# Floating Point Overflow while computing the Kaplan-Meier estimator in SAS

I try to estimate survival curves based on a Kaplan-Meier estimator using proc lifetest. However, SAS outputs an error message which I do not manage to circumvent. Can you help me? The data set is available here.

My code

filename work_di "TO BE COMPLETED";
data data;
infile work_di("data.txt") dlm="09"x firstobs=2;
input id centre time event x \$;
run;

proc lifetest data=data method=KM plots=none;
time time * event(0);
by centre x;
ods output ProductLimitEstimates = surv;
run;


Error message

NOTE: The LOGLOG transform is used to compute the confidence limits for the quartiles  of the
survivor distribution. To suppress using this transform, specify CONFTYPE=LINEAR in the
PROC LIFETEST statement.
ERROR: Floating Point Overflow.
NOTE: The data set WORK.SURV has 77 observations and 10 variables.
ERROR: Termination due to Floating Point Exception
NOTE: The SAS System stopped processing this step because of errors.
NOTE: PROCEDURE LIFETEST used (Total process time):
real time           0.09 seconds
cpu time            0.03 seconds


There are quite a few 1e+308 as time to event in your dataset (variable time). I'd say that your problem is due to that.

                                 The MEANS Procedure
Analysis Variable : time

N            Mean         Std Dev         Minimum         Maximum
--------------------------------------------------------------------
2250               .               .       0.3301358           1E308
--------------------------------------------------------------------

• Yes... indeed ! This should not occur of course! Moreover, those guys are not censored :-S. I will contact the data manager ! Thanks for pointing that out. – ocram Mar 14 '12 at 9:47
• Also note that the log-log transformation (which I used to use until Terry Therneau pointed this out) creates instability in the very early part of the survival curve. For most problems basing confidence intervals on log S(t) is better. – Frank Harrell Mar 14 '12 at 12:10
• @Frank Harrell: Thanks for the advise. Actually, I first though that the problem arised from this log-log transform, and I tried to prevent SAS to compute any confidence interval but it seems that this cannot be done... – ocram Mar 14 '12 at 12:18