Dear list,I am quite new to the world of biostatistics and I encounter some
issues in the precise understanding of the coxph function of the survival
package.I have a set of survival data (patient who had (or died from) a
breast cancer) I'd like to see which are the variables that might cause dead
or not.When trying> summary(coxph(Surv(Time_to_distant_recurrence_yrs,
!Distant_recurrence)~ as.factor(Herceptincat) + as.factor(nodeCat_all) ,
data = her2.matrix))I obtain
Call:coxph(formula = Surv(Time_to_distant_recurrence_yrs,
!Distant_recurrence) ~ as.factor(Herceptincat) + as.factor(nodeCat_all),
data = her2.matrix) n= 231, number of events= 53
coef exp(coef) se(coef) z Pr(>|z|) as.factor(Herceptincat)1 -0.5891
0.5548 0.2805 -2.100 0.03570 * as.factor(nodeCat_all)1 0.9718 2.6426
0.6195 1.569 0.11672 as.factor(nodeCat_all)2 1.9713 7.1803 0.6101
3.231 0.00123 **---Signif. codes: 0 â***â 0.001 â**â 0.01 â*â
0.05 â.â 0.1
â â 1 exp(coef) exp(-coef) lower .95 upper
.95as.factor(Herceptincat)1 0.5548 1.8025 0.3202
0.9614as.factor(nodeCat_all)1 2.6426 0.3784 0.7847
8.8989as.factor(nodeCat_all)2 7.1803 0.1393 2.1720
23.7368Concordance= 0.699 (se = 0.039 )Rsquare= 0.096 (max possible=
0.911 )Likelihood ratio test= 23.33 on 3 df, p=3.444e-05Wald test
= 21.76 on 3 df, p=7.312e-05Score (logrank) test = 24.46 on 3 df,
p=2.001e-05
I think, this means that the two variables I tested are of interest to build
the model with a p-value of about 10e-5. This also mean that having the
Herceptincat =1 is significantly different from having a Herceptincat = 0.
Moreover, having a nodecat_all iof 2 is significantly different from having
a nodecat equals to 0.However, this does not tell me much about the
variables alone.
summary(coxph(Surv(Time_to_distant_recurrence_yrs, !Distant_recurrence)~
as.factor(Herceptincat) , data = her2.matrix))Call:coxph(formula =
Surv(Time_to_distant_recurrence_yrs, !Distant_recurrence) ~
as.factor(Herceptincat), data = her2.matrix) n= 231, number of events= 53
coef exp(coef) se(coef) z Pr(>|z|)as.factor(Herceptincat)1 -0.4326
0.6488 0.2789 -1.551 0.121 exp(coef) exp(-coef)
lower .95 upper .95as.factor(Herceptincat)1 0.6488 1.541 0.3756
1.121Concordance= 0.568 (se = 0.035 )Rsquare= 0.011 (max possible= 0.911
)Likelihood ratio test= 2.45 on 1 df, p=0.1177Wald test = 2.41
on 1 df, p=0.1208Score (logrank) test = 2.44 on 1 df, p=0.118
So, if I only consider one variable (Herceptincat). This does not seem very
interesting. However, nodecat_all is far more interesting.Does it mean that
nodecat_all is enough for me to build a model and that I don't have to take
(Herceptincat). into account? I am bit lost between the global p-values and
those that accont for only one variables. It is even worse with factorized
variables because, you miss a p-value for the first category!I thank you all
for your help!
Call:coxph(formula = Surv(Time_to_distant_recurrence_yrs,
!Distant_recurrence) ~ as.factor(nodeCat_all), data = her2.matrix) n=
231, number of events= 53 coef exp(coef) se(coef)
z Pr(>|z|) as.factor(nodeCat_all)1 0.8525 2.3455 0.6173 1.381
0.16726 as.factor(nodeCat_all)2 1.8288 6.2264 0.6068 3.014 0.00258
**---Signif. codes: 0 â***â 0.001 â**â 0.01 â*â 0.05 â.â 0.1
â â 1
exp(coef) exp(-coef) lower .95 upper .95as.factor(nodeCat_all)1 2.346
0.4263 0.6995 7.865as.factor(nodeCat_all)2 6.226 0.1606
1.8956 20.452Concordance= 0.661 (se = 0.036 )Rsquare= 0.078 (max
possible= 0.911 )Likelihood ratio test= 18.84 on 2 df, p=8.111e-05Wald
test = 17.26 on 2 df, p=0.0001787Score (logrank) test = 19.9
on 2 df, p=4.785e-05
--
View this message in context:
http://r.789695.n4.nabble.com/coxph-and-variables-tp4663087.html
Sent from the R help mailing list archive at Nabble.com.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
