Calcite | Level 5

## Evaluation of the Cox regression model in the Test Set with PROC PHREG

Hi all,

I have a dataset that includes about 10 variables and follow-up for death (n= ~2,000).

I want to evaluation the Cox PH model derived from the train set (70%) in the test set (30%).

For this, I have done this so far to try with age (cont.) sex (categorical)

proc phreg=train;
model duration*death(0)=age sex;
run;
/*regression coefficient age: 0.07 sex: 0.3 */

How would I use the coefficient derived from the test set and evaluate the performance of the model in the test set?

In theory, I think the regression coefficient values are found by maximizing the partial likelihood,

so simply,

hi(t)/h0(t)=exp(0.07*age+0.3*sex)

I want to evaluate this model with Harrell's C-index

proc phreg=test concordance;
/*<----what should be here with regression coefficients from the train set?*/
run;

2 REPLIES 2
Calcite | Level 5

## SAS Evaluation of the Cox PH Model (derived from the Train set) in the Test Set with Proc Phreg

Hi all,

I have a dataset that includes about 10 variables and follow-up for death (n= ~2,000).

I want to evaluation the Cox PH model derived from the train set (70%) in the test set (30%).

For this, I have done this so far to try with age (cont.) sex (categorical)

proc phreg=test;
model duration*death(0)=age sex;
run;
/*regression coefficient age: 0.07 sex: 0.3 */

How would I use the coefficient derived from the test set and evaluate the performance of the model in the test set?

In theory, I think the regression coefficient values are found by maximizing the partial likelihood,

so simply,

hi(t)/h0(t)=exp(0.07*age+0.3*sex)

I want to evaluate this model with Harrell's C-index

proc phreg=test concordance;
/*<----what should be here with regression coefficients from the train set?*/
run;

any help would be appreciated.

Calcite | Level 5

## Re: SAS Evaluation of the Cox PH Model (derived from the Train set) in the Test Set with Proc Phreg

/*Typo: train, NOT test*/

proc phreg=train;
model duration*death(0)=age sex;
run;
/*regression coefficient age: 0.07 sex: 0.3 *
Discussion stats
• 2 replies
• 527 views
• 0 likes
• 1 in conversation