I'm using SPSS to draw a ROC Curve for some Tumor Suppressor genes. The problem is some of the curves are below the Standard Line (the diameter). The question is, how can I interpret these results in cancer diagnosis?
Hello,
The area under the ROC curve (AUROC) should be between 0.5 and 1.0. This area is a measure of the predictive accuracy of a model. An AUROC equal to 0.5 (i.e. coinciding with the diagonal) indicates a random classification model. As a matter of fact, this area should be greater than 0.5 for a model to be acceptable; a model with AUROC of 0.5 or less is worthless.
Now if your AUROC is below 0.5, you will need to reverse all your predictions (make zeroes into ones and ones into zeroes). That should fix the curve and your AUC should go above 0.5. As for why this would occur in the first place, the most likely explanation is that you inadvertently reversed your labels at some point during the coding process (check your syntax). Hope this helps.
Dear Roghayyeh,
The important thing is AUC (area under the ROC curve). It should be statistically significant (the column Asymptotic Sig. in one of the tables included in SPSS output). If it is significant and below the reference (diagonal) line, this means that the correlation between a particupar independent variable and the dependent one is negative.
https://en.wikipedia.org/wiki/Receiver_operating_characteristic
Not too good a source but the basics are there in this wiki article.
Hello,
The area under the ROC curve (AUROC) should be between 0.5 and 1.0. This area is a measure of the predictive accuracy of a model. An AUROC equal to 0.5 (i.e. coinciding with the diagonal) indicates a random classification model. As a matter of fact, this area should be greater than 0.5 for a model to be acceptable; a model with AUROC of 0.5 or less is worthless.
Now if your AUROC is below 0.5, you will need to reverse all your predictions (make zeroes into ones and ones into zeroes). That should fix the curve and your AUC should go above 0.5. As for why this would occur in the first place, the most likely explanation is that you inadvertently reversed your labels at some point during the coding process (check your syntax). Hope this helps.
Hi,
I also agree with Ghassen Soussi. If you are using SPSS this can be easily achieved by changing "state variable" from 1 to 0 (or the other way round) when performing the ROC curve analysis.
(Alternatively, you can leave the coding as is and simply subtract the result from 1)
- Badges
- Science topic
- Similar topics
- Medicine
- Oncology
- Cancer Research
- Cancer Biology