AUC_boot {dmetatools}

Confidence interval for AUC of summary ROC curve

Description

Calculating the confidence interval for AUC of summary ROC curve by parametric bootstrap.

Usage

AUC_boot(TP, FP, FN, TN, B=2000, alpha=0.95)

Arguments

TP

A vector of the number of true positives (TP)

FP

A vector of the number of false positives (FP)

FN

A vector of the number of false negatives (FN)

TN

A vector of the number of true negatives (TN)

B

The number of bootstrap resampling (default: 2000)

alpha

The confidence level (default: 0.95)

Value

The confidence interval for AUC of summary ROC curve is calculated.

• AUC: The AUC of the summary ROC curve.

• AUC_CI: The 95% confidence interval for the AUC of the summary ROC curve (the confidence level can be changed by alpha).

Author(s)

Hisashi Noma <[e-mail]>

References

Noma, H., Matsushima, Y., and Ishii, R. (2021). Confidence interval for the AUC of SROC curve and some related methods using bootstrap for meta-analysis of diagnostic accuracy studies. Communications in Statistics: Case Studies and Data Analysis 7: 344-358. doi:10.1080/23737484.2021.1894408

Examples

require(mada)

data(cervical)

CT <- cervical[cervical$method==1,]
LAG <- cervical[cervical$method==2,]
MRI <- cervical[cervical$method==3,]

fit1 <- reitsma(CT)    # DTA meta-analysis using the Reitsma model
summary(fit1)
fit2 <- reitsma(LAG)
summary(fit2)
fit3 <- reitsma(MRI)
summary(fit3)

plot(fit1)    # Plot the SROC curves
lines(sroc(fit2), lty=2, col="blue")
ROCellipse(fit2, lty=2, pch=2, add=TRUE, col="blue")
lines(sroc(fit3), lty=3, col="red")
ROCellipse(fit3, lty=3, pch=3, add=TRUE, col="red")
points(fpr(CT), sens(CT), cex = .5)
points(fpr(LAG), sens(LAG), pch = 2, cex = 0.5, col="blue")
points(fpr(MRI), sens(MRI), pch = 3, cex = 0.5, col="red")
legend("bottomright", c("CT", "LAG", "MRI"), pch = 1:3, lty = 1:3, col=c("black","blue","red"))

AUC_boot(CT$TP,CT$FP,CT$FN,CT$TN,B=5)
AUC_boot(LAG$TP,LAG$FP,LAG$FN,LAG$TN,B=5)
AUC_boot(MRI$TP,MRI$FP,MRI$FN,MRI$TN,B=5)
# These are example commands for illustration. B should be >= 1000.