I'm sorry if this question is too simple, I really did my best to research it ahead of time but please point me in the right direction if its easy.

I've done a power calculation a few different ways and keep getting differing results, would love to know why

What I want to do:
-We are designing a prospective case control study.
-We'd like to compare a new biomarker test to the established bacterial culture test for a type of infection.
-Based on previous studies we think this new biomarker has about 95% sensitivity, the old test had about 75% sensitivity and we think 95% specificity for both.
-We think the infection will happen in about 10% of these patients
-We'd just like to show that this new biomarker is more sensitive than the old culture

Based on this answer here: Previous question

I did this code:

h1<-ES.h(0.97,0.75) pwr.p.test(h=h1,n=NULL,sig.level = 0.05,power = 0.9,alternative = "greater")

And it said I need about 18 people.

But based on this link here which mentioned expected prevalence: other method

I did this code:

power.diagnostic.test(sens = 0.95, delta = 0.20, prev = 0.1, power = 0.9)

And its suggesting 29 cases and 261 controls.

Why the massive difference?
Moreover a coworker used STATA and got an answer of 48 people needed. Beyond explaining the difference, which would be the best method for estimating in your opinion?

  • $\begingroup$ "We'd just like to show that this new biomarker is more sensitive than the old culture". So the true negative subject will be excluded? Need to specify how will you collect data, how to analyze the data, then you can calculate the sample size. $\endgroup$
    – user158565
    Dec 1, 2018 at 2:12
  • $\begingroup$ @user158565 we were currently planning on the cases being patients we suspect to have an infection and controls being people we suspect don’t. So currently we suspect there will be much more positive cultures AND bio markers in the cases, and vice versa in the controls. $\endgroup$ Dec 1, 2018 at 12:24
  • $\begingroup$ Suppose you finished data collection, what statistic analysis will you perform (suppose the sample size is large)? $\endgroup$
    – user158565
    Dec 1, 2018 at 14:57
  • $\begingroup$ @ user158565 To compare the sensitivities and specificity would that be a mcnemar test? $\endgroup$ Dec 1, 2018 at 16:32
  • $\begingroup$ Yes, you can use McNemar test. But for specificity, I do not know how to calculate the sample size because you assume no difference. $\endgroup$
    – user158565
    Dec 5, 2018 at 5:25


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.