mirror of https://github.com/msberends/AMR.git
471 lines
18 KiB
R
471 lines
18 KiB
R
# ==================================================================== #
|
|
# TITLE #
|
|
# Antimicrobial Resistance (AMR) Data Analysis for R #
|
|
# #
|
|
# SOURCE #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# LICENCE #
|
|
# (c) 2018-2021 Berends MS, Luz CF et al. #
|
|
# Developed at the University of Groningen, the Netherlands, in #
|
|
# collaboration with non-profit organisations Certe Medical #
|
|
# Diagnostics & Advice, and University Medical Center Groningen. #
|
|
# #
|
|
# This R package is free software; you can freely use and distribute #
|
|
# it for both personal and commercial purposes under the terms of the #
|
|
# GNU General Public License version 2.0 (GNU GPL-2), as published by #
|
|
# the Free Software Foundation. #
|
|
# We created this package for both routine data analysis and academic #
|
|
# research and it was publicly released in the hope that it will be #
|
|
# useful, but it comes WITHOUT ANY WARRANTY OR LIABILITY. #
|
|
# #
|
|
# Visit our website for the full manual and a complete tutorial about #
|
|
# how to conduct AMR data analysis: https://msberends.github.io/AMR/ #
|
|
# ==================================================================== #
|
|
|
|
#' Antibiotic Class Selectors
|
|
#'
|
|
#' These functions help to filter and select columns with antibiotic test results that are of a specific antibiotic class, without the need to define the columns or antibiotic abbreviations. \strong{\Sexpr{ifelse(as.double(R.Version()$major) + (as.double(R.Version()$minor) / 10) < 3.2, paste0("NOTE: THESE FUNCTIONS DO NOT WORK ON YOUR CURRENT R VERSION. These functions require R version 3.2 or later - you have ", R.version.string, "."), "")}}
|
|
#' @inheritSection lifecycle Stable Lifecycle
|
|
#' @param ab_class an antimicrobial class, like `"carbapenems"`. The columns `group`, `atc_group1` and `atc_group2` of the [antibiotics] data set will be searched (case-insensitive) for this value.
|
|
#' @param only_rsi_columns a [logical] to indicate whether only columns of class `<rsi>` must be selected (defaults to `FALSE`), see [as.rsi()]
|
|
#' @details \strong{\Sexpr{ifelse(as.double(R.Version()$major) + (as.double(R.Version()$minor) / 10) < 3.2, paste0("NOTE: THESE FUNCTIONS DO NOT WORK ON YOUR CURRENT R VERSION. These functions require R version 3.2 or later - you have ", R.version.string, "."), "")}}
|
|
#'
|
|
#'
|
|
#' These functions can be used in data set calls for selecting columns and filtering rows, see *Examples*. They support base R, but work more convenient in dplyr functions such as [`select()`][dplyr::select()], [`filter()`][dplyr::filter()] and [`summarise()`][dplyr::summarise()].
|
|
#'
|
|
#' All columns in the data in which these functions are called will be searched for known antibiotic names, abbreviations, brand names, and codes (ATC, EARS-Net, WHO, etc.) in the [antibiotics] data set. This means that a selector like e.g. [aminoglycosides()] will pick up column names like 'gen', 'genta', 'J01GB03', 'tobra', 'Tobracin', etc.
|
|
#'
|
|
#' The group of betalactams consists of all carbapenems, cephalosporins and penicillins.
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @name antibiotic_class_selectors
|
|
#' @export
|
|
#' @inheritSection AMR Reference Data Publicly Available
|
|
#' @inheritSection AMR Read more on Our Website!
|
|
#' @examples
|
|
#' # `example_isolates` is a data set available in the AMR package.
|
|
#' # See ?example_isolates.
|
|
#'
|
|
#' # Base R ------------------------------------------------------------------
|
|
#'
|
|
#' # select columns 'IPM' (imipenem) and 'MEM' (meropenem)
|
|
#' example_isolates[, carbapenems()]
|
|
#'
|
|
#' # select columns 'mo', 'AMK', 'GEN', 'KAN' and 'TOB'
|
|
#' example_isolates[, c("mo", aminoglycosides())]
|
|
#'
|
|
#' # filter using any() or all()
|
|
#' example_isolates[any(carbapenems() == "R"), ]
|
|
#' subset(example_isolates, any(carbapenems() == "R"))
|
|
#'
|
|
#' # filter on any or all results in the carbapenem columns (i.e., IPM, MEM):
|
|
#' example_isolates[any(carbapenems()), ]
|
|
#' example_isolates[all(carbapenems()), ]
|
|
#'
|
|
#' # filter with multiple antibiotic selectors using c()
|
|
#' example_isolates[all(c(carbapenems(), aminoglycosides()) == "R"), ]
|
|
#'
|
|
#' # filter + select in one go: get penicillins in carbapenems-resistant strains
|
|
#' example_isolates[any(carbapenems() == "R"), penicillins()]
|
|
#'
|
|
#'
|
|
#' # dplyr -------------------------------------------------------------------
|
|
#'
|
|
#' if (require("dplyr")) {
|
|
#'
|
|
#' # this will select columns 'IPM' (imipenem) and 'MEM' (meropenem):
|
|
#' example_isolates %>%
|
|
#' select(carbapenems())
|
|
#'
|
|
#' # this will select columns 'mo', 'AMK', 'GEN', 'KAN' and 'TOB':
|
|
#' example_isolates %>%
|
|
#' select(mo, aminoglycosides())
|
|
#'
|
|
#' # any() and all() work in dplyr's filter() too:
|
|
#' example_isolates %>%
|
|
#' filter(any(aminoglycosides() == "R"),
|
|
#' all(cephalosporins_2nd() == "R"))
|
|
#'
|
|
#' # also works with c():
|
|
#' example_isolates %>%
|
|
#' filter(any(c(carbapenems(), aminoglycosides()) == "R"))
|
|
#'
|
|
#' # not setting any/all will automatically apply all():
|
|
#' example_isolates %>%
|
|
#' filter(aminoglycosides() == "R")
|
|
#' #> i Assuming a filter on all 4 aminoglycosides.
|
|
#'
|
|
#' # this will select columns 'mo' and all antimycobacterial drugs ('RIF'):
|
|
#' example_isolates %>%
|
|
#' select(mo, ab_class("mycobact"))
|
|
#'
|
|
#'
|
|
#' # get bug/drug combinations for only macrolides in Gram-positives:
|
|
#' example_isolates %>%
|
|
#' filter(mo_is_gram_positive()) %>%
|
|
#' select(mo, macrolides()) %>%
|
|
#' bug_drug_combinations() %>%
|
|
#' format()
|
|
#'
|
|
#'
|
|
#' data.frame(some_column = "some_value",
|
|
#' J01CA01 = "S") %>% # ATC code of ampicillin
|
|
#' select(penicillins()) # only the 'J01CA01' column will be selected
|
|
#'
|
|
#'
|
|
#' # with dplyr 1.0.0 and higher (that adds 'across()'), this is all equal:
|
|
#' # (though the row names on the first are more correct)
|
|
#' example_isolates[carbapenems() == "R", ]
|
|
#' example_isolates %>% filter(carbapenems() == "R")
|
|
#' example_isolates %>% filter(across(carbapenems(), ~.x == "R"))
|
|
#' }
|
|
ab_class <- function(ab_class,
|
|
only_rsi_columns = FALSE) {
|
|
ab_selector(ab_class, function_name = "ab_class", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
aminoglycosides <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("aminoglycoside", function_name = "aminoglycosides", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
betalactams <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("carbapenem|cephalosporin|penicillin", function_name = "betalactams", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
carbapenems <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("carbapenem", function_name = "carbapenems", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporin", function_name = "cephalosporins", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins_1st <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporins.*1", function_name = "cephalosporins_1st", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins_2nd <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporins.*2", function_name = "cephalosporins_2nd", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins_3rd <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporins.*3", function_name = "cephalosporins_3rd", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins_4th <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporins.*4", function_name = "cephalosporins_4th", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
cephalosporins_5th <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("cephalosporins.*5", function_name = "cephalosporins_5th", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
fluoroquinolones <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("fluoroquinolone", function_name = "fluoroquinolones", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
glycopeptides <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("glycopeptide", function_name = "glycopeptides", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
macrolides <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("macrolide", function_name = "macrolides", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
oxazolidinones <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("oxazolidinone", function_name = "oxazolidinones", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
penicillins <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("penicillin", function_name = "penicillins", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
#' @rdname antibiotic_class_selectors
|
|
#' @export
|
|
tetracyclines <- function(only_rsi_columns = FALSE) {
|
|
ab_selector("tetracycline", function_name = "tetracyclines", only_rsi_columns = only_rsi_columns)
|
|
}
|
|
|
|
ab_selector <- function(ab_class,
|
|
function_name,
|
|
only_rsi_columns) {
|
|
meet_criteria(ab_class, allow_class = "character", has_length = 1, .call_depth = 1)
|
|
meet_criteria(function_name, allow_class = "character", has_length = 1, .call_depth = 1)
|
|
meet_criteria(only_rsi_columns, allow_class = "logical", has_length = 1, .call_depth = 1)
|
|
|
|
if (current_R_older_than(3.2)) {
|
|
warning_("antibiotic class selectors such as ", function_name,
|
|
"() require R version 3.2 or later - you have ", R.version.string,
|
|
call = FALSE)
|
|
return(NULL)
|
|
}
|
|
|
|
vars_df <- get_current_data(arg_name = NA, call = -3)
|
|
|
|
# improve speed here so it will only run once when e.g. in one select call
|
|
if (!identical(pkg_env$ab_selector, unique_call_id())) {
|
|
ab_in_data <- get_column_abx(vars_df, info = FALSE, only_rsi_columns = only_rsi_columns, sort = FALSE)
|
|
pkg_env$ab_selector <- unique_call_id()
|
|
pkg_env$ab_selector_cols <- ab_in_data
|
|
} else {
|
|
ab_in_data <- pkg_env$ab_selector_cols
|
|
}
|
|
|
|
if (length(ab_in_data) == 0) {
|
|
message_("No antimicrobial agents found.")
|
|
return(NULL)
|
|
}
|
|
|
|
ab_reference <- subset(antibiotics,
|
|
group %like% ab_class |
|
|
atc_group1 %like% ab_class |
|
|
atc_group2 %like% ab_class)
|
|
ab_group <- find_ab_group(ab_class)
|
|
if (ab_group == "") {
|
|
ab_group <- paste0("'", ab_class, "'")
|
|
examples <- ""
|
|
} else {
|
|
examples <- paste0(" (such as ", find_ab_names(ab_class, 2), ")")
|
|
}
|
|
# get the columns with a group names in the chosen ab class
|
|
agents <- ab_in_data[names(ab_in_data) %in% ab_reference$ab]
|
|
|
|
if (message_not_thrown_before(function_name)) {
|
|
if (length(agents) == 0) {
|
|
message_("No antimicrobial agents of class ", ab_group, " found", examples, ".")
|
|
} else {
|
|
agents_formatted <- paste0("'", font_bold(agents, collapse = NULL), "'")
|
|
agents_names <- ab_name(names(agents), tolower = TRUE, language = NULL)
|
|
need_name <- tolower(gsub("[^a-zA-Z]", "", agents)) != tolower(gsub("[^a-zA-Z]", "", agents_names))
|
|
agents_formatted[need_name] <- paste0(agents_formatted[need_name],
|
|
" (", agents_names[need_name], ")")
|
|
message_("For `", function_name, "(", ifelse(function_name == "ab_class", paste0("\"", ab_class, "\""), ""), ")` using ",
|
|
ifelse(length(agents) == 1, "column: ", "columns: "),
|
|
vector_and(agents_formatted, quotes = FALSE))
|
|
}
|
|
remember_thrown_message(function_name)
|
|
}
|
|
|
|
if (!is.null(attributes(vars_df)$type) &&
|
|
attributes(vars_df)$type %in% c("dplyr_cur_data_all", "base_R") &&
|
|
!any(as.character(sys.calls()) %like% paste0("(across|if_any|if_all)\\((c\\()?[a-z(), ]*", function_name))) {
|
|
structure(unname(agents),
|
|
class = c("ab_selector", "character"))
|
|
} else {
|
|
# don't return with "ab_selector" class if method is a dplyr selector,
|
|
# dplyr::select() will complain:
|
|
# > Subscript has the wrong type `ab_selector`.
|
|
# > It must be numeric or character.
|
|
unname(agents)
|
|
}
|
|
}
|
|
|
|
#' @method c ab_selector
|
|
#' @export
|
|
#' @noRd
|
|
c.ab_selector <- function(...) {
|
|
structure(unlist(lapply(list(...), as.character)),
|
|
class = c("ab_selector", "character"))
|
|
}
|
|
|
|
all_any_ab_selector <- function(type, ..., na.rm = TRUE) {
|
|
cols_ab <- c(...)
|
|
result <- cols_ab[toupper(cols_ab) %in% c("R", "S", "I")]
|
|
if (length(result) == 0) {
|
|
message_("Filtering ", type, " of columns ", vector_and(font_bold(cols_ab, collapse = NULL), quotes = "'"), ' to contain value "R", "S" or "I"')
|
|
result <- c("R", "S", "I")
|
|
}
|
|
cols_ab <- cols_ab[!cols_ab %in% result]
|
|
df <- get_current_data(arg_name = NA, call = -3)
|
|
|
|
if (type == "all") {
|
|
scope_fn <- all
|
|
} else {
|
|
scope_fn <- any
|
|
}
|
|
|
|
x_transposed <- as.list(as.data.frame(t(df[, cols_ab, drop = FALSE]), stringsAsFactors = FALSE))
|
|
vapply(FUN.VALUE = logical(1),
|
|
X = x_transposed,
|
|
FUN = function(y) scope_fn(y %in% result, na.rm = na.rm),
|
|
USE.NAMES = FALSE)
|
|
}
|
|
|
|
#' @method all ab_selector
|
|
#' @export
|
|
#' @noRd
|
|
all.ab_selector <- function(..., na.rm = FALSE) {
|
|
# this is all() for
|
|
all_any_ab_selector("all", ..., na.rm = na.rm)
|
|
}
|
|
|
|
#' @method any ab_selector
|
|
#' @export
|
|
#' @noRd
|
|
any.ab_selector <- function(..., na.rm = FALSE) {
|
|
all_any_ab_selector("any", ..., na.rm = na.rm)
|
|
}
|
|
|
|
|
|
#' @method all ab_selector_any_all
|
|
#' @export
|
|
#' @noRd
|
|
all.ab_selector_any_all <- function(..., na.rm = FALSE) {
|
|
# this is all() on a logical vector from `==.ab_selector` or `!=.ab_selector`
|
|
# e.g., example_isolates %>% filter(all(carbapenems() == "R"))
|
|
# so just return the vector as is, only correcting for na.rm
|
|
out <- unclass(c(...))
|
|
if (na.rm == TRUE) {
|
|
out <- out[!is.na(out)]
|
|
}
|
|
out
|
|
}
|
|
|
|
#' @method any ab_selector_any_all
|
|
#' @export
|
|
#' @noRd
|
|
any.ab_selector_any_all <- function(..., na.rm = FALSE) {
|
|
# this is any() on a logical vector from `==.ab_selector` or `!=.ab_selector`
|
|
# e.g., example_isolates %>% filter(any(carbapenems() == "R"))
|
|
# so just return the vector as is, only correcting for na.rm
|
|
out <- unclass(c(...))
|
|
if (na.rm == TRUE) {
|
|
out <- out[!is.na(out)]
|
|
}
|
|
out
|
|
}
|
|
|
|
#' @method == ab_selector
|
|
#' @export
|
|
#' @noRd
|
|
`==.ab_selector` <- function(e1, e2) {
|
|
calls <- as.character(match.call())
|
|
fn_name <- calls[2]
|
|
# keep only the ... in c(...)
|
|
fn_name <- gsub("^(c\\()(.*)(\\))$", "\\2", fn_name)
|
|
if (is_any(fn_name)) {
|
|
type <- "any"
|
|
} else if (is_all(fn_name)) {
|
|
type <- "all"
|
|
} else {
|
|
type <- "all"
|
|
if (length(e1) > 1) {
|
|
message_("Assuming a filter on ", type, " ", length(e1), " ", gsub("[\\(\\)]", "", fn_name),
|
|
". Wrap around `all()` or `any()` to prevent this note.")
|
|
}
|
|
}
|
|
structure(all_any_ab_selector(type = type, e1, e2),
|
|
class = c("ab_selector_any_all", "logical"))
|
|
}
|
|
|
|
#' @method != ab_selector
|
|
#' @export
|
|
#' @noRd
|
|
`!=.ab_selector` <- function(e1, e2) {
|
|
calls <- as.character(match.call())
|
|
fn_name <- calls[2]
|
|
# keep only the ... in c(...)
|
|
fn_name <- gsub("^(c\\()(.*)(\\))$", "\\2", fn_name)
|
|
if (is_any(fn_name)) {
|
|
type <- "any"
|
|
} else if (is_all(fn_name)) {
|
|
type <- "all"
|
|
} else {
|
|
type <- "all"
|
|
if (length(e1) > 1) {
|
|
message_("Assuming a filter on ", type, " ", length(e1), " ", gsub("[\\(\\)]", "", fn_name),
|
|
". Wrap around `all()` or `any()` to prevent this note.")
|
|
}
|
|
}
|
|
# this is `!=`, so turn around the values
|
|
rsi <- c("R", "S", "I")
|
|
e2 <- rsi[rsi != e2]
|
|
structure(all_any_ab_selector(type = type, e1, e2),
|
|
class = c("ab_selector_any_all", "logical"))
|
|
}
|
|
|
|
is_any <- function(el1) {
|
|
syscall <- paste0(trimws(deparse(sys.calls()[[1]])), collapse = " ")
|
|
el1 <- gsub("(.*),.*", "\\1", el1)
|
|
syscall %like% paste0("[^_a-zA-Z0-9]any\\(", "(c\\()?", el1)
|
|
}
|
|
is_all <- function(el1) {
|
|
syscall <- paste0(trimws(deparse(sys.calls()[[1]])), collapse = " ")
|
|
el1 <- gsub("(.*),.*", "\\1", el1)
|
|
syscall %like% paste0("[^_a-zA-Z0-9]all\\(", "(c\\()?", el1)
|
|
}
|
|
|
|
|
|
find_ab_group <- function(ab_class) {
|
|
ab_class[ab_class == "carbapenem|cephalosporin|penicillin"] <- "betalactam"
|
|
ab_class <- gsub("[^a-zA-Z0-9]", ".*", ab_class)
|
|
ifelse(ab_class %in% c("aminoglycoside",
|
|
"betalactam",
|
|
"carbapenem",
|
|
"cephalosporin",
|
|
"fluoroquinolone",
|
|
"glycopeptide",
|
|
"macrolide",
|
|
"oxazolidinone",
|
|
"tetracycline"),
|
|
paste0(ab_class, "s"),
|
|
antibiotics %pm>%
|
|
subset(group %like% ab_class |
|
|
atc_group1 %like% ab_class |
|
|
atc_group2 %like% ab_class) %pm>%
|
|
pm_pull(group) %pm>%
|
|
unique() %pm>%
|
|
tolower() %pm>%
|
|
sort() %pm>%
|
|
paste(collapse = "/")
|
|
)
|
|
}
|
|
|
|
find_ab_names <- function(ab_group, n = 3) {
|
|
ab_group <- gsub("[^a-zA-Z|0-9]", ".*", ab_group)
|
|
|
|
# try popular first, they have DDDs
|
|
drugs <- antibiotics[which((!is.na(antibiotics$iv_ddd) | !is.na(antibiotics$oral_ddd)) &
|
|
antibiotics$name %unlike% " " &
|
|
antibiotics$group %like% ab_group &
|
|
antibiotics$ab %unlike% "[0-9]$"), ]$name
|
|
if (length(drugs) < n) {
|
|
# now try it all
|
|
drugs <- antibiotics[which((antibiotics$group %like% ab_group |
|
|
antibiotics$atc_group1 %like% ab_group |
|
|
antibiotics$atc_group2 %like% ab_group) &
|
|
antibiotics$ab %unlike% "[0-9]$"), ]$name
|
|
}
|
|
vector_or(ab_name(sample(drugs, size = min(n, length(drugs)), replace = FALSE),
|
|
tolower = TRUE,
|
|
language = NULL),
|
|
quotes = FALSE)
|
|
}
|