Estimating Specificity at Controlled Sensitivity, or Vice Versa

In SenSpe: Estimating Specificity at Controlled Sensitivity, or Vice Versa

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Package `SenSpe' considers biomarker evaluation and comparison in terms of specificity at a controlled sensitivity level, or sensitivity at a controlled specificity level. Point estimation and exact bootstrap of Huang, Parakati, Patil, and Sanda (2023) for the one- and two-biomarker problems are implemented.

Installation

`SenSpe' is available on CRAN:

install.packages("SenSpe")

Estimating specificity at a controlled sensitivity level (or sensitivity at a controlled specificity level) with a single biomarker

Perform point estimation and exact bootstrap-based inference, with a simulated dataset:

library("SenSpe")
## simulate biomarkers of 100 cases and 100 controls
set.seed(1234)
n1 <- 100
n0 <- 100
mk <- c(rnorm(n1,1,1),rnorm(n0,0,1))
## estimate specificity at controlled 0.95 sensitivity
snsp1m(mk, n1=n1, s0=0.95)

Function snsp1m outputs estimated threshold (threshold), estimated specificity at controlled sensitivity (or sensitivity at controlled specificity) (hss), exact bootstrap variance estimate for the performance metric (hvar) along with its components (hvar1 and hvar2), exact bootstrap distribution (btpdf), Wald confidence intervals (wald_ci), percentile confidence interval (pct_ci), score confidence intervals (scr_ci), and exact bootstrap version of the BTII interval in Zhou and Qin (2005, Statistics in Medicine 24, pp 465–477) (zq_ci).

Two-biomarker paired comparison in specificity at a controlled sensitivity level (or sensitivity at a controlled specificity level)

## simulate paired biomarkers X and Y, with correlation 0.5, 100 cases and 100 controls
n1 <- 100
n0 <- 100
rho <- 0.5
set.seed(1234)
mkx <- rnorm(n1+n0,0,1)
mky <- rho*mkx + sqrt(1-rho^2)*rnorm(n1+n0,0,1)
mkx <- mkx + c(rep(2,n1),rep(0,n0))
mky <- mky + c(rep(1,n1),rep(0,n0))
mk <- rbind(mkx,mky)
## compare specificity at controlled 0.95 sensitivity
snsp2mp(mk, 100, 0.95)

Function snsp2mp outputs estimated differences (diff), exact bootstrap mean of the empirical difference (btmn), exact bootstrap variance estimate for the empirical difference (btva), exact bootstrap distribution of the empirical difference (btdist), Wald confidence intervals (wald_ci), percentile confidence interval (pct_ci), score confidence intervals (scr_ci), and an extension of the BTII interval in Zhou and Qin (2005, Statistics in Medicine 24, pp 465–477) (zq_ci).

Two-biomarker unpaired comparison in specificity at a controlled sensitivity level (or sensitivity at a controlled specificity level)

set.seed(1234)
## simulate biomarker X with 100 cases and 100 controls
mkx <- c(rnorm(100,2,1),rnorm(100,0,1))
## simulate biomarker Y with 100 cases and 100 controls
mky <- c(rnorm(100,1,1),rnorm(100,0,1))
## compare specificity at controlled 0.95 sensitivity
snsp2mup(mkx, 100, mky, 100, 0.95)

Function snsp2mup outputs estimated differences (diff), exact bootstrap variance estimate for the empirical difference (hvar), Wald confidence intervals (wald_ci), percentile confidence interval (pct_ci), score confidence intervals (scr_ci), and an extension of the BTII interval in Zhou and Qin (2005, Statistics in Medicine 24, pp 465–477) (zq_ci).

References

Huang, Y., Parakati, I., Patil, D. H.,and Sanda, M. G. (2023). Interval estimation for operating characteristic of continuous biomarkers with controlled sensitivity or specificity, Statistica Sinica 33, 193–214.

SenSpe documentation built on May 29, 2024, 9:21 a.m.