Add AUCs

#3
You're right. Let me rephrase the question:
How can I add AUCs from multiple studies into one summarized AUC. This would indicate the accracy for the diagnostic test used in these studies (assuming no bias I suppose). The aim is to compare diagnostic tests by summing studies grouped by test method.
 

hlsmith

Omega Contributor
#4
Lots of questions.


Area for under the curve is based on a threshold, did all studies have the same threshold or was the threshold based on a composite of variables?


Do you suspect heterogeneity between studies (random effects)? Was disease risk (prevalence) comparable?


Side note, I believe that I have read that the pooling of AUC values is not always recommended, many times people will focus on pooled sensitivities and specificities.


http://methods.cochrane.org/sites/methods.cochrane.org.sdt/files/public/uploads/MetaDAS%20Quick%20Reference%20v1.3%20May%202012.pdf


http://methods.cochrane.org/sites/m...blic/uploads/MetaDAS Readme v1.3 May 2012.pdf
 
#5
But is it not so that AUC is considering alla possible tresholds?
Heterogeneity might be a problem. I haven't considered that yet. So, yes I need to address heterogeneity sooner or later.
Thanks, I will look into what is recommended regarding AUC pooling.
 

hlsmith

Omega Contributor
#6
The receiver operator characteristic curve plots the sensitivity (y-axis) against 1-Specificity (x-axis) for a continuous variable. So yes it has all cut-offs for all continuous values in the dataset. Though, I have never seen a study make all that information available. You would have to have their raw datasets to do that. Most of the time studies present just the optimal cut-off value that maximized the AUC. If multiple studies do this, you may need to make sure they are all using the same cut-off (threshold) and that some of the study aren't controlling for other variables than others.