Time Zero Alignment: When Your Cohort Starts Counting Before Treatment Does
Anas H. Alzahrani, MD PhD MPH
Department of Preventive Medicine and Public Health
Faculty of Medicine, King Abdulaziz University
Real-world evidence papers often fail in a surprisingly primitive way: the clock starts at one moment, treatment is assigned at another, and the analysis pretends those are the same event. Then everyone acts shocked when the treated group looks healthier. The calendar has been doing causal work behind the curtain.
Time zero alignment is the discipline of making three things coincide: when a patient becomes eligible for the comparison, when the treatment strategy is defined, and when follow-up begins. If those clocks drift apart, the analysis inherits bias before the model even opens its eyes.
The Core Design Rule
Define cohort entry at the first moment a patient could reasonably enter either comparison strategy, and start follow-up there unless you can defend an alternative in protocol language rather than after-the-fact convenience.
Decision rule:
If treatment status is determined with information that becomes available after follow-up begins, you do not have a baseline exposure. You have a time-alignment problem that needs explicit design handling.
Immortal time bias is one famous consequence of getting this wrong. It is not the whole story. Misaligned time zero also distorts confounding control, baseline covariate measurement, and the very meaning of the estimand.
The Three Clocks You Need to Align
Eligibility clock
The moment a patient satisfies your inclusion criteria and could plausibly be compared with either strategy.
Assignment clock
The point at which the treatment strategy is defined, whether that means immediate start, delayed start, or no start within a window.
Follow-up clock
The first instant outcomes begin counting toward the effect estimate you plan to report.
Serious causal design tries to make those clocks click together. Weak observational design lets them drift, then hopes regression adjustment will repair the damage. It usually will not.
A Classic Bad Example
Imagine a hospital cohort of patients newly diagnosed with severe infection. The paper defines the treated group as those who receive the antibiotic combination within 14 days of diagnosis. Everyone else is the comparison group. Follow-up for mortality starts on the diagnosis date.
What the methods section says
Exposure was defined as treatment received within 14 days in order to reflect real clinical initiation patterns.
What the design actually did
It required treated patients to survive, remain hospitalized, and stay observable long enough to earn the treated label.
Why reviewers should flinch
Early deaths are pushed into the comparison arm or disappear from treatment qualification entirely, flattering the treatment before biology gets a vote.
That is not a subtle modeling issue. It is a design decision that quietly lets the future decide baseline group membership.
What Goes Wrong When Time Zero Drifts
Immortal time enters through the side door
The treated arm gets event-free waiting time that patients had to accumulate before the treatment actually began.
Baseline covariates become contaminated or stale
Some variables are measured after treatment has effectively started, while others no longer reflect the patient state at the real treatment decision point.
The estimand quietly changes
You may think you estimated immediate treatment versus no treatment, but the design really estimated survival among patients who managed to initiate later.
Confounding control starts at the wrong moment
Adjustment sets built at diagnosis may miss the clinical deterioration, stabilization, or contraindications that shaped delayed treatment initiation.
Interactive time-zero alignment explorer
Let treatment start later and the naive treated group quietly borrows survival from the calendar
This toy model compares two ways of analyzing the same delayed-treatment question. The design-consistent version counts early risk before treatment starts. The naive baseline-treated version simply deletes that early risk from the treated group by requiring patients to survive long enough to qualify.
Untreated comparison risk
46.7%
The cumulative outcome risk across the full follow-up horizon if no one receives treatment.
Design-consistent treated risk
39.2%
Counts early risk before treatment actually starts, then applies the treatment effect only after initiation.
Naive baseline-treated risk
32.9%
Starts by pretending treated patients were already in their post-treatment state at baseline.
| Diagnostic | Expected value | Why it matters |
|---|---|---|
| Patients lost before qualifying as treated | 9.3 per 100 | These are the patients who experienced the outcome before treatment could start and disappear from the naive treated group. |
| Guaranteed event-free days credited to 100 treated patients | 1400 person-days | A baseline-treated analysis gives every treated patient a block of impossible pre-treatment survival credit. |
| Apparent risk ratio from the naive analysis | 0.70 | This is the flattering answer the manuscript reports when the future decides who counts as treated. |
| Risk ratio under aligned time zero | 0.84 | This version still uses a toy model, but at least the calendar is no longer doing causal work in secret. |
How to read the toy model
Increase the treatment delay and the gap between the aligned and naive estimates widens, because more early outcomes get edited out of the treated arm. Increase the pre-treatment daily risk and the same bias grows again, because qualifying as treated becomes a stronger survival test.
The point is not that every real study behaves exactly like this. The point is that once treatment status is defined by surviving into the future, the apparent effect can improve before the treatment itself has done anything useful.
Decision rule
If treatment is identified by what happens during follow-up, assume the design needs explicit time-zero alignment until the authors show otherwise.
If the treated group had to survive, stay event-free, or remain observable before earning its label, a baseline-treated comparison is not a shortcut. It is a bias mechanism.
Common Situations That Need Adult Supervision
| Scenario | What authors often do | Design-consistent move |
|---|---|---|
| Treatment can start days after eligibility | Label patients treated from baseline if they initiate within a future window | Use time-varying treatment coding or emulate baseline strategies with a justified grace-period design |
| Sustained treatment strategies | Compare eventual adherers with non-adherers from time zero | Specify strategies at baseline, then handle deviation with cloning, censoring, and weighting if needed |
| Hospital interventions after stabilization | Ignore that some patients never stabilized long enough to become eligible for treatment | Redefine eligibility at the true decision point or analyze treatment as time-varying from admission onward |
| Registry or claims studies with delayed fills | Start follow-up at diagnosis and classify by prescription filled later | Anchor cohort entry and baseline covariates to the dispensing decision or define trial-compatible initiation strategies |
Reviewer Red Flags
1. Exposure is defined with phrases like “received treatment within 30 days”
That wording usually signals that group membership depends on future information rather than baseline assignment.
2. The paper never names time zero explicitly
If eligibility, assignment, and follow-up start are scattered across paragraphs, the authors may not have aligned them at all.
3. Early outcomes are excluded without a strategy-level justification
Deleting inconvenient early deaths is not clinical realism. It is calendar laundering unless the strategy truly starts later.
4. Baseline covariates are measured in a window that overlaps treatment initiation
Once the covariates and exposure clock overlap, “adjustment” can become post-treatment contamination dressed as rigor.
What Better Design Looks Like
Authors should show
- A one-line definition of time zero tied to eligibility and strategy assignment
- Whether treatment is immediate, time-varying, or defined through baseline strategies with a grace period
- How early outcomes before treatment initiation were handled and why
- Covariate measurement windows anchored to the actual decision point
- Sensitivity analyses that test shorter or longer initiation windows when those windows matter clinically
Reviewers should ask
- Could a patient experience the outcome before becoming eligible for the treated label?
- Does the reported effect compare treatment strategies or survivors who later managed to start treatment?
- Are baseline covariates truly pre-treatment for both groups on the same timeline?
- Would a time-varying exposure analysis or target trial emulation answer the question more honestly?
- Is the paper defending a clinical protocol or merely rationalizing an administratively convenient clock?
The Practical Takeaway
Time zero is not a bookkeeping detail. It is the hinge that decides whether the comparison is about treatment strategies or about which patients lived long enough to look treated. Once the latter slips in, the effect estimate can become impressively precise and fundamentally wrong at the same time.
When the treatment label depends on the future, skepticism is not cynicism. It is basic study-design hygiene.
Use Aqrab when the timeline is doing more persuasion than the treatment
Aqrab is built for exactly this genre of paper: the methods section sounds tidy, the effect estimate looks confident, and the real problem is that the clocks never agreed. If you want structured critique of eligibility definitions, grace periods, time-varying exposure, or reviewer red flags, start at /try. Teams wiring those checks into internal evidence review can explore /developers.
Keep reading
Don't stop at one method.
Good methods judgment comes from contrast. Read the neighboring guides, see where the assumptions diverge, and avoid treating every observational problem like it needs the same hammer.
Channeling Bias: When the Newer Treatment Inherits the Easier Patients
A practical guide to channeling bias for clinical researchers. Covers preferential prescribing, formulary-era drift, specialist selection, and what reviewers should demand before trusting observational comparisons of newer therapies.
Confounding by Contraindication: When the Untreated Group Is Too Fragile for the Therapy
A practical guide to confounding by contraindication for clinical researchers. Covers how treatment avoidance in high-risk patients can make therapies look safer or more effective than they are, and what reviewers should demand instead.
Informative Visit Processes: When Who Shows Up Starts Writing the Results
A practical guide to informative visit processes for clinical researchers. Covers endogenous follow-up, unequal observation schedules, visit-triggered outcome capture, inverse-intensity thinking, and what reviewers should demand before trusting longitudinal real-world results.