TG.ROC: Two-Graphs Receiving Operating Characteristics.

In UncertainInterval: Uncertain Interval Methods for Three-Way Cut-Point Determination in Test Results

Description

The function supports the determination and plot of the sensitivity and specificity against the possible thresholds and shows an intermediate range of test results that is considered as less accurate.

Usage

TG.ROC(
  ref,
  test,
  Se.criterion = 0.9,
  Sp.criterion = 0.9,
  model = c("none", "binormal"),
  plot = FALSE,
  position.legend = "left",
  cex.legend = 1
)

Arguments

ref	The reference standard. A column in a data frame or a vector indicating the classification by the reference test. The reference standard must be coded either as 0 (absence of the condition; controls) or 1 (presence of the condition; deviation from controls).
test	The numeric test scores under evaluation. When mean(test[ref == 0]) > mean(test[ref == 1]) it is assumed that higher test scores indicate presence of the condition, otherwise that lower test scores indicate presence of the condition.
Se.criterion	Default = .95. Minimum desired value of Se.
Sp.criterion	Default = .95. Minimum desired value of Sp.
model	Default = 'none'. Model to use, either bi-normal or none (non-parametric)
plot	Defaults= FALSE. Whether a plot is shown for Se and Sp against the thresholds.
position.legend	Default: 'left'. Position of the legend. Most used values: "left", "right".
cex.legend	Default: 1. Relative size of the legend.

Details

This function implements a non-parametric and a bi-normal model. See Landsheer(2018) for an evaluative description. When model='none' and the data have a limited number of values, the upper and lower threshold show the first values which comply with the criteria.

Warning 1: Whn using test scores where higher test scores indicate presence of a disease, the whole range of test scores starting at the lowest test score have perfect Sensitivity (1.00), at the cost of a maximal number of false positives; the sensitivity is 0.00. When moving to higher test scores, the value of sensitivity decreases and the value of specificity increases. Therefore, the lowest test scores are best used for negative classifications, but these are precisely the test scores with the highest sensitivity. In TG-ROC the test scores <= the lower limit are interpreted for negative classifications. However, the whole range of test values >= lower limit provides the minimal desired positive accuracy (Se.criterion) (at the cost of a large number of false positives). Similarly, the test scores >= the upper limit are interpreted for positive classifications, while the whole range of test values <= upper limit provides the minimal desired negative accuracy (Sp.criterion) (at the cost of a large number of false negatives). Of course, this is also true for tests where the lowest scores indicate the presence of the disease, but only reversed.

Warning 2: The Intermediate range can cover a relatively small part of the area of overlap between the two distributions. In that case test scores with relative low number of false classifications are considered as intermediate.

Please note that the definition of the intermediate interval deviates substantially from the definition of an uncertain interval.

The TG-ROC (Two Graphs Receiver Operating Characteristics) plot shows the diminishing values of Se and increasing values of Sp against the possible thresholds.

Value

Thresholds for the intermediate zone. Lower threshold < Test scores < Upper threshold is the intermediate range. The range of test values >= lower limit provides the desired positive accuracy (Se.criterion), while the range of test values <= upper limit provides the desired negative accuracy (Sp.criterion).

References

Greiner, M. (1995). Two-graph receiver operating characteristic (TG-ROC): A Microsoft-EXCEL template for the selection of cut-off values in diagnostic tests. Journal of Immunological Methods, 185(1), 145-146.

Greiner, M. (1996). Two-graph receiver operating characteristic (TG-ROC): Update version supports optimisation of cut-off values that minimise overall misclassification costs. Journal of Immunological Methods, 191(1), 93-94.

Landsheer, J. A. (2018). The Clinical Relevance of Methods for Handling Inconclusive Medical Test Results: Quantification of Uncertainty in Medical Decision-Making and Screening. Diagnostics, 8(2), 32. https://doi.org/10.3390/diagnostics8020032

Examples

ref = c(rep(0,100), rep(1,100))
test = c(rnorm(100, 0, 1), rnorm(100, 1, 1))
TG.ROC(ref, test, model='binormal', plot=TRUE)
TG.ROC(ref, test, model='none', plot=TRUE)

UncertainInterval documentation built on March 3, 2021, 1:10 a.m.