Raja, Dr. Edwin Amalraj
2018-Mar-21 09:11 UTC
[R] selectFGR vs weighted coxph for internal validation and calibration curve- competing risks model
Dear Geskus, I want to develop a prediction model. I followed your paper and analysed thro' weighted coxph approach. I can develop nomogram based on the final model also. But I do not know how to do internal validation of the model and subsequently obtain calibration plot. Is it possible to use Wolbers et al Epid 2009 approach 9 (R code for internal validation and calibration) . It is possible to get these measures after using R function 'crr' or 'FGR'. That is why I wanted to go in that route. At the same time, I had this doubt because their approach assume a record per individual whereas weight coxph creates two or more records per individual. I am new to R and could not modify the R code easily. Any suggestion? Has anyone done internal validation and calibration after using weighted coxph approach? Can you kindly refer me to the reference which has R code? Thank you very much for all your inputs and suggestions Regards Amalraj raja -----Original Message----- From: Ronald Geskus [mailto:statistics at inter.nl.net] Sent: 21 March 2018 04:01 To: r-help at r-project.org Cc: Raja, Dr. Edwin Amalraj <amalraj.raja at abdn.ac.uk> Subject: Re: [R] selectFGR - variable selection in fine gray model for competing risks Dear Raja, A Fine and Gray model can be fitted using the standard coxph function with weights that correct for right censoring and left truncation. Hence I guess any function that allows to perform stepwise regression with coxph should work. See e.g. my article in Biometrics https://doi.org/10.1111/j.1541-0420.2010.01420.x, or the vignette "Multi-state models and competing risks" in the survival package. best regards, Ronald Geskus, PhD head of biostatistics group Oxford University Clinical Research unit Ho Chi Minh city, Vietnam associate professor University of Oxford http://www.oucru.org/dr-ronald-b-geskus/ "Raja, Dr. Edwin Amalraj" <amalraj.raja at abdn.ac.uk> writes:> Dear All, > > I would like to use R function 'selectFGR' of fine gray model in > competing risks model. I used the 'Melanoma' data in 'riskRegression' > package. Some of the variables are factor. I get solution for full > model but not in variable selection model. Any advice how to use > factor variable in 'selectFGR' function. The following R code is > produced below for reproducibility. > > library(riskRegression) > library(pec) > dat <-data(Melanoma,package="riskRegression") > Melanoma$logthick <- log(Melanoma$thick) > f1 <- Hist(time,status)~age+sex+epicel+ulcer > df1 <-FGR(f1,cause=1, data=Melanoma) > df1 > df <-selectFGR(f1, data=Melanoma, rule ="BIC", direction="backward") > > Thanks in advice for your suggestion. Is there any alternative solution ? > > Regards > Amalraj raja > > > The University of Aberdeen is a charity registered in Scotland, NoSC013683.> Tha Oilthigh Obar Dheathain na charthannas cl?raichte ann an Alba, ?ir.SC013683. The University of Aberdeen is a charity registered in Scotland, No SC013683. Tha Oilthigh Obar Dheathain na charthannas cl?raichte ann an Alba, ?ir. SC013683.
Ronald Geskus
2018-Mar-27 02:40 UTC
[R] selectFGR vs weighted coxph for internal validation and calibration curve- competing risks model
Dear Raja, I don't know of any out-of-the-box function to perform internal validation with a Fine & Gray model. My suggestion is to tweak the validate.cph function from the rms package. Within that function, you will need to incorporate the crprep or finegray function to compute the weights and then use a weighted Cox model. best regards, Ronald Geskus Raja, Dr. Edwin Amalrajwrote:> Dear Geskus, > > I want to develop a prediction model. I followed your paper and analysed > thro' weighted coxph approach. I can develop nomogram based on the final > model also. But I do not know how to do internal validation of the model > and subsequently obtain calibration plot. Is it possible to use Wolbers > et al Epid 2009 approach 9 (R code for internal validation and > calibration) . It is possible to get these measures after using R > function 'crr' or 'FGR'. That is why I wanted to go in that route. At the > same time, I had this doubt because their approach assume a record per > individual whereas weight coxph creates two or more records per > individual. I am new to R and could not modify the R code easily. Any > suggestion? Has anyone done internal validation and calibration after > using weighted coxph approach? Can you kindly refer me to the reference > which has R code? > > Thank you very much for all your inputs and suggestions > > Regards > Amalraj raja > > -----Original Message----- > From: Ronald Geskus [mailto:statistics at inter.nl.net] > Sent: 21 March 2018 04:01 > To: r-help at r-project.org > Cc: Raja, Dr. Edwin Amalraj <amalraj.raja at abdn.ac.uk> > Subject: Re: [R] selectFGR - variable selection in fine gray model for > competing risks > > Dear Raja, > > A Fine and Gray model can be fitted using the standard coxph function with > weights that correct for right censoring and left truncation. Hence I > guess any function that allows to perform stepwise regression with coxph > should work. See e.g. my article in Biometrics > https://doi.org/10.1111/j.1541-0420.2010.01420.x, or the vignette > "Multi-state models and competing risks" in the survival package. > > best regards, > > Ronald Geskus, PhD > head of biostatistics group > Oxford University Clinical Research unit Ho Chi Minh city, Vietnam > associate professor University of Oxford > http://www.oucru.org/dr-ronald-b-geskus/ > > "Raja, Dr. Edwin Amalraj" <amalraj.raja at abdn.ac.uk> writes: > >> Dear All, >> >> I would like to use R function 'selectFGR' of fine gray model in >> competing risks model. I used the 'Melanoma' data in 'riskRegression' >> package. Some of the variables are factor. I get solution for full >> model but not in variable selection model. Any advice how to use >> factor variable in 'selectFGR' function. The following R code is >> produced below for reproducibility. >> >> library(riskRegression) >> library(pec) >> dat <-data(Melanoma,package="riskRegression") >> Melanoma$logthick <- log(Melanoma$thick) >> f1 <- Hist(time,status)~age+sex+epicel+ulcer >> df1 <-FGR(f1,cause=1, data=Melanoma) >> df1 >> df <-selectFGR(f1, data=Melanoma, rule ="BIC", direction="backward") >> >> Thanks in advice for your suggestion. Is there any alternative solution >> ? >> >> Regards >> Amalraj raja >> >> >> The University of Aberdeen is a charity registered in Scotland, No > SC013683. >> Tha Oilthigh Obar Dheathain na charthannas cl?raichte ann an Alba, ?ir. > SC013683. > > > > The University of Aberdeen is a charity registered in Scotland, No > SC013683. > Tha Oilthigh Obar Dheathain na charthannas cl?raichte ann an Alba, ?ir. > SC013683. >-- R.B. Geskus, PhD head of biostatistics group Oxford University Clinical Research unit Ho Chi Minh city, Vietnam associate professor University of Oxford http://www.oucru.org/dr-ronald-b-geskus/