How to correctly interpret log-rank test results

#1
I have question regarding the log rank test. I understand that this test is used to know whether two or more groups have different survival distributions..
I used this test to compare the survival distribution for 2 cohorts and my p-value result is 0.2, does this mean that there is not enough evidence to reject the null hypothesis: the survival distributions are equal and that it means that the survival distributions are indeed different?

From this link, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC403858/ it seems to suggest that the log-rank test can be a poor indicator if there are overlaps between the survival curves. is that correct?
 

obh

Active Member
#2
Hi red,

Correct 0.2 is not significance (for common α=0.05 or even bigger ...) if you would reject the h0: equal survival, the chance of error would be too big 0.2.

"It is unlikely to detect a difference when survival curves cross"
I assume it unlikely in this case since for part of the cases the probability of an event is bigger for group1 and on other cases the probability of an event is higher for group2.
I only guess that if the cross is, in the beginning, the above statement is not correct, but if it is toward the end it is probably correct.

I wouldn't judge that the log-rank test is not good
 

ondansetron

TS Contributor
#3
Also need to consider whether the curves cross because the survival functions are different and non-proportional OR the survival/hazard functions are actually the same and the proportional hazards assumption is not an issue (since the hazard functions are identical, i.e. the observed crossing is just artifact/random variation).
 
#4
Hi red,

Correct 0.2 is not significance (for common α=0.05 or even bigger ...) if you would reject the h0: equal survival, the chance of error would be too big 0.2.

"It is unlikely to detect a difference when survival curves cross"
I assume it unlikely in this case since for part of the cases the probability of an event is bigger for group1 and on other cases the probability of an event is higher for group2.
I only guess that if the cross is, in the beginning, the above statement is not correct, but if it is toward the end it is probably correct.

I wouldn't judge that the log-rank test is not good
thanks for the replies. regarding the log rank test again, I read somewhere that it is also important to note the follow up time between the two groups observed. So my question is if I have one group being observed say between 2010 until 2013 and another group between 2014 and 2016, should they have the same intervals between the follow up times? in my case the intervals could vary but most of it are done almost once a month.