Software

R package CompRisksROC for Competing Risks ROC analysis (Saha and Heagerty, 2010, Biometrics):

Unix/Mac version

Windows version

Installation information is similar to the risksetROC/survivalROC package and can be found here (change to the appropriate package name). Your comments are most welcome!

I co-wrote and maintain R packages for analysis of time-dependent predictive accuracy of survival models.

  • survivalROC - for estimating ROC curve based on Cumulative/Dynamic version of ROC curve. The package can be downloaded from here. (Reference: Heagerty PJ, Lumley T, Pepe MS: Time dependent ROC curves for censored survival data and a diagnostic marker. Biometrics 56:337-344, 2000.)

  • risksetROC - for estimating ROC curve based on Incident/Dynamic version of ROC curve. The package can be downloaded from here. (Reference: Heagerty PJ, Zheng Y: Survival model predictive accuracy and ROC curves. Biometrics 61(1): 92-105, 2005.)

  • Additional help files are here.

  • The documentation with example are here:

The software (R-codes) for the methods discussed in "Time-Dependent Predictive Accuracy in the Presence of Competing Risks" (Biometrics, 2010, available here) is available (see top of page). I have not yet written the help files, but there are comments in the files regarding the purpose of the functions and their arguments – they are very similar to the arguments of survivalROC and risksetROC packages that I co-wrote. Here are some details for the codes:

  1. In the paper, we discuss two approaches – cumulative/dynamic (C/D) approach and incident/dynamic (I/D) approach.

  2. The C/D approach dichotomizes all subjects at t – cases are those who had an event before or at t (T<=t) and the controls are those who survived past t (T>t). The approach is non-parametric and based on cumulative incidence. The codes in the file crCROC.R (competing risks Cumulative ROC) are for C/D approach. Here I have a function to construct ROC only (function crCROC()).

  3. In contrast, the I/D approach dichotomizes the subjects in the riskset and the cases are those who had an event at t (T=t) and the controls are those with T>t. The approach is semi-parametric and based on Cox model and hazard. The codes in the file crRROC.R (competing risks Riskset ROC) are for I/D approach. Here I have functions to construct to construct both ROC (function crRROC()) and AUC (function crRAUC()).

Your comments are most welcome.

Your comments are most welcome.