# p values for hazard ratios

#### msarin

##### New Member
Using a Cox multivariate proportional hazard model in SAS, how does one get the p values for the revision risk (not the p values for the coefficients)? Below are fake data:

Variable Revision Risk p value
Age group in years
35-49 vs. 18-34 0.637 (0.200-2.026)
50-59 vs. 18-34 1.229 (0.883-1.711)
60-69 vs. 18-34 1.076 (0.794-1.458)
>=70 vs. 18-34 1.444 (1.079-1.933)

#### hlsmith

##### Less is more. Stay pure. Stay poor.
If those are estimates plus confidence intervals you don't need pvalues. P values are continuously misunderstood/misinterpreted and wouldn't provide any additional information. If someone tells you that you need pvalues, they are stupid.

#### msarin

##### New Member
Actually, I ended up doing this:

* Calculate p value for hazard ratio;
%macro hazard_p;
est=log(hazardratio);
l=log(HRLowerCL);
u=log(HRUpperCL);
SE=(u-l)/(2*1.96);
z=abs(est/SE);
p=exp(-.717*z-.416*z**2);
drop est--z;
%mend hazard_p;

#### hlsmith

##### Less is more. Stay pure. Stay poor.
I believe your SE calculation is off. So now what does the p value contribute, dependence on black and white decisions.

#### msarin

##### New Member
What would you say the SE is supposed to be?

#### hlsmith

##### Less is more. Stay pure. Stay poor.
No, it is right. My error.

Which shouldn't matter, the U = EST + SE * 1.96 and L = EST - SE * 1.96. So U - L would include the integer for EST, but that is trivial and comparable to likely rounding the critical value to 1.96 twice.

#### msarin

##### New Member
I got this approach from a few on-line sources. Would you suggest more decimal places?

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Nah, it is good enough.