optimal_roc: Optimal ROC Threshold

In mattwarkentin/precogs: Risk Prediction Model Assessment

Description

Functions to identify the optimal ROC threshold according to various optimization criteria.

Usage

index_of_union(sens = NULL, spec = NULL, thr = NULL, auc = NULL)

youden(sens = NULL, spec = NULL, thr = NULL)

top_left(sens = NULL, spec = NULL, thr = NULL)

min_pval(time = NULL, status = NULL, pred = NULL)

conc_prob(sens = NULL, spec = NULL, thr = NULL)

Arguments

sens	Sensitivities
spec	Specificities
thr	Threshold values for ROC
auc	Area under the ROC curve
time	Time
status	Censoring status
pred	Predictions

Details

• Youden's J Index: This function will compute Youden's J index. This is computed by finding the threshold which maximizes the value (sensitivity + specificity - 1)

• Index of Union: This function will compute the Index of Union statistic which represents the ideal ROC threshold according to maximizing the statistic (sensitivity - auc) + (specificity - auc)

• Top Left: Provides the receiver operating characteristic (ROC) threshold (x-axis) that minimizes the distance from the ROC curve to the to the top left corner of the plot (1,1)

• Minimum P-value: Provides the receiver operating characteristic (ROC) threshold that minimizes the P-value according to Unal (2017), Article ID 3762651

• Concordance Probability: Provides the receiver operating characteristic (ROC) threshold that maximizes the concordance probability. The concordance probability is the product of the sensitivity and specificity.

Value

A scalar containing the chosen threshold.

Examples

# Simulate some data
sens <- runif(100,0,1)
spec <- runif(100,0,1)
thr <- runif(100,0,1)
auc <- runif(1,0,1)

index_of_union(sens, spec, thr, auc)
youden(sens, spec, thr)
top_left(sens, spec, thr)
conc_prob(sens, spec, thr)

mattwarkentin/precogs documentation built on Jan. 12, 2020, 6:24 p.m.