Hi,

I'm currently attempting to analyse some data obtained for a couple of new diagnostic tests (of TFBUT in eyes) against an existing test (STT). I'm trying to use an ROC curve for each test of TFBUT with a state variable of 0 = healthy control eyes (with high TFBUT) and 1 = diseased eyes (with low TFBUT) (based on the accepted cut off value of STT). 
When I plot this in SPSS, however, it's coming out significantly below the diagonal (reference?) line of the ROC curve. This really isn't right as plotting the values as histograms shows that these tests correlate extremely well with STT and definitely have extremely good sensitivity and specificity. 
When I reverse my state variable and have 0 as the diseased eyes (with low TFBUT) and 1 as the control eyes (with high TFBUT) the curve comes out looking exactly as I would expect with a really high AUC - however, I don't think the values of specificity and sensitivity given for different thresholds are appropriate as sensitivity is then higher for lower values of the test which isn't appropriate.
How should I address this?

Thank you very much for any help! I'm on a really tight deadline so I am extremely grateful to anyone who can help me work this out.

------------------------------
Primrose Moss
------------------------------

#SPSSStatistics

For anyone else looking into this in future - turns out when you select ROC curve under options you can tell it whether the disease state is correlated with high or low values of your variable.

------------------------------
Primrose Moss
------------------------------

Original Message:
Sent: Wed June 09, 2021 06:26 AM
From: Primrose Moss
Subject: ROC Curves

Hi,

I'm currently attempting to analyse some data obtained for a couple of new diagnostic tests (of TFBUT in eyes) against an existing test (STT). I'm trying to use an ROC curve for each test of TFBUT with a state variable of 0 = healthy control eyes (with high TFBUT) and 1 = diseased eyes (with low TFBUT) (based on the accepted cut off value of STT). 
When I plot this in SPSS, however, it's coming out significantly below the diagonal (reference?) line of the ROC curve. This really isn't right as plotting the values as histograms shows that these tests correlate extremely well with STT and definitely have extremely good sensitivity and specificity. 
When I reverse my state variable and have 0 as the diseased eyes (with low TFBUT) and 1 as the control eyes (with high TFBUT) the curve comes out looking exactly as I would expect with a really high AUC - however, I don't think the values of specificity and sensitivity given for different thresholds are appropriate as sensitivity is then higher for lower values of the test which isn't appropriate.
How should I address this?

Thank you very much for any help! I'm on a really tight deadline so I am extremely grateful to anyone who can help me work this out.

------------------------------
Primrose Moss
------------------------------

#SPSSStatistics