Scribd
Upload
17 views

Chapter 7 Models For Survival Data: (I) Motivation

1. The document introduces survival analysis and concepts like survival functions, hazard functions, and censoring. It provides an example of survival data from a clinical study with censoring. 2. Key relationships between the survival function S(t), hazard function λ(t), and density function f(t) are defined. The hazard function represents the instantaneous failure rate. 3. Common distributions for the hazard function are presented, including the exponential distribution for constant hazard, the Weibull distribution for monotonic hazards, and a "bathtub" hazard function.

Uploaded by

juntujuntu
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
17 views

Chapter 7 Models For Survival Data: (I) Motivation

1. The document introduces survival analysis and concepts like survival functions, hazard functions, and censoring. It provides an example of survival data from a clinical study with censoring. 2. Key relationships between the survival function S(t), hazard function λ(t), and density function f(t) are defined. The hazard function represents the instantaneous failure rate. 3. Common distributions for the hazard function are presented, including the exponential distribution for constant hazard, the Weibull distribution for monotonic hazards, and a "bathtub" hazard function.

Uploaded by

juntujuntu
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Chapter 7 Models for Survival Data

7.1 Introduction
(I) Motivation
Motivating example:

There are 8 patients (I~VIII) in a 12-months clinical study (D: death;

L: loss information).
I II III IV V VI VII VIII 11 (D) 2 7 (D) 7 (L) 2 2.5 3 10 (D) 6 (D) 9 (L)

12

In this data, the exact failure times of some patients are unknown either because the patients withdraw from the study or because the patients were still alive at the end of the study. We refer the above situation as censoring. Censoring is so common in medical experiments that the statistical methods must allow for it.
1

For the above example, we have the following data:


6

9.5

10

11

where * stands for censoring. Further, if we have the following information about these patients: Subject I II III IV V VI VII VIII Survival Censor time indicator 6 1 7 0 9.5 0 7 1 10 0 7 1 6 0 11 1 Group T T T T C C C C # of cigarette 20 30 5 40 3 40 60 10 Gender 1 1 0 1 0 0 1 0 Age 45 20 38 26 42 17 25 29

where Censor indicator (1: death; 0: censoring) and Group (T:

treatment group; C: control group). Objective:


For the above typical survival data, we are concerned with 1. the survival function 2. the comparison of two groups of survival data 3. which factors (# of cigarette, gender or age) are important in deciding the failure rate

(II) Survival functions and hazard functions


Let T be the positive random variable representing failure time. Then,

S (t ) = P (T t ) =
2

f ( x )dx

is called the survival function, where density function. The hazard function is

f (t ) is the probability

(t ) = lim 0

P (t T < t + | T t ) + 1 P (t T < t + ) = lim + 0 P (T t ) 1 = lim + 0

t+

f ( x )dx S (t ) f ( x )dx

1 = lim + S (t ) 0 f (t ) = S (t )
The relationship among

t+

(t ), f (t ), S (t )

1.

(t ) =

d log [S (t )] . dt

2.

S (t ) = exp

t 0

( x )dx

3.

f (t ) = (t ) exp

t 0

( x )dx

[derivation:]

1.

d log [S (t )] 1 dS (t ) = dt S (t ) dt
3

d 1 = S (t ) =

f dt

( x )dx

= = =
2. By 1,

d 1 t f ( x )dx 0 1 S (t ) dt 1 ( 1 ) f (t ) S (t ) f (t ) S (t ) (t )

d log [S (t )] dt t t d log [S ( x )] ( x )dx = dx = 0 0 dx

(t ) =

d log [S ( x )]

= log [S (t )] | t0 = log [S (t )] log [S (0 )] = log [S (t )] log [P (T 0 )] = log [S (t )] log [1 ] = log [S (t )]

S (t ) = exp
3. By 2, since

t 0

( x )dx

(t ) =

f (t ) S (t ) , then

f (t ) = (t )S (t ) = (t ) exp
Example:

t 0

( x )dx

(i) Constant hazards:

(t ) =
4

hazard function

Then,

S (t ) = exp
and

( x )dx = exp

dx = exp ( t )

t t f (t ) = (t ) exp (x )dx = exp dx = exp ( t ) 0 0

is the exponential distribution.

(ii)

(t ) = p ( t ) p 1

hazard function

p>1

p=1 p<1

Then,

S (t ) = exp = exp
and

0 t

( x )dx = exp

p 1

p ( x )
t 0

p 1

p ( x )
0

d x = exp ( t )

dx

t p 1 p f (t ) = (t ) exp ( x )dx = p (t ) exp (t ) 0

is the Weibull distribution. (iii)

(t ) = t +

1 + t
is bathtub hazard function.

hazard function

You might also like