Brant Inman
2008-Feb-04 06:12 UTC
[R] Likelihood ratio test for competing risks regression
R-helpers: I have a question regarding the crr function of the cmprsk package for performing competing risks regression. Specifically, I was wondering if the standard likelihood ratio test for a categorical covariate applies. For example: ---------------------------------------- # Make up a fake example using the PBC dataset in the survival package attach(pbc) stage2 <- ifelse(stage==2, 1, 0) stage3 <- ifelse(stage==3, 1, 0) stage4 <- ifelse(stage==4, 1, 0) newstatus <- ifelse(status==1, 1, ifelse(status==0 & trt==1, 2, 0)) fit1 <- crr(ftime=time, fstatus=newstatus, cencode=0, failcode=1, cov1=(cbind(age, stage2, stage3, stage4))) fit1 fit2 <- crr(ftime=time, fstatus=newstatus, cencode=0, failcode=1, cov1=(cbind(age))) 1 - pchisq(2*(fit1$loglik - fit2$loglik), df=3) ------------------------ If I was interested in the overall impact of "stage" as a predictor in the fake model in fit1 (stage being represented by the 3 dummy variables stage2, stage3 and stage4), would the LR test calculated in the last line of code be valid? Given the pseudolikelihood used in the crr fit, I am not sure if this is a valid statistical test. Thanks for any comments in advance, Brant Inman Mayo Clinic ----------------------------------------------------------> sessionInfo()R version 2.6.1 (2007-11-26) i386-pc-mingw32 locale: LC_COLLATE=English_United States.1252;LC_CTYPE=English_United States.1252;LC_MONETARY=English_United States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252 attached base packages: [1] splines grid stats graphics grDevices utils datasets [8] methods base other attached packages: [1] cmprsk_2.1-7 mitools_1.0 lme4_0.99875-9 [4] Matrix_0.999375-3 nlme_3.1-86 latticeExtra_0.3-1 [7] RColorBrewer_1.0-2 survival_2.34 MASS_7.2-39 [10] foreign_0.8-23 RGraphics_1.0-6 lattice_0.17-4 loaded via a namespace (and not attached): [1] rcompgen_0.1-17>