Accessing Concordance Statistics • demcrit

The purpose of this vignette is to provide access to the raw statistical results for all concordance indexes from pairwise comparisons of algorithms for Parkinson’s Disease Dementia (PDD) discussed in the source article.

Set-up

To explore these results, you first need to install and load a local instance of the demcrit R package:

# If you do not have the devtools package,
# install it by uncommenting the following line:
#install.packages(devtools)
devtools::install_github("josefmana/demcrit")

Data

Once installed, the concordance statistics can be loaded. The following code chunk loads the demcrit package (all subsequent examples assume the package is already loaded) and reads the concordance data into the object concords:

library(demcrit)
data("concords")

You can find a detailed description of the dataset in its documentation:

help("concords")

The concords object is a tibble, meaning it can be easily manipulated using packages from the tidyverse ecosystem.

Example Uses

For example, the following code filters the data to retain only those pairwise comparisons where algorithm "Lvl.II (1)" served as the reference:

library(tidyverse) # load the package

# Extract the subdata:
lvlII1_reference <- concords |>
  filter(reference == "Lvl.II (1)")

For more nuanced data operations, you can also load the rates dataset, which contains algorithm specifications and estimated rates of probable PDD in the sample:

data("rates")
help("rates")

Using both datasets, it is possible, for instance, to retain only comparisons where "Lvl.II (1)" served as the reference and where the predictor used the same operationalisation of IADL deficit. The resulting data can then be sorted, for example, by the predictor’s specificity (from highest to lowest):

# Extract operationalisation of IADL deficit
# used by the "Lvl.II (1)" algorithm:
IADL_oper <- subset(rates, type == "Lvl.II (1)")$IADL

# Find algorithms using the same operationalisation:
algos_of_interest <- subset(rates, IADL == IADL_oper)$type

# Keep only pairs of interest:
lvlII1_reference_2 <- concords |>
  filter(reference == "Lvl.II (1)") |>
  filter(predictor %in% algos_of_interest) |>
  arrange(desc(Specificity))

Export

Finally, you can export the data to CSV or Excel format for use in other software. The following examples use the readr and writexl packages, though any preferred method can be used:

# Export to CSV:
readr::write_csv(concords, "demcrit_concordance_stats.csv")

# Export to Excel (single sheet):
writexl::write_xlsx(concords, "demcrit_concordance_stats.xlsx")

# Export to Excel (two sheets: concordance statistics and algorithm details):
writexl::write_xlsx(
  list(concordance = concords, algorithms = rates),
  path = "demcrit_concordance_stats_with_algos.xlsx"
)

# Note: Adjust the second argument to whatever
# "path/to/your/data" you find suitable.