i*****c
发帖数: 1322 | 1 From the data analysis, I found that a measurement is associated with the
disease. I use the value that identify 60% patients and 20% non patients as
the cut-off value. Is there better way to express this cut-off value other
than OR? Thanks. | D**g
发帖数: 739 | 2 Did you use logistic regression? Why not use ROC curve to see if you have
satisfied AUC, and then use max(specificity+sensitivity) to define your cut-
off value? | i*****c
发帖数: 1322 | 3 I used logistic regression and ROC. But the measurement has high false
positive rate and my boss wants something else to make it looks better.
Unfortunately she doesn't know statistics while I'm still learning, so I don
't know what to do. | D**g
发帖数: 739 | 4 Did you use SAS? If you use SAS, you should be able to get AUC (area under
curve) from output. If AUC is lower than 0.7, then the test may not be a
good discrimination tool. http://gim.unmc.edu/dxtests/ROC3.htm )
In addition, getting a significant association using logistic regression
doesn't guarantee a good accuracy of diagnosis, especial when your data came
from a design which is not for the purpose of diagnosis.A significant
effect and a satisfied AUC are two different things.
By the way, if your have a ROC curve with good AUC, the optimal cutoff can
be determined by the maximum of the Youden index, or x value corresponding
to maximum of sum of specificity and sensitivity. Sometimes, depending on
the purpose such as screen or diagnosis, you can have arbitrary preference
on either specificity or sensitivity.
don
【在 i*****c 的大作中提到】
: I used logistic regression and ROC. But the measurement has high false
: positive rate and my boss wants something else to make it looks better.
: Unfortunately she doesn't know statistics while I'm still learning, so I don
: 't know what to do.
| i*****c
发帖数: 1322 | 5 I used R but got ROC using Origin.The measurement is for screening and the
AUC is larger than 0.7.
You help me a lot! Baozi to thank you:) |
|
