Comparing Two Diagnostic Tests with Dichotomous Results using Paired Data.
testCompareR
Test metrics like sensitivity, specificity, the predictive values and the likelihood ratios are common ways to measure the performance of a diagnostic test. The goal of testCompareR is to make comparing the test metrics from two diagnostic tests with dichotomous outcomes easy. Really easy.
We want clinical researchers to be able to quickly access the statistical methods with the best performance, without having to trawl through the literature or learn how to operate a complicated R package. testCompareR does all of the hard work so you don't have to.
Installation
testCompareR can be installed directly from CRAN using:
install.packages("testCompareR")
You can install the development version of testCompareR from GitHub with:
if (!require("devtools", quietly = TRUE))
install.packages("devtools")
devtools::install_github("Kajlinko/testCompareR")
Examples
testCompareR has two principal functions: compareR() and interpretR().
This example with simulated data demonstrates how they work. The data is provided as a data frame or matrix with three columns in the order test1, test2 and gold. It is sensible to code positive and negative results systematically, but real research can sometimes be messy. The compareR() function will try to take care of that for you.
library(testCompareR)
# simulate some data
test1 <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
test2 <- rep(c("negative", "neg", "no", "n", "-", "0", "2"), 10)
gold <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
df <- data.frame(test1, test2, gold)
# run the tests
results <- compareR(df)
# interpret the results (optional)
interpretR(results)
With two function calls you can calculate the test metrics, including the confidence intervals with the best coverage, and compare the test metrics using the hypothesis tests with the best asymptotic performance. interpretR() even provides a plain English summary of what your results mean in the console.
Additional functions
If you only have one test and the gold standard you can summarise the descriptive statistics quickly with summariseR(). Here the data should be presented as a data frame or matrix with two columns.
library(testCompareR)
# simulate some data
test1 <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
gold <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
df <- data.frame(test1, gold)
# run the tests
summariseR(df)
One final function has been defined to help clinical researchers who want to perform pooled meta-analysis of data that is already available. dataframeR() constructs a data frame that can be provided to compareR() from figures commonly quoted in the literature. dataframeR() has eight parameters: s11, s10, s01, s00, r11, r10, r01, r00.
Understanding the parameter names: s & r represent positive and negative results for the gold standard test, respectively. The first digit represents a positive (1) or negative (0) result for Test 1. The second digit represents a positive (1) or negative (0) result for Test 2.
dataframeR(70, 5, 11, 40, 11, 2, 3, 120)
This data frame can be combined with other data frames to produce a master data frame ready for analysis.
If you have any ideas about additional functionality that you think should be added, please get in touch or, better still, make a pull request and see what you can do with the code.
Contributors
The idea for this package was conceived and developed by Kyle Wilson.
The project was helped greatly by Marc Henrion GitHub.
Additionally, the statistical methods underlying this package and the source code upon which it is based are provided by José Antonio Roldán Nofuentes. If you use the package, please consider referencing his paper when describing your statistical methods. The paper is available here.
License
This work is licensed under the General Public License v3.0 (2007). See LICENSE.md for more details.
References
Yu, Guo & Xu (2014) JSCS. 2014; 84:5,1022-1038 doi:10.1080/00949655.2012.738211
Martín Andrés & Álvarez Hernández (2014) Stat Comput. 2014; 24,65–75 doi:10.1007/s11222-012-9353-5
Roldán-Nofuentes & Sidaty-Regad (2019) JSCS. 2019; 89:14,2621-2644 doi:10.1080/00949655.2019.1628234
Roldán-Nofuentes, Luna del Castillo & Montero-Alonso (2012) Comput Stat Data Anal. 2012; 6,1161–1173. doi:10.1016/j.csda.2011.06.003
Kosinski (2012) Stat Med. 2012; 32,964-977 doi:10.1002/sim.5587
Roldán-Nofuentes, Luna del Castillo (2007) Stat Med. 2007; 26:4179–201. doi:10.1002/sim.2850
Roldán-Nofuentes (2020) BMC Med Res Methodol. 2020; 20,143 doi:10.1186/s12874-020-00988-y.