CHAPTER 2 ST 745, Daowen Zhang

2 Right Censoring and Kaplan-Meier Estimator

In biomedical applications, especially in clinical trials, two important issues arise when studying

“time to event” data (we will assume the event to be “death”. It can be any event of interest):

1. Some individuals are still alive at the end of the study or analysis so the event of interest,

namely death, has not occurred. Therefore we have right censored data.

2. Length of follow-up varies due to staggered entry. So we cannot observe the event for those

individuals with insufficient follow-up time.

Note: It is important to distinguish calendar time and patient time

Figure 2.1: Illustration of censored data

x x x

x x o

x x o

x o o

Study Calendar Study

0 Patient time
starts time ends (measured from entry to study)

In addition to censoring because of insufficient follow-up (i.e., end of study censoring due to

staggered entry), other reasons for censoring includes

• loss to follow-up: patients stop coming to clinic or move away.

• deaths from other causes: competing risks.

Censoring from these types of causes may be inherently different from censoring due to

staggered entry. We will discuss this in more detail later.

PAGE 11
CHAPTER 2 ST 745, Daowen Zhang

Censoring and differential follow-up create certain difficulties in the analysis for such data

as is illustrated by the following example taken from a clinical trial of 146 patients treated after

they had a myocardial infarction (MI).

The data have been grouped into one year intervals and all time is measured in terms of

patient time.

Table 2.1: Data from a clinical trial on myocardial infarction (MI)

Number alive and under

Year since observation at beginning Number dying Number censored

entry into study of interval during interval or withdrawn

[0, 1) 146 27 3

[1, 2) 116 18 10

[2, 3) 88 21 10

[3, 4) 57 9 3

[4, 5) 45 1 3

[5, 6) 41 2 11

[6, 7) 28 3 5

[7, 8) 20 1 8

[8, 9) 11 2 1

[9, 10) 8 2 6

Question: Estimate the 5 year survival rate, i.e., S(5) = P [T ≥ 5].

Two naive and incorrect answers are given by

1. Fb (5) = P [T < 5] = 76 146

deaths in 5 years = 52.1%, S(5)
individuals
b = 1 − Fb (5) = 47.9%.

2. Fb (5) = P [T < 5] = 76 deaths in 5 years b

= 65%, S(5) = 1 − Fb (5) = 35%.
146 -29 (withdrawn in 5 years)

Obviously, we can observe the following

PAGE 12
CHAPTER 2 ST 745, Daowen Zhang

1. The first estimate would be correct if all censoring occurred after 5 years. Of cause, this

was not the case leading to overly optimistic estimate (i.e., overestimates S(5)).

2. The second estimate would be correct if all individuals censored in the 5 years were censored

immediately upon entering the study. This was not the case either, leading to overly

pessimistic estimate (i.e., underestimates S(5)).

Our clinical colleagues have suggested eliminating all individuals who are censored and use

the remaining “complete” data. This would lead to the following estimate

76 deaths in 5 years
Fb (5) = P [T ≤ 5] = = 88.4%, b
S(5) = 1 − Fb (5) = 11.6%.
146 -60 (censored)

This is even more pessimistic than the estimate given by (2).

Life-table Estimate

More appropriate methods use life-table or actuarial method. The problem with the above

two estimates is that they both ignore the fact that each one-year interval experienced censoring

(or withdrawing). Obviously we need to take this information into account in order to reduce

bias. If we can express S(5) as a function of quantities related to each interval and get a very

good estimate for each quantity, then intuitively, we will get a very good estimate of S(5). By

the definition of S(5), we have:

S(5) = P [T ≥ 5] = P [(T ≥ 5) ∩ (T ≥ 4)] = P [T ≥ 4] · P [T ≥ 5|T ≥ 4]

= P [T ≥ 4] · {1 − P [4 ≤ T < 5|T ≥ 4]} = P [T ≥ 4] · q5

= P [T ≥ 3] · P [T ≥ 4|T ≥ 3] · q5 = P [T ≥ 3] · {1 − P [3 ≤ T < 4|T ≥ 3]} · q5

= P [T ≥ 3] · q4 · q5

= = q1 · q2 · q3 · q4 · q5

where qi = 1 − P [i − 1 ≤ T < i|T ≥ i − 1], i = 1, 2, ..., 5. So if we can estimate qi well, then

we will get a very good estimate of S(5). Note that 1 − qi is the mortality rate m(x) at year

x = i − 1 by our definition.

PAGE 13
CHAPTER 2 ST 745, Daowen Zhang

Table 2.2: Life-table estimate of S(5) assuming censoring occurred at the end of interval
bR (t ) = Q
duration [ti−1 , ti ) c(x) =
n(x) d(x) w(x) m d(x)
n(x)
1−m
c(x) S i (1 − m
c(x))

[0, 1) 146 27 3 0.185 0.815 0.815

[1, 2) 116 18 10 0.155 0.845 0.689

[2, 3) 88 21 10 0.239 0.761 0.524

[3, 4) 57 9 3 0.158 0.842 0.441

[4, 5) 45 1 3 0.022 0.972 0.432

Case 1: Let us first assume that anyone censored in an interval of time is censored at the

end of that interval. Then we can estimate each qi = 1 − m(i − 1) in the following way:

d(0) 27
d(0) ∼ Bin(n(0), m(0)) =⇒ m
c(0) = = = 0.185, qb1 = 1 − m
c(0) = 0.815
n(0) 146
d(1) 18
d(1)|H ∼ Bin(n(1), m(1)) =⇒ m
c(1) = = = 0.155, qb2 = 1 − m
c(1) = 0.845
n(1) 116
···

where H means data history (i.e, data before the second interval).

The life table estimate would be computed as shown in Table 2.2. So the 5 year survival

probability estimate SbR (5) = 0.432. (If the assumption that anyone censored in an in-

terval of time is censored at the end of that interval is true, then the estimator SbR (5) is

approximately unbiased to S(5).)

Of course, this estimate SbR (5) will have variation since it was calculated from a sample. We

need to estimate its variation in order to make inference on S(5) (for example, construct a

95% CI for S(5)).

However, SbR (5) is a product of 5 estimates (qb1 – qb5 ), whose variance is not easy to find.

But we have

log(SbR (5)) = log(qb1 ) + log(qb2 ) + log(qb3 ) + log(qb4 ) + log(qb5 ).

So if we can find out the variance of each log(qbi ), we might be able to find out the variance

PAGE 14
CHAPTER 2 ST 745, Daowen Zhang

of log(SbR (5)) and hence the variance of SbR (5).

For this purpose, let us first introduce a very popular method in statistics: delta method:
Delta Method:

θb ∼ N(θ, σ 2 )
a
If

then b ∼ N(f (θ), [f 0 (θ)]2 σ 2 )

f (θ)
a

Proof of delta method: If σ 2 is small, θb will be close to θ with high probability. We hence
b about θ using Taylor expansion:
can expand f (θ)

b ≈ f (θ) + f 0 (θ)(θb − θ).

f (θ)

b from this expansion.

We immediately get the (asymptotic) distribution of f (θ)

Returning to our problem. Let φbi = log(qbi ). Using the delta method, the variance of φbi

is approximately equal to
à !2
1
var(φbi ) = var(qbi ).
qi
Therefore we need to find out and estimate var(qbi ). Of course, we also need to find out the

covariances among φbi and φbj (i 6= j). For this purpose, we need the following theorem:

Double expectation theorem (Law of iterated conditional expectation and variance): If X and

Y are any two random variables (or vectors), then

E(X) = E[E(X|Y )]

Var(X) = Var[E(X|Y )] + E[Var(X|Y )]

Since qbi = 1 − m
c(i − 1), we have

c(i − 1))
var(qbi ) = var(m

PAGE 15
CHAPTER 2 ST 745, Daowen Zhang

c(i − 1)|H)] + var[E(m

= E[var(m c(i − 1)|H)]
" #
m(i − 1)[1 − m(i − 1)]
= E + var[m(i − 1)]
n(i − 1)
" #
1
= m(i − 1)[1 − m(i − 1)]E ,
n(i − 1)

which can be estimated by

c(i − 1)[1 − m
m c(i − 1)]
.
n(i − 1)

Hence the variance of φbi = log(qbi ) can be approximately estimated by

à !2
1 c(i − 1)[1 − m
m c(i − 1)] c(i − 1)
m d
= = .
qbi n(i − 1) [1 − m
c(i − 1)]n(i − 1) (n − d)n

Now let us look at the covariances among φbi and φbj (i 6= i). It is very amazing that they

are all approximately equal to zero!

For example, let us consider the covariance between φb1 and φb2 . Since φb1 = log(qb1 ) and

φb2 = log(qb2 ), using the same argument for the delta method, we know that we only need

to find out the covariance between qb1 and qb2 , or equivalently, the covariance between m
c(0)

c(1). This can be seen from the following:

and m

c(0)m
E[m c(1)] = E[E[m
c(0)m
c(1)|n(0), d(0), w(0)]]

c(0)E[m
= E[m c(1)|n(0), d(0), w(0)]]

c(0)m(1)]
= E[m

c(0)]
= m(1)E[m

c(0)]E[m
= m(1)m(0) = E[m c(1)].

c(0) and m
Therefore, the covariance between m c(1) is zero. Similarly, we can show other

covariances are zero. Hence,

var(log(SbR (5))) = var(φb1 ) + var(φb2 ) + var(φb3 ) + var(φb4 ) + var(φb5 ).

Let θb = log(SbR (5)). Then SbR (5) = eθb. So

var(SbR (5)) = (eθ )2 var(log(SbR (5))) = (S(5))2 [var(φb1 )+var(φb2 )+var(φb3 )+var(φb4 )+var(φb5 )],

PAGE 16
CHAPTER 2 ST 745, Daowen Zhang

which can be estimated by

"
d(0) d(1) d(2)
ar(SbR (5))
vd = (SbR (5))2 + +
(n(0) − d(0))n(0) (n(1) − d(1))n(1) (n(2) − d(2))n(2)
#
d(3) d(4)
+ +
(n(3) − d(3))n(3) (n(4) − d(4))n(4)
4
X d(i)
= (SbR (5))2 . (2.1)
i=0 [n(i) − d(i)]n(i)

Case 2: Let us assume that anyone censored in an interval of time is censored right at the

beginning of that interval. Then the life table estimate would be computed as shown in

Table 2.3. So the 5 year survival probability estimate = 0.400. (In this case, the estimator

SbL (5) is approximately unbiased to S(5).)

The variance estimate of SbL (5) is similar to that of SbR (5) except that we need to change

the “sample size” for each mortality estimate to n − w in equation (2.1).

Table 2.3: Life-table estimate of S(5) assuming censoring occurred at the beginning of interval
bL (t ) = Q
duration [ti−1 , ti ) c(x) =
n(x) d(x) w(x) m d(x)
n(x)−w(x)
1−m
c(x) S i (1 − m
c(x))

[0, 1) 146 27 3 0.189 0.811 0.811

[1, 2) 116 18 10 0.170 0.830 0.673

[2, 3) 88 21 10 0.269 0.731 0.492

[3, 4) 57 9 3 0.167 0.833 0.410

[4, 5) 45 1 3 0.024 0.976 0.400

The naive estimates range from 35% to 47.9% for the five year survival probability with the

“complete case” (i.e., eliminating anyone censored) estimator giving an estimate of 11.6%.

The life-table estimate ranged from 40% to 43.2% depending on whether we assume censoring

occurred at the left (i.e., beginning) or right (i.e., end) of each interval.

More than likely censoring occurs during the interval. Thus SbL and SbR are not correct. A

compromise is to use the following modification:

PAGE 17
CHAPTER 2 ST 745, Daowen Zhang

Table 2.4: Life-table estimate of S(5) assuming censoring occurred during the interval

bLT (t ) = Q
duration [ti−1 , ti ) n(x) c(x) =
d(x) w(x) m d(x)
n(x)−w(x)/2
1−m
c(x) S i (1 − m
c(x))

[0, 1) 146 27 3 0.187 0.813 0.813

[1, 2) 116 18 10 0.162 0.838 0.681

[2, 3) 88 21 10 0.253 0.747 0.509

[3, 4) 57 9 3 0.162 0.838 0.426

[4, 5) 45 1 3 0.023 0.977 0.417

That is, when calculating the mortality estimate in each interval, we use (n(x) − w(x)/2) as

the “sample size”. This number is often referred to as the effective sample size.

So the 5 year survival probability estimate SbLT (5) = 0.417, which is between SbL = 0.400 and

SbR = 0.432.

Figure 2.2: Life-table estimate of the survival probability for MI data

1.0
0.8
Survival probability

0.6
0.4
0.2

0 2 4 6 8 10

Time (years)

Figure 2.2 shows the life-table estimate of the survival probability assuming censoring oc-

curred during interval. Here the estimates were connected using straight lines. No special

significance should be given to this. From this figure, the median survival time is estimated to

PAGE 18
CHAPTER 2 ST 745, Daowen Zhang

be about 3 years.

The variance estimate of the life-tabble estimate SbLT (5) is similar to equation (2.1) except

that the sample size n(i) is changed to n(i) − w(i)/2. That is

4
X d(i)
ar(SbLT (5)) = (SbLT (5))2
vd . (2.2)
i=0 [n(i) − w(i)/2 − d(i)][n(i) − w(i)/2]

Of course, we can also use the above formula to calculate the variance of SbLT (t) at other

time points. For example:

( )
d(0)
ar(SbLT (1))
vd = (SbLT (1))2
[n(0) − w(0)/2 − d(0)][n(0) − w(0)/2]
27
= 0.8132 × = 0.8132 × 0.001590223 = 0.001051088.
(146 − 3/2 − 27)(146 − 3/2)
√
Therefore SE(SbLT (1)) = 0.001051088 = 0.0324.

The calculation presented in Table 2.4 can be implemented using Proc Lifetest in SAS:

options ls=72 ps=60;

Data mi;
input survtime number status;
cards;
0 27 1
0 3 0
1 18 1
1 10 0
2 21 1
2 10 0
3 9 1
3 3 0
4 1 1
4 3 0
5 2 1
5 11 0
6 3 1
6 5 0
7 1 1
7 8 0
8 2 1
8 1 0
9 2 1
9 6 0
;

proc lifetest method=life intervals=(0 to 10 by 1);

time survtime*status(0);
freq number;
run;

PAGE 19
CHAPTER 2 ST 745, Daowen Zhang

Note that the number of observed events and withdrawals in [ti−1 , ti ) were entered after ti−1

instead of ti . Part of the output of the above SAS program is

The LIFETEST Procedure

Life Table Survival Estimates

Effective Conditional
Interval Number Number Sample Probability
[Lower, Upper) Failed Censored Size of Failure
0 1 27 3 144.5 0.1869
1 2 18 10 111.0 0.1622
2 3 21 10 83.0 0.2530
3 4 9 3 55.5 0.1622
4 5 1 3 43.5 0.0230
5 6 2 11 35.5 0.0563
6 7 3 5 25.5 0.1176
7 8 1 8 16.0 0.0625
8 9 2 1 10.5 0.1905
9 10 2 6 5.0 0.4000

Conditional
Probability Survival Median
Interval Standard Standard Residual
[Lower, Upper) Error Survival Failure Error Lifetime
0 1 0.0324 1.0000 0 0 3.1080
1 2 0.0350 0.8131 0.1869 0.0324 4.4265
2 3 0.0477 0.6813 0.3187 0.0393 5.2870
3 4 0.0495 0.5089 0.4911 0.0438 .
4 5 0.0227 0.4264 0.5736 0.0445 .
5 6 0.0387 0.4166 0.5834 0.0446 .
6 7 0.0638 0.3931 0.6069 0.0450 .
7 8 0.0605 0.3469 0.6531 0.0470 .
8 9 0.1212 0.3252 0.6748 0.0488 .
9 10 0.2191 0.2632 0.7368 0.0558 .

Here the numbers in the column under Conditional Probability of Failure are the es-
c(x) = d(x)/(n(x) − w(x)/2).
timated mortality m

The above lifetable estimation can also be implemented using R. Here is the R code:

> tis <- 0:10

> ninit <- 146
> nlost <- c(3,10,10,3,3,11,5,8,1,6)
> nevent <- c(27,18,21,9,1,2,3,1,2,2)
> lifetab(tis, ninit, nlost, nevent)

PAGE 20
CHAPTER 2 ST 745, Daowen Zhang

The output from the above R function is

nsubs nlost nrisk nevent surv pdf hazard se.surv

0-1 146 3 144.5 27 1.0000000 0.186851211 0.20610687 0.00000000
1-2 116 10 111.0 18 0.8131488 0.131861966 0.17647059 0.03242642
2-3 88 10 83.0 21 0.6812868 0.172373775 0.28965517 0.03933747
3-4 57 3 55.5 9 0.5089130 0.082526440 0.17647059 0.04382194
4-5 45 3 43.5 1 0.4263866 0.009801991 0.02325581 0.04452036
5-6 41 11 35.5 2 0.4165846 0.023469556 0.05797101 0.04456288
6-7 28 5 25.5 3 0.3931151 0.046248831 0.12500000 0.04503654
7-8 20 8 16.0 1 0.3468662 0.021679139 0.06451613 0.04699173
8-9 11 1 10.5 2 0.3251871 0.061940398 0.21052632 0.04879991
9-10 8 6 5.0 2 0.2632467 NA NA 0.05579906

se.pdf se.hazard
0-1 0.032426423 0.03945410
1-2 0.028930638 0.04143228
2-3 0.033999501 0.06254153
3-4 0.026163333 0.05859410
4-5 0.009742575 0.02325424
5-6 0.016315545 0.04097447
6-7 0.025635472 0.07202769
7-8 0.021195209 0.06448255
8-9 0.040488466 0.14803755
9-10 NA NA

Note: Here the numbers in the column of hazard are the estimated hazard rates at the

midpoint of each interval by assuming the true survival function S(t) is a straight line in each

interval. You can find an explicit expression for this estimator using the relation
f (t)
λ(t) = ,
S(t)
and the assumption that the true survival function S(t) is a straight line in [ti−1 , ti ):
S(ti ) − S(ti−1 )
S(t) = S(ti−1 ) + (t − ti−1 ), for t ∈ [ti−1 , ti ).
ti − ti−1
These estimates are very close to the mortality estimates we obtained before (the column under

Conditional Probability of Failure in the SAS output.)

Kaplan-Meier Estimator

The Kaplan-Meier or product limit estimator is the limit of the life-table estimator when

intervals are taken so small that only at most one distinct observation occurs within an interval.

Kaplan and Meier demonstrated in a paper in JASA (1958) that this estimator is “maximum

likelihood estimate”.

PAGE 21
CHAPTER 2 ST 745, Daowen Zhang

Figure 2.3: An illustrative example of Kaplan-Meier estimator

1.0

0.8

0.6

0.4

0.2

0.0 x x o x o x x o x o

4.5 7.5 11.5 13.5 15.516.5 19.5

Patient time (years)

9 8 6 4 3 1
1−m
c(x) : 1 1 1 1 10
1 1 9
1 1 1 7
1 1 1 5 4
1 1 2
1 1
b
S(t) : 1 1 1 1 9
. . 8
. . . 48
. . . 192 144
. . 144
. .
10 10 70 350 350 700

We will illustrate through a simple example shown in Figure 2.3 how the Kaplan-Meier

estimator is constructed.

By convention, the Kaplan-Meier estimate is a right continuous step function which takes

jumps only at the death time.

The calculation of the above KM estimate can be implemented using Proc Lifetest in SAS

as follows:

Data example;
input survtime censcode;
cards;
4.5 1
7.5 1
8.5 0
11.5 1
13.5 0
15.5 1
16.5 1
17.5 0
19.5 1
21.5 0
;

Proc lifetest;

PAGE 22
CHAPTER 2 ST 745, Daowen Zhang

time survtime*censcode(0);
run;

And part of the output from the above program is

The LIFETEST Procedure

Product-Limit Survival Estimates

Survival
Standard Number Number
SURVTIME Survival Failure Error Failed Left
0.0000 1.0000 0 0 0 10
4.5000 0.9000 0.1000 0.0949 1 9
7.5000 0.8000 0.2000 0.1265 2 8
8.5000* . . . 2 7
11.5000 0.6857 0.3143 0.1515 3 6
13.5000* . . . 3 5
15.5000 0.5486 0.4514 0.1724 4 4
16.5000 0.4114 0.5886 0.1756 5 3
17.5000* . . . 5 2
19.5000 0.2057 0.7943 0.1699 6 1
21.5000* . . . 6 0
* Censored Observation

The above Kaplan-Meier estimate can also be obtained using R function survfit(). The
code is given in the following:

> survtime <- c(4.5, 7.5, 8.5, 11.5, 13.5, 15.5, 16.5, 17.5, 19.5, 21.5)
> status <- c(1, 1, 0, 1, 0, 1, 1, 0, 1, 0)
> fit <- survfit(Surv(survtime, status), conf.type=c("plain"))

Then we can use R function summary() to see the output:

> summary(fit)
Call: survfit(formula = Surv(survtime, status), conf.type = c("plain"))
time n.risk n.event survival std.err lower 95% CI upper 95% CI
4.5 10 1 0.900 0.0949 0.7141 1.000
7.5 9 1 0.800 0.1265 0.5521 1.000
11.5 7 1 0.686 0.1515 0.3888 0.983
15.5 5 1 0.549 0.1724 0.2106 0.887
16.5 4 1 0.411 0.1756 0.0673 0.756
19.5 2 1 0.206 0.1699 0.0000 0.539

Let d(x) denote the number of deaths at time x. Generally d(x) is either zero or one, but we

allow the possibility of tied survival times in which case d(x) may be greater than one. Let n(x)

PAGE 23
CHAPTER 2 ST 745, Daowen Zhang

denote the number of individuals at risk just prior to time x; i.e., number of individuals in the

sample who neither died nor were censored prior to time x. Then Kaplan-Meier estimate can be

expressed as
à !
Y d(x)
KM (t) = 1− .
x≤t n(x)

Note: In the notation above, the product changes only at times x where d(x) ≥ 1; , i.e.,

only at times where we observed deaths.

Non-informative Censoring

In order that the life-table estimates give unbiased results there is an important assumption

that individuals who are censored are at the same risk of subsequent failure as those who are still

alive and uncensored. The risk set at any time point (the individuals still alive and uncensored)

should be representative of the entire population alive at the same time. If this is the case, the

censoring process is called non-informative. Statistically, if the censoring process is indepen-

dent of the survival time, then we will automatically have non-informative censoring. Actually,

we almost always mean independent censoring by non-informative censoring.

If censoring only occurs because of staggered entry, then the assumption of non-informative

censoring seems plausible. However, when censoring results from loss to follow-up or death from

a competing risk, then this assumption is more suspect. If at all possible censoring from these

later situations should be kept to a minimum.

Greenwood’s Formula for the Variance of the Life-table Estimator

The derivation given below is heuristic in nature but will try to capture some of the salient

feature of the more rigorous treatments given in the theoretical literature on survival analysis.

For this reason, we will use some of the notation that is associated with the “counting process”

approach to survival analysis. In fact we have seen it when we discussed the life-table estimator.

PAGE 24
CHAPTER 2 ST 745, Daowen Zhang

It is useful when considering the product limit estimator to partition time into many small

intervals, say, with interval length equal to ∆x where ∆x is small.

Figure 2.4: Partition of time axis

x
Patient time

Let “x” denote some arbitrary time point on the grid above and define

• Y (x) = number of individuals at risk (i.e., alive and uncensored) at time point x.

• dN (x) = number of observed deaths occurring in [x, x + ∆x).

Recall: Previously, Y (x) was denoted by n(x) and dN (x) was denoted by d(x).

It should be straightforward to see that “w(x)”, the number of censored individuals in [x, x +

∆x), is equal to {[Y (x) − Y (x + ∆x)] − dN (x)}.

Note: In theory, we should be able to choose ∆x small enough so that {dN (x) > 0 and

w(x) > 0} should never occur. In practice, however, data may not be collected in that fashion,

in which case, approximations such as those given with life-table estimators may be necessary.

With these definitions, the Kaplan-Meier estimator can be written as

( )
Y dN (x)
KM (t) = 1− , as ∆x → 0,
Y (x)
all grid points x such that x + ∆x ≤ t

which can be modified if “∆x” is not chosen small enough to be

( )
Y dN (x)
LT (t) = 1− ,
Y (x) − w(x)/2
all grid points x such that x + ∆x ≤ t

where LT (t) means life-table estimator.

dN (x)
If the sample size is large and ∆x is small, then Y (x)
is a small number (i.e., close to zero)

and as long as x is not close to the right hand tail of the survival distribution (where Y (x) may

PAGE 25
CHAPTER 2 ST 745, Daowen Zhang

be very small). If this is the case, then

( ) ( )
dN (x) dN (x)
exp − ≈ 1− .
Y (x) Y (x)

Here we used the approximation ex ≈ 1 + x when x is close to zero. This approximation is exact
dN (x)
when Y (x)
= 0.

Therefore, the Kaplan-Meier estimator can be approximated by

( ) ( )
Y dN (x) X dN (x)
KM (t) ≈ exp − = exp − ,
Y (x) Y (x)
all grid points x such that x + ∆x ≤ t x<t

here and thereafter, {x < t} means {all grid points x such that x + ∆x ≤ t}.

If ∆x is taken to be small enough so that all distinct times (either death times or withdrawal
P dN (x)
times) are represented at most once in any time interval, then the estimator x<t Y (x) will be

uniquely defined and will not be altered by choosing a finer partition for the grid of time points.
P dN (x)
In such a case the quantity x<t Y (x) is sometimes represented as
Z t dN (x)
.
0 Y (x)

1. Basically, this estimator take the sum over all the distinct death times before time t of the

number of deaths divided by the number at risk at each of those distinct death times.

P dN (x)
2. The estimator x<t Y (x) is referred to as the Nelson-Aalen estimator for the cumulative
Rt
hazard function Λ(t) = 0 λ(x)dx. That is

X dN (x)
b
Λ(t) = .
x<t Y (x)

Recall that S(t) = exp(−Λ(t)).

By the definition of an integral,

Z t X
Λ(t) = λ(x)dx ≈ λ(x)∆x.
0
grid points x such that x + ∆x ≤ t

PAGE 26
CHAPTER 2 ST 745, Daowen Zhang

By the definition of a hazard function,

λ(x)∆x ≈ P [x ≤ T < x + ∆x|T ≥ x].

With independent censoring, it would seem reasonable to estimate λ(x)∆x, i.e., “the con-
dN (x)
ditional probability of dying in [x, x + ∆x) given being alive at time x” by Y (x)
. Therefore we

obtain the Nelson-Aalen estimator

X dN (x)
b
Λ(t) = .
x<t Y (x)

We will now show how to estimate the variance of the Nelson-Aalen estimator and then show

how this will be used to estimate the variance of the Kaplan-Meier estimator.

For a grid point x, let H(x) denote the history of all deaths and censoring occurring up to

time x.

H(x) = {dN (u), w(u); for all values u on our grid of points for u < x}.

Note the following

1. Conditional on H(x), we would know the value of Y (x) (i.e., the number of risk at time x)

and that dN (x) would follow a binomial distribution denoted as

dN (x)|H(x) ∼ Bin(Y (x), π(x)),

where π(x) is the Conditional probability of an individual dying in [x, x + ∆x) given that

the individual was at risk at time x (i.e., π(x) = P [x ≤ T < x + ∆x|T ≥ x]). Recall that

this probability can be approximated by π(x) ≈ λ(x)∆x.

2. The following are standard results for a binomially distributed random variable.

(a) E[dN (x)|H(x)] = Y (x)π(x),

PAGE 27
CHAPTER 2 ST 745, Daowen Zhang

(b) Var[dN (x)|H(x)] = Y (x)π(x)[1 − π(x)],

" ¯ #
dN (x) ¯¯
(c) E ¯ H(x) = π(x),
Y (x) ¯
(" #" #" #¯ )
Y (x) dN (x) Y (x) − dN (x) ¯¯
(d) E ¯ H(x) = π(x)[1 − π(x)].
Y (x) − 1 Y (x) Y (x) ¯

b P dN (x)
Consider the Nelson-Aalen estimator Λ(t) = x<t Y (x)
. We have
" # " #
X dN (x) X
dN (x)
b
E[Λ(t)] = E = E
x<t Y (x) x<t Y (x)
" " ¯ ##
X dN (x) ¯¯ X
= E E ¯ H(x) = π(x)
x<t Y (x) ¯ x<t
X Z t
≈ λ(x)∆x ≈ λ(x)dx = Λ(t).
x<t 0

Hence

b P
• E[Λ(t)] = x<t π(x).

P b
• If we take ∆x smaller and smaller, then in the limit x<t π(x) goes to Λ(t). Namely Λ(t)

is nearly unbiased to Λ(t).

b
How to Estimate the Variance of Λ(t)

The definition of variance is given by

b
Var(Λ(t)) b
= E[Λ(t) b
− E(Λ(t))]2
" #2
X dN (x) X
= E − π(x)
x<t Y (x) x<t
" ( )#2
X dN (x)
= E − π(x) .
x<t Y (x)

Note: The square of a sum of terms is equal to the sum of the squares plus the sum of all

cross product terms. So the above expectation is equal to

 ( )2 ( )( )
X dN (x) X dN (x) dN (x0 )
E − π(x) + − π(x) − π(x0 ) 
x<t Y (x) x6=x0 <t
Y (x) Y (x0 )
" #2 "( )( )#
X dN (x) X dN (x) dN (x0 )
= E − π(x) + E − π(x) 0
− π(x0 )
x<t Y (x) x6=x0 <t
Y (x) Y (x )

PAGE 28
CHAPTER 2 ST 745, Daowen Zhang

We will first demonstrate that the cross product terms have expectation equal to zero. Let

us take one such term and let us say, without loss of generality, that x < x0 .
"( )( )#
dN (x) dN (x0 )
E − π(x) 0
− π(x0 )
Y (x) Y (x )
" "( )( )¯ ##
dN (x) dN (x0 ) ¯
0 ¯ 0
= E E − π(x) − π(x ) ¯¯ H(x )
Y (x) Y (x0 )

Note: Conditional on H(x0 ), dN (x), Y (x) and π(x) are constant since x < x0 . Therefore the

above expectation is equal to

"( ) "( )¯ ##
dN (x) dN (x0 ) ¯
¯
E − π(x) E − π(x0 ) 0
¯ H(x )
¯
Y (x) Y (x0 )

The inner conditional expectation is zero since

( ¯ )
dN (x0 ) ¯¯
E ¯ H(x0 ) = π(x0 )
Y (x ) ¯
0

by (2.c). Therefore we show that

"( )( )#
dN (x) dN (x0 )
E − π(x) 0
− π(x0 ) = 0.
Y (x) Y (x )

Since the cross product terms have expectation equal to zero, this implies that
" #2
X
dN (x)
b
Var(Λ(t)) = E − π(x)
x<t Y (x)

Using the double expectation again, we get that

" #2
dN (x)
E − π(x)
Y (x)
  )2 ¯¯ 
 dN (x) ¯
= E E  − π(x) ¯¯ H(x)
Y (x) ¯
" " ¯ ##
dN (x) ¯¯
= E Var ¯ H(x)
Y (x) ¯
" #
π(x)[1 − π(x)]
= E .
Y (x)

Therefore, we have that

" #
b
X π(x)[1 − π(x)]
Var(Λ(t)) = E .
x<t Y (x)

PAGE 29
CHAPTER 2 ST 745, Daowen Zhang
π(x)[1−π(x)]
If we wanted to estimate Y (x)
, then using (2.d) we might think that
h i
dN (x) Y (x)−dN (x)
Y (x) Y (x)
Y (x) − 1

b
may be reasonable. In fact, we would then use as an estimate for Var(Λ(t)) the following

estimator; summing the above estimator over all grid points x such that x + ∆x ≤ t.
 dN (x) h Y (x)−dN (x) i 
X Y (x) Y (x)
d Λ(t))
Var( b =  .
x<t Y (x) − 1

b
In fact, the above variance estimator is unbiased for Var(Λ(t)), which can be seen using the

following argument:
 h i
dN (x) Y (x)−dN (x)
X Y (x) Y (x)
E 
x<t Y (x) − 1
 dN (x) h Y (x)−dN (x) i 
X Y (x) Y (x)
= E 
x<t Y (x) − 1
  n o¯ 
X
dN (x) Y (x)−dN (x) ¯
Y (x) Y (x) ¯
= E E  ¯ H(x) (double expectation again)
¯
x<t Y (x) − 1 ¯
" #
X π(x)[1 − π(x)]
= E (by (2.d))
x<t Y (x)
b
= Var[Λ(t)].

b
What this last argument shows is that an unbiased estimator for Var[Λ(t)] is given by
 dN (x) h Y (x)−dN (x) i 
X Y (x) Y (x)
 .
x<t Y (x) − 1

Note: If the survival data are continuous (i.e., no ties) and ∆x is taken small enough, then

dN (x) would take on the values 0 or 1 only. In this case

h i
dN (x) Y (x)−dN (x)
Y (x) Y (x) dN (x)
= ,
Y (x) − 1 Y 2 (x)

and

X dN (x)
d Λ(t))
Var( b = ,
x<t Y 2 (x)

PAGE 30
CHAPTER 2 ST 745, Daowen Zhang

which is also written as

Z t dN (x)
.
0 Y 2 (x)

Remark:

P dN (x)
• We proved that the Nelson-Aalen estimator x<t Y (x) is an unbiased estimator for
P
x<t π(x). We argued before that in the limit as ∆x goes to zero,

X dN (x) Z t dN (x)
becomes .
x<t Y (x) 0 Y (x)

• We also argued that π(x) ≈ λ(x)∆x, hence as ∆x goes to zero, then

X Z t
π(x) goes to λ(x)dx.
x<t 0

These two arguments taken together imply that

Z t dN (x)
0 Y (x)

is an unbiased estimator of the cumulative hazard function

Z t
Λ(t) = λ(x)dx,
0

namely,
"Z #
t dN (x)
E = Λ(t).
0 Y (x)

b P dN (x)
• Since Λ(t) = x<t Y (x)
is made up of a sum of random variables that are conditionally

uncorrelated, they have a “martingale” structure for which there exists a body of theory

that enables us to show that

PAGE 31
CHAPTER 2 ST 745, Daowen Zhang

b
Λ(t) b
is asymptotically normal with mean Λ(t) and variance Var[Λ(t)], which can be estimated

unbiasedly by
 dN (x) h Y (x)−dN (x) i 
X Y (x) Y (x)
d Λ(t))
Var( b =  ;
x<t Y (x) − 1

and in the case of no ties, by

X dN (x)
d Λ(t))
Var( b = .
x<t Y 2 (x)

b
Let us refer to the estimated standard error of Λ(t) by
  dN (x) h Y (x)−dN (x) i 1/2
X Y (x) Y (x)
b
se[Λ(t)] =   .
x<t Y (x) − 1

b
The unbiasedness and asymptotic normality of Λ(t) about Λ(t) allow us to form confidence

intervals for Λ(t) (at time t). Specifically, the (1 − α)th confidence interval for Λ(t) is given by

b
Λ(t) b
± zα/2 ∗ se(Λ(t)),

where zα/2 is the (1 − α/2)th quantile of a standard normal distribution. That is, the random

interval

b
[Λ(t) b
− zα/2 ∗ se(Λ(t)), b
Λ(t) b
+ zα/2 ∗ se(Λ(t))]

covers the true value Λ(t) with probability 1 − α.

This result could also be used to construct confidence intervals for the survival function S(t).

This is seen by realizing that

S(t) = e−Λ(t) ,

in which case the confidence interval is given by

b b b b
[e−Λ(t)−zα/2 ∗se(−Λ(t)) , e−Λ(t)+zα/2 ∗se(Λ(t)) ],

PAGE 32
CHAPTER 2 ST 745, Daowen Zhang

meaning that this random interval will cover the true value S(t) with probability 1 − α.

An example: We will use the hypothetical data shown in Figure 2.3 to illustrate the calcu-
b
lation of Λ(t), d Λ(t),
Var b and confidence intervals for Λ(t) and S(t). For illustration, let us take

t = 17. Note that there are no ties in this example. So

X dN (x) Z t
b dN (x) 1 1 1 1 1
Λ(t) = = = + + + + = 0.804,
x<t Y (x) 0 Y (x) 10 9 7 5 4
X dN (x) Z t dN (x)
d Λ(t)]
b 1 1 1 1 1
Var[ = 2
= 2
= 2 + 2 + 2 + 2 + 2 = 0.145,
x<t Y (x) 0 Y (x) 10 9 7 5 4
√
se[ b
b Λ(t)] = 0.145 = 0.381.

So the 95% confidence interval for Λ(t) is

0.804 ± 1.96 · 0.381 = [0.0572, 1.551].

and the Nelson-Aalen estimate of S(t) is

b b
S(t) = e−Λ(t) = e−0.804 = 0.448.

The 95% confidence interval for S(t) is

[e−1.551 , e−0.0572 ] = [0.212, 0.944].

b
Note The above Nelson-Aalen estimate S(t) = 0.448 is different from (but close to) the

Kaplan-Meier estimate KM (t) = 0.411. It should also be noted that above confidence interval
b
for the survival probability S(t) is not symmetric about the estimator S(t). Another way of

getting approximate confidence intervals for S(t) = e−Λ(t) is by using the delta method. This

method guarantees symmetric confidence intervals.

Hence a (1 − α)th confidence interval for f (θ) is given by

b ± z |f 0 (θ)|
f (θ) b σ b.
α/2

b
In our case, Λ(t) takes on the role of θ, Λ(t) b f (θ) = e−θ so that
takes on the role of θ,

S(t) = f {Λ(t)}. Since

b b
|f 0 (θ) = | − e−θ | = e−θ , and S(t) = e−Λ(t) .

PAGE 33
CHAPTER 2 ST 745, Daowen Zhang

Consequently, using the delta method we get

b
S(t) b
∼ N(S(t), [S(t)]2 Var[Λ(t)]),
a

and a (1 − α)th confidence interval for S(t) is given by

b
S(t) b
± zα/2 {S(t) b
∗ se[Λ(t)]}.

b
Remark: Note that [S(t)]2 Var[Λ(t)] b
is an estimate of Var[S(t)], b
where S(t) b
= exp[−Λ(t)].

Previously, we showed that the Kaplan-Meier estimator

" #
Y dN (x)
KM (t) = 1−
x<t Y (x)
b
was well approximated by S(t) b
= exp[−Λ(t)].

b
Thus a reasonable estimator of Var(KM (t)) would be to use the estimator of Var[exp(−Λ(t))],

or (by using the delta method)

X dN (x)
b
[S(t)]2d b b
Var[Λ(t))] = [S(t)]2
.
x<t Y 2 (x)

This is very close (asymptotically the same) as the estimator for the variance of the Kaplan-

Meier estimator given by Greenwood. Namely

" #
X dN (x)
d
Var{KM (t)} = {KM (t)}2 .
x<t [Y (x) − w(x)/2][Y (x) − dN (x) − w(x)/2]

Note: SAS uses the above formula to calculate the estimated variance for the life-table estimate

of the survival function, by replacing KM (t) on both sides by LT (t).

Note: The summation in the above equation can be viewed as the variance estimate for the
b
cumulative hazard estimator defined by Λ KM (t) = −log[KM (t)]. Namely,

X dN (x)
b
Var{Λ KM (t)} = .
x<t [Y (x) − w(x)/2][Y (x) − dN (x) − w(x)/2]

In the example shown in Figure 2.3, using the delta-method approximation for getting a

confidence interval with the Nelson-Aalen estimator, we get that a 95% CI for S(t) (where t=17)

PAGE 34
CHAPTER 2 ST 745, Daowen Zhang

is

b b b
e−Λ(t) ± 1.96 ∗ e−Λ(t) se[Λ(t)] = e−0.801 ± 1.96 ∗ e−0.801 ∗ 0.381 = [0.114, 0.784].

b
The estimated se[S(t)] = 0.171.

If we use the Kaplan-Meier estimator, together with Greenwood’s formula for estimating the

variance, to construct a 95% confidence interval for S(t), we would get

· ¸· ¸· ¸· ¸· ¸
1 1 1 1 1
KM (t) = 1− 1− 1−= 0.411 1− 1−
10 9 4 7 5
½ ¾
d 2 1 1 1 1 1
Var[KM (t)] = 0.411 + + + + = 0.03077
10 ∗ 9 9 ∗ 8 7 ∗ 6 5 ∗ 4 4 ∗ 3
√
b
se[KM (t)] = 0.03077 = 0.175
d Λ b 1 1 1 1 1
Var[ KM (t)] = + + + + = 0.182
10 ∗ 9 9 ∗ 8 7 ∗ 6 5 ∗ 4 4 ∗ 3
b
se[Λ KM (t)] = 0.427.

Thus a 95% confidence interval for S(t) is given by

KM (t) ± 1.96 ∗ se[KM

b (t)] = 0.411 ± 1.96 ∗ 0.175 = [0.068, 0.754],

which is close to the confidence interval using delta method, considering the sample size is only 10.
b
In fact the estimated standard errors for S(t) and KM (t) using delta method and Greenwood’s

formula are 0.171 and 0.175 respectively, which agree with each other very well.

Note: If we want to use R function survfit() to construct a confidence interval for S(t) with

the form KM (t) ± zα/2 ∗ se[KM

b (t)], we have to specify the argument conf.type=c("plain")

in survfit(). The default constructs the confidence interval for S(t) by exponentiating the

confidence interval for the cumulative hazard using the Kaplan-Meier estimator. For example,

a 95% CI for S(t) is KM (t) ∗ [e−1.96∗se[ΛbKM (t)] , e1.96∗se[ΛbKM (t)] ] = 0.411 ∗ [e−1.96∗0.427 , [e1.96∗0.427 ] =

[0.178, 0.949].

Comparison of confidence intervals for S(t)

1. exponentiating the 95% CI for cumulative hazard using Nelson-Aalen estimator: [0.212, 0.944].

PAGE 35
CHAPTER 2 ST 745, Daowen Zhang

2. Delta-method using Nelson-Aalen estimator: [0.114, 0.784].

3. exponentiating the 95% CI for cumulative hazard using Kaplan-Meier estimator: [0.178, 0.949].

4. Kaplan-Meier estimator together with Greenwood’s formula for variance: [0.068, 0.754].

These are relatively close and the approximations become better with larger sample sizes.

Of the different methods for constructing confidence intervals, “usually” the most accurate

is based on exponentiating the confidence intervals for the cumulative hazard function based on

Nelson-Aalen estimator. We don’t feel that symmetry is necessarily an important feature that

confidence interval need have.

Summary

Q ³ ´
1. We first estimate S(t) by KM (t) = 1− d(x) b
, then estimate Λ(t) by Λ
x<t n(x) KM (t) =

−log[KM (t)]. Their variance estimates are

X dN (x)
d Λ
Var{ b
KM (t)} =
x<t [Y (x) − w(x)/2][Y (x) − dN (x) − w(x)/2]
d
Var{KM d Λ
(t)} = {KM (t)}2 ∗ Var{ b
KM (t)}.

The confidence intervals for S(t) can be constructed in two ways:

b b b
KM (t) ± zα/2 ∗ se[KM (t)], or e−ΛKM (t)±zα/2 ∗se[ΛKM (t)] = KM (t) ∗ e±zα/2 ∗se[ΛKM (t)]

b P dN (x)
2. We first estimate Λ(t) by Nelson-Aalen estimator Λ(t) = x<t Y (x) , then estimate S(t) by
b
S(t) = e−Λb(t) . Their variance estimates are given by
 dN (x) h Y (x)−dN (x) i 
X Y (x) Y (x)
d Λ(t)}
Var{ b =  
x<t Y (x) − 1
Var{ b
d S(t)} b
= {S(t)}2 d Λ(t)}.
∗ Var{ b

The confidence intervals for S(t) can also be constructed in two ways:

b b b b b b
S(t) ± zα/2 ∗ se[S(t)], or e−Λ(t)±zα/2 ∗se[Λ(t)] = S(t) ∗ e±zα/2 ∗se[Λ(t)] .

PAGE 36
CHAPTER 2 ST 745, Daowen Zhang

Estimators of quantiles (such as median, first and third quartiles) of a distribution can be

obtained by inverse relationships. This is most easily illustrated through an example.

Suppose we want to estimate the median S −1 (0.5) or any other quantile ϕ = S −1 (θ); 0 <

θ < 1. Then the point estimate of ϕ is obtained (using the Kaplan-Meier estimator of S(t))

ϕb = KM −1 (θ), i .e., KM (ϕ)

b = θ.

An approximate (1 − α)th confidence interval for ϕ if given by [ϕbL , ϕbU ], where ϕbL satisfies

KM (ϕbL ) − zα/2 ∗ se[KM (ϕbL )] = θ

and ϕbU satisfies

KM (ϕbU ) + zα/2 ∗ se[KM (ϕbU )] = θ.

b
Proof: We prove this argument for a general estimator S(t). So if we use the Kaplan-Meier
b
estimator, then S(t) is KM (t). It can also be the Nelson-Aalen estimator. Then

P [ϕbL < ϕ < ϕbU ] = P [S(ϕbU ) < θ < S(ϕbL )] (note that S(t) is decreasing and S(ϕ) = θ)

= 1 − (P [S(ϕbU ) > θ] + P [S(ϕbL ) < θ]).

Denote ϕU the solution to the equation

b
S(ϕU ) + zα/2 ∗ se[S(ϕ U )] = θ.

Then ϕbU will be close to ϕU . Therefore,

b ϕ
P [S(ϕbU ) > θ] = P [S(ϕbU ) > S( b ϕ
bU ) + zα/2 ∗ se[S( bU )]]
" #
b ϕ
S( bU ) − S(ϕbU )
= P b ϕ
< −zα/2
se[S( bU )]
" #
b
S(ϕ U ) − S(ϕU )
≈ P b
< −zα/2
se[S(ϕ U )]

≈ P [Z < −zα/2 ] (Z ∼ N(0, 1))

α
= .
2

PAGE 37
CHAPTER 2 ST 745, Daowen Zhang

Similarly, we can show that

α
P [S(ϕbL ) < θ] ≈ .
2

Therefore,
µ ¶
α α
P [ϕbL < ϕ < ϕbU ] ≈ 1 − + = 1 − α.
2 2

We illustrate this practice using a simulated data set generated using the following R com-
mands

> survtime <- rexp(50, 0.2)

> censtime <- rexp(50, 0.1)
> status <- (survtime <= censtime)
> obstime <- survtime*status + censtime*(1-status)
> fit <- survfit(Surv(obstime, status))
> summary(fit)
Call: survfit(formula = Surv(obstime, status))

time n.risk n.event survival std.err lower 95% CI upper 95% CI

0.0747 50 1 0.980 0.0198 0.9420 1.000
0.0908 49 1 0.960 0.0277 0.9072 1.000
0.4332 46 1 0.939 0.0341 0.8747 1.000
0.4420 45 1 0.918 0.0392 0.8446 0.998
0.5454 44 1 0.897 0.0435 0.8161 0.987
0.6126 43 1 0.877 0.0472 0.7887 0.974
0.7238 42 1 0.856 0.0505 0.7622 0.961
1.1662 40 1 0.834 0.0536 0.7356 0.946
1.2901 39 1 0.813 0.0563 0.7097 0.931
1.3516 38 1 0.791 0.0588 0.6843 0.915
1.4490 37 1 0.770 0.0609 0.6594 0.899
1.6287 35 1 0.748 0.0630 0.6342 0.882
1.8344 34 1 0.726 0.0649 0.6094 0.865
1.9828 33 1 0.704 0.0666 0.5850 0.847
2.1467 32 1 0.682 0.0680 0.5610 0.829
2.3481 31 1 0.660 0.0693 0.5373 0.811
2.4668 30 1 0.638 0.0704 0.5140 0.792
2.5135 29 1 0.616 0.0713 0.4910 0.773
2.5999 28 1 0.594 0.0721 0.4683 0.754
2.9147 27 1 0.572 0.0727 0.4459 0.734
2.9351 25 1 0.549 0.0733 0.4228 0.713
3.2168 24 1 0.526 0.0737 0.3999 0.693
3.4501 22 1 0.502 0.0742 0.3762 0.671
3.5620 21 1 0.478 0.0744 0.3528 0.649
3.6795 20 1 0.455 0.0744 0.3298 0.627
3.8475 18 1 0.429 0.0744 0.3056 0.603
4.8888 16 1 0.402 0.0745 0.2800 0.578
5.3910 15 1 0.376 0.0742 0.2551 0.553
6.1186 14 1 0.349 0.0736 0.2307 0.527
6.1812 13 1 0.322 0.0726 0.2069 0.501
6.1957 12 1 0.295 0.0714 0.1837 0.474
6.2686 10 1 0.266 0.0701 0.1584 0.445
6.3252 9 1 0.236 0.0682 0.1340 0.416

PAGE 38
CHAPTER 2 ST 745, Daowen Zhang

6.5206 7 1 0.202 0.0663 0.1065 0.385

7.1127 6 1 0.169 0.0632 0.0809 0.352
9.3017 3 1 0.112 0.0623 0.0379 0.333
11.1589 1 1 0.000 NA NA NA

The true survival time has an exponential distribution with λ = 0.2/year (so the true mean

is 5 years and median is 5 ∗ log(2) ≈ 3.5 years). The (potential) censoring time is independent

from the survival time and has an exponential distribution with λ = 0.1/year (so it is stochas-

tically larger than the survival time). The Kaplan estimate (solid line) and its 95% confidence

intervals (dotted lines) are shown in Figure 2.5, which is generated using R function plot(fit,

xlab="Patient time (years)", ylab="survival probability"). Note that these CIs are

constructed by exponentiating the CIs for Λ(t). From this figure, the median survival time is

estimated to be 3.56 years, with its 95% confidence interval [2.51, 6.20].

Figure 2.5: Illustration for constructing 95% CI for median survival time
1.0
0.8
survival probability

0.6
0.4
0.2
0.0

2.51 3.56 6.20

0 2 4 6 8 10

Patient time (years)

If we use symmetric confidence intervals of S(t) to construct the confidence interval for the

median of the true survival time, then we need to specify conf.type=c("plain") in survfit()

as follows

> fit <- survfit(Surv(obstime, status), conf.type=c("plain"))

We get the following output using summary()

PAGE 39
CHAPTER 2 ST 745, Daowen Zhang

> summary(fit)
Call: survfit(formula = Surv(obstime, status), conf.type = c("plain"))

time n.risk n.event survival std.err lower 95% CI upper 95% CI

0.0747 50 1 0.980 0.0198 0.9412 1.000
0.0908 49 1 0.960 0.0277 0.9057 1.000
0.4332 46 1 0.939 0.0341 0.8723 1.000
0.4420 45 1 0.918 0.0392 0.8414 0.995
0.5454 44 1 0.897 0.0435 0.8121 0.983
0.6126 43 1 0.877 0.0472 0.7839 0.969
0.7238 42 1 0.856 0.0505 0.7567 0.955
1.1662 40 1 0.834 0.0536 0.7292 0.939
1.2901 39 1 0.813 0.0563 0.7025 0.923
1.3516 38 1 0.791 0.0588 0.6763 0.907
1.4490 37 1 0.770 0.0609 0.6506 0.890
1.6287 35 1 0.748 0.0630 0.6245 0.872
1.8344 34 1 0.726 0.0649 0.5988 0.853
1.9828 33 1 0.704 0.0666 0.5736 0.835
2.1467 32 1 0.682 0.0680 0.5487 0.815
2.3481 31 1 0.660 0.0693 0.5242 0.796
2.4668 30 1 0.638 0.0704 0.5001 0.776
2.5135 29 1 0.616 0.0713 0.4763 0.756
2.5999 28 1 0.594 0.0721 0.4528 0.735
2.9147 27 1 0.572 0.0727 0.4296 0.715
2.9351 25 1 0.549 0.0733 0.4055 0.693
3.2168 24 1 0.526 0.0737 0.3818 0.671
3.4501 22 1 0.502 0.0742 0.3570 0.648
3.5620 21 1 0.478 0.0744 0.3326 0.624
3.6795 20 1 0.455 0.0744 0.3087 0.600
3.8475 18 1 0.429 0.0744 0.2834 0.575
4.8888 16 1 0.402 0.0745 0.2565 0.548
5.3910 15 1 0.376 0.0742 0.2302 0.521
6.1186 14 1 0.349 0.0736 0.2046 0.493
6.1812 13 1 0.322 0.0726 0.1796 0.464
6.1957 12 1 0.295 0.0714 0.1552 0.435
6.2686 10 1 0.266 0.0701 0.1283 0.403
6.3252 9 1 0.236 0.0682 0.1024 0.370
6.5206 7 1 0.202 0.0663 0.0724 0.332
7.1127 6 1 0.169 0.0632 0.0447 0.293
9.3017 3 1 0.112 0.0623 0.0000 0.235
11.1589 1 1 0.000 NA NA NA

The Kaplan estimate (solid line) and its symmetric 95% confidence intervals (dotted lines) are

shown in Figure 2.6. Note that the Kaplan estimate is the same as before. From this figure, the

median survival time is estimated to be 3.56 years, with its 95% confidence interval [2.51, 6.12].

Note: If we treat the censored data obstime as uncensored and fit an exponential model

to it, then the “best” estimate of the median survival time is 2.5, with 95% confidence interval

[1.8, 3.2] (using the methodology to be presented in next chapter). These estimates severely

underestimate the true median survival time 3.5 years.

PAGE 40
CHAPTER 2 ST 745, Daowen Zhang

Figure 2.6: Illustration for constructing 95% CI for median survival time using symmetric CIs

of S(t)

1.0
0.8
survival probability

0.6
0.4
0.2
0.0

2.51 3.56 6.12

0 2 4 6 8 10

Patient time (years)

Note:

If we want a CI for the quantile such as the median survival time with a different confidence

level, say, 90%, then we need to construct 90% confidence intervals for S(t). This can be done

by specifying conf.int=0.9 in the R function survfit().

If we use Proc Lifetest in SAS to compute the Kaplan-Meier estimate, it will produce 95%

confidence intervals for 25%, 50% (median) and 75% quantiles of the true survival time.

Other types of censoring and truncation:

• Left censoring: This kind of censoring occurs when the event of interest is only known to

happen before a specific time point. For example, in a study of time to first marijuana use

(example 1.17, page 17 of Klein & Moeschberger) 191 high school boys were asked “when

did you first use marijuana?”. Some answers were “I have used it but cannot recall when

the first time was”. For these boys, their time to first marijuana use is left censored at

their current age. For the boys who never used marijuana, their time to first marijuana use

is right censored at their current age. Of course, we got their exact time to first marijuana

PAGE 41
CHAPTER 2 ST 745, Daowen Zhang

use for those boys who remembered when they first used it.

• Interval censoring occurs when the event of interest is only known to take place in an

interval. For example, in a study to compare time to cosmetic deterioration of breasts

for breast cancer patients treated with radiotherapy and radiotherapy + chemotherapy,

patients were examined at each clinical visit for breast retraction and the breast retraction

is only known to take place between two clinical visits or right censored at the end of the

study. See example 1.18 on page 18 of Klein & Moeschberger.

• Left truncation occurs when the time to event of interest in the study sample is greater

than a (left) truncation variable. For example, in a study of life expectancy (survival time

measured from birth to death) using elderly residents in a retirement community (example

1.16, page 15 of Klein & Moeschberger), the individuals must survive to a sufficient age to

enter the retirement community. Therefore, their survival time is left truncated by their

age entering the community. Ignoring the truncation will lead to a biased sample and the

survival time from the sample will over estimate the underlying life expectancy.

• Right truncation occurs when the time to event of interest in the study sample is less

than a (right) truncation variable. A special case is when the study sample consists of

only those individuals who have already experienced the event. For example, to study the

induction period (also called latency period or incubation period) between infection with

AIDS virus and the onset of clinical AIDS, the ideal approach will be to collect a sample

of patients infected with AIDS virus and then follow them for some period of time until

some of them develop clinical AIDS. However, this approach may be too lengthy and costly.

An alternative approach is to study those patients who were infected with AIDS from a

contaminated blood transfusion and later developed clinical AIDS. In this case, the total

number of patients infected with AIDS is unknown. A similar approach can be used to

study the induction time for pediatric AIDS. Children were infected with AIDS in utero or

at birth and later developed clinical AIDS. But the study sample consists of children only

known to develop AIDS. This sampling scheme is similar to the case-control design. See

PAGE 42
CHAPTER 2 ST 745, Daowen Zhang

example 1.19 on page 19 of Klein & Moeschberger for more description and the data.

Note: The K-M survival estimation approach cannot be directly applied to the data with the

above censorings and truncations. Modified K-M approach or others have to be used. Similar to

right censoring case, the censoring time and truncation time are often assumed to be independent

of the time to event of interest (survival time). Since right censoring is the most common

censoring scheme, we will focus on this special case most of the time in this course. Nonparametric

estimation of the survival function (or the cumulative distribution function) for the data with

other censoring or truncation schemes can be found in Chapters 4 and 5 of Klein & Moeschberger.

PAGE 43