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.
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")
::install_github("Kajlinko/testCompareR") devtools
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
<- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
test1 <- rep(c("negative", "neg", "no", "n", "-", "0", "2"), 10)
test2 <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
gold
<- data.frame(test1, test2, gold)
df
# run the tests
<- compareR(df)
results
# 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.
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
<- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
test1 <- rep(c("positive", "pos", "p", "yes", "y", "+", "1"), 10)
gold
<- data.frame(test1, gold)
df
# 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.
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.
This work is licensed under the General Public License v3.0 (2007). See LICENSE.md for more details.
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