A Confidence Interval For The Median Survival Time

A Confidence Interval for the Median Survival Time
Author(s): Ron Brookmeyer and John Crowley

Source: Biometrics, Vol. 38, No. 1 (Mar., 1982), pp. 29-41
Published by: International Biometric Society
Stable URL: http://www.jstor.org/stable/2530286 .
Accessed: 27/03/2013 14:49
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
International Biometric Society is collaborating with JSTOR to digitize, preserve and extend access to
Biometrics.
http://www.jstor.org
This content downloaded from 129.25.131.235 on Wed, 27 Mar 2013 14:49:55 PM

All use subject to JSTOR Terms and Conditions
BIOMETRICS 38v 29-41
March 1982
A Confidence Interval for the Median SurvivalTime

Ron Broohneyerl alld John Crowley2
Departmentsof Statisticsand HumanOncology,Universityof Wisconsin-Madison,
1120 West JohnsonStreet, Madison,Wisconsin53715, U.S.A.
SUMMARY
A nonparametricasymptoticconfidenceintervalfor the median survivaltime is developed for the
case where data are subject to arbitraryright censoring. This is accomplishedby inverting a
generalizationof the sign test for censoreddata. A simulationstudy shows that this nonparametric
confidenceintervalperformswell for a variety of underlyingsurvivalfunctions.The procedureis
appliedto data from a clinicaltrial that comparedf'ourdosage regimensof 5-uorouracil.
1. Introduction
It has become a commonpracticein the medicalliteratureto give a point estimatefor the
mediansurvivaltime. Peto et al. (1977) observed that this can be very misleading:if the
survivalcurve is relativelyflat in the neighborhoodof 50/Osurvival,there can be a great
deal of variability in the estimated median. It would be more appropriateto cite a
confidenceintervalfor the median.
Confidenceintervalscan be derived with the assumptionof a parametricform for the
underlyingsurvivaldistribution.The results of Bartholomew(1957), for example, lead to
an interval estimate for the median when the survivalcurve is exponential. However, a
distribution-freeconfidenceintervalis often desirable.
In the one-sample problem without censoring,a nonparametricconfidenceintervalfor
the median is developed by inversion of the sign test. Similarly,for the problem with
censoring, a nonparametricconfidence interval for the median can be obtained by
generalizationof the sign test to enable censoreddata to be handled.This generalizedsign
test is given in 3, and the resultingconfidenceintervalis presented in 4. Some Monte
Carloresultsof the proposed confidenceintervalare given in 5, followed by an example
using data from a clinical trial of treatmentsfor colorectal cancer.
2. The SurvivalFunction and its Quanfiles
Let X1, X2, . . ., X be the true survivaltimes of a sample of size n. They are assumedto
be irldependent,identically distributedrandom variables with survival function S(t)=
pr(X:>t) and cumulativedistributionfunction F(t)=1-S(t). When observationsare
subject to arbitraryright censoring, the period of followup for the ith individual is
restricted to an amount Ti. Then the observed survival time for the ith individual is
Xi = min(X, Ti). One also observes bi, which will indicate if Xi is censored or not. Thus,
if Xi <X, the observationis said to be censored and we set bi= 0. On the other hand, if
Xi = X, the death is observed and we set bi = 1.
Present address:Department of Biostatistics,School of Hygiene and Public Health, The Johns

Hopkins University,615 N. Wolfe Street, Baltimore,Maryland21205, U.S.A.
2 Present address: Fred Hutchinson Cancer Research Center, 1124 Columbia Street, Seattle?
Washington98104, U.S.A.
Key words: Censoring;Confidenceintervals;Medians;Sign test; Survivalanalysis.
29

30 Biometrics,Murch1982
Kaplan and Meier (1958) introduced the product-limitestimator S0(t) of a survival

function. It is defined as
1-F(t) - S(t)- rI (ni -di)

distinct tti
Xist
The product is taken over all distinct observed survival times {Xi};ni is the numberof
observationsgreaterthan or equal to Xi (that is, ni is the numberof individualsat risk at
time Xi), and di is the numberof observeddeaths that occurredat time Xi In case of a tie
between a censored observationand an observed deathSthe convention is adopted that
the observed death is ranked before the censored observation.If the largest observation,
say X(n),is censored, then S(t) is defined to be O for tX(n).
The estimator S(t) is a right-continuousstep function which jumps at the observed
death points. It has some desirable properties. Kaplan and Meier showed that it is the
distributionfunction which maximizesthe likelihood of the observationswithin the class
of all cumulative distributionfunctions which have probabilitymass at each observed
death time. Efron (1967) showed that the Kaplan-Meier estimate S(t) of a survival
distributionS(t) satisfies the self-consistencyrelation
raS(t) = NX(t)+ E (1-bi) S(X) ' (1)
where Nx(t) is the numberof Xt greaterthan t, it is assumedthat there are no ties among
the observed survival times; note that Efron originally worked with left-continuous
survivalfuIlctions.
The pth quantile of the distributionF is defined to be (F)-l(p)=inf{t:F(t)p}.
J. Sander,in TechnicalReport No. 5, Biostatistics,StanfordUniversity(1975), studiedthe
large sample propertiesof (F0)1(p) under the model of randomcensorship,and showed
under suitable regularityconditions that n2{(Ft')-l(p)-(F)-l(p)} converges weakly to a
zero-mean Gaussianprocess. As in the uncensoredcase, the expressionfor the varianceof
the pth sample quantile depends explicitly on f0{(F0)-1(p)},where f is the continuous
density correspondingto F. The problem with the use of this variance to derive a
confidenceinterval is the difficultyin estimatingthe density at the median.
3. A Sign Test for Censored Data

As discussed in 2, estimates of a survival distribution,and thus its quantiles, can be
readily obtained. A point estimate should be accompaniedby a confidenceinterval.This
will be accomplishedby inversion of a generalizedsign test for censored data.
First, we eXtendthe sign test to the censored-dataproblem. Given a sample of size n,
we are interestedin the hypothesis-testingproblem with Ho median of survivalfunction
S = M; H1: median of survival function S + M. Consider the scoring function
Q(Xi, bi)- pr(Xj>M | Xi, bi,S). This conditionalprobabilityis to be interpreted,in the
spirit of Efron (1967), as if Xi actuallyhad the distributionS. Then,
rl Xi>M,
Q(Xi, bi)- t S(M)/S(Xi), Xi < M, bi = ,
tO, Xi<M, bi-1.

ConfidenceIntervalfor Median Survival 31
Define the test statistic U= (l/n) 1 Q(Xi, bi). Then we have the followingresult:
Theorem.The test statistic U (used to test Ho median=M) is equal to S(M), the
Kaplan-Meierestimate evaluated at M.
The proof of the theorem, which uses the propertyof self-consistency,is contained in
Appendix 1.
The asymptoticdistributionof the Kaplan-Meierestimate is well known. Breslow and
Crowley (1974) showed that
S(M)-S(M)
r
where a2=tS0(M)}2lo (l-F)-2dF. Here F is the cumulativedistributionof observed

lifetimes, i.e. 1-F(t) = pr(Xi> t), and F(t) is the distributionof true observed lifetimes,
i.e. F(t) = pr(XiS t, bi = 1). A consistent estimate of Cr2 (Greenwood, 1926) is given by
distinct Nx(Xi){Nx(Xi)+ di} (2)

XisM
where di is the numberof observeddeaths at Xi and NX(Xi)is the numberof patientswith

observed survivaltimes larger than Xi. This estimate can be used when there are ties in
the data, as well as in the no-ties situation.
Then
{g (M)-S 0(M)}2/{S(M)} di L t NX(Xi){NX(Xi) + di}

XiSM
is approximatelyX2(l)
Under the null hypothesis S(M)=2, and an approximateoe-leveltest is not to reject
Ho when
{S(M)-2}2< c{S(M)} iE t NX(Xi){NX(Xi)+4}

XiSM
where c,xis such that pr{x2(1)> c,x}= oe.
4. A ConfidenceIntervalfor the Median

An asymptoticl -oe confidenceregion R,xfor the median is immediatelyobtained as the
set of all parametervalues not rejected by the sign test at level o. That is,
R ={ml{s(m)-2}26 ct{s (m)} disEnctNx(Xi){Nx(Xi)+di}}

XiAM
To find the region Rs,e,it is necessary only to check if observed death times are in the
region.This is because the Kaplan-Meierestimate and its estimatedvariancejump only at
observed death times. Thus, if tl and t2 are two consecutive observed death times with
tl<t2 and tleR,X, then the interval [tl, t2) is contained in R,,. One would expect the
confidence region Rs, to be an interval which includes the estimated median. This is
generallytrue, as the next results show.

32 Biometrics,March 1982
We shall assumethat the n observedsurvivaltimes X1, . . ., Xn are distinct,i.e. that no

ties occur. Let tl and t2 be consecutive observed death times with tl < t2 and associated
Kaplan-Meier estimates S1 and S2, respectively. The following results are proved in
Appendix 2:
PropertyA. If S2 > .5, then tl E R,x implies t2E R.
PropertyB. If S1<.5 and tl XR,x,then t24 Ra if ca < 1.
In practice, however, we are usually interested in values of c,x>3. Unfortunately,
PropertyB does not hold for arbitraryc,x,as seen from the following example. Let ni be
the number of patients at risk at observed death times t(ni). Consider the following
sequence {ni}:
4500, 499, 498, . . ., 201, 200, 6, 5, 4, . . ., 1}.

A large numberof patientswere censored at death time t(200). A calculationwith c,x= 8
shows that S200= .398, t(200) Ro,but t(6) E Ro,.However, such examplesare rare and are
characterizedby an extreme censoringpattern.
Simulationresults suggest that the confidenceregion is almost always an interval (see
5). Thus, it is reasonableto consideronly the intervalpart of the confidenceregion R,x.
Let {ti} be the ordered observed distinct death times. Define the confidence interval
l<X= [ti, t>),where ti is the smallestobserved death time in R,xwith S(ti)> .5, and tj is the
smallest observed death time not in R,xwith S(tj)<.5.
Occasionallyit happens that an upper confidencelimit cannot be obtained. If the last
observeddeath time is in I,x,then I,, becomes a one-sided confidenceintervalof the form
[ti,oo). Furthermore, if the Kaplan-Meier survival curve does not reach the median
because of extensive censoring, only a lower confidence limit (or perhaps an empty
interval)can be obtained.
An alternativeconfidenceregion Re can be derivedby replacingS(M) in the estimate
Cr2 by 2, the value of S(M) under Hov but this often leads to a one-sided confidence
interval of the form [ti,oo). This can be verified analytically. Simulation results also
supportedthe conclusionthat the region Re was far too frequentlyone-sided. In addition,
we are not guaranteedan interval, as the example given above still applies.
5. SimulationResults
In order to investigate the coverage probabilityand length of the nonparametricconfi-
dence interval I,x,a computer simulationwas performed.The nonparametricinterval I,x
was comparedto two parametricintervalswhich assume an exponentialsurvivaldistribu-
tion.
The results of Bartholomew (1957) can be used to derive the maximum likelihood
estimate of the median and its asymptotic variance. For an exponential(A)survival
distributiongiven by S(t)=exp(-At),A>0, the maximum likelihood estimate of the
median M is
^ ln 2 E (observedsurvivaltimes)
M= d '
where d is the numberof observeddeaths. lDheestimatedvarianceof M is M/ Pi, where
i P( M )'

Confidence Interval for Median Survival 33
Ti being the censoringtime of the ith patient. Here, we have used the expected Fisher
informationto obtain the estimate of the variance. Thus an approximate(1-oe)-level
confidenceinterval,called here the 'Bartholomewconfidence interval',is given by
{ (Pi)2'M+(Pi)24' (3)
where z2cy the 100(1-2oe) percentage point of an N(0, 1) distribution.In order to
iS
calculatethe Bartholomewconfidenceintervalgiven by (3), the censoringtimes {Ti}of all

patientsmust be known. This is often not the case; in particular,the censoringtimes are
usually not known for those patients who actuallydied.
The 'variance-stabilizedconfidence interval' is derived for an exponential survival
distribution by transformationof the confidence interval for log A. The asymptotic
varianceof the maximumlikelihood estimate A is estimatedby A2/d,so l/d estimatesthe
A
variance of logA. Here, we have used the observed Fisher informationto obtain the
estimate of variance. The confidence limits for log A are transformed to provide a
confidenceintervalfor the median.The variance-stabilized1-oe confidenceintervalfor M
is then
[ ln2 ln2 )
lexp(ln A+ d-2z2,x)' exp(ln A-d-2Z-,x) J '
where A = d/ (observedsurvivaltimes).
Three experimentswere performed,each with a differentunderlyingsurvivaldistribu-
tion (exponential, Weibull and Rayleigh). Figure 1 gives a plot of the three hazard
functionscorrespondingto the three distributions.For each experiment,400 simulations
were completed.For every simulation,survivaland censoringtimes were generatedfor 50
patients. The censoring distributionwas assumed to be uniform on [0, T]. For each
simulation a 1- oe confidence interval for the median survival time was computed
(oe=.01,.05,.10,.20 and .25). Then the observed coverageprobabilitywas recorded,i.e.
20
1S _
A(t) Weibull/ /
(X 103) 10 * / /
/ / Exponentis
5 / /
X Rayleigh
I x I I . .
20 40 60 80 100
TIME
Figure 1. Hazardsfor three survivaldistributions.Exponential:A(t) .01. Weibull: A(t)-.0018t2.

Rayleigh:A(t)=.0002t.

Biometrics,March 1982
34
the fractionof the 400 computedconfidenceintervalswhich containedthe true median. If
the interval estimates were performing adequately, the observed coverage probability
would be close to 1- oe.Several values of T were used (T - 650, 500, 250 and 150), and
for each the percentageof patients with censored survivaltimes was recorded.
In order to compare interval lengths for the three types of confidence intervals
(nonparametric,Bartholomew and variance-stabilized),average interval lengths were
computed.If the censoringwas extensive (roughlygreaterthan 40%), the nonparametric
method occasionallyyielded only one-sided intervals.In these instances,average lengths
could not be comparedfairly. In addition,when average intervallengths were computed,
so were averagevalues of the Lehmannloss function. For an intervalfor M of the form
[ti,tj), this is defined (Lehmann,1959, p. 82) to be
[tj-ti if ti<M6tj,
L(M,ti,tj)=itj-M if M<ti,
(M-ti if tj>M.
A check was performedto see if the confidenceregionswere actuallyintervals.This was
the case in all of the simulatedsamples.
Survivaltimes for Experiment1 were generatedfrom an exponential(.01) distribution.
The observed coverage probabilities are given in Table 1. For the three types of
confidence intervalsconsidered,the observed coverage probabilitieswere approximately
the same. However, the average lengths of the confidence intervals were somewhat
smaller for the parametricintervals than for the nonparametricintervals (see Table 2).
Average values of the Lehmannloss function were similarto average intervallengths.
Survivaltimes for Experiment2 were generatedfrom a Weibull(.0012, 1.5) distribution
[S(t)=exp(-AtY), with A=.0012,=1.5]. Coverageprobabilitiesare given inTable 3.
The observed coverage probabilities for the nonparametricinterval were consistently
good. However, when there was little censoring the observed coverage probabilitiesfor
the parametricintervalswere poor. The averageconfidenceintervallengthswere approxi-
mately the same for the three types of intervals(Table 4), as were average values of the
Lehmannloss function.
Table1
l-oe confidence intervals: observed coverage probabilities. Exponential (.01) survival
distribution
oe
Confidence interval % censored

.01 .05 .10 .20 .25
Nonparametric 15 .015 .065 .100 .230 .270

20 .012 .060 .120 .220 .255
37 .010 .057 .120 .227 .272
52 .012 .055 .107 .220 .270
Bartholomew 15 .022 .055 .102 .202 .237
20 .022 .065 .102 .235 .275
37 .025 .063 .115 .220 .260
52 .022 .063 .095 .217 .297
Variance-stabilized 15 .010 .060 .112 .200 .262
20 .010 .060 .122 .212 .267
37 .012 .067 .12S .217 .265
52 .007 .060 .107 .235 .313

ConfidenceIntervalforMedian Survival 35
Table 2
Average confidence interval length. Exponential (.01) survival
distribution
15% censored 20% censored
a .ol .10 .01 .10
Nonparametric 74.4 48.0 75.9 48.6

Bartholomew 55.5 35.5 57.1 36.5
Variance-stabilized 57.1 35.9 58.8 37.0
Table 3
1- oe confidence intervals:observed coverage probabilities.Weibull (0.0012, 1.5)
survivaldistribution
.
0e

.01 .05 .10 .20 .25
Nonparametric 12 .012 .057 .110 .227 .262

16 .010 .055 .102 .215 .252
32 .012 .057 .117 .235 .287
51 .005 .063 .102 .217 .272
Bartholomew 12 .102 .245 .365 .510 .587
16 .090 .200 .297 .465 .522
32 .032 .095 .160 .245 .300
51 .005 .030 .060 .135 .195
16 .032 .125 .227 .405 .472
32 .012 .057 .055 .165 .272
51 .005 .027 .055 .165 .217
Table 4
Averageconfidenceintervallength.Weibull(.0012, 1.5) survivaldistribution
12% censored 16% censored 32% censored
oz: .01 .10 .01 .10 .01 .10
Nonparametric 48.4 32.0 49.1 32.5 53.7 34.8

Bartholomew 45.3 28.9 47.0 30.0 57.4 36.7
Variance-stabilized 46.4 29.2 48.1 30.3 58.6 36.8

Table 5
l-ot confidence intervals: observed coverage probabilities. Rayleigh (.0001) survival
distribution
of

.01 .05 .10 .20 .25
Nonparametric 14 .010 .060 .105 .220 .265

18 .007 .057 .120 .215 .280
36 .007 .060 .132 .232 .280
57 .015 .055 .117 .205 .240
Bartholomew 14 .167 .422 .570 .775 .840
18 .117 .335 .477 .670 .737
36 .020 .080 .155 .247 .290
57 .002 .007 .020 .085 .130
18 .032 .192 372 .600 .688
36 .005 .035 .097 .217 .265
57 .002 .020 .057 .167 .212
Survivaltimes for Experiment 3 were generated from a Rayleigh (.0001) distribution

[S(t) = exp(-At2), with A = .0001]. Results are given in Tables S and 6. The nonparamet-
ric confidence interval performedwell for all degrees of censoring. However, observed
coverage probabilitiesfor the parametricconfidence intervalswere very poor when the
censoringwas less than 20%.As the censoring increased, the observed coverage prob-
abilitiesfor the parametricintervalsimproved,but the intervallengths increaseddramati-
cally. Average parametric interval lengths were larger than the nonparametricones.
Similarresults were obtained for the Lehmannloss function.
The simulationresultsindicatethat the parametricmethodsperformwell when the data
arise from an exponential distribution. When the data are not generated from an
exponential model, the observed coverage probabilities for the parametric intervals
fluctuatewildly with the degree of censoring.The nonparametricmethod performedwell
for the three survivaldistributionsinvestigated. In addition, the nonparametricinterval
length and the Lehmann loss function were reasonable when compared against the
parametriccompetitors.Although the nonparametricmethod performedwell in moderate
sample sizes, the normal approximationto the distributionof S may be inadequate in
small samples. Suitable transformationsof S may improve the approximation;however
they have not been consideredhere.
Emerson (1982) related the generalizedsign test discussedhere to a binomialdistribu-
tion for small samples. Following a similarsimulationdesign to comparethe two studies,
Table 6
Average confidence interval length. Rayleigh (.0001) survival distribution
l
14% censored- 18% censored 36% censored

o: .01 .10 .01 .10 .01 .10
Nonparametric 43.8 29.0 44.4 29.2 49.1 32.1

Bartholomew 51.5 33.0 54.1 34.6 70^0 44.8
Variance-stabilized 52.7 33.2 55.1 34.7 70.9 44.4

99 /O [37,88)
ConfidenceIntervalfor MedianSurvival
37
he obtained results which appear quite good. Efron (1981) developed a confidence
interval by the method of bootstrappingcensored samples; that is, the distributionof
the sample median is estimatedby Monte Carlo simulationof samplesgeneratedfrom S.
These techniquesare summarizedand reviewed by Reid (1981).
6. An Example
The confidenceintervalprocedurediscussedearlier was applied to data from a Phase III
colore,ctalcance,rclinicaltrial (Ansfieldet al., 1977). [The data analyzedwere an updated
versionreportedin the CentralOncologyGroupFinal Report (COG 7030), Spring1977.]
Four dosage regimensof S-fluorouracilwere compared.Table 7 summarizesthe data for
each of the four groups. The 95/Oconfidenceintervalsfor each of the four medians are
also shown. Treatment1, an intravenousloading-courseschedule, had the largest median
survival time of 61 weeks. Although the medians 'appear' to be different, all four
confidence intervals overlap. In fact, the Gehan (Breslow) test shows no significant
difference among the groups (P=.22). Since there were ties in the data, the Kaplan-
Meier variancewas estimatedby (2). Table 8 details the calculationsfor Treatment 1. A
numerical check shows that all the confidence regions are indeed proper intervals
(R,,,--I,,,). Figure 2 shows'the Kaplan-Meierestimate for Treatment 1 and several 1-oe
confidenceintervals (oc=.01,.05,.10,.20 and .25).
75 /O C43,64)
80% [4C)64)
100 90% [39 73)
90-q%, 95% [38,73)
80- \
70- tbo
> 60- 50 ^
(t) 50 _ i3'o 4t
a} 40- O
980 O
0 95 '-_
10 - I 99 . 1 -0vv
O ' I I I 1 1 1 , I I
O 25 50 75 100 125 150 175 200 225
Weeks
Figure2. Confidenceintervalsfor Treatment1.

38 Biometrics, March 1982
Table 7
Samplesizes, mediansand 95% confidenceintervalsfor four treatments
_ , . . . .
Treatment
1 2 3 4
Censored 16 8 14 7
Observed deaths 37 48 44 45
Sample size 53 56 58 52
Median survival time (weeks) 61 41 47 29
95% nonparametric confidence
interval [38,73) [31,51) [28,60) [25,46)
95% variance-stabilized
confidence interval (38,73) (28,50) (30,55) (27,48)
Table 8
Calculations for 95% confidence interval for median of
Treatment1
Observed x
death time if
(weeks) S0 SE(S) (S _ 1)2 < 3 84{SE(S0)2}
6 .962 .027
11 .943 .032
13 .925 .037
14 .906 .040
15 .887 .044
16 .868 .047
21 .829 .052
22 .810 .055
23 .791 .057
25 .771 .058
26 .752 .060
27 .732 .062
28 .713 .063
31 .692 .064
34 .671 .066
37 .649 .067
38 .627 .069 x
39 .604 ? x
40 .582 .071 x
43 .559 .072 x
44 .534 .072 x
54 .508 .074 x
61 .423 .080 x
64 .395 .079 x
73 .338 .079
83 .308 .077
88 .277 .076
90 .246 .073
91 .215 .070
106 .179 .067
128 .144 .062
144 .096 .057

ConfidenceIntervalfor MedianSurvival
39
ACKNOWLEDGEMENTS
This researchwas supportedin part by NIH grantCA-18332. We would like to thank Dr
John Emersonfor some helpful commentson an earliermanuscript,and Dr Fred Ansfield
for permissionto use his data.
RESUME
Un intervalle de confiance asymptotiquenon parametriquepour le temps de survie median est
developpe dans le cas ou les donnees sont arbitrairementtronquees a droite. Ceci est realise a
l'aide d'une generalisationdu test des signes pour les donnees tronquees. Une etude de simulation
montre que cet intervallenon parametriqueconvient bien pour une variete de fonctions de survie
sous jacentes. La procedureest appliquee aux donnees d'une experience clinique comparantdes
regimessuivantquatredosages de 5-fluorouracil.
REFERENCES
Ansfield, F., Klotz, J. and the Central Oncology Group (1977). A Phase III study comparingthe
clinicalutility of four dosage regimensof 5-fluorouracil.Cancer39, 34-40.
Bartholomew,D. J. (1957). A problemin life testing.Joumalof theAmericanStatisticalAssociation
52, 350-355.
Breslow, N. and Crowley, J. (1974). A large sample study of the life table and product limit
estimates under randomcensorship.Annals of Statistics2, 437-453.
Efron, B. (1967). The two-sampleproblemwith censoreddata. In Proceedingsof the Fifth Berkeley
Symposiumon MathematicalStatisticsand Probability,Vol. 4, L. LeCam and J. Neyman (eds),
831-854. Berkeleyt Universityof CaliforniaPress.
Efron, B. (1981). Censoreddata and the bootstrap.Journalof the AmericanStatisticalAssociation
76, 312-319.
Emerson, J. (1982). Nonparametricconfidence intervalsfor the median in the presence of right
censoring.Biometrics38, 17-27.
Gehan, E. A. (1965). A generalized Wilcoxon test for comparing arbitrarilysingly-censored
samples. Biometrika52, 203-223.
Greenwood, M. (1926). The Natural Durationof Cancer.Reports on Public Health and Medical
Subjects,No. 33, 1-26. London: HMSO.
Kaplan, E. L. and Meier, P. (1958). Nonparametricestimation from incomplete observations.
Journalof the AmericanStatisticalAssociation 53, 457-481.
Lehmann,E. L. (1959). TestingStatisticalHypotheses.New York: Wiley.
Peto, R., Pike, M. C., Armitage, P., Breslow, N. E., Cox, D. R., Howard, S. V., Mantel, N.,
McPherson,K., Peto, J. and Smith, P. G. (1977). Design and analysisof randomizedclinical
trials requiring prolonged observation of each patient. II. Analysis and examples. British
Joumal of Cancer35, 1-39.
Reid, N. (1981). Estimatingthe median survivaltime. Biometrika68, 601-608.
ReceivedJuly 1980; revisedJanuary1981
APPENDIX1
Proofof the Theorem
To prove that the test statistic U is the Kaplan-Meierestimate at M, write
1 lt
U - - E Q(Xi, bi)
n i=l
1 n
--E pr(Xi> M | Xi, Ai,;)
n i=l
=-{ E; pr(Xi > M | Xi, bi, S) + E pr(Xi > M | Xi, bi, S) }.

n {xixi=1} {xixi=o}
We have
E pr(Xi> M | Xi, bi,S) = N1(M),
{X,: Xi = 1}

where NX(1\4)
= numberof Xi > M with bi = 1. Also,
E pr(Xt'>M|Xi,Ai,S)=NX(M+ E (1-bi)S (M)

{Xi: Xi =} X, SM S ( i)
where NX(M)
= numberof Xi >M with Ai= O. Thus,
U=-{Nx(M)+Nx(M)+ , S(X) }
=-{NX (M) + E ( A i )X ( )} '

n Xi<M S ( i)
Comparisonof the term in bracketswith the definitionof selfconsistencygiven by (1) shows that
U-(l/n){nS(M)}= S(M)-
APPENDIX2
Proofs of PropertyA and PropertyB

Proof of PropertyA. Let
i XEtj NX (Xi ){NX (Xi ) + 1} 1
and let N be the number of patients at risk at t2. Then tleRa implies (Sl-2)2<c,,(S)2V1 and
{NS2/(N-1)-2}2 < {N/(N-1)}2(S 2)2c,,V1, since g = NS 2/(N-1). Simplifying,
(g 0)2 _ N-1 S 2 + 1 (N ) < (S2)2c<xVl

.
Adding (S2)2c,,/{(N-1)N} to both sides of the inequalitygives
(S2)2- N S2+4- ( N ) +(N 1)tN<(S2)2C>V2' (4)

since V2 = V1 + 1/{(N- 1)N}. Thus it suices to show that (S2_2)2 is less than the left-handside of
(4). We have
+ cA 1 +1(l 2+ 1 )+ c,,
< S2 + 1 {N-102 C<X(S2)2 S2 1 {N-182 c (S0)2

N2 4\ N / N2 N 4\ N J N(N-1)
since S(9) -20
This implies that
(S 2)2-S 2 + 1 < (S 2)2-S 2+ g 2 + 1 (N 1) + C<X

(g2)
Comparisonwith (4) gives the desired result:

(g2-2) <(g2) C<xv2
Proof of PropertyB. Since tl + Ra, we have (4) with the inequalitysign reversed:
(g2)2_ N 1 g2+1 (_) +(N 2)l)N>(92) c<,V2.

Confidence Interval for Median Survival 41
It sufficesto show that (S2-2)2 is greater than the left-hand side of the above inequality.Since
C(X < 1,
_>__ 1 + ct =N-1+l (N-1)2+ c<,

4 4 4N2 4N2 2N2 4 N 4N2
,S2+1 (N-1)2 c<,
N 4 N 4N2
S 1 (N-1)2+ cXx(g2)
N 4 N N(N-1)
since S <2 which implies g2 <2(N- 1)/N. This gives
(S 2-2) = (S2) -S 2+4 > (S2) cxx
V2


A Confidence Interval For The Median Survival Time

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

A Confidence Interval For The Median Survival Time

Diunggah oleh

Hak Cipta:

Format Tersedia

A Confidence Interval for the Median Survival Time

Author(s): Ron Brookmeyer and John Crowley

This content downloaded from 129.25.131.235 on Wed, 27 Mar 2013 14:49:55 PM