How to calculate 10 year survival probability with proc phreg

[Maartje]

Hello all,

For my school project I'm developing a risk prediction model for overall mortality. The overall equation to do so is: 1 - 10yearsurival probability^(exp(regression coefficient x1.. x2.. x3.. x3). Where the 10 year survival probability has to be calculated with proc phreg.

The syntax for my model is:

   proc phreg data=a.predict;

model time*died15(0)= age cur former mod med crp ;

run;

I'm not sure how to calculate the 10 year surival probability.. I know I should use the BASELINE statement and that I have to calculate it for my covariates set to zero?

I hope you can help me out :)!

Maartje

[Reeza]

I don’t know if you necessarily need a BASELINE statement, does the output dataset not have the survival estimates at various points in time? You can definitely use BASELINE to specify the values you’re interested in as well. 

Its not 0 though, it’s usually the mean for continuous variables and the reference level for categorical variables. 

[Maartje]

Ohh thanks for your response. I didnt specify the reference levels for my categorical variables and thats why I didnt get the survival probability.

I still have a small question about filling in the equation for the 10-year risk prediction for mortality. Now I've calculated the survival probability with the baseline statement where the mean for continuous variables and the reference level for categorical variables was used. If I want to calculate the 10 year risk for mortality for a person of age 70 and the mean age used to calculate the baseline survival probability was 60, should i use the difference in age between the mean used for baseline survival calculations and the actual age (in this case age = 10? Or should i just fill in the actual age of this person?

Thanks again 🙂

[Reeza]

The model uses the actual age so you would use that. 

[Maartje]

Hmm, thanks and sorry to bother u again.. But my predicted 10 year mortality risks are really high (also compared to the number of events) and does not vary much in between participants and I can't seem to find out why..

Should I calculate the baseline survival probability for ALL variables available in the data set? Now Ive calculated the baseline survival probability with the predictors which are in the end model.

Thanks!

[Reeza]

---

@Maartje wrote:

Hmm, thanks and sorry to bother u again.. But my predicted 10 year mortality risks are really high (also compared to the number of events) and does not vary much in between participants and I can't seem to find out why..

Should I calculate the baseline survival probability for ALL variables available in the data set? Now Ive calculated the baseline survival probability with the predictors which are in the end model.

Thanks!

---

No, you first build a model with the variables that you think are relevant. Once your model is built you then would look at the 10 year survival. If you N is small, especially for the subgroup analysis you're results will be all over the place.