ST5212: Survival Analysis

2008/2009: Semester II
Tutorial 1
1. In an experiment to evaluate the eect of diets on mammary carcinogenesis risk of rates.
A sample of rates were treated with the drug DMBA. Starting 6 weeks after DMBA adminis-
tration, each rat was examined once weekly for 14 weeks. The survival time of concern is the
time until a tumor is detected after the DMBA administration. Describe in detail the types of
censoring that are represented by the following rats, and provide the survival time or censoring
time data in days:
(a) A rat that had a tumor at the rst examination at 6 weeks.
Left censoring, censoring time = 6.
(b) A rat that survived the study without having any tomors.
Type I censoring, censoring time = 14.
(c) A rat that did not have tumor at week 12 but had tumor at week 13.
Interval censoring, (12, 13]
(d) A rat that died (without tumor present and death was unrelated to the occurance of
cancer) at day 37.
Random censoring, censoring time = 37/7.
2. A large number of disease-free individuals were enrolled in a study beginning Janury 1,
1970, and were followed for 30 years to assess the age at which they developed breast cancer.
Individuals had clinical exams every 3 years after enrollment. For for selected individuals
described below, discuss the types of censoring and truncation that are represented and provide
their data (survival time or censoring time).
(a) A healthy woman, enrolled in the study at age 30, never developed breast cancer during
the study.
Type I censoring, censoring time = 60.
(b) A healthy woman, enrolled in the study at age 40, was diagnosed with breast cancer at
the fth exam after enrollment.
Interval censoring, censoring time = (52, 55].
1
(c) A healthy woman, enrolled in the study at age 50, died from a cause unrelated to the
disease at age 61.
Random censoring, censoring time = 61.
(d) A healthy woman, enrolled in the study at age 42, moved away from the community at
age 55 and was never diagnosed with breast cancer during the period of observation.
Random censoring, censoring time = 55.
The observasions are left truncated respectively at age 30, 40, 50 and 42.
3. To estimate the distribution of the ages at which postmenopausal woman develop breast
cancer, a sample of eight 50-year old women were given yearly mammograms for a period of 10
years. At each exam, the presence or absence of breast cancer was recorded. In the study, no
tumors were detected by the women by self-examination between two scheduled yearly exams.
Four of the eight women were not detected with breast cancer during the study period. The
ages at onset of breast cancer for the eight women were in the following intervals:
(55, 56], (58, 59], (52, 53], (59, 60], 60, 60, 60, 60.
What type of censoring or truncation is represented in this example?
The rst four are interval censored, the second four are Type I right censored. No truncation
involved, since the study only concerns the population of 50-year old women at the time of study.
4. In the survey concerning the age of rst use of manujuana, data was collected by asking the
question: when did you rst use manujuana? The anwsers of four persons are in the following:
(a) Person A, aged 18: I rst used manujuana at age 16.
Time to event, 16.
(b) Person B, aged 17: I used manujuana but I forget at what age.
Left censoring time = 17.
(c) Person C, aged 17: I never use manujuana.
Right censoring time = 17.
(d) Pdrson D, aged 19: I rst used manujuana in secondary school but I forget which year.
Interval censoring time = (12, 15].
2
For each of the four persons, discuss whether there is a censoring or truncation, if there is, what
type? Provide the survey data.
5. Suppose that the time to death X has an exponetial distribution with hazard rate and
that the right-censoring time C is exponential with hazard rate . Let T = min(X, C) and
= 1 if X C; 0, if X > C. Assume that X and C are indepedent.
(a) Find P( = 1).
P( = 1) = P(X C)
=
_

0
_

x
e
x
e
c
dcdx
=
_

0
e
x
e
x
dx
=

+
.
(b) Find the distribution of T.
P(T t) = P(min(X, C) t)
= 1 P(min(X, C) > t)
= 1 P(X > t, C > t)
= 1 P(X > t)P(C > t)
= 1 e
t
e
t
,
i.e., T E( + ).
(c) Show that and T are independent.
It suces to show P( = 1, T t) = P( = 1)P(T t).
P( = 1, T t) = P(X t, X C)
=
_
t
0
_

x
e
x
e
c
dcdx
=
_
t
0
e
x
e
x
dx
=

+
[1 e
(+)t
]
= P( = 1)P(T t).
3
(d) Let (T
1
,
1
), . . . , (T
n
,
n
) be a random sample from this model. Show that the maximum
likelihood estimator of is

n
i=1

i
/

n
i=1
T
i
. Use part (a)-(c) to nd the mean and
variance of

.
The likelihood function is given by
L(, ) =
n

i=1
_
_

+
_

i
_

+
_
1
i
( + )e
(+)T
i
_
.
Maximizing the likelihood w.r.t. and yields

n
i=1

n
i=1
T
i
.
By independence of T
i
and
i
,
E

= E
n

i=1

i
E
_
1

n
i=1
T
i
_
= n

+

+
n 1
=
n
n 1
.
Note that

n
i=1
T
i
Gamma(, ) with = n, = 1/( +), and the pdf of the Gamma
distribution is given by
f(t) =
t
1
()

e
t/
.
Using the pdf it can be easily obtained that
E
_
1

n
i=1
T
i
_
=
+
n 1
, E
_
1

n
i=1
T
i
_
2
=
( + )
2
(n 1)(n 2)
.
Finally, it is obtained that
Var(

) = E

2
(E

)
2
= E(
n

i=1

i
)
2
E
_
1

n
i=1
T
i
_
2
(E
n

i=1

i
)
2

_
E
_
1

n
i=1
T
i
__
2
=
(n)
2
+ n(n 1)
(n 1)
2
(n 2)
.
4