Bonferroni correction in Proc Phreg

[kayeee_]

Hi everyone,

I am currently analyzing the association between several biomarkers (continuous variables) and a disease (a binary variable, 0 or 1) along with confounders using Cox proportional hazard models.

Given the abundance of biomarkers, I should multiple testing with corrected p-values for each biomarker.

I have obtained the estimate of the effective number of independent tests, which is 50. I understand that the significance threshold for p-values should be set at 0.05/50.

Now, my question is: should I divide the calculated p-values by 50 as well?

Also, if it should be is there any option to transform "<0.001" to detailed value?

For example as code:

proc phreg data=mydata;
class confounders;
model time*status(0)=confounders biomarker1/RL;
ods select ParameterEstimates;ods output ParameterEstimates=result;
run;

data result; set result;
corrected_p=ProbChiSq/50; /*This is my question*/
run;

/*repeat this as the number of biomarekrs */

Thanks in advance!

[StatDave]

Your code shows a response variable called TIME. This does not seem to be the binary indicator of disease that you mention. If the binary disease variable is actually what your model is supposed to predict, then use PROC LOGISTIC, not PROC PHREG. 

However, regardless of the models generating the p-values, adjustment of p-values for multiple testing can be done using PROC MULTTEST which offers many modern p-value adjustment methods. The Bonferroni method is extremely conservative (many would say much too conservative). Using a method such as the Holm method in MULTTEST is more powerful and controls for the family-wise error rate.  You can use MULTTEST by gathering all of the p-values for your biomarkers into a single variable in a single data set and then using MULTTEST to adjust the p-values with the adjustment method of your choice, such as Holm. This is illustrated in the example titled "Inputting Raw p-Values" in the MULTTEST documentation.