comprisk.AUC.CI: bootstrap confidence interval for AUC under competing risk...

In mdbrown/survCompetingRisk: Evaluating Prognostic Markers under Competing Risk

Description

This function calculates bootstrap confidence intervals for AUC for time to event data when there are two causes of failure (competing risks). Two different ways of defining cases for these data can be accounted for in the analysis (see 'type' variable below). Confidence bands for the AUC is derived from the empirical bootstrap distribution.

Usage

comprisk.AUC.CI(times, status1, status2, x, Z = NULL, predict.time, type, smooth=FALSE, alpha = 0.05, bootstraps = 500, sigma)

Arguments

times	vector of min(time to event, censoring time) for each individual.
status1	vector of indicators specifying that event 1 was the cause of failure (should equal 1 for individuals where event 1 occured)
status2	vector of indicators specifying that event 2 was the cause of failure (should equal 1 for individuals where event 2 occured)
x	vector of marker values
Z	vector of (discrete) covariate values
predict.time	The prediction time to use to calculate the AUC. Should be a single numeric value.
type	Can take on two values. 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. See 'references' for more information on type.
smooth	TRUE/FALSE (default). When set to TRUE non-parametric smoothing is used instead of Cox proportional hazards model. Note, when smooth = TRUE, Z is ignored.
alpha	(1-alpha)% CI's are returned. Default is alpha = .05 (95% CI's given).
bootstraps	number of bootstraps to use to obtain CI's. Default is set to 500.
sigma	Tuning parameter to be set when smooth = TRUE. The default value of sigma = sd(x)/length(x)^1/3

Value

A dataframe with the following values: Z, lower bound CI for AUC, upper bound CI for AUC. Each row is a confidence interval for the specific Z strata. If no Z is specified, Z is defined as equalling 1 for all individuals, and can be ignored here.

If type = 1, only one confidence interval for AUC is given. If type = 2, we stratify on event type, and so confidence intervals are given for both event 1 and for event 2.

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.

Examples

data(crdata)

#type 1 
myAUC.type1.CI <- comprisk.AUC.CI( times = crdata$times, 
                                status1 = crdata$status1, 
                                status2 = crdata$status2, 
                                x = crdata$x, 
                                Z = crdata$Z, 
                                predict.time = 10, 
                                type = 1, 
                                alpha = .05, 
                                bootstraps = 100) 
#confidence iterval                                
myAUC.type1.CI

myROC.type1 <- comprisk.ROC( times = crdata$times,
                             status1 = crdata$status1, 
                             status2 = crdata$status2, 
                             x = crdata$x, 
                             Z = crdata$Z, 
                             predict.time = 10, 
                             type = 1)
#estimate                                                    
myROC.type1$AUC

#type 2
myAUC.type2.CI <- comprisk.AUC.CI( times = crdata$times, 
                                status1 = crdata$status1, 
                                status2 = crdata$status2, 
                                x = crdata$x, 
                                Z = crdata$Z, 
                                predict.time = 10, 
                                type = 2, 
                                alpha = .05, 
                                bootstraps = 100) 
                                
#confidence iterval                                
myAUC.type2.CI

myROC.type2 <- comprisk.ROC( times = crdata$times,
                             status1 = crdata$status1, 
                             status2 = crdata$status2, 
                             x = crdata$x, 
                             Z = crdata$Z, 
                             predict.time = 10, 
                             type = 2)
#estimate                                                    
myROC.type2$AUC

mdbrown/survCompetingRisk documentation built on May 22, 2019, 3:23 p.m.