survCompetingRisk
This package aims to help to evaluate the prognostic accuracy of a marker with multiple competing risk events. Functions to calculate the AUC, ROC, PPV, and NPV are provided. A discrete covariate Z, if available, can be included.
There are five main functions in this package:
comprisk.ROC: Calculate the values needed to plot the ROC curve, along with the AUC.comprisk.ROC.CI: Calculate bootstrap confidence intervals for a given set of TPR or FPR values.comprisk.AUC.CI: Calculate bootstrap confidence intervals for the AUC.comprisk.PPV: Calculate PPV and NPV values.comprisk.PPV.CI: Calculate bootstrap confidence intervals for a given set of PPV, NPV, or v (marker quantile).
Tutorial
library(survCompetingRisk)#simulated data for illustration
data(crdata)
#take a look
head(crdata)## times status1 status2 x Z
## 1 9.84883036 0 1 -1.3663182 1
## 2 1.72881288 1 0 0.1148226 0
## 3 0.49552382 1 0 -0.2008914 1
## 4 6.50569196 0 1 0.2233128 0
## 5 0.09322555 1 0 2.3786242 1
## 6 0.95367818 1 0 1.0352069 0##ROC curve
#Set type = 1 if case is defined by the event of interest, and controls are all the rest. Set type = 2 if case is defined by stratifying on event type, and controls are those who have not experienced any events.
myROC.type1 <- comprisk.ROC( times = crdata$times,
status1 = crdata$status1,
status2 = crdata$status2,
x = crdata$x,
Z = crdata$Z,
predict.time = 10,
type = 1)
myROC.type1## $AUC
## Z_0 Z_1
## 1 0.8489155 0.8413245
##
## $ROC
## TPR FPR Z
## 1 1.0000000 1.0000000 0
## 2 0.9995985 0.9851011 0
## 3 0.9989487 0.9704038 0
## 4 0.9982518 0.9557448 0
## 5 0.9974051 0.9412073 0
## 6 0.9965563 0.9266715 0
## 7 0.9955183 0.9122893 0
## 8 0.9944637 0.8979207 0
## 9 0.9932859 0.8836519 0
## 10 0.9920935 0.8693951 0
##
## ...(only first ten rows displayed)...#plot
tmp <- myROC.type1$ROC
plot(tmp$FPR[tmp$Z==0], tmp$TPR[tmp$Z==0], type="l",lwd=2, main="Type 1", xlab="FPR", ylab="TPR")
lines(tmp$FPR[tmp$Z==1], tmp$TPR[tmp$Z==1], lty=2, lwd=2)
legend(x="bottomright", c("Z = 0", "Z = 1"), lty=c(1,2))
lines(c(0, 1), c(0,1), col="lightgrey")
## type 2
myROC.type2 <- comprisk.ROC( times = crdata$times,
status1 = crdata$status1,
status2 = crdata$status2,
x = crdata$x,
Z = crdata$Z,
predict.time = 10,
type = 2)
myROC.type2## $AUC
## event1 event2
## Z_0 0.8537762 0.5425009
## Z_1 1.6909566 1.1010442
##
## $ROC
## TPR.event1 FPR.event1 TPR.event2 FPR.event2 Z
## 1 1.0000000 1.0000000 1.0000000 1.0000000 0
## 2 0.9995985 0.9855387 0.9849776 0.9855387 0
## 3 0.9989487 0.9706154 0.9703441 0.9706154 0
## 4 0.9982518 0.9556401 0.9557743 0.9556401 0
## 5 0.9974051 0.9405470 0.9413937 0.9405470 0
## 6 0.9965563 0.9254528 0.9270156 0.9254528 0
## 7 0.9955183 0.9102868 0.9128547 0.9102868 0
## 8 0.9944637 0.8951176 0.8987120 0.8951176 0
## 9 0.9932859 0.8799377 0.8847005 0.8799377 0
## 10 0.9920935 0.8647580 0.8707043 0.8647580 0
##
## ...(only first ten rows displayed)...comprisk.AUC.CI( times = crdata$times,
status1 = crdata$status1,
status2 = crdata$status2,
x = crdata$x,
Z = crdata$Z,
predict.time = 10,
type = 1,
bootstraps = 25) #set to 500-1000 in practice!## Z AUC_0.025 AUC_0.975
## 1 0 0.8296305 0.8719347
## 2 1 0.8162288 0.8549955See help files for individual examples ?comprisk.ROC, ?comprisk.ROC.CI, ...
Direct any questions to mdbrown@fhcrc.org.
References
For more information regarding estimation procedures see the following references:
Zheng Y, Cai T, Jin Y, Feng Z. Evaluating prognostic accuracy of biomarkers under competing risk. Biometrics. 2012 Jun;68(2):388-96.
Zheng Y, Cai T, Feng Z, and Stanford J. Semiparametric Models of Time-dependent Predictive Values of Prognostic Biomarkers. Biometrics. 2010, 66: 50-60.