I analyze the psychometric qualities of a screening instrument to predict depression after stroke using ROC analysis.
In this context, I read several papers which report an AUC without a p-value or an AUC with a 95% confidence interval including 0.50, for example: 0.48-0.72. However, the authors discuss the sensitivity and specificity of the instrument in these papers and conclude that e.g. the sensitivity is moderate/good.
Isn't it necessary for the AUC to be significantly different from 0.50, ascertained either by p < .05 or a CI not including .50, in order to conclude that there is any psychometric accuracy to the instrument? Or is it possible that an instrument shows good sensitivity and/or specificity while the AUC is not different from .50?
https://www.youtube.com/watch?v=egTNM8NSa2k
https://www.youtube.com/watch?v=ju2nku4Rz78
Cristian Ramos-Vera, thank you very much for your reply. Dr. Shaneyfelt states that a test with an AUC = 0.5 is useless, however he does not spefically adress the issue of it being significantly different from 0.5 but I figure that is implied?
Unfortunately, my Spanish is not sufficient for the second video.
Hi Julian,
my question refers to original research we conducted at our clinic. Cosmin seems to focus on reviews? So they are probably not the most suitable to address. Thank you anyways!
Simon, I agree with your interpretation. The AUC being greater than .5 is similar to wanting a correlation to be significantly greater than zero. I suppose AUC =.50 is the null hypothesis and we would want to reject it.
Mark
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments