Performing power analysis to find sample size for cox regression analysis experiment

errrrram — Mon, 22 Jan 2024 00:13:40 GMT

I'm trying to calculate the sample size for a fictional experiment for class. The experiment is similar to measuring cholesterol until time of first CVD incident, and we would analyze it with cox regression analysis. We would have two groups of subjects that exercise and a group that does not. When I try to calculate the sample size I get high numbers, I think because I'm misunderstanding something about the variables. I wasn't told to do coxreg, so I might be mistaken about that.

A paper I looked at found that rate of CVD for subjects similar to mine was 23.1 person-years. So I calculated that Eventprob=0.115 (=[23.1/1000][5 years]).

Another paper I found suggested that exercise decreased rate of cvd by 23%, so I used HAZARDRATIO=0.77.

I wasn't sure what to put for stddev.

proc power;
coxreg
power=0.8
ntotal=.
Eventprob=0.023
HAZARDRATIO=0.77
STDDEV= 0.23;
run;

Re: Performing power analysis to find sample size for cox regression analysis experiment

ballardw — Mon, 22 Jan 2024 16:03:40 GMT

Your description does not match the code shown: You show " I calculated that Eventprob=0.115 (=[23.1/1000][5 years])." in the text but the code contains Eventprob = 0.023.

I am not sure what a CVD=23% translates too in this but it appears that you might have used that as the STDDEV and is likely a very small STDDEV. I might suggest trying a list of values for one of those parameters and see the impact.

When I use STDDEV= 2 1 0.23 then the Ntotals are 1249, 4996 and 94435 respectively. Changing the Eventprob to 0.115 drops those to 250, 1000 and 18887 respectively.

topic Re: Performing power analysis to find sample size for cox regression analysis experiment in Statistical Procedures

Performing power analysis to find sample size for cox regression analysis experiment

Re: Performing power analysis to find sample size for cox regression analysis experiment