Time zero, index date and follow-up
Why a date can change the answer
An index date determines what counts as baseline, who enters the study and when outcomes can occur. If treated patients must survive longer to enter than their comparators, the groups start with different opportunities to experience the outcome.
When to use it
Make the timeline explicit for longitudinal cohorts, target-trial emulations and treatment-effect studies. Prediction studies also need a precise prediction time: a value recorded later cannot be treated as if it was available when the decision was made.
When not to impose a follow-up comparison
A cross-sectional prevalence estimate may use a survey reference period without a treatment-assignment date. Do not invent a causal follow-up design when the data only provide a snapshot.
Define the windows before assembling the cohort
- Eligibility: what must be true at entry, and which records establish it?
- Baseline: which prior period measures treatment history and confounders, and how much observation is required?
- Assignment: when are the compared treatment strategies defined without using later outcomes?
- Follow-up: when does outcome counting begin and end, including administrative end dates and loss of observation?
- Events: distinguish the outcome, competing events and censoring; record the assumptions needed for the chosen analysis.
A common timing error
Suppose a mortality study starts at discharge, but the treated group is defined by attending a visit within 30 days. A patient dying before the visit can never enter that group. Counting the preceding days as treated credits survival required to attend the visit to the treatment. The comparison needs a strategy aligned at discharge, or a justified method for treatment that changes over time.
Aligned and misaligned time zero
Alternatives answer different questions
A new-user comparison can align groups at initiation when both treatment options are suitable. A landmark analysis instead starts among those eligible and event-free at a later specified point, changing the target population. Strategies permitting a grace period may require cloning, censoring and weighting or another explicit method; simply dropping early events does not repair the original question.
Causal assumptions and incomplete records
Aligned dates do not remove confounding, guarantee treatment overlap or make censoring independent. Missing exposure dates or eligibility variables can create selection if those records are excluded. State what missingness or censoring assumptions support the proposed method and examine sensitivity to plausible departures.
What to watch out for
Eligibility, treatment assignment and follow-up must share a defensible starting point. Requiring future survival or treatment receipt can create immortal time or selection bias; adding baseline covariates does not repair a misaligned timeline.
Before using this in your workspace
For each longitudinal aim, write a small timeline from the baseline window through assignment to follow-up. Walk through one person treated early, one treated late and one with an early outcome. If their group membership requires knowing the future, revisit the design before fitting a model.
- Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2016.04.014
- Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available (opens in a new tab) · Am J Epidemiol · doi 10.1093/aje/kwv254