Concept Architecture
Concept
Informative Censoring is a form of censoring in which the probability of an individual being censored is related to the underlying event process or prognosis. Unlike independent censoring, informative censoring violates a fundamental assumption of conventional survival analysis because censored individuals differ systematically from those remaining under observation. In health economics, informative censoring can bias estimates of survival, treatment effectiveness, quality-adjusted life-years and cost-effectiveness if not appropriately addressed.
Mathematically, informative censoring occurs when the event time and censoring time are statistically dependent. If T denotes the event time and C denotes the censoring time, informative censoring is represented by T ?? C. Standard Kaplan?Meier estimation and Cox proportional hazards modelling no longer provide unbiased estimates under this condition, requiring alternative approaches such as inverse probability of censoring weighting, joint modelling or sensitivity analyses.
In practice, informative censoring is identified by examining the reasons for loss to follow-up, treatment discontinuation or withdrawal and determining whether these are associated with disease severity or prognosis. Statistical methods that explicitly account for the censoring mechanism are applied to reduce bias before survival estimates are incorporated into health economic models.
Purpose
Used to identify and adjust for censoring mechanisms that are related to patient prognosis, thereby improving the validity of survival estimates used in clinical and health economic evaluation.
Mathematical Formulae
Primary Formula
T ?? C
Supporting Formulae
Independent censoring assumption:
T ? C
Inverse Probability of Censoring Weight:
w? = 1 / P(C? > t � X?)
Weighted estimating equation:
?w?U?(?) = 0
Related Mathematical Methods
- Inverse probability of censoring weighting (IPCW)
- Joint modelling
- Kaplan?Meier estimation
- Cox proportional hazards model
- Multiple imputation
- Sensitivity analysis
Example
In a heart failure trial, patients with rapidly worsening disease are more likely to withdraw because of deteriorating health. Since the probability of censoring is related to prognosis, the censoring mechanism is informative. An inverse probability of censoring weighting analysis is performed before estimating survival for a cost-effectiveness model.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| IF | =IF(C2="Censored",1,0) | Creates a censoring indicator for analysis. |
| COUNTIFS | =COUNTIFS(C:C,"Censored")/COUNTA(C:C) | Calculates the proportion of censored observations. |
| LOGEST | =LOGEST(Y2:Y201,X2:X201) | Supports estimation of censoring probabilities in simplified analyses. |
| SUMPRODUCT | =SUMPRODUCT(Weights,Outcomes) | Calculates weighted outcome estimates using IPCW weights. |
VBA (Optional)
Automate estimation of censoring weights and generation of adjusted survival datasets for subsequent health economic modelling.
Sources
- Robins JM, Finkelstein DM. Correcting for Noncompliance and Dependent Censoring in an AIDS Clinical Trial with Inverse Probability of Censoring Weighted Log-Rank Tests.
- Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations.
- Cox DR. Regression Models and Life-Tables.
- Klein JP, Moeschberger ML. Survival Analysis: Techniques for Censored and Truncated Data.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
Related Concepts (2)
Frequently Asked Questions (6)
What is informative censoring?
A form of censoring in which the reason observation ends is related to a participant's underlying risk, violating standard survival analysis assumptions.
Source: Kalbfleisch & Prentice 2002
Why does informative censoring bias survival estimates?
Informative censoring occurs when the reason a patient's observation ends is linked to their risk of the event, as when patients who are deteriorating withdraw from a study. Because those censored differ in prognosis from those who remain, their loss is not neutral, and standard methods, which assume censoring carries no information, then over- or under-estimate survival. It resembles dependent censoring in violating that key assumption. The censoring itself signals something about the outcome. Collett (2015) describes this problem.
Source: Collett 2015
How does informative censoring arise?
Informative censoring arises when censoring is linked to prognosis, for example when patients who are deteriorating withdraw from a study or are lost to follow-up, or when those with worse outcomes are censored for reasons related to their condition. In such cases, the individuals censored are not representative of those still at risk, since their censoring is connected to their likelihood of the event. This dependence between the reason for censoring and the event risk, rather than censoring occurring for unrelated reasons, is what makes the censoring informative.
Source: Kalbfleisch & Prentice 2002
Why is informative censoring a problem?
Informative censoring is a problem because standard survival methods assume censoring is unrelated to the event risk, so that censored individuals represent those still at risk, and when this fails, the methods can give biased estimates, for example overestimating survival if higher-risk individuals are preferentially censored. The bias is systematic and not corrected by standard analysis. Because informative censoring distorts survival estimates in ways that ordinary methods do not account for, recognising and addressing it is important for valid conclusions from time-to-event data.
Source: Collett 2015
How can informative censoring be addressed?
Informative censoring can be addressed by minimising informative loss of follow-up through good study conduct, by measuring and adjusting for factors related to both censoring and the outcome, and by using specialised methods, such as inverse probability of censoring weighting or sensitivity analyses that explore its effect. Where the dependence relates to measured covariates, conditioning on them can reduce bias. Because it cannot always be fully corrected, sensitivity analyses assessing how much informative censoring could affect results are important, alongside efforts to prevent censoring that depends on prognosis.
Source: Kalbfleisch & Prentice 2002
How does informative censoring differ from administrative censoring?
Informative censoring occurs when the reason observation ends is related to a participant's risk of the event, so censored individuals differ in prognosis and the analysis can be biased, whereas administrative censoring occurs because a study reaches its planned end while a participant is event-free, determined by the study timeline and unrelated to prognosis, so it is typically non-informative and unproblematic. Loss to follow-up related to how a patient is doing is informative, while censoring at a study's fixed end is administrative. So the two differ in whether censoring carries information about the outcome.
Source: Kalbfleisch & Prentice 2002
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 14 Nov 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-ES-CTM-043
Stable URI · Machine-readable · Resolvable · CC BY 4.0