I think this works:
95% CI for bias-corrected performance stats
aucs.boot<-NULL
slopes.boot<-NULL
ints.boot<-NULL
dxy.boot<-NULL
R2.boot<-NULL
B.boot<-NULL
for (i in 1:1000){
data_boot=data[sample(1:nrow(data), 15000 , replace=TRUE),]
model.boot <- lrm(outcome ~ predictor.vector
, data=data_boot, y=TRUE, x=TRUE, linear.predictors=T)
val.boot<-validate(model.boot, B=1000)
res.boot<-list(CalculateAucFromDxy(val.boot))
aucs.boot <- rbind(aucs.boot, res.boot[[1]][“AUC”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
slopes.boot <- rbind(slopes.boot, res.boot[[1]][“Slope”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
ints.boot <- rbind(ints.boot, res.boot[[1]][“Intercept”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
dxy.boot <- rbind(dxy.boot, res.boot[[1]][“Dxy”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
R2.boot <- rbind(R2.boot, res.boot[[1]][“R2”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
B.boot <- rbind(B.boot, res.boot[[1]][“B”, c(“index.orig”,“training”,“test”,“optimism”,“index.corrected”,“n”)])
}
quantile(aucs.boot[,“index.corrected”], c(0.025))
quantile(aucs.boot[,“index.corrected”], c(0.975))