Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Diagnostic accuracy / 95% confidence intervals

    Hello,

    I have a case control study with a binary outcome (disease/no disease) and two clinical diagnosis "tests" which I would like to compare. The first "test" is binary (present/not present), the second is ordinal with a total of 4 categories (0=not present, 1=low suspicion, 2=moderate suspicion, 3=high suspicion). I also have two covariates that I need to adjust for in the analysis.

    I have considered different methods for looking at sensitivity and specificity for these two tests but cannot decide on the best approach as I seem to run into limitations/questions with each method. I have read a lot about diagnostic accuracy and ROC analysis so I understand the concepts but struggling to implement.

    I have considered:

    1. logistic regression followed by estat classification with relevant cut-off --> how do I get 95% CIs for sensitivity and specificity? is there a way to obtain sens/spec (with 95% CIs) assuming covariate1=0, covariate 2=1 etc?

    2. roctab --> is there a way to adjust for covariates?

    3. roccurve/roc reg with covariate adjustments --> is there stata code that allows binary/ordinal classifiers with these commands? I've read some literature (Pepe et al 2009, 2003) which suggest you can do this with binary/ordinal classifiers and installed the relevant stata packages but cannot find stata code that actually allows this even after trolling the associated help files.

    Appreciate any advice on any or all of the above.
    Thank you!
    Last edited by Anne-Marie Rick; 02 Apr 2019, 12:46.
Working...
X