AMR/R/ggplot_rsi.R

192 lines
8.2 KiB
R

# ==================================================================== #
# TITLE #
# Antimicrobial Resistance (AMR) Analysis #
# #
# AUTHORS #
# Berends MS (m.s.berends@umcg.nl), Luz CF (c.f.luz@umcg.nl) #
# #
# LICENCE #
# This program is free software; you can redistribute it and/or modify #
# it under the terms of the GNU General Public License version 2.0, #
# as published by the Free Software Foundation. #
# #
# This program is distributed in the hope that it will be useful, #
# but WITHOUT ANY WARRANTY; without even the implied warranty of #
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the #
# GNU General Public License for more details. #
# ==================================================================== #
#' AMR bar plots with \code{ggplot}
#'
#' Use these functions to create bar plots for antimicrobial resistance analysis. All functions rely on internal \code{\link[ggplot2]{ggplot}} functions.
#' @param data a \code{data.frame} with column(s) of class \code{"rsi"} (see \code{\link{as.rsi}})
#' @param position position adjustment of bars, either \code{"stack"} (default) or \code{"dodge"}
#' @param x variable to show on x axis, either \code{"Antibiotic"} (default) or \code{"Interpretation"} or a grouping variable
#' @param fill variable to categorise using the plots legend, either \code{"Antibiotic"} (default) or \code{"Interpretation"} or a grouping variable
#' @param facet variable to split plots by, either \code{"Interpretation"} (default) or \code{"Antibiotic"} or a grouping variable
#' @param translate_ab a column name of the \code{\link{antibiotics}} data set to translate the antibiotic abbreviations into, using \code{\link{abname}}. Default behaviour is to translate to official names according to the WHO. Use \code{translate_ab = FALSE} to disable translation.
#' @param ... other parameters passed on to \code{\link[ggplot2]{facet_wrap}}
#' @details At default, the names of antibiotics will be shown on the plots using \code{\link{abname}}. This can be set with the option \code{get_antibiotic_names} (a logical value), so change it e.g. to \code{FALSE} with \code{options(get_antibiotic_names = FALSE)}.
#'
#' \strong{The functions}\cr
#' \code{geom_rsi} will take any variable from the data that has an \code{rsi} class (created with \code{\link{as.rsi}}) using \code{\link{portion_df}} and will plot bars with the percentage R, I and S. The default behaviour is to have the bars stacked and to have the different antibiotics on the x axis.
#'
#' \code{facet_rsi} creates 2d plots (at default based on S/I/R) using \code{\link[ggplot2]{facet_wrap}}.
#'
#' \code{scale_y_percent} transforms the y axis to a 0 to 100\% range.
#'
#' \code{scale_rsi_colours} sets colours to the bars: green for S, yellow for I and red for R.
#'
#' \code{theme_rsi} is a \code{\link[ggplot2]{theme}} with minimal distraction.
#'
#' \code{ggplot_rsi} is a wrapper around all above functions that uses data as first input. This makes it possible to use this function after a pipe (\code{\%>\%}). See Examples.
#' @rdname ggplot_rsi
#' @export
#' @examples
#' library(dplyr)
#' library(ggplot2)
#'
#' # get antimicrobial results for drugs against a UTI:
#' ggplot(septic_patients %>% select(amox, nitr, fosf, trim, cipr)) +
#' geom_rsi()
#'
#' # prettify the plot using some additional functions:
#' df <- septic_patients[, c("amox", "nitr", "fosf", "trim", "cipr")]
#' ggplot(df) +
#' geom_rsi() +
#' facet_rsi() +
#' scale_y_percent() +
#' scale_rsi_colours() +
#' theme_rsi()
#'
#' # or better yet, simplify this using the wrapper function - a single command:
#' septic_patients %>%
#' select(amox, nitr, fosf, trim, cipr) %>%
#' ggplot_rsi()
#' \donttest{
#' # it also supports groups (don't forget to use the group on `x` or `facet`):
#' septic_patients %>%
#' select(hospital_id, amox, nitr, fosf, trim, cipr) %>%
#' group_by(hospital_id) %>%
#' ggplot_rsi(x = "hospital_id",
#' facet = "Antibiotic",
#' nrow = 1) +
#' labs(title = "AMR of Anti-UTI Drugs Per Hospital",
#' x = "Hospital")
#'
#' # genuine analysis: check 2 most prevalent microorganisms
#' septic_patients %>%
#' # create new bacterial ID's, with all CoNS under the same group (Becker et al.)
#' mutate(bactid = as.bactid(bactid, Becker = TRUE)) %>%
#' # filter on top 2 bacterial ID's
#' filter(bactid %in% top_freq(freq(.$bactid), 2)) %>%
#' # determine first isolates
#' mutate(first_isolate = first_isolate(.,
#' col_date = "date",
#' col_patient_id = "patient_id",
#' col_bactid = "bactid")) %>%
#' # filter on first isolates
#' filter(first_isolate == TRUE) %>%
#' # join the `microorganisms` data set
#' left_join_microorganisms() %>%
#' # select full name and some antiseptic drugs
#' select(mo = fullname,
#' cfur, gent, cipr) %>%
#' # group by MO
#' group_by(mo) %>%
#' # plot the thing, putting MOs on the facet
#' ggplot_rsi(x = "Antibiotic",
#' facet = "mo") +
#' labs(title = "AMR of Top Two Microorganisms In Blood Culture Isolates",
#' subtitle = "Only First Isolates, CoNS grouped according to Becker et al.",
#' x = "Microorganisms")
#' }
ggplot_rsi <- function(data,
position = "stack",
x = "Antibiotic",
fill = "Interpretation",
facet = NULL,
translate_ab = "official",
...) {
if (!"ggplot2" %in% rownames(installed.packages())) {
stop('this function requires the ggplot2 package.', call. = FALSE)
}
p <- ggplot2::ggplot(data = data) +
geom_rsi(position = position, x = x, fill = fill, translate_ab = translate_ab) +
scale_y_percent() +
theme_rsi()
if (fill == "Interpretation") {
# set RSI colours
p <- p + scale_rsi_colours()
}
if (!is.null(facet)) {
p <- p + facet_rsi(facet = facet, ...)
}
p
}
#' @rdname ggplot_rsi
#' @export
geom_rsi <- function(position = "stack",
x = c("Antibiotic", "Interpretation"),
fill = "Interpretation",
translate_ab = "official") {
x <- x[1]
if (x %in% tolower(c('ab', 'antibiotic', 'abx', 'antibiotics'))) {
x <- "Antibiotic"
} else if (x %in% tolower(c('SIR', 'RSI', 'interpretation', 'interpretations', 'result'))) {
x <- "Interpretation"
}
options(get_antibiotic_names = translate_ab)
ggplot2::layer(geom = "bar", stat = "identity", position = position,
mapping = ggplot2::aes_string(x = x, y = "Percentage", fill = fill),
data = AMR::portion_df, params = list())
}
#' @rdname ggplot_rsi
#' @export
facet_rsi <- function(facet = c("Interpretation", "Antibiotic"), ...) {
facet <- facet[1]
if (facet %in% tolower(c('SIR', 'RSI', 'interpretation', 'interpretations', 'result'))) {
facet <- "Interpretation"
} else if (facet %in% tolower(c('ab', 'antibiotic', 'abx', 'antibiotics'))) {
facet <- "Antibiotic"
}
ggplot2::facet_wrap(facets = facet, scales = "free", ...)
}
#' @rdname ggplot_rsi
#' @export
scale_y_percent <- function() {
ggplot2::scale_y_continuous(name = "Percentage",
breaks = seq(0, 1, 0.1),
limits = c(0, 1),
labels = percent(seq(0, 1, 0.1)))
}
#' @rdname ggplot_rsi
#' @export
scale_rsi_colours <- function() {
ggplot2::scale_fill_brewer(palette = "RdYlGn")
}
#' @rdname ggplot_rsi
#' @export
theme_rsi <- function() {
theme_minimal() +
theme(panel.grid.major.x = element_blank(),
panel.grid.minor = element_blank(),
panel.grid.major.y = element_line(colour = "grey75"))
}