topic Re: Power analysis for Poisson regression in Statistical Procedures

Power analysis for Poisson regression

gtrifan — Mon, 17 Oct 2022 15:19:31 GMT

Hi all, I'm trying to get a power analysis for a health care study that will evaluate incidence rate ratio of an event (predicted variable as counts and few categorical and continuous independent predictors). I tried to find a way to do it in SAS, but I can't seems to find the right code.

I tried Cox Regression, code below, but I'm not sure if this will give me the same power results as those estimating for Poisson regression, as one takes into account time to event and the other the counts or rates only...any suggestions on how to proceed ? Thanks.

ods noproctitle;
ods graphics / imagemap=on;

proc power;
coxreg test=score sides=2 ntotal=13000 eventprob=0.017 alpha=0.05
hazardratio=1.2 1.5 rsquare=0 0.3 stddev=1.1 power=.;
plot x=n;
run;

Re: Power analysis for Poisson regression

John_Castelloe — Fri, 28 Oct 2022 16:53:19 GMT

You can use the CUSTOM statement in PROC POWER, and there's an example in this paper from SAS Global Forum 2018. The main example involves logistic regression, but the section "Steps for a Poisson Regression Alternative" at the bottom of page 12 describes how to adapt it to a Poisson regression.

Re: Power analysis for Poisson regression

gtrifan — Fri, 28 Oct 2022 17:39:43 GMT

Thank you!