DEPARTEMEN AKTUARIA

ANALISIS SURVIVAL
 EVALUASI
• TUGAS : 15%
• QUIZ I : 17.5%
• QUIZ II : 17.5%
• ETS : 25%
• EAS : 25%

CONGRATULATIONS
Korlas A : !
Korlas B :

www.its.ac.id INSTITUT TEKNOLOGI SEPULUH NOPEMBER, Surabaya - Indonesia

 3

Textbooks
1. Kleinbaum, D.G., and Klein, M. (2012). Survival Analysis,
third edition, Springer Science and Bussiness Media, LLC.
2. Hosmer, D.W., Lemeshow, S., and May, S. (2008). Applied
Survival Analysis, John Wiley & Sons, Inc., Hoboken, New
Jersey.
3. Klein, J.P., and Moeschberger, M.L. (2003). Survival
Analysis: Techniques for Censored and Truncated Data,
second edition, Springer, New York.
4. Cox, D.R., and Oakes, D. (1984). Analysis of Survival Data,
University Printing House, Cambridge.
5. Le, C.T. (1997). Applied Survival Analysis, John Wiley &
Sons.
 4

Course Outline
Week 1 – Introduction to Survival Analysis: basic concept
and censored data.
Week 2 – Survival Function: Survival Function
(parametric), Kaplan Meier survival curve, hazard rate.
Week 3 and 4 – Log Rank (LR)Test: LR test for 2 group,
and more than 2 group.
Week 5, 6 and 7 – Parametric survival regression:
exponential, Weibull, and loglogistic regression.
Week 8 – Midterm
 5

Course Outline Cont.

Week 9 – Cox proportional hazard (PH) model: estimation,
hazard ratio, interval estimation
Week 10 – The evaluation of the assumption of PH: graph
and goodness of fit aproaches.
Week 11 – Test for the assumption of PH using covariate
time dependent
Week 12 – Stratified Cox Procedure
Week 13 – Survival Analysis for recurrent event
Week 14 and 15 – Competing Risk Survival Analysis
Week 16 – Final Exam
 7

Course Outline
Week 1 – Introduction to Survival Analysis: basic concept
and censored data.
Week 2 – Survival Function: Survival Function
(parametric), Kaplan Meier survival curve, hazard rate.
Week 3 and 4 – Log Rank (LR)Test: LR test for 2 group,
and more than 2 group.
Week 5, 6 and 7 – Parametric survival regression:
exponential, Weibull, and loglogistic regression.
Week 8 – Midterm
 8

What is Survival Analysis? (1/3)

 Survival analysis is a collection of statistical
procedures for data analysis for which the outcome
variable of interest is time until an event occurs.
 By time, we mean years, months, weeks, or days
from the beginning of follow-up of an individual until
an event occurs, alternatively, time can refer to
the age of an individual when an event occurs.
 By event, we mean death, disease incidence,
relapse from remission, recovery (e.g., return to
work) or any designated experience of interest
that may happen to an individual.
 9

What is Survival Analysis? (2/3)

 Although more than one event may be
considered in the same analysis, we will assume
that only one event is of designated interest.
 When more than one event is considered (e.g.,
death from any of several causes), the statistical
problem can be characterized as either a
recurrent event or a competing risk problem.
 10

What is Survival Analysis? (3/3)

 In a survival analysis, we usually refer to the
time variable as survival time, because it
gives the time that an individual has
“survived” over some follow-up period. We
also typically refer to the event as a failure,
because the event of interest usually is
death, disease incidence, or some other
negative individual experience.
 11

Applications (1/2)
1. Leukemia patients/time in remission (weeks)
2. Elderly (60+) population/time until death (years)
3. Parolees (recidivism study)/time until rearrest
(weeks)
4. Heart transplants/time until death (months)
5. Smoking study/time to first smoking (age)
6. Sociology/birth of the first child
7. Labor economics/time to unemployment
(James Heckman: Nobel prize winner 2000)
8. Industrial statistics – reliability
 12

Application (2/2)
Sociology

T
marriage birth of the
first child

Labor Economy

T
employmen unemploymen
t t
 13

Important Descriptive Measures for T

Key points:
 Understand the meaning and application of
each measure.
 Understand the mathematical relationship
between different measures.
 14

Summary
 Outcome: Time until an event occurs (T>0)
Start follow-up Time Event
 Event : death, disease, relapse, recovery
 Assume 1 event
 >1 event Recurrent event
or
Competing risk
Time ≡ survival time
Event ≡ failure
 15

Censored Data (1/2)

 Censoring occurs when we have some
information about individual survival time, but
we don’t know the survival time exactly.
 16

Censored Data (2/2)

Three reasons why censoring may occur:
1. study ends – no event
2. lost to follow-up (drop out)
3. Withdraws from the study due to the
reason that is not the event of interest
 17

Types of censored data

 Right censoring (type I censoring)
 Left censoring
 Interval censoring
 Double censoring
 18

Right Censoring
 Right censoring occurs when a subject
leaves the study before an event occurs, or
the study ends before the event has
occurred.
 Example: Suppose you’re conducting a study
on pregnancy duration. You’re ready to
complete the study and run your analysis, but
some women in the study are still pregnant, so
you don’t know exactly how long their
pregnancies will last.
 19

Illustration for Right Censoring

study begin study end
January 15, December 20,
2005 2012
C
T
1 x

T
C

2 o x

3 o Lost to follow up

4 o Withdraw


n
 20

Notations (Right Censoring)

 Let C be the censoring variable: time from
beginning to end-if-study
 Observed variables:
X=min(T,C) and δ=I(T≤C)  plot
δ=1  X=T, C>T
δ=0  X=C, X>C
 Objective: recover the information of T based
on a random sample of (X, d)
 Common assumption:⊥ T C (independent
censorship)
 21

Left Censoring
 The “failure” occurred before a particular time.
 Turnbull and Weiss (1978) report part of a study
conducted at the Stanford-Palo Alto Peer
Counseling Program (see Hamburg et al. [1975]
for details of the study). In this study, 191
California high school boys were asked, “When
did you first use marijuana?” The answers were
the exact ages (uncensored observations); or “I
have used it but can not recall just when the first
time was,” which is a left-censored observation
 22

Illustration for Left Censoring

study begin study end
January 15, December 20,
2005 C 2012
T
1 x

2 o x


C

n
 23

Notations (Left Censoring)

 Let C be the censoring variable = the time
from use marijuana to recruitment
 Observed variables:
X=max(T,C) and δ=I(T≥C)
δ=1  X=T, C≥T
δ=0  X=C, X<C
 24

Double Censoring
 Both left and right censoring.
 A rare case.
 Turnbull and Weiss (1978) report part of a study
conducted at the Stanford-Palo Alto Peer
Counseling Program (see Hamburg et al. [1975] for
details of the study). In this study, 191 California
high school boys were asked, “When did you first
use marijuana?” The answers were the exact ages
(uncensored observations); “I never used it,” which
are right-censored observations at the boys’ current
ages; or “I have used it but can not recall just when
the first time was,” which is a left-censored
 25

Illustration for Double Censoring

study begin study end
January 15, December 20,
2005 C 2012
T
1 x

2 o
C
T
C

3 o x

n
 26

Notations (Double Censoring)

 Let (Cl, Cr) be the left and right censoring
variables, respectively.
 Observed variables: X=max{min(T, Cr), Cl}
δ=1 if X=T and Cl<T<Cr  exact
δ=0 if X=Cr and T>Cr  right censoring
δ=-1 if X=Cl and T<Cl  left cencoring
 27

Interval Censoring
 We know the “failure” occurred within some
given time period.
 If we don’t know exactly when some students
used marijuana but we know it was within some
interval of time, these observations would
be interval-censored.
 28

Terminology and Notation

 T = survival time (T ≥ 0); T is random variable
 t = specific value for T
 δ = (0, 1) random variable
 1 if failure
=
 0 censored
cause: study ends, lost to follow-up, withdraws
 S(t) = P(T > t)= survivor function
 h(t) = hazard function
 29

Survival Curve (1/2)

Theoretical S(t):

Properties:
 They are nonincreasing; that is, they head downward as t
increases
 at time t = 0, S(t) = S(0) = 1
 at time t = ∞, S(t) = S(∞) = 0
 30

Survival Curve (2/2)

Sˆ (t ) in practice:

 In practice, when using actual data, we usually obtain graphs

that are step functions.
 31

Hazard Function (1/3)

 The hazard function is given by
P ( t ≤ T < t +t | T ≥ t )
h ( t ) = lim
t →0 t
 This mathematical formula is difficult to explain in practical
terms.
 The hazard function h(t) gives the instantaneous potential per
unit time for the event to occur, given that the individual has
survived up to time t.
 Note that, in contrast to the survivor function, which focuses
on not failing, the hazard function focuses on failing, that is,
on the event occurring (S(t): not failing v.s. h(t): failing)
 32

Hazard Function (2/3)

 The numerator: conditional probability.
 Because of the given sign here, the hazard function is
sometimes called a conditional failure rate.
 Rate: 0 to ∞
 The value obtained will give a different number
depending on the units of time used, and may even give
a number larger than one.
 33

Hazard Function (3/3)

Hazard Function

 h(t)≥0
 h(t) has no upper bound
 It is always nonnegative, that is, equal to or greater than zero
 It has no upper bound
 34

S(t) v.s. h(t)

S(t): directly describes survival
h(t):
 A measure of instantaneous potential
 Identify specific model form
 Math model for survival analysis
 35

Relationship of S(t) and h(t)

 If you know one, you can determine the other.
 General Formulae:
( t ) exp  − ∫0 h ( u ) du 
 t
S=

 dS ( t ) dt 
h (t ) = −  
 S ( ) 
t

 The first of these formulae describes how the survivor

function S(t) can be written in terms of an integral
involving the hazard function.
 36

Goal of Survival Analysis (1/2)

The basic goals of survival analysis:
 To estimate and interpret survivor and/or hazard
functions from survival data
 To compare survivor and/or hazard functions
 To assess the relationship of explanatory variables to
survival time
 37

Goal of Survival Analysis (2/2)

 Note that up to 6 weeks, the survivor function for the

treatment group lies above that for the placebo group, but
thereafter the two functions are at about the same level.
 This dual graph indicates that up to 6 weeks the treatment is
more effective for survival than the placebo but has about the
same effect thereafter.
 38

Basic Data Layout

 39

Example
 40

Descriptive Measure (1/2)

 41

Descriptive Measure (2/2)

 Placebo hazard > treatment hazard: suggests
that treatment is more effective than placebo
 Descriptive measures give overall comparison;
they do not give comparison over time.
 42

Estimated Survivor Curves

 43

Multivariable Example
 44

Measure of Effect:
Linear regression:
regression coefficient β

Logistic regression
odds ratio exp(β)

Survival analysis
hazard ratio exp(β)
 45

Censoring Assumption
Three assumptions about censoring:
 Independent (vs.non-independent) censoring
 Random (vs. non-random) censoring
 Non-informative (vs. informative) censoring
 46

Independent Censoring
 Most useful
 Affects validity
 Independent censoring is random censoring
conditional on each level of covariates.
 47

Non-informative Censoring
 Non-informative censoring occurs if the
distribution of survival times (T) provides no
information about the distribution of
censorship times (C), and vice versa.