topic Re: Help with coding to calculating total # of events required for time-dependent (survival) data in SAS Programming

Help with coding to calculating total # of events required for time-dependent (survival) data

sascrazedpupil — Thu, 02 Dec 2021 19:11:08 GMT

Hello, I attempted this problem in SAS Studio, and am not sure if I input the correct code for it. I'd greatly appreciate any help!

Problem:

"For determining sample size estimates for time dependent (survival) data, usually the number of estimated events (D) (e.g. deaths, rejections, relapse, reops, etc.) must first be calculated. Once that is established, the total sample size (n) may be derived based on the percentage of patients that remain event-free at the end of the study."

The two equations for D and n are found below:

These equations are for the (Parametric) version of sample size determination for Survival Data or Time to Failure data.

D = total # of events required

α = Z critical value - Type I error

Β = Z critical value - Type II error

Δ= hazardratio = λΤ/λc = lnπT / lnπC

n=D/(1-%event-free)

Example:

90% Power

Δ = 2.0

D=4(1.96 + 1.282)^2/[ln2]^2 = 87.5 = 88events

This assumes that all participants will be followed until their event.

Suppose that previous studies suggest that 30% of subjects remain event-free at the end of trial. Then, sample size n = D = 88 / (1-0.30) = 125.7 = 126 1 - %event-free

This number might be further inflated to account for dropout, but we won’t worry about that here.

In SAS write code to calculate D and n for the following scenarios:

When alpha = 0.05 and the hazard ratio may be either 2.0 or 1.5, for power = 0.8 and also for power = 0.90. And either 30% or 20% of subjects remain event-free at the end of trial.

Do you see that you will have 8 pair of estimates for D and n in your output? Please include your SAS code. Your estimate for D and n should be in integer form.

Note:
Power ZB

80% 0.84

90% 1.28

Here is my code:

proc power;
twosamplefreq test=fisher
groupproportions= (0.30 0.20)
npergroup=.
power=0.90;
run;

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

Reeza — Thu, 02 Dec 2021 20:31:18 GMT

Since the data is survival data, I don't think twosamplefreq is appropriate, Twosamplesurvival is likely more appropriate.

https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/4908-2020.pdf

@sascrazedpupil wrote:

Hello, I attempted this problem in SAS Studio, and am not sure if I input the correct code for it. I'd greatly appreciate any help!

Problem:

"For determining sample size estimates for time dependent (survival) data, usually the number of estimated events (D) (e.g. deaths, rejections, relapse, reops, etc.) must first be calculated. Once that is established, the total sample size (n) may be derived based on the percentage of patients that remain event-free at the end of the study."

The two equations for D and n are found below:

These equations are for the (Parametric) version of sample size determination for Survival Data or Time to Failure data.

D = total # of events required

α = Z critical value - Type I error

Β = Z critical value - Type II error

Δ= hazardratio = λΤ/λc = lnπT / lnπC

n=D/(1-%event-free)

Example:

90% Power

Δ = 2.0

D=4(1.96 + 1.282)^2/[ln2]^2 = 87.5 = 88events

This assumes that all participants will be followed until their event.

Suppose that previous studies suggest that 30% of subjects remain event-free at the end of trial. Then, sample size n = D = 88 / (1-0.30) = 125.7 = 126 1 - %event-free
This number might be further inflated to account for dropout, but we won’t worry about that here.

In SAS write code to calculate D and n for the following scenarios:

When alpha = 0.05 and the hazard ratio may be either 2.0 or 1.5, for power = 0.8 and also for power = 0.90. And either 30% or 20% of subjects remain event-free at the end of trial.

Do you see that you will have 8 pair of estimates for D and n in your output? Please include your SAS code. Your estimate for D and n should be in integer form.
Note:
Power ZB

80% 0.84

90% 1.28

Here is my code:
proc power;
twosamplefreq test=fisher
groupproportions= (0.30 0.20)
npergroup=.
power=0.90;
run;

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

sascrazedpupil — Fri, 03 Dec 2021 18:07:20 GMT

Ok, thank you. I checked with my TA, and she said I only have to write SAS code for the formulas specified in the problem. How do I do that?

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

Reeza — Fri, 03 Dec 2021 18:20:47 GMT

I think you can use the formula for D and n with the information in the question.

FYI - a good way to answer these questions is to list the information you have in the question (alpha = 0.5, power=2/1.5 etc), figure out what you need to get to (D/n)
and then if you have the formula's it becomes plug and play. I see the formula for N but not D in your posts.

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

sascrazedpupil — Fri, 03 Dec 2021 18:47:43 GMT

Ok, here's the new code I typed, using the formulas given to me (D is "number of events", and n is "sample size"):

data fep6; set fep6;
Num_events=(4*(1.96+1.282)**2)/log(2)**2;
sample_size= Num_events/(1-0.30);
run;

It's not showing me the answer in my output, how do I fix that?

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

Reeza — Fri, 03 Dec 2021 18:51:18 GMT

@sascrazedpupil wrote:

data fep6; set fep6;

Do not code like this, give it a new data set name such as fep7.


data fep7; 
set fep6;
Num_events=(4*(1.96+1.282)**2)/log(2)**2;
sample_size= Num_events/(1-0.30);
run;

proc print data=fep7;
run;

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

sascrazedpupil — Fri, 03 Dec 2021 19:05:32 GMT

Thank you! Do I need to have observations for it to give me a numeric Output (Under "Notes" in my Log, it says "No observations read from the data set WORK.FEP6.", since I'm not using any datasets with values, but instead just writing SAS code for the formulas specified in the assignment.

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

Reeza — Fri, 03 Dec 2021 19:10:19 GMT

Then remove the SET statement.
SET provides the input data, where there is none in this case it seems.

Re: Help with coding to calculating total # of events required for time-dependent (survival) data

sascrazedpupil — Fri, 03 Dec 2021 19:11:57 GMT

Thank you, it worked! You are a lifesaver.