Survival Analysis

Applications

Engineering

Censoring vs Truncation

Modelling

Censoring

Truncation

Left-right

Astronomy

Medical Studies

Functions

For modelling, it is often assumed that the censoring event is independent of the time

Definition: Some information is known about a part of the data. Ex: It is known that patient survived up to 180 days, but no information known beyond that since 180 is the censoring time

No information is known about a subset of the data

Circumstances

Interval censoring

Malaria example: we know the date between negative and positive, but not a precise date

May lead to problem (biased models), if this is not the case

Right - Left

Right: The real value for the variable of interest could not be observed, this may be due to various factors

Types

Left: the beginning of the variable of interest may have been before the current estimate/value

May not be the case: transplant studies, patient may leave because they need another treatment, patients may leave because they got better, administrative reasons

S(t)

H(t)

h(t)

Hazard of event at point t

1 - F(t) (probability of death, event, whatever the nature is)

Models

Parametric

Non-Parametric

Hazard function

Type I

Type II

Random

the start times and end times are random

All have the same censoring time

Example: Patients followed in a study, 40 patients are followed from beggining to end

All start at the same time, but censoring happens after x failures

Following 40 lightbulbs, finish experiment once 5 of them have failed

Long studies where the entrance gap of time varies widely, and patients leave due to random reasons or at different times