeks before enrollment as a baseline predictor.

That measurement does not necessarily mean follow-up should begin on the assessment date.

Similarly, administrative records may be extracted years after a historical cohort began. The extraction date is not the historical start of follow-up.

The key requirement is that baseline variables are interpreted according to when they were actually measured and whether they precede the relevant outcome and exposure decisions.

How Does Time Zero Differ From the Start of Exposure?

Time zero and exposure initiation may coincide, but they need not always do so.

Consider a cohort of students eligible to begin an AI literacy program. The study may compare strategies such as initiating training within a specified period versus not initiating within that period.

Eligibility and follow-up could begin when students first become eligible, even though actual training may start later.

If researchers instead classify students according to whether they eventually complete training and then assign all earlier observation time to the trained group, they risk using future information to define baseline exposure.

The appropriate solution depends on the causal question. Possible approaches include defining exposure at baseline, using time-varying exposure methods, or specifying intervention strategies that permit a defined initiation period.

These methods are not interchangeable, and some require additional assumptions or specialized analyses.

What Is Immortal Time Bias?

Immortal time bias is a form of bias that can arise when a period during which an individual must remain event-free to satisfy a later classification is incorrectly handled in the study design or analysis.

It commonly occurs when researchers use future exposure information to classify participants into baseline groups.

For example, suppose students are classified as trained if they complete an AI literacy program at any time during their first academic year. Students who discontinue before completing training cannot qualify as trained under that definition.

If the analysis begins at enrollment but classifies eventual completers as trained from the outset, their pre-completion time is incorrectly attributed to the trained group. This can create an artificial advantage for training because the students had to remain enrolled long enough to complete it.

Research on target-trial emulation identifies misalignment among eligibility, intervention classification, and follow-up as a central source of such bias.

Importantly, immortal time bias is not an unavoidable feature of delayed treatment initiation. It arises from inappropriate selection, classification, or allocation of observation time. Hernán and colleagues' methodological analysis explains why synchronizing these design elements is essential.

Can You See How Incorrect Time Zero Distorts a Comparison?

Consider a hypothetical cohort of students eligible for an intervention at enrollment. Some students begin training after several months, while others never receive it.

If the study begins counting outcomes at enrollment but defines the trained group using eventual completion, students who withdraw early may be placed disproportionately in the untrained group.

Even if training has no beneficial effect, this classification can produce a misleading association between training and retention.

A defensible analysis might compare students who initiate training at a clearly defined eligibility point, model training as a time-varying exposure, or emulate specified initiation strategies using appropriate methods.

Simply moving the trained group's starting date to training completion while retaining enrollment as the untrained group's starting date is not necessarily a valid correction. The groups may then have different eligibility histories and opportunities to experience the outcome.

The goal is to establish a coherent comparison, not merely to eliminate inconvenient early observations.

What Happens When Participants Enter at Different Times?

In many cohorts, individuals become eligible on different calendar dates.

For example, doctoral students may enroll in different semesters. Their individual time zero could be their respective enrollment dates.

Researchers may analyze outcomes using time since enrollment as the underlying time scale, while also considering calendar period if institutional conditions change.

Alternatively, an analysis may use calendar time as its time scale and account for participants entering the risk set only after they become eligible.

These choices affect the interpretation of follow-up and the construction of risk sets.

The broader issue of staggered entry into cohort follow-up therefore requires attention to both individual entry dates and the selected analytical time scale.

What Is Delayed Entry or Left Truncation?

Delayed entry occurs when participants begin contributing to the observed risk set after the origin of the analytical time scale.

For example, researchers may use age as the time scale in a cohort study. A participant first observed at age 35 should not contribute observed risk time at ages 20 through 34.

Appropriate left-truncated survival methods allow the participant to enter the risk set at the observed entry age.

However, delayed entry methods generally require assumptions about the relationship between entry and the outcome process, conditional on relevant variables. They cannot automatically correct selection caused by observing only individuals who survived or remained event-free until entry.

It is also important to distinguish delayed entry from ordinary staggered recruitment. Participants entering on different calendar dates do not necessarily require left-truncated analysis when each is followed from their own properly defined time zero using time since entry.

When Does Follow-Up End?

Defining time zero also requires specifying the rules for ending observation.

Follow-up may end when a participant experiences the outcome, reaches the administrative end of observation, is lost to follow-up, or experiences an event that changes their ability to experience the outcome.

The correct classification depends on the estimand.

For example, if the outcome is doctoral withdrawal without completion, graduation may be a competing event rather than ordinary censoring.

If the outcome is departure from a university for any reason, graduation could instead be part of the composite outcome.

Researchers should define these rules before analysis and avoid treating every form of departure as equivalent.

How Is Follow-Up Time Calculated?

For a participant whose observation begins at a defined time zero and ends at an event or censoring date, observed follow-up duration is the elapsed time between those points.

When calculating person-time, researchers should ensure that only qualifying time at risk is included in the denominator.

The length of observation is not necessarily identical to the time during which a participant remains eligible to experience a particular event.

Should Time Zero Be the Same for Every Outcome?

Not necessarily. A cohort may examine multiple outcomes with different eligibility conditions or meaningful starting points.

For example, the risk of first-year academic withdrawal might be measured from enrollment, while the occurrence of a professional licensing examination result may be relevant only after students become eligible to sit the examination.

Using enrollment as the starting point for every outcome could introduce inappropriate observation periods for some analyses.

Researchers should define the relevant time origin for each outcome or estimand, while maintaining comparability within each intended contrast.

The possibility of examining multiple outcomes in one cohort does not remove the need for outcome-specific risk definitions.

How Should Time Zero Be Reported?

The methods section should specify the eligibility criteria, the event or date defining cohort entry, the relationship between time zero and exposure classification, and the rules for ending follow-up.

Where participants enter at different times, researchers should describe the entry process and relevant time scales.

Where exposure can begin after eligibility, the analysis should explain how that timing was represented and whether intervention strategies, time-varying exposures, or other methods were used.

The STROBE reporting recommendations call for clear descriptions of relevant recruitment, exposure, follow-up, and data collection periods, as well as methods of participant selection and follow-up.

Transparent reporting helps readers determine whether the study's risk sets and exposure comparisons correspond to its stated research question.

Defining Follow-Up for AI Literacy Training and Doctoral Retention

Why Would a Simpler Classification Be Misleading?

Suppose the researchers classified everyone who eventually completed training as exposed from enrollment and everyone else as unexposed.

Students who withdrew before having an opportunity to complete training would necessarily appear among the noncompleters. Meanwhile, eventual completers would have demonstrated sufficient persistence to reach completion.

This could make training appear protective even if the difference were partly or entirely produced by the classification procedure.

The methodological solution is to define eligibility, exposure strategies, time zero, and outcome observation as a coherent system before estimating the association.

Common Errors in Defining Cohort Follow-Up

How to Choose a Defensible Starting Point for Follow-Up

Begin by stating exactly which outcome or exposure contrast you want to estimate and identifying when participants first satisfy the relevant eligibility conditions.

For a descriptive cohort, an event such as enrollment may provide a natural starting point. For a causal intervention comparison, the relationship among eligibility, exposure strategies, and follow-up requires more explicit design work.

Before extracting or analyzing records, construct a timeline showing when eligibility begins, when exposure is measured or initiated, when outcomes can occur, and when observation ends.

For studies involving changing exposures, the handling of exposure changes during follow-up should be planned alongside the definition of time zero.

For studies involving different participant entry dates, determine whether the question concerns time since entry, calendar time, age, or another relevant time scale.

The practical objective is to ensure that participants are compared from appropriate starting points and that no group receives an artificial advantage through the way observation time is allocated.

Before Starting Cohort Follow-Up

Frequently Asked Questions About Cohort Time Zero

Time Zero Must Match the Question and the Comparison

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.