VerifiedEvidence: highv1.0.0

Non-Informative Censoring

A form of censoring in which its timing provides no information about a participant's underlying risk, satisfying the assumption standard survival methods require.

Last reviewedDarrin Baines IP Ltd

Concept Architecture

Concept

Theoretically, Non-Informative Censoring is a censoring mechanism in which the probability of an observation being censored is unrelated to the underlying event process or patient prognosis. It is the fundamental assumption underlying conventional survival analysis and ensures that censored individuals are statistically representative of those remaining under observation. In health economics, non-informative censoring is essential for obtaining unbiased estimates of survival, treatment effectiveness and quality-adjusted life-years used in economic evaluation.

Mathematically, non-informative censoring is represented by the statistical independence of the event time and censoring time. If T denotes the event time and C denotes the censoring time, the assumption is expressed as T ? C. Under this assumption, estimators such as the Kaplan?Meier estimator, Cox proportional hazards model and parametric survival models consistently estimate the survival function and treatment effects without requiring adjustment for the censoring mechanism.

In practice, non-informative censoring commonly arises from administrative study closure or random loss to follow-up that is unrelated to disease progression. Investigators assess the plausibility of this assumption by examining reasons for censoring, comparing censored and uncensored participants and performing sensitivity analyses. Health economic models routinely use survival estimates derived under the assumption of non-informative censoring.


Purpose

Used to justify unbiased estimation of survival outcomes by assuming that censoring occurs independently of the underlying event process.


Mathematical Formulae

Primary Formula

T ? C

Supporting Formulae

Survival Function:

S(t) = P(T > t)

Kaplan?Meier estimator:

?(t) = ?(1 ? d? / n?)

Related Mathematical Methods

  • Kaplan?Meier estimation
  • Cox proportional hazards model
  • Parametric survival modelling
  • Log-rank test
  • Survival analysis
  • Sensitivity analysis

Example

A five-year randomised controlled trial concludes according to the planned protocol. Patients who remain alive without experiencing the event at study closure are administratively censored. Because the timing of study closure is unrelated to patient prognosis, the censoring mechanism is considered non-informative, allowing unbiased estimation of survival curves for subsequent cost-effectiveness modelling.


Excel Implementation

FunctionExample FormulaHealth Economics Application
IF=IF(C2="Censored",1,0)Creates a censoring indicator for survival analyses.
COUNTIFS=COUNTIFS(C:C,"Censored")/COUNTA(C:C)Calculates the proportion of censored observations.
FILTER=FILTER(A2:D201,C2:C201="Censored")Identifies censored participants for review.
COUNT=COUNT(B2:B201)Counts observations contributing to survival estimation.

VBA (Optional)

Automate identification and reporting of censored observations before preparation of survival datasets for health economic modelling.


Sources

  • Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. Journal of the American Statistical Association. 1958.
  • Cox DR. Regression Models and Life-Tables. Journal of the Royal Statistical Society Series B. 1972.
  • Klein JP, Moeschberger ML. Survival Analysis: Techniques for Censored and Truncated Data.
  • Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model.
  • Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
  • NICE. Health Technology Evaluation Manual.

Frequently Asked Questions (6)

  • What is non-informative censoring?

    A form of censoring in which its timing provides no information about a participant's underlying risk, satisfying the assumption standard survival methods require.

    Source: Kalbfleisch & Prentice 2002

  • Why does non-informative censoring not bias a survival estimate?

    Censoring is non-informative when the fact and timing of a patient's censoring tell nothing about how likely they were to have the event, so those censored at any point are representative of all still at risk then. Because their departure removes no particular risk group, the survival estimated from the remaining patients stays unbiased. This is the assumption standard survival methods rely on, and it holds for censoring driven by the study calendar rather than by prognosis. Neutral censoring preserves an honest estimate. Collett (2015) describes this assumption.

    Source: Collett 2015

  • Why does non-informative censoring matter?

    Non-informative censoring matters because standard survival methods assume that censoring is unrelated to the event risk, so that those censored represent those still at risk, and this assumption is necessary for unbiased estimation. When censoring is non-informative, removing censored individuals from the risk set at their censoring time does not distort the survival estimate. If the assumption fails, through informative censoring, the estimates can be biased. So non-informative censoring is the condition ensuring valid survival analysis, making the plausibility of this assumption an important consideration when interpreting time-to-event results.

    Source: Kalbfleisch & Prentice 2002

  • What are examples of non-informative censoring?

    Examples of non-informative censoring include administrative censoring, where a participant remains event-free when a study reaches its planned end date, since the censoring is set by the study timeline rather than the participant's prognosis; and censoring from loss of follow-up for reasons unrelated to the outcome, such as a participant relocating for reasons unconnected to their condition. In these cases the censored individuals are representative of those still at risk, so the censoring carries no information about the outcome, satisfying the assumption for valid survival analysis.

    Source: Collett 2015

  • How is non-informative censoring assessed?

    Non-informative censoring cannot be tested directly, since the outcomes of censored individuals after censoring are unobserved, so its plausibility is assessed by reasoning about why censoring occurred: censoring for reasons unrelated to prognosis, such as a study ending, supports the assumption, while censoring related to how patients are doing, such as dropout due to worsening, suggests informative censoring. Comparing characteristics of censored and uncensored individuals and using sensitivity analyses help evaluate the assumption. So its plausibility is judged from the reasons for censoring rather than confirmed empirically.

    Source: Kalbfleisch & Prentice 2002

  • How does non-informative censoring differ from informative censoring?

    Non-informative censoring occurs when the reason observation ends is unrelated to a participant's risk of the event, so censored individuals are representative and standard methods give valid estimates, whereas informative censoring occurs when censoring is related to prognosis, so censored individuals differ in risk and the methods can be biased. Administrative censoring is typically non-informative, while loss to follow-up related to how a patient is doing is informative. So the two differ in whether censoring carries information about the outcome, which determines whether standard survival analysis is valid.

    Source: Kalbfleisch & Prentice 2002

Trust Record

Verified by Dr Darrin Baines

British health economist

Professional identity: darrinbaines.org

Verification date: 17 Nov 2025

Content version: 1.0.0

Canonical Identity

Term code
HE-ES-CTM-057

Stable URI · Machine-readable · Resolvable · CC BY 4.0