Type: Package
Title: Continuous Biomarker Evaluation with Adjustment of Covariates
Version: 0.1.5
Author: Ziyi Li
Maintainer: Ziyi Li <zli16@mdanderson.org>
Description: Compute covariate-adjusted specificity at controlled sensitivity level, or covariate-adjusted sensitivity at controlled specificity level, or covariate-adjust receiver operating characteristic curve, or covariate-adjusted thresholds at controlled sensitivity/specificity level. All statistics could also be computed for specific sub-populations given their covariate values. Methods are described in Ziyi Li, Yijian Huang, Datta Patil, Martin G. Sanda (2021+) "Covariate adjustment in continuous biomarker assessment".
License: GPL-2
Encoding: UTF-8
Depends: R (≥ 4.0), quantreg, RColorBrewer
Suggests: knitr, rmarkdown
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2021-03-31 20:16:31 UTC; zli16
Repository: CRAN
Date/Publication: 2021-04-02 08:20:03 UTC

Covariate-adjusted ROC

Description

Compute covariate-adjusted specificity at controlled sensitivity level, or covariate-adjusted sensitivity at controlled specificity level, or covariate-adjust receiver operating characteristic curve.

Usage

caROC(diseaseData, controlData, userFormula, control_sensitivity = NULL,
control_specificity = NULL, mono_resp_method = "ROC",
whichSE = "sample", global_ROC_controlled_by = "sensitivity",
nbootstrap = 100, CI_alpha = 0.95, logit_CI = TRUE,
verbose = TRUE)

Arguments

diseaseData

Data from patients including dependent (biomarker) and independent (covariates) variables.

controlData

Data from controls including dependent (biomarker) and independent (covariates) variables.

userFormula

A character string to represent the function for covariate adjustment. For example, let Y denote biomarker, Z1 and Z2 denote two covariates. Then userFormula = "Y ~ Z1 + Z2".

control_sensitivity

The level(s) of sensitivity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

control_specificity

The level(s) of specificity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

mono_resp_method

The method used to restore monotonicity of the ROC curve or computed sensitivity/specificity value. It should one from the following: "none", "ROC". "none" is not applying any monotonicity respecting method. "ROC" is to apply ROC-based monotonicity respecting approach. Default value is "ROC".

whichSE

The method used to compute standard error. It should be one from the following: "sample", "bootstrap", meaning to calculate the standard error using sample-based approach or bootstrap. Default is "sample".

global_ROC_controlled_by

Whether sensitivity/specificity is used to control when computing global ROC. It should one from the following: "sensitivity", "specificity". Default is "sensitivity".

nbootstrap

Number of boostrap iterations. Default is 100.

CI_alpha

Percentage of confidence interval. Default is 0.95.

logit_CI

Whether to apply logit-based confidence interval. Logit-transformed CI has been identified to be more robust near border area.

verbose

Whether to print out messages. Default value is true.

Value

If control_sensitivity or control_specificity is provided, compute covariate-adjusted specificity (sensitivity) at controlled sensitivity (specificity) level.

Estimate

Covariate-adjusted sensitivity/specificity.

SE

Estimated standard error.

CI

Estimated confidence intervals.

If both control_sensitivity and control_specificity are null, compuate covariate-adjusted ROC curve.

sensitivity

Estimated sensitivity.

specificity

Estimated specificity.

mono_adj

Monotonicity adjustment method.

Author(s)

Ziyi.li <ziyi.li@emory.edu>

Examples

n1 = n0 = 500

## generate data
Z_D <- rbinom(n1, size = 1, prob = 0.3)
Z_C <- rbinom(n0, size = 1, prob = 0.7)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C == 0) + Y_C_Z1 * (Z_C == 1)
M1 <- Y_D_Z0 * (Z_D == 0) + Y_D_Z1 * (Z_D == 1)

diseaseData <- data.frame(M = M1, Z = Z_D)
controlData <- data.frame(M = M0, Z = Z_C)
userFormula = "M~Z"

## calculate covariate-adjusted specificity at
## controlled sensitivity levels (0.2, 0.8, 0.9)
caROC(diseaseData,controlData,userFormula,
      control_sensitivity = c(0.2,0.8, 0.9),
      control_specificity = NULL,mono_resp_method = "ROC",
      whichSE = "bootstrap",nbootstrap = 100,
      CI_alpha = 0.95, logit_CI = TRUE)

## calculate covariate-adjusted sensitivity at
## controlled specificity levels (0.2, 0.8, 0.9)
caROC(diseaseData,controlData,userFormula,
      control_sensitivity = NULL,
      control_specificity = c(0.7,0.8, 0.9),mono_resp_method = "none",
      whichSE = "sample",nbootstrap = 100,
      CI_alpha = 0.95, logit_CI = TRUE)

## calculate the whole covariate-adjusted ROC curve
ROC1 <- caROC(diseaseData,controlData,userFormula = "M~Z",
                 mono_resp_method = "none")
ROC2 <- caROC(diseaseData,controlData,userFormula = "M~Z",
                 mono_resp_method = "ROC")

Get confidence band for covariate-adjusted ROC curve.

Description

Use this function to compute the confidence band for covariate-adjusted ROC curve, with or without monotonicity respecting methods.

Usage

caROC_CB(diseaseData, controlData, userFormula,
mono_resp_method, global_ROC_controlled_by = "sensitivity",
CB_alpha = 0.95, logit_CB = FALSE, nbootstrap = 100,
nbin = 100, verbose = FALSE)

Arguments

diseaseData

Data from patients including dependent (biomarker) and independent (covariates) variables.

controlData

Data from controls including dependent (biomarker) and independent (covariates) variables.

userFormula

A character string to represent the function for covariate adjustment. For example, let Y denote biomarker, Z1 and Z2 denote two covariates. Then userFormula = "Y ~ Z1 + Z2".

mono_resp_method

The method used to restore monotonicity of the ROC curve or computed sensitivity/specificity value. It should one from the following: "none", "ROC". "none" is not applying any monotonicity respecting method. "ROC" is to apply ROC-based monotonicity respecting approach. Default value is "ROC".

global_ROC_controlled_by

Whether sensitivity/specificity is used to control when computing global ROC. It should one from the following: "sensitivity", "specificity". Default is "sensitivity".

CB_alpha

Percentage of confidence band. Default is 0.95.

logit_CB

Whether to use logit-transformed (then transform back) confidence band. Default is FALSE.

nbootstrap

Number of boostrap iterations. Default is 100.

nbin

Number of bins used for constructing confidence band. Default is 100.

verbose

Whether to print out messages during bootstrap. Default value is FALSE.

Value

If global ROC is controlled by sensitivity, a list will be output including the following

Sensitivity

Vector of sensitivities;

Specificity_upper

Upper confidence band for specificity estimations;

Specificity_lower

Lower confidence band for specificity estimations;

global_ROC_controlled_by

"sensitivity".

If global ROC is controlled by Specificity, a list will be output including the following

Specificity

Vector of specificity;

Sensitivity_upper

Upper confidence band for sensitivity estimations;

Sensitivity_lower

Lower confidence band for sensitivity estimations;

global_ROC_controlled_by

"specificity".

Author(s)

Ziyi.li <ziyi.li@emory.edu>

Examples

n1 = n0 = 500

## generate data
Z_D <- rbinom(n1, size = 1, prob = 0.3)
Z_C <- rbinom(n0, size = 1, prob = 0.7)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C == 0) + Y_C_Z1 * (Z_C == 1)
M1 <- Y_D_Z0 * (Z_D == 0) + Y_D_Z1 * (Z_D == 1)

diseaseData <- data.frame(M = M1, Z = Z_D)
controlData <- data.frame(M = M0, Z = Z_C)
userFormula = "M~Z"

### calculate confidence band by controlling sensitivity
### using different monotonicity respecting methods


ROC_CB1 <- caROC_CB(diseaseData,controlData,userFormula,
                      mono_resp_method = "none",
                      CB_alpha = 0.95,
                      nbin = 100,verbose = FALSE)
ROC_CB2 <- caROC_CB(diseaseData,controlData,userFormula,
                      mono_resp_method = "ROC",
                      CB_alpha = 0.95,
                      nbin = 100,verbose = FALSE)

Calculate covariate-adjusted threshold.

Description

This function is used to calculate covariate-adjusted threshold(s) at controlled sensitivity levels or specificity levels.

Usage

caThreshold(userFormula, new_covariates, diseaseData = NULL,
controlData = NULL, control_sensitivity = NULL,
control_specificity = NULL)

Arguments

userFormula

A character string to represent the function for covariate adjustment. For example, let Y denote biomarker, Z1 and Z2 denote two covariates. Then userFormula = "Y ~ Z1 + Z2".

new_covariates

A data frame containing covariates for new data. For example, if my userFormula is "Y ~ Z1 + Z2", new_covariates could be data.frame(Z1 = rnorm(100), Z2 = rnorm(100)).

diseaseData

Data from patients including dependent (biomarker) and independent (covariates) variables.

controlData

Data from controls including dependent (biomarker) and independent (covariates) variables.

control_sensitivity

The level(s) of sensitivity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

control_specificity

The level(s) of specificity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

Value

A vector of covariate-adjusted threshold for all subjects if a scalar sensitivity/specificity is given. A data matrix of covariate-adjusted thresholds for all subjects if a vector of sensitivity/specificity is given.

Author(s)

Ziyi Li <ziyi.li@emory.edu>

Examples

n1 = n0 = 500

## generate data
Z_D <- rbinom(n1, size = 1, prob = 0.3)
Z_C <- rbinom(n0, size = 1, prob = 0.7)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C == 0) + Y_C_Z1 * (Z_C == 1)
M1 <- Y_D_Z0 * (Z_D == 0) + Y_D_Z1 * (Z_D == 1)

diseaseData <- data.frame(M = M1, Z = Z_D)
controlData <- data.frame(M = M0, Z = Z_C)
userFormula = "M~Z"

### generate new covariates
new_covariates <- data.frame(Z = rbinom(20, size = 1, prob = 0.5))

### calculate covariate-adjusted thresholds at controlled
### sensitivity level 0.7, 0.8, 0.9
caThreshold(userFormula, new_covariates,
            diseaseData = diseaseData,
            controlData = NULL,
            control_sensitivity = c(0.7,0.8,0.9),
            control_specificity = NULL)

### calculate covariate-adjusted thresholds at controlled
### sensitivity level 0.7
caThreshold(userFormula,new_covariates,
            diseaseData = diseaseData,
            controlData = NULL,
            control_sensitivity = 0.7,
            control_specificity = NULL)

### calculate covariate-adjusted thresholds at controlled
### specificity level 0.7, 0.8, 0.9
caThreshold(userFormula,new_covariates,
            diseaseData = NULL,
            controlData = controlData,
            control_sensitivity = NULL,
            control_specificity = c(0.7,0.8,0.9))

### calculate covariate-adjusted thresholds at controlled
### specificity level 0.7
caThreshold(userFormula,new_covariates,
            diseaseData = NULL,
            controlData = controlData,
            control_sensitivity = NULL,
            control_specificity = 0.7)

Plot covariate-adjusted ROC.

Description

Function to plot the ROC curve generated from caROC().

Usage

plot_caROC(myROC, ...)

Arguments

myROC

ROC output from caROC() function.

...

Arguments to tune generated plots.

Details

This function can be used to plot other ROC curve, as long as the input contains two components "sensitivity" and "specificity".

Value

Plot the ROC curve.

Author(s)

Ziyi Li <zli16@mdanderson.org>

Examples

n1 = n0 = 500

## generate data
Z_D <- rbinom(n1, size = 1, prob = 0.3)
Z_C <- rbinom(n0, size = 1, prob = 0.7)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C == 0) + Y_C_Z1 * (Z_C == 1)
M1 <- Y_D_Z0 * (Z_D == 0) + Y_D_Z1 * (Z_D == 1)

diseaseData <- data.frame(M = M1, Z = Z_D)
controlData <- data.frame(M = M0, Z = Z_C)
userFormula = "M~Z"

ROC1 <- caROC(diseaseData,controlData,userFormula,
                 mono_resp_method = "none")
ROC2 <- caROC(diseaseData,controlData,userFormula,
                 mono_resp_method = "ROC")

plot_caROC(ROC1)
plot_caROC(ROC2, col = "blue")

Plot confidence band of covariate-adjusted ROC.

Description

A function to plot the confidence band of covariate-adjusted ROC.

Usage

plot_caROC_CB(myROC_CB, add = TRUE, ...)

Arguments

myROC_CB

Output from caROC_CB() function.

add

Whether to add confidence band to existing plot (TRUE) or draw a new one (FALSE). Default is TRUE.

...

Any parameters related with the plot.

Value

No values will be return. This function is for plotting only.

Author(s)

Ziyi Li<ziyi.li@emory.edu>

Examples

library(caROC)
n1 = n0 = 100

## generate data
Z_D <- rbinom(n1, size = 1, prob = 0.3)
Z_C <- rbinom(n0, size = 1, prob = 0.7)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C == 0) + Y_C_Z1 * (Z_C == 1)
M1 <- Y_D_Z0 * (Z_D == 0) + Y_D_Z1 * (Z_D == 1)

diseaseData <- data.frame(M = M1, Z = Z_D)
controlData <- data.frame(M = M0, Z = Z_C)
formula = "M~Z"

ROC_CB1 <- caROC_CB(diseaseData,controlData,formula,
                       mono_resp_method = "none",
                       CB_alpha = 0.95,
                       nbin = 100,verbose = FALSE)
### plot confidence band individually
plot_caROC_CB(ROC_CB1, add = FALSE, lty = 2, col = "blue")

### plot confidence band together with the ROC curve
ROC1 <- caROC(diseaseData,controlData,formula,
                 mono_resp_method = "none", verbose = FALSE)
plot_caROC(ROC1)
plot_caROC_CB(ROC_CB1, add = TRUE, lty = 2, col = "blue")

Plot covariate-adjusted ROC for specific subpopulations.

Description

Function to plot the ROC curve generated from sscaROC().

Usage

plot_sscaROC(myROC, ...)

Arguments

myROC

ROC output from sscaROC() function.

...

Arguments to tune generated plots.

Details

This function can be used to plot other ROC curve, as long as the input contains two components "sensitivity" and "specificity".

Value

Plot the ROC curve.

Author(s)

Ziyi Li <zli16@mdanderson.org>

Examples

n1 = n0 = 1000

## generate data
Z_D1 <- rbinom(n1, size = 1, prob = 0.3)
Z_D2 <- rnorm(n1, 0.8, 1)

Z_C1 <- rbinom(n0, size = 1, prob = 0.7)
Z_C2 <- rnorm(n0, 0.8, 1)

Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C1 == 0) + Y_C_Z1 * (Z_C1 == 1) + Z_C2
M1 <- Y_D_Z0 * (Z_D1 == 0) + Y_D_Z1 * (Z_D1 == 1) + 1.5 * Z_D2

diseaseData <- data.frame(M = M1, Z1 = Z_D1, Z2 = Z_D2)
controlData <- data.frame(M = M0, Z1 = Z_C1, Z2 = Z_C2)
userFormula = "M~Z1+Z2"

target_covariates = c(1, 0.7, 0.9)


myROC <- sscaROC(diseaseData,
                 controlData,
                 userFormula,
                 target_covariates,
                 global_ROC_controlled_by = "sensitivity",
                 mono_resp_method = "none")
plot_sscaROC(myROC, lwd = 1.6)

Plot confidence band of covariate-adjusted ROC in specific subpopulations.

Description

A function to plot the confidence band of covariate-adjusted ROC in specific subpopulations.

Usage

plot_sscaROC_CB(myROC_CB, add = TRUE, ...)

Arguments

myROC_CB

Output from sscaROC_CB() function.

add

Whether to add confidence band to existing plot (TRUE) or draw a new one (FALSE). Default is TRUE.

...

Any parameters related with the plot.

Value

No values will be return. This function is for plotting only.

Author(s)

Ziyi Li<zli16@mdanderson.org>

Examples

n1 = n0 = 500

## generate data
Z_D1 <- rbinom(n1, size = 1, prob = 0.3)
Z_D2 <- rnorm(n1, 0.8, 1)
Z_C1 <- rbinom(n0, size = 1, prob = 0.7)
Z_C2 <- rnorm(n0, 0.8, 1)
Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C1 == 0) + Y_C_Z1 * (Z_C1 == 1) + Z_C2
M1 <- Y_D_Z0 * (Z_D1 == 0) + Y_D_Z1 * (Z_D1 == 1) + 1.5 * Z_D2

diseaseData <- data.frame(M = M1, Z1 = Z_D1, Z2 = Z_D2)
controlData <- data.frame(M = M0, Z1 = Z_C1, Z2 = Z_C2)

userFormula = "M~Z1+Z2"
target_covariates = c(1, 0.7, 0.9)


# example that takes more than a minute to run
myROC <- sscaROC(diseaseData,
                controlData,
                userFormula,
                target_covariates,
                global_ROC_controlled_by = "sensitivity",
                mono_resp_method = "none")

# default nbootstrap is 100
# set nboostrap as 10 here to improve example speed
myROCband <- sscaROC_CB(diseaseData,
                       controlData,
                       userFormula,
                       mono_resp_method = "none",
                       target_covariates,
                       global_ROC_controlled_by = "sensitivity",
                       CB_alpha = 0.95,
                       logit_CB = FALSE,
                       nbootstrap = 10,
                       nbin = 100,
                       verbose = FALSE)

plot_sscaROC(myROC, lwd = 1.6)
plot_sscaROC_CB(myROCband, col = "purple", lty = 2)

Covariate-adjusted continuous biomarker evaluations for specific population.

Description

Provides evalution for continuous biomarkers at controlled sensitivity/specificity level, or ROC curve in specified sub-population.

Usage

sscaROC(diseaseData, controlData, userFormula, target_covariates,
control_sensitivity = NULL, control_specificity = NULL, mono_resp_method = "ROC",
whichSE = "sample", global_ROC_controlled_by = "sensitivity", nbootstrap = 100,
CI_alpha = 0.95, logit_CI = TRUE, verbose = TRUE)

Arguments

diseaseData

Data from patients including dependent (biomarker) and independent (covariates) variables.

controlData

Data from controls including dependent (biomarker) and independent (covariates) variables.

userFormula

A character string to represent the function for covariate adjustment. For example, let Y denote biomarker, Z1 and Z2 denote two covariates. Then userFormula = "Y ~ Z1 + Z2".

target_covariates

Covariates of the interested sub-population. It could be a vector, e.g. c(1, 0.5, 0.8), or a matrix, e.g. target_covariates = matrix(c(1, 0.7, 0.9, 1, 0.8, 0.8), 2, 3, byrow = TRUE)

control_sensitivity

The level(s) of sensitivity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

control_specificity

The level(s) of specificity to be controlled at. Could be a scalar (e.g. 0.7) or a numeric vector (e.g. c(0.7, 0.8, 0.9)).

mono_resp_method

The method used to restore monotonicity of the ROC curve or computed sensitivity/specificity value. It should one from the following: "none", "ROC". "none" is not applying any monotonicity respecting method. "ROC" is to apply ROC-based monotonicity respecting approach. Default value is "ROC".

whichSE

The method used to compute standard error. It should be one from the following: "sample", "bootstrap", meaning to calculate the standard error using sample-based approach or bootstrap. Default is "sample".

global_ROC_controlled_by

Whether sensitivity/specificity is used to control when computing global ROC. It should one from the following: "sensitivity", "specificity". Default is "sensitivity".

nbootstrap

Number of boostrap iterations. Default is 100.

CI_alpha

Percentage of confidence interval. Default is 0.95.

logit_CI

Whether to apply logit-based confidence interval. Logit-transformed CI has been identified to be more robust near border area.

verbose

Whether to print out messages. Default value is true.

Value

If control_sensitivity or control_specificity is provided, compute covariate-adjusted specificity (sensitivity) at controlled sensitivity (specificity) level.

Estimate

Covariate-adjusted sensitivity/specificity.

SE

Estimated standard error.

CI

Estimated confidence intervals.

If both control_sensitivity and control_specificity are null, compuate covariate-adjusted ROC curve.

sensitivity

Estimated sensitivity.

specificity

Estimated specificity.

mono_adj

Monotonicity adjustment method.

Author(s)

Ziyi.li <zli16@mdanderson.org>

Examples

n1 = n0 = 1000
## generate data
Z_D1 <- rbinom(n1, size = 1, prob = 0.3)
Z_D2 <- rnorm(n1, 0.8, 1)
Z_C1 <- rbinom(n0, size = 1, prob = 0.7)
Z_C2 <- rnorm(n0, 0.8, 1)
Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)
M0 <- Y_C_Z0 * (Z_C1 == 0) + Y_C_Z1 * (Z_C1 == 1) + Z_C2
M1 <- Y_D_Z0 * (Z_D1 == 0) + Y_D_Z1 * (Z_D1 == 1) + 1.5 * Z_D2
diseaseData <- data.frame(M = M1, Z1 = Z_D1, Z2 = Z_D2)
controlData <- data.frame(M = M0, Z1 = Z_C1, Z2 = Z_C2)
userFormula = "M~Z1+Z2"
target_covariates = c(1, 0.7, 0.9)
res <- sscaROC(diseaseData,controlData,
               userFormula = userFormula,
               control_sensitivity = c(0.2,0.8, 0.9),
               target_covariates = target_covariates,
               control_specificity = NULL,
               mono_resp_method = "none",
               whichSE = "sample",nbootstrap = 100,
               CI_alpha = 0.95, logit_CI = TRUE)


## bootstrap-based variance estimation
res <- sscaROC(diseaseData,controlData,
               userFormula = userFormula,
               control_sensitivity = c(0.2,0.8, 0.9),
               target_covariates = target_covariates,
               control_specificity = NULL,
               mono_resp_method = "none",
               whichSE = "bootstrap",nbootstrap = 100,
               CI_alpha = 0.95, logit_CI = TRUE)
## monotonization by ROC-based
res <- sscaROC(diseaseData,controlData,
               userFormula = userFormula,
               control_sensitivity = c(0.2,0.8, 0.9),
               target_covariates = target_covariates,
               control_specificity = NULL,
               mono_resp_method = "ROC",
               whichSE = "bootstrap",nbootstrap = 100,
               CI_alpha = 0.95, logit_CI = TRUE)
## control specificity
res <- sscaROC(diseaseData,controlData,
               userFormula = userFormula,
               control_sensitivity = NULL,
               target_covariates = target_covariates,
               control_specificity = c(0.2,0.8, 0.9),
               mono_resp_method = "ROC",
               whichSE = "bootstrap",nbootstrap = 100,
               CI_alpha = 0.95, logit_CI = TRUE)
### get ROC curves
myROC <- sscaROC(diseaseData,
                 controlData,
                 userFormula,
                 target_covariates,
                 global_ROC_controlled_by = "sensitivity",
                 mono_resp_method = "none")

Get confidence band for covariate-adjusted ROC curve for specified sub-population.

Description

Use this function to compute the confidence band for covariate-adjusted ROC curve, with or without monotonicity respecting methods for sub-population.

Usage

sscaROC_CB(diseaseData, controlData, userFormula, mono_resp_method = "none",
target_covariates, global_ROC_controlled_by = "sensitivity", CB_alpha = 0.95,
logit_CB = FALSE, nbootstrap = 100, nbin = 100, verbose = FALSE)

Arguments

diseaseData

Data from patients including dependent (biomarker) and independent (covariates) variables.

controlData

Data from controls including dependent (biomarker) and independent (covariates) variables.

userFormula

A character string to represent the function for covariate adjustment. For example, let Y denote biomarker, Z1 and Z2 denote two covariates. Then userFormula = "Y ~ Z1 + Z2".

mono_resp_method

The method used to restore monotonicity of the ROC curve or computed sensitivity/specificity value. It should one from the following: "none", "ROC". "none" is not applying any monotonicity respecting method. "ROC" is to apply ROC-based monotonicity respecting approach. Default value is "ROC".

target_covariates

Covariates of the interested sub-population. It could be a vector, e.g. c(1, 0.5, 0.8), or a matrix, e.g. target_covariates = matrix(c(1, 0.7, 0.9, 1, 0.8, 0.8), 2, 3, byrow = TRUE)

global_ROC_controlled_by

Whether sensitivity/specificity is used to control when computing global ROC. It should one from the following: "sensitivity", "specificity". Default is "sensitivity".

CB_alpha

Percentage of confidence band. Default is 0.95.

logit_CB

Whether to use logit-transformed (then transform back) confidence band. Default is FALSE.

nbootstrap

Number of boostrap iterations. Default is 100.

nbin

Number of bins used for constructing confidence band. Default is 100.

verbose

Whether to print out messages during bootstrap. Default value is FALSE.

Value

If global ROC is controlled by sensitivity, a list will be output including the following

Sensitivity

Vector of sensitivities;

Specificity_upper

Upper confidence band for specificity estimations;

Specificity_lower

Lower confidence band for specificity estimations;

global_ROC_controlled_by

"sensitivity".

If global ROC is controlled by Specificity, a list will be output including the following

Specificity

Vector of specificity;

Sensitivity_upper

Upper confidence band for sensitivity estimations;

Sensitivity_lower

Lower confidence band for sensitivity estimations;

global_ROC_controlled_by

"specificity".

Author(s)

Ziyi.li <zli16@mdanderson.org>

Examples

n1 = n0 = 500

## generate data
Z_D1 <- rbinom(n1, size = 1, prob = 0.3)
Z_D2 <- rnorm(n1, 0.8, 1)
Z_C1 <- rbinom(n0, size = 1, prob = 0.7)
Z_C2 <- rnorm(n0, 0.8, 1)
Y_C_Z0 <- rnorm(n0, 0.1, 1)
Y_D_Z0 <- rnorm(n1, 1.1, 1)
Y_C_Z1 <- rnorm(n0, 0.2, 1)
Y_D_Z1 <- rnorm(n1, 0.9, 1)

M0 <- Y_C_Z0 * (Z_C1 == 0) + Y_C_Z1 * (Z_C1 == 1) + Z_C2
M1 <- Y_D_Z0 * (Z_D1 == 0) + Y_D_Z1 * (Z_D1 == 1) + 1.5 * Z_D2

diseaseData <- data.frame(M = M1, Z1 = Z_D1, Z2 = Z_D2)
controlData <- data.frame(M = M0, Z1 = Z_C1, Z2 = Z_C2)

userFormula = "M~Z1+Z2"
target_covariates = c(1, 0.7, 0.9)

# default nbootstrap is 100
# set nboostrap as 10 here to improve example speed

myROCband <- sscaROC_CB(diseaseData,
                       controlData,
                       userFormula,
                       mono_resp_method = "none",
                       target_covariates,
                       global_ROC_controlled_by = "sensitivity",
                       CB_alpha = 0.95,
                       logit_CB = FALSE,
                       nbootstrap = 10,
                       nbin = 100,
                       verbose = FALSE)