Basic Statistics

214 INTRODUCTION TO SURVIVAL ANALYSIS

2 4 6 a 10

Analysis time

Figure 14.7 Kaplan-Meier survival estimate.

14.4 COMPARISON OF CLINICAL LIFE TABLES ANDTHE

KAPLAN-MEIER METHOD

The Kaplan-Meier method is recommended when the sample size is small. It also

has the advantage that the user does not have to decide on the length of the inter-

val as must be done for clinical life tables. Many statistical programs that include

survival analysis in their output will display an estimate of the survival function us-

ing the Kaplan-Meier method. It is also possible to compute a mean survival time

and the variance of the survival time using methods provided by Kaplan and Meier

(see Gross and Clark [1975]) or a median (see Miller [1981]). Note that the mean

survival time is sometimes difficult to interpret because survival distributions can be

highly skewed. Some statistical programs will also print out the standard error of the

estimated survival function at each distinct time of death.

The clinical life table method is useful for larger sample sizes. It also has the

advantage of directly furnishing estimates of the hazard function and the death density

function. It is possible to compute the standard errors of the estimated survival

function, hazard function, and death density function at each interval.

Statistical programs often give the median survival time or it can easily be estimated

from the clinical life table. If the investigators lack access to a survival program that

produces a clinical life table, some spreadsheet programs can be used to perform much

of the work. The time in, say, days between dates may be obtained from spreadsheet

programs for each patient. The times can be sorted so that it is straightforward to obtain

the counts. Actually, it is simply necessary to know only which interval each length