I have a follow up question to Dr Winsemius' post. You can use the AIC criterion against all possible cut off values C to see which minimizes the AIC and then that is the ideal cut off in trying to dichotomize a continuous variable. What I am wondering here is, does the survivalROC package, or any other package in R or function in SAS compute this? I have been reading and this does not seem to be addressed anywhere so please point me in the right direction. Thanks, Alexia -- View this message in context: http://r.789695.n4.nabble.com/HOW-to-use-the-survivalROC-to-get-optimal-cut-off-values-tp3073382p4405992.html Sent from the R help mailing list archive at Nabble.com.
HOW to use the survivalROC to get optimal cut-off values?
3 messages · alexiamelissa, David Winsemius
On Feb 21, 2012, at 12:08 AM, alexiamelissa wrote:
I have a follow up question to Dr Winsemius' post. You can use the AIC criterion against all possible cut off values C to see which minimizes the AIC and then that is the ideal cut off in trying to dichotomize a continuous variable. What I am wondering here is, does the survivalROC package, or any other package in R or function in SAS compute this? I have been reading and this does not seem to be addressed anywhere so please point me in the right direction.
The usual attempts to set a cut-point make the very restrictive and simplistic assumption that the cost of a decision that results in a false positive are the same as the cost of a false positive. This is almost never the case. Furthermore, these studies are often done with case and control populations that are not representative of the populations for which the test will be applied in the future. I think handing off the task to an automatic procedure dressed-up to construct and "ideal" or "scientific" answer is misguided window dressing. They are an effort to avoid thinking carefully about the costs of the alternative outcomes and fail to realize tat their are multiple parties being affected with no meaningful input regarding their respective utilities. I'm not saying that quantitative analysis of these issues is not useful, just that it is unlikely to be done well by one function in a package in R or SAS..
David Winsemius, MD West Hartford, CT
On Feb 21, 2012, at 3:21 AM, David Winsemius wrote:
On Feb 21, 2012, at 12:08 AM, alexiamelissa wrote:
I have a follow up question to Dr Winsemius' post. You can use the AIC criterion against all possible cut off values C to see which minimizes the AIC and then that is the ideal cut off in trying to dichotomize a continuous variable. What I am wondering here is, does the survivalROC package, or any other package in R or function in SAS compute this? I have been reading and this does not seem to be addressed anywhere so please point me in the right direction.
Errors (mostly?) corrected: The usual attempts to set a cut-point make the very restrictive and simplistic assumption that the costs of a decision resulting in a false positive are the same as the costs of a false positive. This is almost never the case. Furthermore, these studies are often done with case and control populations that are not representative of the populations for which the test will be applied in the future. I think handing off the task to an automatic procedure dressed-up to construct an "ideal" or "scientific" answer is misguided. They are an effort to avoid thinking carefully about the costs of the alternative outcomes, and fail to account for the reality that there are multiple parties being affected with no meaningful input regarding their respective utilities. Apologies.
I'm not saying that quantitative analysis of these issues is not useful, just that it is unlikely to be done well by one function in a package in R or SAS.. --
David Winsemius, MD West Hartford, CT