생존 분석의 가정

Censoring occurs when the full survival time for an individual is not observed, such as when participants withdraw from a study or the study ends before an event occurs. Understanding censoring is critical because it affects data completeness and the validity of survival estimates. Proper handling of censoring ensures accurate analysis of time-to-event phenomena.

Independent censoring assumes that the reasons for censoring are unrelated to the likelihood of the event of interest. For example, if participants with severe symptoms are more likely to drop out, survival estimates become biased. Ensuring censoring is independent of health status or outcome risk is essential for reliable analysis and valid conclusions.

The Cox proportional hazards assumption states that the hazard ratio between groups remains constant over time. If one group's risk of an event is twice that of another at the study start, this risk ratio must hold throughout. This assumption is fundamental to Cox models and ensures valid comparison of survival between groups.

Stationarity assumes that the probability of an event changing over time does so similarly across all study groups unless explicitly modeled. External factors influencing survival should impact all groups equally unless accounted for. This assumption ensures that observed differences between groups reflect true treatment effects rather than unequal environmental influences.

A clinically relevant event must be well-defined, clear, and observable to enable accurate measurement and analysis. Ambiguous or misclassified events compromise the validity of survival data. Selecting pivotal events ensures that survival analysis produces meaningful results applicable to clinical practice and research questions.

Adequate follow-up duration and sufficient sample size ensure robust event occurrences for reliable statistical power. Short follow-up periods may miss critical events and lead to incomplete conclusions. Careful determination of these design elements prevents biased results and enables valid comparisons when comparing the survival analysis of two or more groups.

Survival times exhibit positive skewness because events tend to occur more frequently early in the study period, with fewer occurrences as time progresses. This distribution differs from the normal distribution assumed in many other statistical analyses. Recognizing this skewness is essential for selecting appropriate survival models and ensuring valid statistical inference.