Concept Architecture
Concept
Dependent Censoring is a form of informative censoring in which the probability of an observation being censored depends on the underlying event process or the unobserved outcome of interest. It violates the assumption of independent censoring that underpins conventional survival analysis and may introduce bias into estimates of survival, hazard functions and treatment effects. In health economics, dependent censoring is an important methodological consideration because biased survival estimates can lead to inaccurate estimates of life expectancy, quality-adjusted life-years and cost-effectiveness.
Mathematically, dependent censoring occurs when the censoring time and event time are statistically dependent rather than independent. Standard Kaplan?Meier estimation and Cox proportional hazards models assume T ? C, where T denotes event time and C denotes censoring time. When this assumption is violated, specialised methods such as inverse probability of censoring weighting (IPCW), joint modelling or sensitivity analyses are required to obtain unbiased estimates.
In practice, dependent censoring is assessed by examining reasons for censoring, comparing patient characteristics between censored and uncensored individuals and conducting sensitivity analyses. Statistical methods that explicitly model the censoring mechanism or jointly model longitudinal and survival data are commonly applied when informative censoring is suspected. Health economic models frequently use adjusted survival estimates derived from these methods.
Purpose
Used to identify and address informative censoring that may bias survival analyses, thereby improving the validity of clinical effectiveness estimates used in 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)
- Kaplan?Meier estimation
- Cox proportional hazards model
- Joint modelling
- Multiple imputation
- Sensitivity analysis
Example
A cancer trial experiences substantial patient withdrawal because individuals with worsening disease are more likely to discontinue follow-up. Since the probability of censoring depends on prognosis, the censoring mechanism is informative. An IPCW analysis is performed to adjust the survival estimates before they are incorporated into a cost-effectiveness model.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| IF | =IF(C2="Censored",1,0) | Creates a censoring indicator. |
| COUNTIFS | =COUNTIFS(C:C,"Censored")/COUNTA(C:C) | Calculates the censoring proportion. |
| LOGEST | =LOGEST(Y2:Y201,X2:X201) | Supports modelling of censoring probabilities in simplified analyses. |
| SUMPRODUCT | =SUMPRODUCT(Weights,Outcomes) | Calculates weighted summary estimates using IPCW weights. |
VBA (Optional)
Automate calculation of inverse probability censoring weights and generation of adjusted 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.
- Robins JM, Finkelstein DM. Correcting for Noncompliance and Dependent Censoring in an AIDS Clinical Trial with Inverse Probability of Censoring Weighted Log-Rank Tests.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
Related Concepts (2)
Library
Publications
1
Economic Evaluation in Clinical Trials — Glick, Doshi, Sonnad & Polsky, 2nd Edition ed., 2015 (Oxford University Press)
Practical guidance on conducting cost-effectiveness analyses alongside controlled trials, covering trial design, measurement of costs and quality-adjusted life years, handling censored and missing data, and reporting stochastic uncertainty. Volume 4 in the Handbooks in Health Economic Evaluation series.
BookView source →
Frequently Asked Questions (6)
What is dependent censoring?
A survival analysis situation in which the reason for censoring is related to a participant's underlying risk of the event, violating standard assumptions.
Source: Kalbfleisch & Prentice 2002
Why does dependent censoring bias a survival estimate?
Standard survival methods assume that patients who are censored are, at that moment, no more or less likely to have the event than those who remain, so their loss carries no information about risk. Dependent censoring breaks this assumption, because the reason for censoring is tied to the patient's prognosis, as when the sickest patients drop out. Their departure then removes high-risk patients selectively, making survival look better than it is. The censoring itself leaks information about the outcome. Collett (2015) describes this problem.
Source: Collett 2015
How does dependent censoring arise?
Dependent censoring arises when the reason for censoring is linked to a participant's risk of the event, for example when patients who are doing poorly withdraw from a study or are lost to follow-up, or when those with worse prognosis are censored for reasons related to their condition. In such cases, the individuals censored differ systematically in risk from those remaining, so censoring carries information about the outcome. This dependence between censoring and prognosis, rather than censoring occurring for reasons unrelated to risk, is what produces dependent censoring.
Source: Kalbfleisch & Prentice 2002
Why is dependent censoring a problem?
Dependent censoring is a problem because standard survival methods, such as the Kaplan-Meier estimator and the Cox model, assume censoring is independent of the event risk, so that censored individuals are representative of those still at risk. When censoring depends on prognosis, this assumption fails, and the methods can give biased survival estimates, for example overestimating survival if higher-risk individuals are preferentially censored. Because the bias is systematic and not corrected by standard analysis, dependent censoring threatens the validity of survival estimates, making its recognition and handling important.
Source: Collett 2015
How can dependent censoring be addressed?
Dependent censoring can be addressed by trying to prevent 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 account for or explore the effect of dependent censoring. Where the dependence relates to measured covariates, conditioning on them can help. Because it cannot always be fully corrected, sensitivity analyses assessing how much dependent censoring could bias results are important, alongside efforts to minimise informative censoring.
Source: Kalbfleisch & Prentice 2002
How does dependent censoring differ from independent censoring?
Dependent censoring occurs when the reason observation ends is related to a participant's underlying risk of the event, so censored individuals differ in prognosis from those remaining, violating standard assumptions and biasing results, whereas independent censoring occurs for reasons unrelated to risk, so censored individuals are representative and standard methods give valid estimates. Administrative censoring, ending because a study closes, is typically independent, while loss to follow-up related to prognosis is dependent. So the two differ in whether censoring is informative 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: 13 Nov 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/dependent-censoring
- Term code
- HE-ES-CTM-026
Stable URI · Machine-readable · Resolvable · CC BY 4.0