AFAICS the curves don't actually cross in your graph.
You have 4 events in the smaller group and 1 in the larger group,
and a weakly significant p-value looks plausible to me. Of course, there's
no way for anyone to check without having the data.
-thomas
On Fri, 9 Mar 2007, carol white wrote:
Dear All,
It seems that p-value generated by survdiff is too low for my plot. for such KM
curves that cross each other (see attached), I didn't expect to get p-val =
0.03 (see result of survdiff below). Could you see where the discrepancy comes
from?
thanks
----------------------------------------------------------------------------------------------------------
metInd = which (ds[,3] == 0)
print (survdiff(Surv (dsSurv[metInd], ds[metInd,4])~ifelse(pred.ds[metInd] < 0,
0, 1)))
Call:
survdiff(formula = Surv(dsSurv[metInd], ds[metInd, 4]) ~
ifelse(pred.ds[metInd] < 0, 0, 1))
N Observed Expected (O-E)^2/E
ifelse(pred.ds[metInd] < 0, 0, 1)=0 125 1 3.22 1.53
ifelse(pred.ds[metInd] < 0, 0, 1)=1 69 4 1.78 2.78
(O-E)^2/V
ifelse(pred.ds[metInd] < 0, 0, 1)=0 4.31
ifelse(pred.ds[metInd] < 0, 0, 1)=1 4.31
Chisq= 4.3 on 1 degrees of freedom, p= 0.0379
plot (survfit(Surv (dsSurv[metInd], ds[metInd,4])~ifelse(pred.ds[metInd] < 0,
0, 1)))
---------------------------------
Sucker-punch spam with award-winning protection.
Try the free Yahoo! Mail Beta.
Thomas Lumley Assoc. Professor, Biostatistics
tlumley@u.washington.edu University of Washington, Seattle
|