+ Reply to Thread
Results 1 to 3 of 3

Thread: Power calculation for accuracy tests

  1. #1
    Points: 4, Level: 1
    Level completed: 7%, Points required for next Level: 46

    Thanked 0 Times in 0 Posts

    Power calculation for accuracy tests

    Excuse my ignorance

    I would like to do a simple study. I have 3 screening tests and a gold standard test which will put patients into +ve and -ve (binary) groups. I expect that 22% of patients recruited will be +ve and the rest negative.

    I want to establish the sensitivity and specificity of the screening tests with ROC analysis. How do I know how many patients i need to recruit?

    Thanks so much in advance.


  2. #2
    Points: 3,015, Level: 33
    Level completed: 77%, Points required for next Level: 35

    Thanked 7 Times in 7 Posts

    Re: Power calculation for accuracy tests

    Have these tests already been derived and you are trying to validate them?

    If so, the following reference has a normogram which should answer your question:
    Carley S, Dosman S, Jones SR, Harrison M. Simple nomograms to calculate sample size in diagnostic studies. Emerg Med J. 2005;22:180-1.

  3. #3
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Not Ames, IA
    Thanked 1,185 Times in 1,146 Posts

    Re: Power calculation for accuracy tests

    Well, like many times when you attempt to determine a sample size or power, you first have to start with potential projections for results in the study (if you are not already there). As mentioned, are there available data elements (in the literature) to help you come up with the components needed in the calculations (prevalence, SEN, SPEC, etc. for the tests)?

    I would recommend looking at Ch. 5 & 6 in:

    Statistical Methods in Diagnostiic Medicine by Zhou, Obuchowski, and McClish.

    You can run a power calculation for each planned pairwise comparison, you can also just run it on the two most comparable tests under the rationale that if that comparison is well powered the others will be as well (just an assumption). I will also point out that you will likely want to address the paired data and multiple planned tests (test 1 v 2, 1 v 3, 2 v 3), so a bonferroni correction would have you change your 0.05 alpha to 0.0167.
    Stop cowardice, ban guns!

+ Reply to Thread


Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts

Advertise on Talk Stats