How to interpret these results !! Hazard Ratio

[AminSaad]

Parameter              DF   Parameter   Standard    Chi-Square   Pr > ChiSq      Hazard

Ratio                                  Estimate       Error

RF_CATEGORY   1  1       0.39888         0.54005       0.5455               0.4602             1.490                                       RF_CATEGORY 1

RF_CATEGORY   2  1       0.99090         0.52381       3.5785               0.0585             2.694                                       RF_CATEGORY 2

RF_CATEGORY   3  1       1.32557         0.55011       5.8063               0.0160             3.764                                       RF_CATEGORY 3

Effect                      DF  Wald Chi-Square      Pr > ChiSq

RF_CATEGORY   3       12.8814            0.0049

Hi everyone,

My background in statistic is not very solid. I have received this report and I'm not sure how to figure out whether the HR is significant or not? Could anyone explain. thanks.

[Ksharp]

If I was right check Pr > ChiSq

if alpha=0.05 , then only the last one is significant .

[AminSaad]

Ok I know this part about the 0.05, anything lower is statistically significant. What got me confused is the second table with the wald chi square ?? 0.0049 it reports this for only DF = 3,,, in other tables where the HR was significant for all there categories. I only got one p value for the wald  chi-square? what does it signify?

[JacobSimonsen · Accepted solution]

The p-value of 0.0049 test the hypothesis that all three parameters are zero.

DF=3 because there are three less in a model where all the four Groups are equal (four Groups because there is also the reference Group).

[AminSaad]

for the same variables I got the KM cureves. and there I have four curves with the reference and then I got this table? does that mean that the 3 curves are statistically different from the reference curve? and why do we have more that one test ?

test	chi square	df	pr>chisquare
Log-Rank	13.7304	3	0.0033
Wilcoxon	13.7798	3	0.0032
-2Log(LR)	13.8084	3	0.0032

[JacobSimonsen]

When the wald test (or log-rank or -log(LR)) are significant it can be due to only one big contrast between two Groups, and not neccesaryly that all four Groups differ from each other. It is correct as you say that four Groups can not be assumed to be equal.

Proc phreg produce Wald, LogRank and -2Log(LR), they are asymptotic equivalent. You can write "type3(wald)" as option in the modelstatement if you only want the wald test.