Survival Analysis and Time-to-Event Methods
Survival analysis is the branch of statistics concerned with the time until an event of interest occurs — death, relapse, recovery, device failure, or any other clearly defined endpoint. Its distinguishing feature is that for some subjects the event has not happened by the end of observation, so their event times are only partially known (censored). The field develops methods that use this incomplete information correctly rather than discarding it.
Definition
Survival analysis comprises statistical methods for analysing the expected duration of time until one or more events occur, accommodating censored observations in which the event time is known only to exceed (or fall within) some interval.
Scope
This area orients the reader to the core ideas that unify time-to-event methods: the survival and hazard functions, censoring and follow-up, nonparametric estimation of survival curves, comparison of groups, and regression modelling of the hazard. It links to the detailed topics — censoring and follow-up data, Kaplan-Meier curves, the proportional hazards assumption, Cox regression, and competing risks — and treats them as methodological reference material rather than clinical guidance.
Sub-topics
Core questions
- How long until an event of interest occurs, and how is that distribution described by the survival and hazard functions?
- How can censored observations contribute information without biasing the analysis?
- How are survival curves estimated and compared between groups?
- How is the effect of covariates on the rate of events modelled, and what assumptions does that modelling require?
- What changes when more than one type of event can occur (competing risks)?
Key concepts
- Survival function S(t)
- Hazard function and cumulative hazard
- Censoring and truncation
- Risk set
- Nonparametric estimation (Kaplan-Meier)
- Log-rank comparison
- Proportional hazards regression
- Competing risks and cumulative incidence
Mechanisms
Time-to-event data are described by the survival function S(t), the probability of being event-free beyond time t, and equivalently by the hazard function, the instantaneous rate of the event among those still at risk. Because follow-up is finite and subjects enter and leave observation at different times, the data are typically right-censored: a subject's event time is known only to exceed their last observed time. Methods such as the Kaplan-Meier estimator and the Cox proportional hazards model are built on the risk set — the subjects under observation and event-free just before each event time — so that each event contributes only the information that is actually available. This treatment of censored and time-varying follow-up is what separates survival analysis from ordinary regression of a continuous outcome (Clark et al., 2003; Leung et al., 1997).
Clinical relevance
Time-to-event methods underlie most reporting of prognosis and treatment effect in clinical research, including survival curves, hazard ratios, and median survival. Understanding them supports critical appraisal of how such evidence is generated; the area is descriptive of analytic methods and is not a source of diagnostic or treatment recommendations.
Epidemiology
Survival methods are pervasive in oncology, cardiology, infectious disease, transplantation, and public-health cohort studies, wherever the timing of an event — not merely whether it occurred — is informative. Their adoption grew rapidly after the Kaplan-Meier estimator (1958) and Cox regression (1972) provided practical tools for censored data.
Evidence & guidelines
There are no clinical practice guidelines for survival analysis itself; the methodological reference standards are seminal statistical papers and biostatistics texts. The Kaplan-Meier estimator (Kaplan & Meier, 1958) and Cox's proportional hazards model (Cox, 1972) are the foundational methods, with tutorials and textbooks (Clark et al., 2003; Collett, 2015; Putter et al., 2007) consolidating practice for medical research.
History
Actuarial life-table methods predate the field by centuries, but modern survival analysis took shape in the mid-twentieth century. Kaplan and Meier's 1958 product-limit estimator gave a rigorous nonparametric survival curve for censored data; the log-rank family of tests followed for group comparison; and Cox's 1972 proportional hazards model brought covariate-adjusted regression to time-to-event outcomes without specifying the baseline hazard. Later work on competing risks and multi-state models extended the framework to settings with several event types (Putter et al., 2007).
Key figures
- Edward L. Kaplan
- Paul Meier
- David R. Cox
- Nathan Mantel
Related topics
Seminal works
- kaplan-meier-1958
- cox-1972
Frequently asked questions
- How does survival analysis differ from ordinary regression?
- It models time until an event while correctly handling censored observations, where the event has not yet occurred when observation ends; such partial information cannot be accommodated by standard regression of a continuous outcome.
- What are the two functions that describe time-to-event data?
- The survival function S(t), the probability of remaining event-free beyond time t, and the hazard function, the instantaneous event rate among those still at risk; either fully determines the other.