How Do You Define the Starting Point for Prognosis?

When Should the Clock Start in a Prognostic Study?

You want to investigate what happens to patients after a diagnosis, students after enrollment, or participants after entering a defined condition or state. You have identified the outcome and collected longitudinal data, but an important design decision remains: when should follow-up begin?

Should the starting point be the date of diagnosis, the first consultation, treatment initiation, discharge, or the date the participant entered your database?

These choices may seem administrative, yet they can change the population being studied, the outcomes counted, and the meaning of the resulting prognosis. A study that begins at diagnosis does not necessarily answer the same question as one that begins after treatment or among individuals who have already survived several months.

Choose Time Zero to Match the Prognostic Question and Intended Decision

How the Starting Point Shapes Prognostic Study Design and Interpretation

What Is Time Zero in Prognostic Research?

Time zero is the origin of the follow-up timescale used to investigate subsequent outcomes. It identifies when participants become eligible for the specified prognostic analysis and when their outcome experience begins to contribute to the study.

In a study of survival after diagnosis, time zero may be the date of diagnosis. In a study of readmission after hospital discharge, it may be the discharge date. In educational research, a study of first-year withdrawal may begin on the date students enter the academic program.

The starting point should be defined by an observable and reproducible rule rather than a convenient date chosen after examining the data.

Time zero is also called the index date, cohort entry date, or origin of follow-up, although these terms may have somewhat different meanings in particular study designs.

Its purpose is to establish a coherent temporal relationship between the population, the available information, and the outcome being studied.

Why Does Time Zero Change the Research Question?

Consider a condition for which some patients deteriorate shortly after diagnosis while others remain stable and eventually receive treatment.

A study beginning at diagnosis includes the early period during which deterioration may occur. A study beginning at treatment initiation excludes, or otherwise handles differently, events occurring before treatment.

A study beginning at hospital discharge includes only individuals who reached discharge and met the study's discharge-related eligibility criteria.

These populations have different starting states and may experience different subsequent risks.

Consequently, a six-month outcome probability measured from diagnosis cannot be interpreted as equivalent to a six-month probability measured from discharge.

The time origin defines the question, not merely the position of zero on a statistical graph.

How Do Common Starting Points Compare?

There is no universally correct starting point. The appropriate choice follows from the target population and the decision the study is intended to inform.

What Is an Inception Cohort?

An inception cohort includes participants at a common, meaningful stage in the development or recognition of a condition, such as a new diagnosis or the beginning of a treatment episode.

Its advantage is that participants enter the study at a relatively comparable point in the course being investigated.

For example, a cohort of newly diagnosed patients may provide a more direct estimate of prognosis from diagnosis than a cohort containing a mixture of newly diagnosed patients and long-term survivors.

However, the term inception cohort does not guarantee validity. The starting event must be defined consistently, the recruitment process should be appropriate, and relevant outcomes must be ascertained.

Some conditions have uncertain onset dates, making the date of diagnosis a practical but imperfect proxy for the beginning of the disease process.

Researchers should explain what their chosen inception point represents.

Why Must Eligibility Be Aligned With Time Zero?

Eligibility criteria should establish whether a participant belongs to the target population at the chosen starting point.

Suppose researchers want to estimate mortality after a first diagnosis. They should not require participants to have completed six months of follow-up before being considered eligible for a cohort that supposedly begins at diagnosis.

Such a requirement would exclude people who died during those six months and change the population to survivors who remained observable.

Similarly, requiring future treatment completion when defining a baseline cohort can create selection based on events occurring after time zero.

Eligibility conditions involving future events must be handled through an appropriately defined conditional population or longitudinal analysis rather than being silently imposed on the original baseline cohort.

What Is Immortal Time Bias?

Immortal time bias can arise when a period during which participants must remain alive or event-free to satisfy a later classification is incorrectly allocated or interpreted in the analysis.

For example, imagine classifying patients as having received a treatment if they start it at any point during the first 90 days after diagnosis.

Patients who receive treatment on day 60 must have remained alive and eligible long enough to receive it. If researchers classify them as treated from the original diagnosis date, the preceding period may be incorrectly attributed to the treated group.

This can create an artificial apparent advantage for treatment.

The problem is especially prominent in observational comparisons of treatment effects, but the underlying timing issue is also relevant when prognostic groups are defined using future information.

Researchers should distinguish baseline characteristics from exposures or states that arise during follow-up.

Time-dependent methods, appropriately defined landmark analyses, or redesigned cohort entry rules may address particular timing problems, depending on the question.

Does Every Participant Need the Same Calendar Date?

No. Participants may enter a prognostic cohort on different calendar dates while sharing the same conceptual time origin.

For example, patients diagnosed during a three-year recruitment period can each have time zero defined as their own diagnosis date.

The analysis may then measure elapsed time since diagnosis rather than time since the study's administrative opening date.

However, calendar time can still matter when treatment practices, diagnostic criteria, or background risks change during recruitment.

Researchers may need to account for such changes, particularly in studies spanning substantial periods.

The requirement is conceptual consistency, not identical dates on the calendar.

When Should Baseline Prognostic Factors Be Measured?

Baseline prognostic factors should ordinarily be measured at or before the intended prediction time.

For example, a study estimating risk at hospital admission should use information available at admission if its findings are intended to inform admission decisions.

Laboratory results obtained several days later should not be treated as though they were available at admission.

Doing so can introduce information leakage, particularly when later measurements reflect clinical deterioration or treatment response.

When a factor is measured within a permitted window around time zero, researchers should specify that window and consider whether measurement timing affects interpretation.

The question of whether prognostic factors should be measured before treatment is particularly important when treatment can alter their values.

What If Important Information Becomes Available Only Later?

Information obtained after baseline may still be useful, but the research question must accommodate its timing.

For example, a model intended to predict outcomes on the third day of hospitalization may legitimately use measurements obtained during the first three days, provided it predicts outcomes after the third-day prediction point.

This differs from a model intended to predict the same outcome at admission.

Researchers may use landmark prediction, time-updated covariates, or dynamic prediction methods to incorporate later information appropriately.

These approaches should not be confused with simply adding future measurements to a baseline regression model.

The prediction time and outcome horizon must remain explicit.

What Is a Landmark Analysis?

A landmark analysis selects a prespecified time after an initial event and evaluates subsequent outcomes among participants who remain eligible at that landmark.

For example, researchers might examine one-year prognosis among patients who are alive and event-free 30 days after diagnosis.

The landmark becomes the relevant prediction origin, and the analysis is conditional on reaching that point.

Information available by the landmark may be used to describe prognostic groups or develop predictions for subsequent outcomes.

However, the resulting estimates do not describe prognosis from the original diagnosis date for all initially eligible patients.

Researchers should state this conditional interpretation clearly.

What Is Delayed Entry or Left Truncation?

Delayed entry occurs when participants become observable or enter the risk set after the origin of the timescale being studied.

For example, a registry may begin observing patients some time after their diagnosis while researchers wish to use time since diagnosis as the analysis timescale.

Patients who experienced the event before entering observation may be absent from the available data.

Survival methods accommodating left truncation can handle appropriately recorded delayed entry under assumptions concerning the entry process.

However, these methods do not automatically recover the prognosis of all newly diagnosed patients when the registry contains only selected survivors.

Researchers must distinguish the risk set that can be estimated from the broader population they may wish to describe.

What Happens When Participants Have Already Experienced the Outcome?

The answer depends on whether the outcome can recur and on the intended estimand.

For a study of time to first occurrence, participants who experienced the event before time zero generally cannot be treated as still at risk of their first event.

For recurrent outcomes, such as repeated hospitalizations, a different event-history structure may be appropriate.

Researchers should specify whether the study concerns first events, recurrent events, or transitions between states.

Failure to distinguish these structures can make outcome rates and time-to-event estimates difficult to interpret.

How Does Time Zero Affect Competing Events?

Competing events are defined relative to the outcome and starting state.

For example, death before hospital readmission may prevent a subsequent readmission, while death before discharge means a patient never entered a discharge-based cohort.

These are different situations.

The first is a competing event after time zero; the second concerns eligibility for the chosen starting population.

Researchers should therefore define time zero before deciding how competing events affect the prognostic analysis.

Outcome definitions, competing events, and risk sets should form a coherent description of the process being studied.

How Does Time Zero Affect the Length of Follow-Up?

A follow-up horizon has meaning only relative to a defined origin.

Six-month prognosis after diagnosis is not equivalent to six-month prognosis after treatment initiation, even when both analyses use the same nominal duration.

The appropriate horizon should reflect when the outcome is expected to occur and when the information would be useful for decisions.

Researchers should distinguish elapsed follow-up time, calendar observation periods, and the intended prediction horizon.

These distinctions are central to deciding how long participants should be followed in prognostic research.

How Should Time Zero Be Documented?

The study protocol should state the exact event or rule defining time zero, how its date is determined, and what happens when records contain multiple plausible dates.

Researchers should specify the eligibility criteria assessed at that point, the baseline measurement window, and the start of outcome ascertainment.

For existing datasets, the availability and accuracy of dates should be evaluated rather than assumed.

Where alternative plausible definitions exist, prespecified sensitivity analyses may examine whether conclusions depend materially on the selected origin.

Transparent reporting should allow another researcher to reconstruct cohort entry and follow-up from the documented procedures.

How Diagnosis and Discharge Produce Different Prognostic Questions

Neither starting point is inherently incorrect. They define different populations and different prognostic questions.

The methodological error would be to calculate an outcome proportion using one starting population and interpret it as though it described another.

Common Mistakes When Defining the Starting Point for Prognosis

How to Select a Defensible Time Zero for Your Study

Begin with the decision the prognostic information should support. Identify when the person making that decision would need the estimate and what information would be available at that moment.

Then define the population eligible at that time and establish how outcomes will be measured afterward.

For a protocol, write an explicit timeline identifying eligibility, time zero, baseline measurement, subsequent events, censoring, and the outcome horizon.

For retrospective studies, verify that dates and event definitions can be reconstructed consistently from the available records.

A defensible starting point is one that another researcher could reproduce and that corresponds to the population and future outcomes the study claims to describe.

Before Finalizing Time Zero in a Prognostic Study

Frequently Asked Questions About Time Zero in Prognostic Studies

Time Zero Determines Whose Prognosis You Are Actually Estimating

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.