How to Get information From Proc LIFETEST and PHREG

[SASuserlot]

Hi

Thank you for your inputs and helps on this topic.

I was able to run the proc lifetest and phreg but confused, which datasets need to be used to get the information since outputs produces so much information . I need to get the following information. I was providing dataset I have.

/* Create the original dataset */
*Information in data
AVAL ->Number of Days  after the Treatment that  Adverse Event Occurred"
CNSR->  If the event occurred, then CNSR =0 otherwise, 1
ARM-> IND and Placebo;
data have;
    input USUBJID ARM AVAL CNSR;
    datalines;
1 1 1 1
2 1 2 0
3 1 10 1
4 2 5 1
5 2 6 0
6 2 8 1
7 2 10 0
8 2 11 0
9 1 12 1
10 2 13 1
11 2 15 1
12 1 10 0
13 1 18 1
14 2 19 0
15 2 21 1
16 1 23 1
17 1 5 1
18 2 25 1
;
run;

data have;
	set have;
	if arm = 1 the TRT = 'Placebo';
	if arm =2 then TRT = 'IND';
run;

** for Quartiles and estimates/ Not using any from Estimates data**
proc lifetest method=KM data=have   plots=none;*timelist=(5,10,15); 
strata arm;
time aval*cnsr(1); 
ods output 	productlimitestimates = estimates
			Quartiles = quart;
run;
** for HR and 95 %CI/ Not sure where to get P- value**;
proc phreg data=have;
    class arm (ref='2'); 
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
run;

*** for  adverse event free probability and 95% CI at day 10**;
proc lifetest method=KM data=have   plots=none timelist=(5,10,15); 
strata arm;
time aval*cnsr(1); 
ods output 	productlimitestimates = estiTime;
run;

From PHREG: Cox Model

1 . Probability (IND Vs Placebo) - P-value

2. Hazard Ratio (IND Vs Placebo)

3 . 95% CI  (IND Vs Placebo)

From PROC LIFETEST:

Quartiles:

1. Q1 with 95%CI

2. Q3 with 95% CI

3. Median with 95% CI

From PROC LIFETEST:

Adverse Event Free probability at Day 10

[quickbluefish]

I don't know much about quartile estimates from PROC LIFETEST, but for your PHREG question, you need to set ref='1' to get the HR for IND vs. placebo.  The way you currently have it has IND set as the ref - this is not what you want.  As for the p-value, it's just the thing in the parameter estimates output labeled "Pr > ChiSq".  In this case, you do NOT need the HAZARDRATIO statement.  

For the LIFETEST question about AE-free survival at day 10 - you can get this from the column labeled SURVIVAL in your output dataset (EstiTime) -- i.e., survival at day 10.  When I run your code, I get 68.57% for ARM 1 and 76.19% for ARM 2.

Good luck.

[SASuserlot]

Thank you very much for your inputs.

[Ksharp · Accepted solution]

** for HR and 95 %CI/ Not sure where to get P- value**;
proc phreg data=have;
    class arm (ref='2'); 
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
run;

There four way to get P-Value:

1)

proc phreg data=have;
    class arm (ref='2')/param=glm; 
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
run;

2)

proc phreg data=have;
    class arm (ref='2')/param=glm; 
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
	estimate 'arm1 vs arm2' arm 1 -1/exp cl;
/*	lsmeans arm/diff exp cl;*/
run;

3)

proc phreg data=have;
    class arm (ref='2')/param=glm; 
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
/*	estimate 'arm1 vs arm2' arm 1 -1/exp cl;*/
	lsmeans arm/diff exp cl;
run;

4)

proc phreg data=have;
    class arm (ref='2')/param=glm;  
    model aval*cnsr(1) = arm  / risklimits ties=efron rl;
    hazardratio arm / diff=ref; 
    ods output  ParameterEstimates = hR;
	lsmestimate  arm 'arm1 vs arm2' 1 -1/exp cl; 
run;

/*****************************************************************************/

*** for  adverse event free probability and 95% CI at day 10**;
proc lifetest method=KM data=have outs=outs  plots=none timelist=(5,10,15); 
strata arm;
time aval*cnsr(1); 
ods output 	productlimitestimates = estiTime;
run;
proc sql;
create table temp as
select distinct arm,10 as aval from have;
quit;
data want;
 set outs temp;
 by arm aval;
run;
data want;
 set want;
 by arm;
 retain point lower upper;
 if first.arm then call missing(point, lower, upper) ;
 if not missing(SURVIVAL) then point=SURVIVAL;
 if not missing(SDF_LCL) then lower=SDF_LCL;
 if not missing(SDF_UCL) then upper=SDF_UCL;
run;
proc sql ;
select distinct aval, arm,point,lower,upper
 from want 
  where aval=10;
quit;

[SASuserlot]

Thank you very much @Ksharp for your detailed Response. 

I think I got 95% of what I wanted. However, I have questions regarding choosing which variable in Proc life test estimates. This is for my education, and I always read it, but I need to remember it.

In my data, If Adverse events occur, I was given CNSR =0; otherwise, it's 1.

When we run the proc life test, we get variables  'SURVIVAL' and sometimes "FAILURE' ( depending on the code).

My question is in this Scenario: 

'Survival'  Variable indicates the  Adverse Event Free Probability or  Probability of an Adverse Event  can happen? because in some places we used to see "1- SURVIVAL"

[quickbluefish]

again, in PHREG, you need to set (ref='1') instead of (ref='2'). The way you have it is backwards.

[SASuserlot]

Thanks, you are correct, I changed in my program, forgot to mention in the msg.

[Ksharp]

"If Adverse events occur, I was given CNSR =0; otherwise, it's 1."

So the "interesting" thing in survivial analysis is "Adverse events", the censored flag is 1 (the "interesting" thing not happened), therefore the following statement is right:

time aval*cnsr(1); 

'Survival'  Variable is the survival probability(a.k.a the probability of  "interesting" thing not happened),

I think it should be "Adverse Event Free Probability" (I am not familiar with concept of Free Probability).

"Probability of an Adverse Event  can happen"

I think it should be Failure Probability, you could get this by the following code:

data have;
    input USUBJID ARM AVAL CNSR;
    datalines;
1 1 1 1
2 1 2 0
3 1 10 1
4 2 5 1
5 2 6 0
6 2 8 1
7 2 10 0
8 2 11 0
9 1 12 1
10 2 13 1
11 2 15 1
12 1 10 0
13 1 18 1
14 2 19 0
15 2 21 1
16 1 23 1
17 1 5 1
18 2 25 1
;
run;
proc lifetest data=have  plots=survival(failure atrisk=5 to 15 by 5);;
strata arm;
time aval*cnsr(1); 
run;

[SASuserlot]

Thank you for your explanation.  I think I got a better picture now.