Clone-censor-weight analysis
Clone-censor-weight analysis compares strategies by copying eligible people, censoring copies when they deviate, and weighting the remaining follow-up.
Fit
When to use it, and when not to
Use it when
- At time zero, a person’s observed history is compatible with several treatment strategies, including strategies with a grace period.
Not when
- Strategies are already distinct and assigned at baseline. A simpler target-trial analysis may suffice. Read: Target trial emulation
- You cannot measure what predicts strategy deviation and the outcome. Censoring weights cannot remove that unmeasured selection. Read: Sensitivity analyses
Before you start
What you need
- A target-trial protocoltargetAlign eligibility, strategy assignment, time zero, outcome and follow-up.
- Rules for strategy compatibilityDefine grace periods and exactly when a copy first deviates.
- Time-ordered treatment and health recordsUse measured history to model remaining compatible with the assigned strategy.
- Measured history explains informative strategy deviation.How to check: Artificial censoring must be conditionally independent of potential outcomes. Unmeasured reasons for deviation can bias the contrast.
- Some compatible follow-up remains for every relevant history.How to check: Inspect survival of clones and weight tails within each strategy. Sparse support cannot be repaired by duplication.
- Strategies are well-defined and observable.How to check: Specify dose, initiation window and subsequent treatment rules. Do not use future survival to define eligibility.
- Copies of one person remain dependent.How to check: Cluster uncertainty by the original person, or bootstrap original people and repeat cloning and weight estimation.
- Choose a missing-data method because of a stated assumption.How to check: Missing completely at random (MCAR) means gaps depend on neither observed nor unseen data. Missing at random (MAR) means recorded data explain gaps. Under MAR, suitable multiple imputation may help; it is not inherently better. Missing not at random (MNAR) means unseen values still predict gaps. Under plausible MNAR, add a delta or tipping-point sensitivity analysis. Dropping people with missing exposure can cause selection bias.
Plain language
How it works
Clone each eligible person into each compatible strategy. For “start within a grace period,” censor a copy at the deadline if treatment never starts. For “remain untreated,” censor a copy when treatment starts. Outcomes during the grace period count in all copies still compatible at that time.
Weight uncensored follow-up by the inverse probability of remaining compatible, given measured history. Combine with other needed censoring weights and fit the prespecified outcome model. Compare strategy-specific risks or other chosen summaries. Cloning prevents future treatment from defining baseline groups; the causal interpretation still needs exchangeability, positivity and consistency.
Worked example
A worked example
- Setting
- Eligible adults could start treatment within a grace period or remain untreated during it.
- Question
- How are compatible histories represented and weighted?
- Approach
- Copy 100 people into 2 strategies. For one compatible copy, conditional probabilities of remaining uncensored are 0.80 then 0.50.
- Result
- There are 200 initial clones. Remaining uncensored has probability 0.40; the unstabilized censoring weight is 1 / 0.40 = 2.5.
- What it means
- The original sample still has 100 people. A copy contributes only while compatible; weighting addresses measured selection from artificial censoring.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Cloning does not randomize treatment or add independent people. Artificial censoring creates selection that needs valid weights, measured history and support for each strategy.
- Requiring survival through the grace period to enter.Instead: Start follow-up at eligibility. Count early outcomes in every strategy still compatible with the observed history.
- Stopping after cloning and censoring.Instead: Estimate censoring weights and inspect support by strategy.
- Treating copies as independent patients.Instead: Use the original patient identifier for uncertainty.
Inside Mallard
How Mallard uses this
The method registry links this method to its published analysis article and code. You can find it in Learn and the Atlas.
target-trial-clone-censor-weight).analysis.clone_censor_weight, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- How to estimate the effect of treatment duration on survival outcomes using observational data (opens in a new tab) · BMJ · doi 10.1136/bmj.k182
- 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
- De-Mystifying the Clone-Censor-Weight Method for Causal Research Using Observational Data: A Primer for Cancer Researchers (opens in a new tab) · Cancer Med · doi 10.1002/cam4.70461