AMR/R/eucast_rules.R

1095 lines
57 KiB
R
Executable File

# ==================================================================== #
# 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/ #
# ==================================================================== #
format_eucast_version_nr <- function(version, markdown = TRUE) {
# for documentation - adds title, version number, year and url in markdown language
lst <- c(EUCAST_VERSION_BREAKPOINTS, EUCAST_VERSION_EXPERT_RULES)
version <- format(unique(version), nsmall = 1)
txt <- character(0)
for (i in seq_len(length(version))) {
v <- version[i]
if (markdown == TRUE) {
txt <- c(txt, paste0("[", lst[[v]]$title, " ", lst[[v]]$version_txt, "](", lst[[v]]$url, ")",
" (", lst[[v]]$year, ")"))
} else {
txt <- c(txt, paste0(lst[[version]]$title, " ", lst[[v]]$version_txt,
" (", lst[[v]]$year, ")"))
}
}
vector_and(txt, quotes = FALSE)
}
#' Apply EUCAST Rules
#'
#' @description
#' Apply rules for clinical breakpoints and intrinsic resistance as defined by the European Committee on Antimicrobial Susceptibility Testing (EUCAST, <https://eucast.org>), see *Source*. Use [eucast_dosage()] to get a [data.frame] with advised dosages of a certain bug-drug combination, which is based on the [dosage] data set.
#'
#' To improve the interpretation of the antibiogram before EUCAST rules are applied, some non-EUCAST rules can applied at default, see *Details*.
#' @inheritSection lifecycle Stable Lifecycle
#' @param x data with antibiotic columns, such as `amox`, `AMX` and `AMC`
#' @param info a [logical] to indicate whether progress should be printed to the console, defaults to only print while in interactive sessions
#' @param rules a [character] vector that specifies which rules should be applied. Must be one or more of `"breakpoints"`, `"expert"`, `"other"`, `"custom"`, `"all"`, and defaults to `c("breakpoints", "expert")`. The default value can be set to another value, e.g. using `options(AMR_eucastrules = "all")`. If using `"custom"`, be sure to fill in argument `custom_rules` too. Custom rules can be created with [custom_eucast_rules()].
#' @param verbose a [logical] to turn Verbose mode on and off (default is off). In Verbose mode, the function does not apply rules to the data, but instead returns a data set in logbook form with extensive info about which rows and columns would be effected and in which way. Using Verbose mode takes a lot more time.
#' @param version_breakpoints the version number to use for the EUCAST Clinical Breakpoints guideline. Can be either `r vector_or(names(EUCAST_VERSION_BREAKPOINTS), reverse = TRUE)`.
#' @param version_expertrules the version number to use for the EUCAST Expert Rules and Intrinsic Resistance guideline. Can be either `r vector_or(names(EUCAST_VERSION_EXPERT_RULES), reverse = TRUE)`.
#' @param ampc_cephalosporin_resistance a [character] value that should be applied to cefotaxime, ceftriaxone and ceftazidime for AmpC de-repressed cephalosporin-resistant mutants, defaults to `NA`. Currently only works when `version_expertrules` is `3.2`; '*EUCAST Expert Rules v3.2 on Enterobacterales*' states that results of cefotaxime, ceftriaxone and ceftazidime should be reported with a note, or results should be suppressed (emptied) for these three agents. A value of `NA` (the default) for this argument will remove results for these three agents, while e.g. a value of `"R"` will make the results for these agents resistant. Use `NULL` or `FALSE` to not alter results for these three agents of AmpC de-repressed cephalosporin-resistant mutants. Using `TRUE` is equal to using `"R"`. \cr For *EUCAST Expert Rules* v3.2, this rule applies to: `r vector_and(gsub("[^a-zA-Z ]+", "", unlist(strsplit(eucast_rules_file[which(eucast_rules_file$reference.version == 3.2 & eucast_rules_file$reference.rule %like% "ampc"), "this_value"][1], "|", fixed = TRUE))), quotes = "*")`.
#' @param ... column name of an antibiotic, see section *Antibiotics* below
#' @param ab any (vector of) text that can be coerced to a valid antibiotic code with [as.ab()]
#' @param administration route of administration, either `r vector_or(dosage$administration)`
#' @param only_rsi_columns a [logical] to indicate whether only antibiotic columns must be detected that were transformed to class `<rsi>` (see [as.rsi()]) on beforehand (defaults to `FALSE`)
#' @param custom_rules custom rules to apply, created with [custom_eucast_rules()]
#' @inheritParams first_isolate
#' @details
#' **Note:** This function does not translate MIC values to RSI values. Use [as.rsi()] for that. \cr
#' **Note:** When ampicillin (AMP, J01CA01) is not available but amoxicillin (AMX, J01CA04) is, the latter will be used for all rules where there is a dependency on ampicillin. These drugs are interchangeable when it comes to expression of antimicrobial resistance. \cr
#'
#' The file containing all EUCAST rules is located here: <https://github.com/msberends/AMR/blob/master/data-raw/eucast_rules.tsv>. **Note:** Old taxonomic names are replaced with the current taxonomy where applicable. For example, *Ochrobactrum anthropi* was renamed to *Brucella anthropi* in 2020; the original EUCAST rules v3.1 and v3.2 did not yet contain this new taxonomic name. The file used as input for this `AMR` package contains the taxonomy updated until [`r CATALOGUE_OF_LIFE$yearmonth_LPSN`][catalogue_of_life()].
#'
#' ## Custom Rules
#'
#' Custom rules can be created using [custom_eucast_rules()], e.g.:
#'
#' ```
#' x <- custom_eucast_rules(AMC == "R" & genus == "Klebsiella" ~ aminopenicillins == "R",
#' AMC == "I" & genus == "Klebsiella" ~ aminopenicillins == "I")
#'
#' eucast_rules(example_isolates, rules = "custom", custom_rules = x)
#' ```
#'
#'
#' ## 'Other' Rules
#'
#' Before further processing, two non-EUCAST rules about drug combinations can be applied to improve the efficacy of the EUCAST rules, and the reliability of your data (analysis). These rules are:
#'
#' 1. A drug **with** enzyme inhibitor will be set to S if the same drug **without** enzyme inhibitor is S
#' 2. A drug **without** enzyme inhibitor will be set to R if the same drug **with** enzyme inhibitor is R
#'
#' Important examples include amoxicillin and amoxicillin/clavulanic acid, and trimethoprim and trimethoprim/sulfamethoxazole. Needless to say, for these rules to work, both drugs must be available in the data set.
#'
#' Since these rules are not officially approved by EUCAST, they are not applied at default. To use these rules, include `"other"` to the `rules` argument, or use `eucast_rules(..., rules = "all")`. You can also set the option `AMR_eucastrules`, i.e. run `options(AMR_eucastrules = "all")`.
#' @section Antibiotics:
#' To define antibiotics column names, leave as it is to determine it automatically with [guess_ab_col()] or input a text (case-insensitive), or use `NULL` to skip a column (e.g. `TIC = NULL` to skip ticarcillin). Manually defined but non-existing columns will be skipped with a warning.
#'
#' The following antibiotics are used for the functions [eucast_rules()] and [mdro()]. These are shown below in the format 'name (`antimicrobial ID`, [ATC code](https://www.whocc.no/atc/structure_and_principles/))', sorted alphabetically:
#'
#' `r create_ab_documentation(c("AMC", "AMK", "AMP", "AMX", "APL", "APX", "ATM", "AVB", "AVO", "AZD", "AZL", "AZM", "BAM", "BPR", "CAC", "CAT", "CAZ", "CCP", "CCV", "CCX", "CDC", "CDR", "CDZ", "CEC", "CED", "CEI", "CEM", "CEP", "CFM", "CFM1", "CFP", "CFR", "CFS", "CFZ", "CHE", "CHL", "CIC", "CID", "CIP", "CLI", "CLM", "CLO", "CLR", "CMX", "CMZ", "CND", "COL", "CPD", "CPI", "CPL", "CPM", "CPO", "CPR", "CPT", "CPX", "CRB", "CRD", "CRN", "CRO", "CSL", "CTB", "CTC", "CTF", "CTL", "CTS", "CTT", "CTX", "CTZ", "CXM", "CYC", "CZA", "CZD", "CZO", "CZP", "CZX", "DAL", "DAP", "DIC", "DIR", "DIT", "DIX", "DIZ", "DKB", "DOR", "DOX", "ENX", "EPC", "ERY", "ETP", "FEP", "FLC", "FLE", "FLR1", "FOS", "FOV", "FOX", "FOX1", "FUS", "GAT", "GEM", "GEN", "GRX", "HAP", "HET", "IPM", "ISE", "JOS", "KAN", "LEN", "LEX", "LIN", "LNZ", "LOM", "LOR", "LTM", "LVX", "MAN", "MCM", "MEC", "MEM", "MET", "MEV", "MEZ", "MFX", "MID", "MNO", "MTM", "NAC", "NAF", "NAL", "NEO", "NET", "NIT", "NOR", "NOV", "NVA", "OFX", "OLE", "ORI", "OXA", "PAZ", "PEF", "PEN", "PHE", "PHN", "PIP", "PLB", "PME", "PNM", "PRC", "PRI", "PRL", "PRP", "PRU", "PVM", "QDA", "RAM", "RFL", "RID", "RIF", "ROK", "RST", "RXT", "SAM", "SBC", "SDI", "SDM", "SIS", "SLF", "SLF1", "SLF10", "SLF11", "SLF12", "SLF13", "SLF2", "SLF3", "SLF4", "SLF5", "SLF6", "SLF7", "SLF8", "SLF9", "SLT1", "SLT2", "SLT3", "SLT4", "SLT5", "SLT6", "SMX", "SPI", "SPX", "SRX", "STR", "STR1", "SUD", "SUL", "SUT", "SXT", "SZO", "TAL", "TAZ", "TCC", "TCM", "TCY", "TEC", "TEM", "TGC", "THA", "TIC", "TIO", "TLT", "TLV", "TMP", "TMX", "TOB", "TRL", "TVA", "TZD", "TZP", "VAN"))`
#' @aliases EUCAST
#' @rdname eucast_rules
#' @export
#' @return The input of `x`, possibly with edited values of antibiotics. Or, if `verbose = TRUE`, a [data.frame] with all original and new values of the affected bug-drug combinations.
#' @source
#' - EUCAST Expert Rules. Version 2.0, 2012.\cr
#' Leclercq et al. **EUCAST expert rules in antimicrobial susceptibility testing.** *Clin Microbiol Infect.* 2013;19(2):141-60; \doi{https://doi.org/10.1111/j.1469-0691.2011.03703.x}
#' - EUCAST Expert Rules, Intrinsic Resistance and Exceptional Phenotypes Tables. Version 3.1, 2016. [(link)](https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Expert_Rules/Expert_rules_intrinsic_exceptional_V3.1.pdf)
#' - EUCAST Intrinsic Resistance and Unusual Phenotypes. Version 3.2, 2020. [(link)](https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Expert_Rules/2020/Intrinsic_Resistance_and_Unusual_Phenotypes_Tables_v3.2_20200225.pdf)
#' - EUCAST Breakpoint tables for interpretation of MICs and zone diameters. Version 9.0, 2019. [(link)](https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Breakpoint_tables/v_9.0_Breakpoint_Tables.xlsx)
#' - EUCAST Breakpoint tables for interpretation of MICs and zone diameters. Version 10.0, 2020. [(link)](https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Breakpoint_tables/v_10.0_Breakpoint_Tables.xlsx)
#' - EUCAST Breakpoint tables for interpretation of MICs and zone diameters. Version 11.0, 2021. [(link)](https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Breakpoint_tables/v_11.0_Breakpoint_Tables.xlsx)
#' @inheritSection AMR Reference Data Publicly Available
#' @inheritSection AMR Read more on Our Website!
#' @examples
#' \donttest{
#' a <- data.frame(mo = c("Staphylococcus aureus",
#' "Enterococcus faecalis",
#' "Escherichia coli",
#' "Klebsiella pneumoniae",
#' "Pseudomonas aeruginosa"),
#' VAN = "-", # Vancomycin
#' AMX = "-", # Amoxicillin
#' COL = "-", # Colistin
#' CAZ = "-", # Ceftazidime
#' CXM = "-", # Cefuroxime
#' PEN = "S", # Benzylpenicillin
#' FOX = "S", # Cefoxitin
#' stringsAsFactors = FALSE)
#'
#' a
#' # mo VAN AMX COL CAZ CXM PEN FOX
#' # 1 Staphylococcus aureus - - - - - S S
#' # 2 Enterococcus faecalis - - - - - S S
#' # 3 Escherichia coli - - - - - S S
#' # 4 Klebsiella pneumoniae - - - - - S S
#' # 5 Pseudomonas aeruginosa - - - - - S S
#'
#'
#' # apply EUCAST rules: some results wil be changed
#' b <- eucast_rules(a)
#'
#' b
#' # mo VAN AMX COL CAZ CXM PEN FOX
#' # 1 Staphylococcus aureus - S R R S S S
#' # 2 Enterococcus faecalis - - R R R S R
#' # 3 Escherichia coli R - - - - R S
#' # 4 Klebsiella pneumoniae R R - - - R S
#' # 5 Pseudomonas aeruginosa R R - - R R R
#'
#'
#' # do not apply EUCAST rules, but rather get a data.frame
#' # containing all details about the transformations:
#' c <- eucast_rules(a, verbose = TRUE)
#' }
#'
#' eucast_dosage(c("tobra", "genta", "cipro"), "iv")
eucast_rules <- function(x,
col_mo = NULL,
info = interactive(),
rules = getOption("AMR_eucastrules", default = c("breakpoints", "expert")),
verbose = FALSE,
version_breakpoints = 11.0,
version_expertrules = 3.2,
ampc_cephalosporin_resistance = NA,
only_rsi_columns = FALSE,
custom_rules = NULL,
...) {
meet_criteria(x, allow_class = "data.frame")
meet_criteria(col_mo, allow_class = "character", has_length = 1, is_in = colnames(x), allow_NULL = TRUE)
meet_criteria(info, allow_class = "logical", has_length = 1)
meet_criteria(rules, allow_class = "character", has_length = c(1, 2, 3, 4, 5), is_in = c("breakpoints", "expert", "other", "all", "custom"))
meet_criteria(verbose, allow_class = "logical", has_length = 1)
meet_criteria(version_breakpoints, allow_class = c("numeric", "integer"), has_length = 1, is_in = as.double(names(EUCAST_VERSION_BREAKPOINTS)))
meet_criteria(version_expertrules, allow_class = c("numeric", "integer"), has_length = 1, is_in = as.double(names(EUCAST_VERSION_EXPERT_RULES)))
meet_criteria(ampc_cephalosporin_resistance, allow_class = c("logical", "character", "rsi"), has_length = 1, allow_NA = TRUE, allow_NULL = TRUE)
meet_criteria(only_rsi_columns, allow_class = "logical", has_length = 1)
meet_criteria(custom_rules, allow_class = "custom_eucast_rules", allow_NULL = TRUE)
if ("custom" %in% rules & is.null(custom_rules)) {
warning_("No custom rules were set with the `custom_rules` argument",
call = FALSE,
immediate = TRUE)
rules <- rules[rules != "custom"]
if (length(rules) == 0) {
if (info == TRUE) {
message_("No other rules were set, returning original data", add_fn = font_red, as_note = FALSE)
}
return(x)
}
}
x_deparsed <- deparse(substitute(x))
if (length(x_deparsed) > 1 || any(x_deparsed %unlike% "[a-z]+")) {
x_deparsed <- "your_data"
}
check_dataset_integrity()
breakpoints_info <- EUCAST_VERSION_BREAKPOINTS[[which(as.double(names(EUCAST_VERSION_BREAKPOINTS)) == version_breakpoints)]]
expertrules_info <- EUCAST_VERSION_EXPERT_RULES[[which(as.double(names(EUCAST_VERSION_EXPERT_RULES)) == version_expertrules)]]
# support old setting (until AMR v1.3.0)
if (missing(rules) & !is.null(getOption("AMR.eucast_rules", default = NULL))) {
rules <- getOption("AMR.eucast_rules")
}
if (interactive() & verbose == TRUE & info == TRUE) {
txt <- paste0("WARNING: In Verbose mode, the eucast_rules() function does not apply rules to the data, but instead returns a data set in logbook form with extensive info about which rows and columns would be effected and in which way.",
"\n\nThis may overwrite your existing data if you use e.g.:",
"\ndata <- eucast_rules(data, verbose = TRUE)\n\nDo you want to continue?")
showQuestion <- import_fn("showQuestion", "rstudioapi", error_on_fail = FALSE)
if (!is.null(showQuestion)) {
q_continue <- showQuestion("Using verbose = TRUE with eucast_rules()", txt)
} else {
q_continue <- utils::menu(choices = c("OK", "Cancel"), graphics = FALSE, title = txt)
}
if (q_continue %in% c(FALSE, 2)) {
message_("Cancelled, returning original data", add_fn = font_red, as_note = FALSE)
return(x)
}
}
# try to find columns based on type
# -- mo
if (is.null(col_mo)) {
col_mo <- search_type_in_df(x = x, type = "mo", info = info)
stop_if(is.null(col_mo), "`col_mo` must be set")
}
decimal.mark <- getOption("OutDec")
big.mark <- ifelse(decimal.mark != ",", ",", ".")
formatnr <- function(x, big = big.mark, dec = decimal.mark) {
trimws(format(x, big.mark = big, decimal.mark = dec))
}
warned <- FALSE
warn_lacking_rsi_class <- character(0)
txt_ok <- function(n_added, n_changed, warned = FALSE) {
if (warned == FALSE) {
if (n_added + n_changed == 0) {
cat(font_subtle(" (no changes)\n"))
} else {
# opening
if (n_added > 0 & n_changed == 0) {
cat(font_green(" ("))
} else if (n_added == 0 & n_changed > 0) {
cat(font_blue(" ("))
} else {
cat(font_grey(" ("))
}
# additions
if (n_added > 0) {
if (n_added == 1) {
cat(font_green("1 value added"))
} else {
cat(font_green(formatnr(n_added), "values added"))
}
}
# separator
if (n_added > 0 & n_changed > 0) {
cat(font_grey(", "))
}
# changes
if (n_changed > 0) {
if (n_changed == 1) {
cat(font_blue("1 value changed"))
} else {
cat(font_blue(formatnr(n_changed), "values changed"))
}
}
# closing
if (n_added > 0 & n_changed == 0) {
cat(font_green(")\n"))
} else if (n_added == 0 & n_changed > 0) {
cat(font_blue(")\n"))
} else {
cat(font_grey(")\n"))
}
}
warned <<- FALSE
}
}
cols_ab <- get_column_abx(x = x,
soft_dependencies = c("AMC",
"AMP",
"AMX",
"CIP",
"ERY",
"FOX1",
"GEN",
"MFX",
"NAL",
"NOR",
"PEN",
"PIP",
"TCY",
"TIC",
"TOB"),
hard_dependencies = NULL,
verbose = verbose,
info = info,
only_rsi_columns = only_rsi_columns,
...)
if (!"AMP" %in% names(cols_ab) & "AMX" %in% names(cols_ab)) {
# ampicillin column is missing, but amoxicillin is available
if (info == TRUE) {
message_("Using column '", cols_ab[names(cols_ab) == "AMX"], "' as input for ampicillin since many EUCAST rules depend on it.")
}
cols_ab <- c(cols_ab, c(AMP = unname(cols_ab[names(cols_ab) == "AMX"])))
}
# data preparation ----
if (info == TRUE & NROW(x) > 10000) {
message_("Preparing data...", appendLF = FALSE, as_note = FALSE)
}
# Some helper functions ---------------------------------------------------
get_antibiotic_columns <- function(x, cols_ab) {
x <- strsplit(x, ", *")[[1]]
x_new <- character()
for (val in x) {
if (toupper(val) %in% ls(envir = asNamespace("AMR"))) {
# antibiotic group names, as defined in data-raw/_internals.R, such as `CARBAPENEMS`
val <- eval(parse(text = toupper(val)), envir = asNamespace("AMR"))
} else if (toupper(val) %in% AB_lookup$ab) {
# separate drugs, such as `AMX`
val <- as.ab(val)
} else {
stop_("antimicrobial agent (group) not found in EUCAST rules file: ", val, call = FALSE)
}
x_new <- c(x_new, val)
}
cols_ab[match(x_new, names(cols_ab))]
}
get_antibiotic_names <- function(x) {
x <- x %pm>%
strsplit(",") %pm>%
unlist() %pm>%
trimws() %pm>%
vapply(FUN.VALUE = character(1), function(x) if (x %in% antibiotics$ab) ab_name(x, language = NULL, tolower = TRUE, fast_mode = TRUE) else x) %pm>%
sort() %pm>%
paste(collapse = ", ")
x <- gsub("_", " ", x, fixed = TRUE)
x <- gsub("except CAZ", paste("except", ab_name("CAZ", language = NULL, tolower = TRUE)), x, fixed = TRUE)
x <- gsub("except TGC", paste("except", ab_name("TGC", language = NULL, tolower = TRUE)), x, fixed = TRUE)
x <- gsub("cephalosporins (1st|2nd|3rd|4th|5th)", "cephalosporins (\\1 gen.)", x)
x
}
format_antibiotic_names <- function(ab_names, ab_results) {
ab_names <- trimws(unlist(strsplit(ab_names, ",")))
ab_results <- trimws(unlist(strsplit(ab_results, ",")))
if (length(ab_results) == 1) {
if (length(ab_names) == 1) {
# like FOX S
x <- paste(ab_names, "is")
} else if (length(ab_names) == 2) {
# like PEN,FOX S
x <- paste(paste0(ab_names, collapse = " and "), "are both")
} else {
# like PEN,FOX,GEN S (although dependency on > 2 ABx does not exist at the moment)
# nolint start
# x <- paste(paste0(ab_names, collapse = " and "), "are all")
# nolint end
}
return(paste0(x, " '", ab_results, "'"))
} else {
if (length(ab_names) == 2) {
# like PEN,FOX S,R
paste0(ab_names[1], " is '", ab_results[1], "' and ",
ab_names[2], " is '", ab_results[2], "'")
} else {
# like PEN,FOX,GEN S,R,R (although dependency on > 2 ABx does not exist at the moment)
paste0(ab_names[1], " is '", ab_results[1], "' and ",
ab_names[2], " is '", ab_results[2], "' and ",
ab_names[3], " is '", ab_results[3], "'")
}
}
}
as.rsi_no_warning <- function(x) {
if (is.rsi(x)) {
return(x)
}
suppressWarnings(as.rsi(x))
}
# Preparing the data ------------------------------------------------------
verbose_info <- data.frame(rowid = character(0),
col = character(0),
mo_fullname = character(0),
old = as.rsi(character(0)),
new = as.rsi(character(0)),
rule = character(0),
rule_group = character(0),
rule_name = character(0),
rule_source = character(0),
stringsAsFactors = FALSE)
old_cols <- colnames(x)
old_attributes <- attributes(x)
x <- as.data.frame(x, stringsAsFactors = FALSE) # no tibbles, data.tables, etc.
rownames(x) <- NULL # will later be restored with old_attributes
# create unique row IDs - combination of the MO and all ABx columns (so they will only run once per unique combination)
x$`.rowid` <- vapply(FUN.VALUE = character(1),
as.list(as.data.frame(t(x[, c(col_mo, cols_ab), drop = FALSE]),
stringsAsFactors = FALSE)),
function(x) {
x[is.na(x)] <- "."
paste0(x, collapse = "")
})
# save original table, with the new .rowid column
x.bak <- x
# keep only unique rows for MO and ABx
x <- x %pm>%
pm_arrange(`.rowid`) %pm>%
# big speed gain! only analyse unique rows:
pm_distinct(`.rowid`, .keep_all = TRUE) %pm>%
as.data.frame(stringsAsFactors = FALSE)
x[, col_mo] <- as.mo(as.character(x[, col_mo, drop = TRUE]))
# rename col_mo to prevent interference with joined columns
colnames(x)[colnames(x) == col_mo] <- ".col_mo"
col_mo <- ".col_mo"
# join to microorganisms data set
x <- left_join_microorganisms(x, by = col_mo, suffix = c("_oldcols", ""))
x$gramstain <- mo_gramstain(x[, col_mo, drop = TRUE], language = NULL)
x$genus_species <- trimws(paste(x$genus, x$species))
if (info == TRUE & NROW(x) > 10000) {
message_(" OK.", add_fn = list(font_green, font_bold), as_note = FALSE)
}
if (any(x$genus == "Staphylococcus", na.rm = TRUE)) {
all_staph <- MO_lookup[which(MO_lookup$genus == "Staphylococcus"), ]
all_staph$CNS_CPS <- suppressWarnings(mo_name(all_staph$mo, Becker = "all", language = NULL))
}
if (any(x$genus == "Streptococcus", na.rm = TRUE)) {
all_strep <- MO_lookup[which(MO_lookup$genus == "Streptococcus"), ]
all_strep$Lancefield <- suppressWarnings(mo_name(all_strep$mo, Lancefield = TRUE, language = NULL))
}
n_added <- 0
n_changed <- 0
# Other rules: enzyme inhibitors ------------------------------------------
if (any(c("all", "other") %in% rules)) {
if (info == TRUE) {
cat("\n")
cat(word_wrap(
font_bold(paste0("Rules by this AMR package (",
font_red(paste0("v", utils::packageDescription("AMR")$Version, ", ",
format(as.Date(utils::packageDescription("AMR")$Date), format = "%Y"))), "), see ?eucast_rules\n"))))
}
ab_enzyme <- subset(antibiotics, name %like% "/")[, c("ab", "name")]
colnames(ab_enzyme) <- c("enzyme_ab", "enzyme_name")
ab_enzyme$base_name <- gsub("^([a-zA-Z0-9]+).*", "\\1", ab_enzyme$enzyme_name)
ab_enzyme$base_ab <- antibiotics[match(ab_enzyme$base_name, antibiotics$name), "ab", drop = TRUE]
ab_enzyme <- subset(ab_enzyme, !is.na(base_ab))
# make ampicillin and amoxicillin interchangable
ampi <- subset(ab_enzyme, base_ab == "AMX")
ampi$base_ab <- "AMP"
ampi$base_name <- ab_name("AMP", language = NULL)
amox <- subset(ab_enzyme, base_ab == "AMP")
amox$base_ab <- "AMX"
amox$base_name <- ab_name("AMX", language = NULL)
# merge and sort
ab_enzyme <- rbind(ab_enzyme, ampi, amox)
ab_enzyme <- ab_enzyme[order(ab_enzyme$enzyme_name), ]
for (i in seq_len(nrow(ab_enzyme))) {
# check if both base and base + enzyme inhibitor are part of the data set
if (all(c(ab_enzyme$base_ab[i], ab_enzyme$enzyme_ab[i]) %in% names(cols_ab), na.rm = TRUE)) {
col_base <- unname(cols_ab[ab_enzyme$base_ab[i]])
col_enzyme <- unname(cols_ab[ab_enzyme$enzyme_ab[i]])
# Set base to R where base + enzyme inhibitor is R ----
rule_current <- paste0(ab_enzyme$base_name[i], " ('", font_bold(col_base), "') = R if ",
tolower(ab_enzyme$enzyme_name[i]), " ('", font_bold(col_enzyme), "') = R")
if (info == TRUE) {
cat(word_wrap(rule_current,
width = getOption("width") - 30,
extra_indent = 6))
}
run_changes <- edit_rsi(x = x,
to = "R",
rule = c(rule_current, "Other rules", "",
paste0("Non-EUCAST: AMR package v", utils::packageDescription("AMR")$Version)),
rows = which(as.rsi_no_warning(x[, col_enzyme, drop = TRUE]) == "R"),
cols = col_base,
last_verbose_info = verbose_info,
original_data = x.bak,
warned = warned,
info = info,
verbose = verbose)
n_added <- n_added + run_changes$added
n_changed <- n_changed + run_changes$changed
verbose_info <- run_changes$verbose_info
x <- run_changes$output
warn_lacking_rsi_class <- c(warn_lacking_rsi_class, run_changes$rsi_warn)
# Print number of new changes
if (info == TRUE) {
# print only on last one of rules in this group
txt_ok(n_added = n_added, n_changed = n_changed, warned = warned)
# and reset counters
n_added <- 0
n_changed <- 0
}
# Set base + enzyme inhibitor to S where base is S ----
rule_current <- paste0(ab_enzyme$enzyme_name[i], " ('", font_bold(col_enzyme), "') = S if ",
tolower(ab_enzyme$base_name[i]), " ('", font_bold(col_base), "') = S")
if (info == TRUE) {
cat(word_wrap(rule_current,
width = getOption("width") - 30,
extra_indent = 6))
}
run_changes <- edit_rsi(x = x,
to = "S",
rule = c(rule_current, "Other rules", "",
paste0("Non-EUCAST: AMR package v", utils::packageDescription("AMR")$Version)),
rows = which(as.rsi_no_warning(x[, col_base, drop = TRUE]) == "S"),
cols = col_enzyme,
last_verbose_info = verbose_info,
original_data = x.bak,
warned = warned,
info = info,
verbose = verbose)
n_added <- n_added + run_changes$added
n_changed <- n_changed + run_changes$changed
verbose_info <- run_changes$verbose_info
x <- run_changes$output
warn_lacking_rsi_class <- c(warn_lacking_rsi_class, run_changes$rsi_warn)
# Print number of new changes
if (info == TRUE) {
# print only on last one of rules in this group
txt_ok(n_added = n_added, n_changed = n_changed, warned = warned)
# and reset counters
n_added <- 0
n_changed <- 0
}
}
}
} else {
if (info == TRUE) {
cat("\n")
message_("Skipping inheritance rules defined by this AMR package, such as setting trimethoprim (TMP) = R where trimethoprim/sulfamethoxazole (SXT) = R. Add \"other\" or \"all\" to the `rules` argument to apply those rules.")
}
}
if (!any(c("all", "custom") %in% rules) & !is.null(custom_rules)) {
if (info == TRUE) {
message_("Skipping custom EUCAST rules, since the `rules` argument does not contain \"custom\".")
}
custom_rules <- NULL
}
# Official EUCAST rules ---------------------------------------------------
eucast_notification_shown <- FALSE
if (!is.null(list(...)$eucast_rules_df)) {
# this allows: eucast_rules(x, eucast_rules_df = AMR:::eucast_rules_file %>% filter(is.na(have_these_values)))
eucast_rules_df <- list(...)$eucast_rules_df
} else {
# otherwise internal data file, created in data-raw/_internals.R
eucast_rules_df <- eucast_rules_file
}
# filter on user-set guideline versions ----
if (any(c("all", "breakpoints") %in% rules)) {
eucast_rules_df <- subset(eucast_rules_df,
reference.rule_group %unlike% "breakpoint" |
(reference.rule_group %like% "breakpoint" & reference.version == version_breakpoints))
}
if (any(c("all", "expert") %in% rules)) {
eucast_rules_df <- subset(eucast_rules_df,
reference.rule_group %unlike% "expert" |
(reference.rule_group %like% "expert" & reference.version == version_expertrules))
}
# filter out AmpC de-repressed cephalosporin-resistant mutants ----
# cefotaxime, ceftriaxone, ceftazidime
if (is.null(ampc_cephalosporin_resistance) || isFALSE(ampc_cephalosporin_resistance)) {
eucast_rules_df <- subset(eucast_rules_df,
reference.rule %unlike% "ampc")
} else {
if (isTRUE(ampc_cephalosporin_resistance)) {
ampc_cephalosporin_resistance <- "R"
}
eucast_rules_df[which(eucast_rules_df$reference.rule %like% "ampc"), "to_value"] <- as.character(ampc_cephalosporin_resistance)
}
# Go over all rules and apply them ----
for (i in seq_len(nrow(eucast_rules_df))) {
rule_previous <- eucast_rules_df[max(1, i - 1), "reference.rule", drop = TRUE]
rule_current <- eucast_rules_df[i, "reference.rule", drop = TRUE]
rule_next <- eucast_rules_df[min(nrow(eucast_rules_df), i + 1), "reference.rule", drop = TRUE]
rule_group_previous <- eucast_rules_df[max(1, i - 1), "reference.rule_group", drop = TRUE]
rule_group_current <- eucast_rules_df[i, "reference.rule_group", drop = TRUE]
# don't apply rules if user doesn't want to apply them
if (rule_group_current %like% "breakpoint" & !any(c("all", "breakpoints") %in% rules)) {
next
}
if (rule_group_current %like% "expert" & !any(c("all", "expert") %in% rules)) {
next
}
if (isFALSE(info) | isFALSE(verbose)) {
rule_text <- ""
} else {
if (is.na(eucast_rules_df[i, "and_these_antibiotics", drop = TRUE])) {
rule_text <- paste0("always report as '", eucast_rules_df[i, "to_value", drop = TRUE], "': ", get_antibiotic_names(eucast_rules_df[i, "then_change_these_antibiotics", drop = TRUE]))
} else {
rule_text <- paste0("report as '", eucast_rules_df[i, "to_value", drop = TRUE], "' when ",
format_antibiotic_names(ab_names = get_antibiotic_names(eucast_rules_df[i, "and_these_antibiotics", drop = TRUE]),
ab_results = eucast_rules_df[i, "have_these_values", drop = TRUE]), ": ",
get_antibiotic_names(eucast_rules_df[i, "then_change_these_antibiotics", drop = TRUE]))
}
}
if (i == 1) {
rule_previous <- ""
rule_group_previous <- ""
}
if (i == nrow(eucast_rules_df)) {
rule_next <- ""
}
if (info == TRUE) {
# Print EUCAST intro ------------------------------------------------------
if (rule_group_current %unlike% "other" & eucast_notification_shown == FALSE) {
cat(
paste0("\n", font_grey(strrep("-", 0.95 * options()$width)), "\n",
word_wrap("Rules by the ", font_bold("European Committee on Antimicrobial Susceptibility Testing (EUCAST)")), "\n",
font_blue("https://eucast.org/"), "\n"))
eucast_notification_shown <- TRUE
}
# Print rule (group) ------------------------------------------------------
if (rule_group_current != rule_group_previous) {
# is new rule group, one of Breakpoints, Expert Rules and Other
cat(font_bold(
ifelse(
rule_group_current %like% "breakpoint",
paste0("\n",
word_wrap(
breakpoints_info$title, " (",
font_red(paste0(breakpoints_info$version_txt, ", ", breakpoints_info$year)), ")\n")),
ifelse(
rule_group_current %like% "expert",
paste0("\n",
word_wrap(
expertrules_info$title, " (",
font_red(paste0(expertrules_info$version_txt, ", ", expertrules_info$year)), ")\n")),
""))), "\n")
}
# Print rule -------------------------------------------------------------
if (rule_current != rule_previous) {
# is new rule within group, print its name
cat(italicise_taxonomy(word_wrap(rule_current,
width = getOption("width") - 30,
extra_indent = 6),
type = "ansi"))
warned <- FALSE
}
}
# Get rule from file ------------------------------------------------------
if_mo_property <- trimws(eucast_rules_df[i, "if_mo_property", drop = TRUE])
like_is_one_of <- trimws(eucast_rules_df[i, "like.is.one_of", drop = TRUE])
mo_value <- trimws(eucast_rules_df[i, "this_value", drop = TRUE])
# be sure to comprise all coagulase-negative/-positive staphylococci when they are mentioned
if (mo_value %like% "coagulase" && any(x$genus == "Staphylococcus", na.rm = TRUE)) {
if (mo_value %like% "negative") {
eucast_rules_df[i, "this_value"] <- paste0("^(", paste0(all_staph[which(all_staph$CNS_CPS %like% "negative"),
"fullname",
drop = TRUE],
collapse = "|"),
")$")
} else {
eucast_rules_df[i, "this_value"] <- paste0("^(", paste0(all_staph[which(all_staph$CNS_CPS %like% "positive"),
"fullname",
drop = TRUE],
collapse = "|"),
")$")
}
like_is_one_of <- "like"
}
# be sure to comprise all beta-haemolytic Streptococci (Lancefield groups A, B, C and G) when they are mentioned
if (mo_value %like% "group [ABCG]" && any(x$genus == "Streptococcus", na.rm = TRUE)) {
eucast_rules_df[i, "this_value"] <- paste0("^(", paste0(all_strep[which(all_strep$Lancefield %like% "group [ABCG]"),
"fullname",
drop = TRUE],
collapse = "|"),
")$")
like_is_one_of <- "like"
}
if (like_is_one_of == "is") {
# so e.g. 'Enterococcus' will turn into '^Enterococcus$'
mo_value <- paste0("^", mo_value, "$")
} else if (like_is_one_of == "one_of") {
# so 'Clostridium, Actinomyces, ...' will turn into '^(Clostridium|Actinomyces|...)$'
mo_value <- paste0("^(",
paste(trimws(unlist(strsplit(mo_value, ",", fixed = TRUE))),
collapse = "|"),
")$")
} else if (like_is_one_of != "like") {
stop("invalid value for column 'like.is.one_of'", call. = FALSE)
}
source_antibiotics <- eucast_rules_df[i, "and_these_antibiotics", drop = TRUE]
source_value <- trimws(unlist(strsplit(eucast_rules_df[i, "have_these_values", drop = TRUE], ",", fixed = TRUE)))
target_antibiotics <- eucast_rules_df[i, "then_change_these_antibiotics", drop = TRUE]
target_value <- eucast_rules_df[i, "to_value", drop = TRUE]
if (is.na(source_antibiotics)) {
rows <- tryCatch(which(x[, if_mo_property, drop = TRUE] %like% mo_value),
error = function(e) integer(0))
} else {
source_antibiotics <- get_antibiotic_columns(source_antibiotics, cols_ab)
if (length(source_value) == 1 & length(source_antibiotics) > 1) {
source_value <- rep(source_value, length(source_antibiotics))
}
if (length(source_antibiotics) == 0) {
rows <- integer(0)
} else if (length(source_antibiotics) == 1) {
rows <- tryCatch(which(x[, if_mo_property, drop = TRUE] %like% mo_value
& as.rsi_no_warning(x[, source_antibiotics[1L]]) == source_value[1L]),
error = function(e) integer(0))
} else if (length(source_antibiotics) == 2) {
rows <- tryCatch(which(x[, if_mo_property, drop = TRUE] %like% mo_value
& as.rsi_no_warning(x[, source_antibiotics[1L]]) == source_value[1L]
& as.rsi_no_warning(x[, source_antibiotics[2L]]) == source_value[2L]),
error = function(e) integer(0))
# nolint start
# } else if (length(source_antibiotics) == 3) {
# rows <- tryCatch(which(x[, if_mo_property, drop = TRUE] %like% mo_value
# & as.rsi_no_warning(x[, source_antibiotics[1L]]) == source_value[1L]
# & as.rsi_no_warning(x[, source_antibiotics[2L]]) == source_value[2L]
# & as.rsi_no_warning(x[, source_antibiotics[3L]]) == source_value[3L]),
# error = function(e) integer(0))
# nolint end
} else {
stop_("only 2 antibiotics supported for source_antibiotics")
}
}
cols <- get_antibiotic_columns(target_antibiotics, cols_ab)
# Apply rule on data ------------------------------------------------------
# this will return the unique number of changes
run_changes <- edit_rsi(x = x,
to = target_value,
rule = c(rule_text, rule_group_current, rule_current,
ifelse(rule_group_current %like% "breakpoint",
paste0(breakpoints_info$title, " ", breakpoints_info$version_txt, ", ", breakpoints_info$year),
paste0(expertrules_info$title, " ", expertrules_info$version_txt, ", ", expertrules_info$year))),
rows = rows,
cols = cols,
last_verbose_info = verbose_info,
original_data = x.bak,
warned = warned,
info = info,
verbose = verbose)
n_added <- n_added + run_changes$added
n_changed <- n_changed + run_changes$changed
verbose_info <- run_changes$verbose_info
x <- run_changes$output
warn_lacking_rsi_class <- c(warn_lacking_rsi_class, run_changes$rsi_warn)
# Print number of new changes ---------------------------------------------
if (info == TRUE & rule_next != rule_current) {
# print only on last one of rules in this group
txt_ok(n_added = n_added, n_changed = n_changed, warned = warned)
# and reset counters
n_added <- 0
n_changed <- 0
}
} # end of going over all rules
# Apply custom rules ----
if (!is.null(custom_rules)) {
if (info == TRUE) {
cat("\n")
cat(font_bold("Custom EUCAST rules, set by user"), "\n")
}
for (i in seq_len(length(custom_rules))) {
rule <- custom_rules[[i]]
rows <- which(eval(parse(text = rule$query), envir = x))
cols <- as.character(rule$result_group)
cols <- c(cols[cols %in% colnames(x)], # direct column names
unname(cols_ab[names(cols_ab) %in% cols])) # based on previous cols_ab finding
cols <- unique(cols)
target_value <- as.character(rule$result_value)
rule_text <- paste0("report as '", target_value, "' when ",
format_custom_query_rule(rule$query, colours = FALSE), ": ",
get_antibiotic_names(cols))
if (info == TRUE) {
# print rule
cat(italicise_taxonomy(word_wrap(format_custom_query_rule(rule$query, colours = FALSE),
width = getOption("width") - 30,
extra_indent = 6),
type = "ansi"))
warned <- FALSE
}
run_changes <- edit_rsi(x = x,
to = target_value,
rule = c(rule_text,
"Custom EUCAST rules",
paste0("Custom EUCAST rule ", i),
paste0("Object '", deparse(substitute(custom_rules)),
"' consisting of ", length(custom_rules), " custom rules")),
rows = rows,
cols = cols,
last_verbose_info = verbose_info,
original_data = x.bak,
warned = warned,
info = info,
verbose = verbose)
n_added <- n_added + run_changes$added
n_changed <- n_changed + run_changes$changed
verbose_info <- run_changes$verbose_info
x <- run_changes$output
warn_lacking_rsi_class <- c(warn_lacking_rsi_class, run_changes$rsi_warn)
# Print number of new changes ---------------------------------------------
if (info == TRUE & rule_next != rule_current) {
# print only on last one of rules in this group
txt_ok(n_added = n_added, n_changed = n_changed, warned = warned)
# and reset counters
n_added <- 0
n_changed <- 0
}
}
}
# Print overview ----------------------------------------------------------
if (info == TRUE | verbose == TRUE) {
verbose_info <- x.bak %pm>%
pm_mutate(row = pm_row_number()) %pm>%
pm_select(`.rowid`, row) %pm>%
pm_right_join(verbose_info,
by = c(".rowid" = "rowid")) %pm>%
pm_select(-`.rowid`) %pm>%
pm_select(row, pm_everything()) %pm>%
pm_filter(!is.na(new) | is.na(new) & !is.na(old)) %pm>%
pm_arrange(row, rule_group, rule_name, col)
rownames(verbose_info) <- NULL
}
if (info == TRUE) {
if (verbose == TRUE) {
wouldve <- "would have "
} else {
wouldve <- ""
}
cat(paste0("\n", font_grey(strrep("-", 0.95 * options()$width)), "\n"))
cat(word_wrap(paste0("The rules ", paste0(wouldve, "affected "),
font_bold(formatnr(pm_n_distinct(verbose_info$row)),
"out of", formatnr(nrow(x.bak)),
"rows"),
", making a total of ",
font_bold(formatnr(nrow(verbose_info)), "edits\n"))))
total_n_added <- verbose_info %pm>% pm_filter(is.na(old)) %pm>% nrow()
total_n_changed <- verbose_info %pm>% pm_filter(!is.na(old)) %pm>% nrow()
# print added values
if (total_n_added == 0) {
colour <- cat # is function
} else {
colour <- font_green # is function
}
cat(colour(paste0("=> ", wouldve, "added ",
font_bold(formatnr(verbose_info %pm>%
pm_filter(is.na(old)) %pm>%
nrow()), "test results"),
"\n")))
if (total_n_added > 0) {
added_summary <- verbose_info %pm>%
pm_filter(is.na(old)) %pm>%
pm_count(new, name = "n")
cat(paste(" -",
paste0(formatnr(added_summary$n), " test result", ifelse(added_summary$n > 1, "s", ""),
" added as ", paste0('"', added_summary$new, '"')), collapse = "\n"))
}
# print changed values
if (total_n_changed == 0) {
colour <- cat # is function
} else {
colour <- font_blue # is function
}
if (total_n_added + total_n_changed > 0) {
cat("\n")
}
cat(colour(paste0("=> ", wouldve, "changed ",
font_bold(formatnr(verbose_info %pm>%
pm_filter(!is.na(old)) %pm>%
nrow()), "test results"),
"\n")))
if (total_n_changed > 0) {
changed_summary <- verbose_info %pm>%
pm_filter(!is.na(old)) %pm>%
pm_mutate(new = ifelse(is.na(new), "NA", new)) %pm>%
pm_count(old, new, name = "n")
cat(paste(" -",
paste0(formatnr(changed_summary$n), " test result", ifelse(changed_summary$n > 1, "s", ""), " changed from ",
paste0('"', changed_summary$old, '"'), " to ", paste0('"', changed_summary$new, '"')), collapse = "\n"))
cat("\n")
}
cat(paste0(font_grey(strrep("-", 0.95 * options()$width)), "\n"))
if (verbose == FALSE & total_n_added + total_n_changed > 0) {
cat("\n", word_wrap("Use ", font_bold("eucast_rules(..., verbose = TRUE)"), " (on your original data) to get a data.frame with all specified edits instead."), "\n\n", sep = "")
} else if (verbose == TRUE) {
cat("\n", word_wrap("Used 'Verbose mode' (", font_bold("verbose = TRUE"), "), which returns a data.frame with all specified edits.\nUse ", font_bold("verbose = FALSE"), " to apply the rules on your data."), "\n\n", sep = "")
}
}
if (length(warn_lacking_rsi_class) > 0) {
warn_lacking_rsi_class <- unique(warn_lacking_rsi_class)
# take order from original data set
warn_lacking_rsi_class <- warn_lacking_rsi_class[order(colnames(x.bak))]
warn_lacking_rsi_class <- warn_lacking_rsi_class[!is.na(warn_lacking_rsi_class)]
warning_("Not all columns with antimicrobial results are of class <rsi>. Transform them on beforehand, with e.g.:\n",
" - ", x_deparsed, " %>% as.rsi(", ifelse(length(warn_lacking_rsi_class) == 1,
warn_lacking_rsi_class,
paste0(warn_lacking_rsi_class[1], ":", warn_lacking_rsi_class[length(warn_lacking_rsi_class)])), ")\n",
" - ", x_deparsed, " %>% mutate_if(is.rsi.eligible, as.rsi)\n",
" - ", x_deparsed, " %>% mutate(across(where(is.rsi.eligible), as.rsi))",
call = FALSE)
}
# Return data set ---------------------------------------------------------
if (verbose == TRUE) {
verbose_info
} else {
# x was analysed with only unique rows, so join everything together again
x <- x[, c(cols_ab, ".rowid"), drop = FALSE]
x.bak <- x.bak[, setdiff(colnames(x.bak), cols_ab), drop = FALSE]
x.bak <- x.bak %pm>%
pm_left_join(x, by = ".rowid")
x.bak <- x.bak[, old_cols, drop = FALSE]
# reset original attributes
attributes(x.bak) <- old_attributes
x.bak
}
}
# helper function for editing the table ----
edit_rsi <- function(x,
to,
rule,
rows,
cols,
last_verbose_info,
original_data,
warned,
info,
verbose) {
cols <- unique(cols[!is.na(cols) & !is.null(cols)])
# for Verbose Mode, keep track of all changes and return them
track_changes <- list(added = 0,
changed = 0,
output = x,
verbose_info = last_verbose_info,
rsi_warn = character(0))
txt_error <- function() {
if (info == TRUE) cat("", font_red_bg(font_white(" ERROR ")), "\n\n")
}
txt_warning <- function() {
if (warned == FALSE) {
if (info == TRUE) cat(" ", font_rsi_I_bg(" WARNING "), sep = "")
}
warned <<- TRUE
}
if (length(rows) > 0 & length(cols) > 0) {
new_edits <- x
if (any(!vapply(FUN.VALUE = logical(1), x[, cols, drop = FALSE], is.rsi), na.rm = TRUE)) {
track_changes$rsi_warn <- cols[!vapply(FUN.VALUE = logical(1), x[, cols, drop = FALSE], is.rsi)]
}
tryCatch(
# insert into original table
new_edits[rows, cols] <- to,
warning = function(w) {
if (w$message %like% "invalid factor level") {
xyz <- vapply(FUN.VALUE = logical(1), cols, function(col) {
new_edits[, col] <<- factor(x = as.character(pm_pull(new_edits, col)),
levels = unique(c(to, levels(pm_pull(new_edits, col)))))
TRUE
})
suppressWarnings(new_edits[rows, cols] <<- to)
warning_("Value \"", to, "\" added to the factor levels of column", ifelse(length(cols) == 1, "", "s"),
" ", vector_and(cols, quotes = "`", sort = FALSE),
" because this value was not an existing factor level.",
call = FALSE)
txt_warning()
warned <- FALSE
} else {
warning_(w$message, call = FALSE)
txt_warning()
}
},
error = function(e) {
txt_error()
stop(paste0("In row(s) ", paste(rows[1:min(length(rows), 10)], collapse = ","),
ifelse(length(rows) > 10, "...", ""),
" while writing value '", to,
"' to column(s) `", paste(cols, collapse = "`, `"),
"`:\n", e$message),
call. = FALSE)
}
)
track_changes$output <- new_edits
if ((info == TRUE | verbose == TRUE) && !isTRUE(all.equal(x, track_changes$output))) {
get_original_rows <- function(rowids) {
as.integer(rownames(original_data[which(original_data$.rowid %in% rowids), , drop = FALSE]))
}
for (i in seq_len(length(cols))) {
verbose_new <- data.frame(rowid = new_edits[rows, ".rowid", drop = TRUE],
col = cols[i],
mo_fullname = new_edits[rows, "fullname", drop = TRUE],
old = x[rows, cols[i], drop = TRUE],
new = to,
rule = font_stripstyle(rule[1]),
rule_group = font_stripstyle(rule[2]),
rule_name = font_stripstyle(rule[3]),
rule_source = font_stripstyle(rule[4]),
stringsAsFactors = FALSE)
colnames(verbose_new) <- c("rowid", "col", "mo_fullname", "old", "new",
"rule", "rule_group", "rule_name", "rule_source")
verbose_new <- verbose_new %pm>% pm_filter(old != new | is.na(old) | is.na(new) & !is.na(old))
# save changes to data set 'verbose_info'
track_changes$verbose_info <- rbind(track_changes$verbose_info,
verbose_new,
stringsAsFactors = FALSE)
# count adds and changes
track_changes$added <- track_changes$added + verbose_new %pm>%
pm_filter(is.na(old)) %pm>%
pm_pull(rowid) %pm>%
get_original_rows() %pm>%
length()
track_changes$changed <- track_changes$changed + verbose_new %pm>%
pm_filter(!is.na(old)) %pm>%
pm_pull(rowid) %pm>%
get_original_rows() %pm>%
length()
}
}
}
return(track_changes)
}
#' @rdname eucast_rules
#' @export
eucast_dosage <- function(ab, administration = "iv", version_breakpoints = 11.0) {
meet_criteria(ab, allow_class = c("character", "numeric", "integer", "factor"))
meet_criteria(administration, allow_class = "character", is_in = dosage$administration[!is.na(dosage$administration)], has_length = 1)
meet_criteria(version_breakpoints, allow_class = c("numeric", "integer"), has_length = 1, is_in = as.double(names(EUCAST_VERSION_BREAKPOINTS)))
# show used version_breakpoints number once per session (pkg_env will reload every session)
if (message_not_thrown_before(paste0("eucast_dosage_v", gsub("[^0-9]", "", version_breakpoints)), entire_session = TRUE)) {
message_("Dosages for antimicrobial drugs, as meant for ",
format_eucast_version_nr(version_breakpoints, markdown = FALSE), ". ",
font_red("This note will be shown once per session."))
remember_thrown_message(paste0("eucast_dosage_v", gsub("[^0-9]", "", version_breakpoints)), entire_session = TRUE)
}
ab <- as.ab(ab)
lst <- vector("list", length = length(ab))
for (i in seq_len(length(ab))) {
df <- AMR::dosage[which(AMR::dosage$ab == ab[i] & AMR::dosage$administration == administration), , drop = FALSE]
lst[[i]] <- list(ab = "",
name = "",
standard_dosage = ifelse("standard_dosage" %in% df$type,
df[which(df$type == "standard_dosage"), ]$original_txt,
NA_character_),
high_dosage = ifelse("high_dosage" %in% df$type,
df[which(df$type == "high_dosage"), ]$original_txt,
NA_character_))
}
out <- do.call("rbind", lapply(lst, as.data.frame, stringsAsFactors = FALSE))
rownames(out) <- NULL
out$ab <- ab
out$name <- ab_name(ab, language = NULL)
out
}