risksetROC (version 1.0.4.1)

CoxWeights: Incident/Dynamic (I/D) ROC curve, AUC and integrated AUC (iAUC) estimation of censored survival data

Description

This function estimates of TP and FP based on a Cox model as discussed in Heagerty and Zheng, 2005, for incident/dynamic ROC curve. TP is estimated as Equation (1) and FP is estimated as Equation (2) of the paper.

Usage

CoxWeights(marker, Stime, status, predict.time, entry)

Arguments

marker

estimated linear predictor from a set of covariates. Note that this linear predictor can be obtained from any model.

Stime

For right censored data, this is the follow up time. For left truncated data, this is the ending time for the interval.

status

Indicator of status, 1 if death or event, 0 otherwise.

predict.time

Time point of the ROC curve.

entry

For left truncated data, this is the entry time of the interval. The default is set to NULL for right censored data.

Value

Returns a list of the following items:

eta

unique marker values for calculation of TP and FP

TP

True Positive values corresponding to unique marker values

FP

False Positive values corresponding to unique marker values

AUC

Area Under (ROC) Curve at time predict.time

Details

Suppose we have censored survival data (right censored or both left-truncated and right censored data) along with a marker value and we want to see how well the marker predicts the survival time for the subjects in the dataset using Incident/dynamic definition of ROC curve. In particular, suppose we have survival times in days and we want to see how well the marker predicts the one-year survival (predict.time=365 days). This function CoxWeights(), returns the unique marker values, TP (True Positive), FP (False Positive) and AUC (Area under (ROC) curve) corresponding to the time point of interest (predict.time). Note that the linear predictor marker can be obtained from any model, specifically, the survival model may be based on either a PH or a time-varying Cox model.

References

Heagerty, P.J., Zheng Y. (2005) Survival Model Predictive Accuracy and ROC curves Biometrics, 61, 92 -- 105

Examples

# NOT RUN {
library(MASS)
data(VA)
survival.time <- VA$stime
survival.status <- VA$status
score <- VA$Karn
cell.type <- factor(VA$cell )
tx <- as.integer( VA$treat==1 )
age <- VA$age
survival.status[VA$stime > 500 ] <- 0
survival.time[VA$stime > 500 ] <- 500
library(survival)
fit0 <- coxph( Surv(survival.time,survival.status)
        ~ score + cell.type + tx + age, na.action=na.omit )
summary(fit0)
eta <- fit0$linear.predictor
AUC <- NULL
out <- CoxWeights(marker=eta, Stime=survival.time, status=survival.status,
predict.time=30)
## to see how well the marker predicts one-month survival
AUC <- out$AUC
# }