Hi R users,
I'd like to draw a nomogram using a competing-risks regression (crr function in R), rather than a cox regression. However, the nomogram function provided in the Design package is not good for this purpose. Do you have any suggestion. I really appreciate your help Many thanks F.Abdollah, MD San-Raffele hospital Milan, Italy |
Replace the Design package with the rms package. Use the ordinary linear regression trick to predict the linear predictor from the competing risk regression, then use nomogram on this new model (that merely represents the fit of interest).
Frank
Frank Harrell
Department of Biostatistics, Vanderbilt University |
Many thanks for the prompt response.
However, I am afraid that it is not completely clear for me. I apologize, I am not a statistician. Sorry, may be what I will say make totally non sense, but what I understood is the following: Let's suppose that I need to predict cancer-specific survival using the variables X and Y. What I need to do is to develop a model that include these variables and predict cancer-specific survival using the competing-risks regression. Then, I shall calculate the "predictions" of this model at a certain time point, then I shall use these prediction as an "endpoint", and predict it using a linear regression model that include the same variables, i.e X and Y. Finally, I use the coefficients of this final model to develop a nomogram. Is that correct? Many thanks again Should I calculate the "prediction" of the competing-risks regression model, and then use this "prediction" as an "endpoint" and predict it using a linear regression by including the same variables as predictors? On Fri, Jun 24, 2011 at 1:27 PM, Frank Harrell [via R] <[hidden email]> wrote: > Replace the Design package with the rms package. Use the ordinary linear > regression trick to predict the linear predictor from the competing risk > regression, then use nomogram on this new model (that merely represents the > fit of interest). > Frank > > Firas Abdollah wrote: > Hi R users, > > I'd like to draw a nomogram using a competing-risks regression (crr function > in R), rather than a cox regression. However, the nomogram function provided > in the Design package is not good for this purpose. > Do you have any suggestion. > I really appreciate your help > > Many thanks > > F.Abdollah, MD > San-Raffele hospital > Milan, Italy > > Frank Harrell > Department of Biostatistics, Vanderbilt University > > ________________________________ > If you reply to this email, your message will be added to the discussion > below: > http://r.789695.n4.nabble.com/Competing-risks-nomogram-tp3621907p3622291.html > To unsubscribe from Competing-risks nomogram, click here. -- Firas Abdollah, MD Dept. of Urology San Raffaele Hospital Vita-Salute University, Via Olgettina 60, 20132, Milan, Italy Tel. +39 02 2643 7286 Fax. +39 02 2643 7298 E-mail: [hidden email] |
Yes you use the linear predictor from your regression as the dependent variable in the rms package's ols function. You will get an R^2 of 1.0. You can depict the ols model with nomogram(). Note that there are so many statistical issues in competing risks that doing this without a statistician is risky.
Frank
Frank Harrell
Department of Biostatistics, Vanderbilt University |
Many thanks
On Sat, Jun 25, 2011 at 10:00 PM, Frank Harrell [via R] <[hidden email]> wrote: > Yes you use the linear predictor from your regression as the dependent > variable in the rms package's ols function. You will get an R^2 of 1.0. > You can depict the ols model with nomogram(). Note that there are so many > statistical issues in competing risks that doing this without a statistician > is risky. > Frank > Frank Harrell > Department of Biostatistics, Vanderbilt University > > ________________________________ > If you reply to this email, your message will be added to the discussion > below: > http://r.789695.n4.nabble.com/Competing-risks-nomogram-tp3621907p3625011.html > To unsubscribe from Competing-risks nomogram, click here. -- Firas Abdollah, MD Dept. of Urology San Raffaele Hospital Vita-Salute University, Via Olgettina 60, 20132, Milan, Italy Tel. +39 02 2643 7286 Fax. +39 02 2643 7298 E-mail: [hidden email] |
In reply to this post by Firas Abdollah
Dear Firas,
Regression on the subdistribution hazard can be performed by fitting a weighted Cox model. See Geskus, Biometrics 67, p. 39-49, 2011. Hence, cph (and coxph) can be used directly; there is no need to use the crr function in cmprsk. The result from the weighted cph fit should allow you to obtain a nomogram for the cause-specific cumulative incidence. with best regards, Ronald Geskus Academic Medical Center Amsterdam |
I will try it out..Thanks a lot
On Mon, Jun 27, 2011 at 10:08 AM, rgeskus [via R] <[hidden email]> wrote: > Dear Firas, > > Regression on the subdistribution hazard can be performed by fitting a > weighted Cox model. See Geskus, Biometrics 67, p. 39-49, 2011. Hence, cph > (and coxph) can be used directly; there is no need to use the crr function > in cmprsk. The result from the weighted cph fit should allow you to obtain a > nomogram for the cause-specific cumulative incidence. > > with best regards, > > Ronald Geskus > Academic Medical Center > Amsterdam > > ________________________________ > If you reply to this email, your message will be added to the discussion > below: > http://r.789695.n4.nabble.com/Competing-risks-nomogram-tp3621907p3627184.html > To unsubscribe from Competing-risks nomogram, click here. -- Firas Abdollah, MD Dept. of Urology San Raffaele Hospital Vita-Salute University, Via Olgettina 60, 20132, Milan, Italy Tel. +39 02 2643 7286 Fax. +39 02 2643 7298 E-mail: [hidden email] |
Powered by Nabble | Edit this page |