1
$\begingroup$

I am performing a prospective cohort study in an existing cohort study. The sample of participants I can use from this existing cohort is 190. I am using a Cox model to analyze the data and to get a Hazard ratio. But now I want to know what the power of the study is. How can I calculate it? I can't find it on the internet...

So I am doing a prospective survival analysis (in an existing cohort) of 190 participants, the exposure is divided either into three categories and continuous and the outcome is yes/no.

After that, I also want to calculate the power of the study when it is stratified for some covariates.

I am using SAS, so maybe there is an option in SAS for this?

I hope some of you can help me. Thanks in advance!

$\endgroup$
  • 1
    $\begingroup$ I haven't much to add beside @frank-harrell 's answer, but would be curious why you want to calculate power. Maybe there is some (mis)conception about power or maybe there is another statistic which is more useful to what you are trying to accomplish. OP, could you elaborate? $\endgroup$ – IWS Jan 2 '18 at 12:55
  • $\begingroup$ Thanks for answering! I did not find any statistically significant result in my analyses.. So my supervisor asked if I calculated the power of the study.. What I thought about power is that it is the chance of a ''real'' result. Or to explain if the power is 80%, the result in the sample had a chance of 80% to be true in the whole population. Or am I wrong? Now I don't know why she wanted me to calculate the power... And she wanted me to especially calculate it for stratified analyses. $\endgroup$ – Laddi Jan 2 '18 at 15:37
1
$\begingroup$

Power is a long-run pre-study concept and has no value once data are in. Compute the 0.95 confidence interval for the hazard ratio of main interest. And note that you don't stratify on covariates (unless one of them is categorical and you suspect non-proportional hazards for that one); you model them in the regression part of the model.

Power involves assuming an effect you would be embarrassed to miss. This effect size is highly subjective and is usually gamed, e.g., assumed to be very large so it's easy to detect, making your power look good.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.