Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- ROC Curve coordinates and cut off points

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

03-02-2011 06:52 AM

Hi All,

I am using proc logistic to create and compare ROC curves for diagnostics. I get great curves and can produce the coordinates of the curve, however, the coordinate table does not tell me at whcih cut off for the test in question the coordinate occurs - so for example, i would know that the point on a curve has sensitivity of x and specificity of y, but i cannot tell at which cut off for the test x and y are acheived. Is there a way of adding the cut off to the table so i can state what suitable cut offs are for particular tests?

I am currently using the following to run the curves and comparisons:

ods graphics on;

proc logistic data=ROCPTT plots=roc(id=prob);

model WEREPIANIMALSFOUND(event='1') = NUMBEROFABPOSITIVEYSINSAMPLE10 INITIALBULKMILKODRATIO NUMBEROFABPOSITIVEYSINSAMPLE10*INITIALBULKMILKODRATIO/ outroc=roc_score nofit;

roc 'ys' NUMBEROFABPOSITIVEYSINSAMPLE10;

roc 'BMab' INITIALBULKMILKODRATIO;

roc 'ys*BMab' NUMBEROFABPOSITIVEYSINSAMPLE10*INITIALBULKMILKODRATIO;

roccontrast reference('YS') / estimate e;

run;

ods graphics off;

outroc=roc_score produces the table of coordinates but with no cut of values mentioned. SPSS will produce the same table with cut offs, but it doesnt seem able to combine the tests to produce a combined ROC curve like SAS and the stats are better in the SAS model.

Thanks in advance Message was edited by: reb

I am using proc logistic to create and compare ROC curves for diagnostics. I get great curves and can produce the coordinates of the curve, however, the coordinate table does not tell me at whcih cut off for the test in question the coordinate occurs - so for example, i would know that the point on a curve has sensitivity of x and specificity of y, but i cannot tell at which cut off for the test x and y are acheived. Is there a way of adding the cut off to the table so i can state what suitable cut offs are for particular tests?

I am currently using the following to run the curves and comparisons:

ods graphics on;

proc logistic data=ROCPTT plots=roc(id=prob);

model WEREPIANIMALSFOUND(event='1') = NUMBEROFABPOSITIVEYSINSAMPLE10 INITIALBULKMILKODRATIO NUMBEROFABPOSITIVEYSINSAMPLE10*INITIALBULKMILKODRATIO/ outroc=roc_score nofit;

roc 'ys' NUMBEROFABPOSITIVEYSINSAMPLE10;

roc 'BMab' INITIALBULKMILKODRATIO;

roc 'ys*BMab' NUMBEROFABPOSITIVEYSINSAMPLE10*INITIALBULKMILKODRATIO;

roccontrast reference('YS') / estimate e;

run;

ods graphics off;

outroc=roc_score produces the table of coordinates but with no cut of values mentioned. SPSS will produce the same table with cut offs, but it doesnt seem able to combine the tests to produce a combined ROC curve like SAS and the stats are better in the SAS model.

Thanks in advance Message was edited by: reb

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

Did you figure this out ever? If so, could you paste your solution / options that you used?

Thanks