Skip to content
My Glossary Project

Informative Censoring

TL;DR

Section titled “TL;DR”
  • Censoring depends on participant characteristics rather than occurring at random, which can bias results.
  • Biased censoring can lead to overestimation or underestimation of effects if not addressed.
  • Analysts can account for informative censoring using methods such as weighted regression analysis or sensitivity analysis.

Definition

Section titled “Definition”

Informative censoring (also called non-random censoring) occurs when the censoring of data is not random and is influenced by the characteristics of the study population.

Explanation

Section titled “Explanation”

When censoring is related to participant characteristics or outcomes, the remaining observed data no longer represent a random sample of the original population. That dependence can introduce bias into estimates and lead to incorrect conclusions about effects or associations. Because censored observations may differ systematically from those who remain under observation, standard analyses that assume random censoring can produce misleading results.

Examples

Section titled “Examples”

Cancer treatment study

Section titled “Cancer treatment study”

A study on the effectiveness of a new cancer treatment follows patients for a specified time, but some patients do not complete the study because of the severity of their cancer. In this case, censoring is not random: patients with more severe cancer are more likely to be censored, which can lead to an overestimation of the treatment’s effectiveness since the censored patients may have had worse outcomes.

Smoking and heart disease study

Section titled “Smoking and heart disease study”

A study examining the relationship between smoking and heart disease follows a group of individuals for a specified time, but some individuals do not complete the study because they die from heart disease. Here, censoring is not random: smokers are more likely to be censored due to their increased risk of heart disease, which can lead to an underestimation of the relationship between smoking and heart disease because the censored individuals may have had worse outcomes.

Notes or pitfalls

Section titled “Notes or pitfalls”
  • Informative censoring can introduce bias and cause incorrect conclusions if not accounted for.
  • To mitigate bias from non-random censoring, researchers can use methods such as weighted regression analysis or sensitivity analysis.
Section titled “Related terms”
  • Non-random censoring
  • Censoring
  • Weighted regression analysis
  • Sensitivity analysis