mirror of https://github.com/msberends/AMR.git
231 lines
13 KiB
R
Executable File
231 lines
13 KiB
R
Executable File
# ==================================================================== #
|
|
# TITLE #
|
|
# Antimicrobial Resistance (AMR) Analysis #
|
|
# #
|
|
# SOURCE #
|
|
# https://gitlab.com/msberends/AMR #
|
|
# #
|
|
# LICENCE #
|
|
# (c) 2019 Berends MS (m.s.berends@umcg.nl), Luz CF (c.f.luz@umcg.nl) #
|
|
# #
|
|
# 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. #
|
|
# #
|
|
# This R package was created for academic research and was publicly #
|
|
# released in the hope that it will be useful, but it comes WITHOUT #
|
|
# ANY WARRANTY OR LIABILITY. #
|
|
# Visit our website for more info: https://msberends.gitab.io/AMR. #
|
|
# ==================================================================== #
|
|
|
|
#' Data set with ~500 antibiotics
|
|
#'
|
|
#' A data set containing all antibiotics with a J0 code and some other antimicrobial agents, with their DDDs. Except for trade names and abbreviations, all properties were downloaded from the WHO, see Source.
|
|
#' @format A \code{\link{data.frame}} with 488 observations and 16 variables:
|
|
#' \describe{
|
|
#' \item{\code{atc}}{ATC code, like \code{J01CR02}}
|
|
#' \item{\code{certe}}{Certe code, like \code{amcl}}
|
|
#' \item{\code{umcg}}{UMCG code, like \code{AMCL}}
|
|
#' \item{\code{abbr}}{Abbreviation as used by many countries, used internally by \code{\link{as.atc}}}
|
|
#' \item{\code{official}}{Official name by the WHO, like \code{"Amoxicillin and beta-lactamase inhibitor"}}
|
|
#' \item{\code{official_nl}}{Official name in the Netherlands, like \code{"Amoxicilline met enzymremmer"}}
|
|
#' \item{\code{trivial_nl}}{Trivial name in Dutch, like \code{"Amoxicilline/clavulaanzuur"}}
|
|
#' \item{\code{trade_name}}{Trade name as used by many countries (a total of 294), used internally by \code{\link{as.atc}}}
|
|
#' \item{\code{oral_ddd}}{Defined Daily Dose (DDD), oral treatment}
|
|
#' \item{\code{oral_units}}{Units of \code{ddd_units}}
|
|
#' \item{\code{iv_ddd}}{Defined Daily Dose (DDD), parenteral treatment}
|
|
#' \item{\code{iv_units}}{Units of \code{iv_ddd}}
|
|
#' \item{\code{atc_group1}}{ATC group, like \code{"Macrolides, lincosamides and streptogramins"}}
|
|
#' \item{\code{atc_group2}}{Subgroup of \code{atc_group1}, like \code{"Macrolides"}}
|
|
#' \item{\code{useful_gramnegative}}{\code{FALSE} if not useful according to EUCAST, \code{NA} otherwise (see Source)}
|
|
#' \item{\code{useful_grampositive}}{\code{FALSE} if not useful according to EUCAST, \code{NA} otherwise (see Source)}
|
|
#' }
|
|
#' @source - World Health Organization (WHO) Collaborating Centre for Drug Statistics Methodology: \url{https://www.whocc.no/atc_ddd_index/}
|
|
#'
|
|
#' EUCAST Expert Rules, Intrinsic Resistance and Exceptional Phenotypes Tables. Version 3.1, 2016: \url{http://www.eucast.org/fileadmin/src/media/PDFs/EUCAST_files/Expert_Rules/Expert_rules_intrinsic_exceptional_V3.1.pdf}
|
|
#'
|
|
#' European Commission Public Health PHARMACEUTICALS - COMMUNITY REGISTER: \url{http://ec.europa.eu/health/documents/community-register/html/atc.htm}
|
|
#' @inheritSection AMR Read more on our website!
|
|
#' @seealso \code{\link{microorganisms}}
|
|
# use this later to further fill AMR::antibiotics
|
|
# drug <- "Ciprofloxacin"
|
|
# url <- xml2::read_html(paste0("https://www.ncbi.nlm.nih.gov/pccompound?term=", drug)) %>%
|
|
# html_nodes(".rslt") %>%
|
|
# .[[1]] %>%
|
|
# html_nodes(".title a") %>%
|
|
# html_attr("href") %>%
|
|
# gsub("/compound/", "/rest/pug_view/data/compound/", ., fixed = TRUE) %>%
|
|
# paste0("/XML/?response_type=display")
|
|
# synonyms <- url %>%
|
|
# read_xml() %>%
|
|
# xml_contents() %>% .[[6]] %>%
|
|
# xml_contents() %>% .[[8]] %>%
|
|
# xml_contents() %>% .[[3]] %>%
|
|
# xml_contents() %>% .[[3]] %>%
|
|
# xml_contents() %>%
|
|
# paste() %>%
|
|
# .[. %like% "StringValueList"] %>%
|
|
# gsub("[</]+StringValueList[>]", "", .)
|
|
|
|
# last two columns created with:
|
|
# antibiotics %>%
|
|
# mutate(useful_gramnegative =
|
|
# if_else(
|
|
# atc_group1 %like% '(fusidic|glycopeptide|macrolide|lincosamide|daptomycin|linezolid)' |
|
|
# atc_group2 %like% '(fusidic|glycopeptide|macrolide|lincosamide|daptomycin|linezolid)' |
|
|
# official %like% '(fusidic|glycopeptide|macrolide|lincosamide|daptomycin|linezolid)',
|
|
# FALSE,
|
|
# NA
|
|
# ),
|
|
# useful_grampositive =
|
|
# if_else(
|
|
# atc_group1 %like% '(aztreonam|temocillin|polymyxin|colistin|nalidixic)' |
|
|
# atc_group2 %like% '(aztreonam|temocillin|polymyxin|colistin|nalidixic)' |
|
|
# official %like% '(aztreonam|temocillin|polymyxin|colistin|nalidixic)',
|
|
# FALSE,
|
|
# NA
|
|
# )
|
|
# )
|
|
#
|
|
# ADD NEW TRADE NAMES FROM OTHER DATAFRAME
|
|
# antibiotics_add_to_property <- function(ab_df, atc, property, value) {
|
|
# if (length(atc) > 1L) {
|
|
# stop("only one atc at a time")
|
|
# }
|
|
# if (!property %in% c("abbr", "trade_name")) {
|
|
# stop("only possible for abbr and trade_name")
|
|
# }
|
|
#
|
|
# value <- gsub(ab_df[which(ab_df$atc == atc),] %>% pull("official"), "", value, fixed = TRUE)
|
|
# value <- gsub("||", "|", value, fixed = TRUE)
|
|
# value <- gsub("[äáàâ]", "a", value)
|
|
# value <- gsub("[ëéèê]", "e", value)
|
|
# value <- gsub("[ïíìî]", "i", value)
|
|
# value <- gsub("[öóòô]", "o", value)
|
|
# value <- gsub("[üúùû]", "u", value)
|
|
# if (!atc %in% ab_df$atc) {
|
|
# message("SKIPPING - UNKNOWN ATC: ", atc)
|
|
# }
|
|
# if (is.na(value)) {
|
|
# message("SKIPPING - VALUE MISSES: ", atc)
|
|
# }
|
|
# if (atc %in% ab_df$atc & !is.na(value)) {
|
|
# current <- ab_df[which(ab_df$atc == atc),] %>% pull(property)
|
|
# if (!is.na(current)) {
|
|
# value <- paste(current, value, sep = "|")
|
|
# }
|
|
# value <- strsplit(value, "|", fixed = TRUE) %>% unlist() %>% unique() %>% paste(collapse = "|")
|
|
# value <- gsub("||", "|", value, fixed = TRUE)
|
|
# # print(value)
|
|
# ab_df[which(ab_df$atc == atc), property] <- value
|
|
# message("Added ", value, " to ", ab_official(atc), " (", atc, ", ", ab_certe(atc), ")")
|
|
# }
|
|
# ab_df
|
|
# }
|
|
#
|
|
"antibiotics"
|
|
|
|
#' Data set with ~20,000 microorganisms
|
|
#'
|
|
#' A data set containing the complete microbial taxonomy of the kingdoms Bacteria, Fungi and Protozoa from ITIS. MO codes can be looked up using \code{\link{as.mo}}.
|
|
#' @inheritSection ITIS ITIS
|
|
#' @format A \code{\link{data.frame}} with 18,833 observations and 15 variables:
|
|
#' \describe{
|
|
#' \item{\code{mo}}{ID of microorganism}
|
|
#' \item{\code{tsn}}{Taxonomic Serial Number (TSN), as defined by ITIS}
|
|
#' \item{\code{genus}}{Taxonomic genus of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{species}}{Taxonomic species of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{subspecies}}{Taxonomic subspecies of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{fullname}}{Full name, like \code{"Echerichia coli"}}
|
|
#' \item{\code{family}}{Taxonomic family of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{order}}{Taxonomic order of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{class}}{Taxonomic class of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{phylum}}{Taxonomic phylum of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{subkingdom}}{Taxonomic subkingdom of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{kingdom}}{Taxonomic kingdom of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{gramstain}}{Gram of microorganism, like \code{"Gram negative"}}
|
|
#' \item{\code{prevalence}}{An integer based on estimated prevalence of the microorganism in humans. Used internally by \code{\link{as.mo}}, otherwise quite meaningless. It has a value of 25 for manually added items and a value of 1000 for all unprevalent microorganisms whose genus was somewhere in the top 250 (with another species).}
|
|
#' \item{\code{ref}}{Author(s) and year of concerning publication as found in ITIS, see Source}
|
|
#' }
|
|
#' @source Integrated Taxonomic Information System (ITIS) public online database, \url{https://www.itis.gov}.
|
|
#' @inheritSection AMR Read more on our website!
|
|
#' @seealso \code{\link{as.mo}} \code{\link{mo_property}} \code{\link{microorganisms.codes}}
|
|
"microorganisms"
|
|
|
|
#' Data set with previously accepted taxonomic names
|
|
#'
|
|
#' A data set containing old (previously valid or accepted) taxonomic names according to ITIS. This data set is used internally by \code{\link{as.mo}}.
|
|
#' @inheritSection as.mo ITIS
|
|
#' @format A \code{\link{data.frame}} with 2,383 observations and 4 variables:
|
|
#' \describe{
|
|
#' \item{\code{tsn}}{Old Taxonomic Serial Number (TSN), as defined by ITIS}
|
|
#' \item{\code{name}}{Old taxonomic name of the microorganism as found in ITIS, see Source}
|
|
#' \item{\code{tsn_new}}{New Taxonomic Serial Number (TSN), as defined by ITIS}
|
|
#' \item{\code{ref}}{Author(s) and year of concerning publication as found in ITIS, see Source}
|
|
#' }
|
|
#' @source [3] Integrated Taxonomic Information System (ITIS) on-line database, \url{https://www.itis.gov}.
|
|
#' @inheritSection AMR Read more on our website!
|
|
#' @seealso \code{\link{as.mo}} \code{\link{mo_property}} \code{\link{microorganisms}}
|
|
"microorganisms.old"
|
|
|
|
#' Translation table for microorganism codes
|
|
#'
|
|
#' A data set containing commonly used codes for microorganisms. Define your own with \code{\link{set_mo_source}}.
|
|
#' @format A \code{\link{data.frame}} with 3,303 observations and 2 variables:
|
|
#' \describe{
|
|
#' \item{\code{certe}}{Commonly used code of a microorganism}
|
|
#' \item{\code{mo}}{Code of microorganism in \code{\link{microorganisms}}}
|
|
#' }
|
|
#' @inheritSection AMR Read more on our website!
|
|
#' @seealso \code{\link{as.mo}} \code{\link{microorganisms}}
|
|
"microorganisms.codes"
|
|
|
|
#' Data set with 2,000 blood culture isolates from septic patients
|
|
#'
|
|
#' An anonymised data set containing 2,000 microbial blood culture isolates with their full antibiograms found in septic patients in 4 different hospitals in the Netherlands, between 2001 and 2017. It is true, genuine data. This \code{data.frame} can be used to practice AMR analysis. For examples, please read \href{https://msberends.gitlab.io/AMR/articles/AMR.html}{the tutorial on our website}.
|
|
#' @format A \code{\link{data.frame}} with 2,000 observations and 49 variables:
|
|
#' \describe{
|
|
#' \item{\code{date}}{date of receipt at the laboratory}
|
|
#' \item{\code{hospital_id}}{ID of the hospital, from A to D}
|
|
#' \item{\code{ward_icu}}{logical to determine if ward is an intensive care unit}
|
|
#' \item{\code{ward_clinical}}{logical to determine if ward is a regular clinical ward}
|
|
#' \item{\code{ward_outpatient}}{logical to determine if ward is an outpatient clinic}
|
|
#' \item{\code{age}}{age of the patient}
|
|
#' \item{\code{gender}}{gender of the patient}
|
|
#' \item{\code{patient_id}}{ID of the patient, first 10 characters of an SHA hash containing irretrievable information}
|
|
#' \item{\code{mo}}{ID of microorganism created with \code{\link{as.mo}}, see also \code{\link{microorganisms}}}
|
|
#' \item{\code{peni:rifa}}{40 different antibiotics with class \code{rsi} (see \code{\link{as.rsi}}); these column names occur in \code{\link{antibiotics}} data set and can be translated with \code{\link{abname}}}
|
|
#' }
|
|
#' @inheritSection AMR Read more on our website!
|
|
"septic_patients"
|
|
|
|
#' Supplementary Data
|
|
#'
|
|
#' These \code{\link{data.table}s} are transformed from the \code{\link{microorganisms}} and \code{\link{microorganisms}} data sets to improve speed of \code{\link{as.mo}}. They are meant for internal use only, and are only mentioned here for reference.
|
|
#' @rdname supplementary_data
|
|
#' @name supplementary_data
|
|
#' @inheritSection AMR Read more on our website!
|
|
# # Renew data:
|
|
# microorganismsDT <- data.table::as.data.table(AMR::microorganisms)
|
|
# # sort on (1) bacteria, (2) fungi, (3) protozoa and then human pathogenic prevalence and then TSN:
|
|
# data.table::setkey(microorganismsDT, kingdom, prevalence, fullname)
|
|
# microorganisms.prevDT <- microorganismsDT[prevalence == 9999,]
|
|
# microorganisms.unprevDT <- microorganismsDT[prevalence != 9999,]
|
|
# microorganisms.oldDT <- data.table::as.data.table(AMR::microorganisms.old)
|
|
# data.table::setkey(microorganisms.oldDT, tsn, name)
|
|
# devtools::use_data(microorganismsDT, overwrite = TRUE)
|
|
# devtools::use_data(microorganisms.prevDT, overwrite = TRUE)
|
|
# devtools::use_data(microorganisms.unprevDT, overwrite = TRUE)
|
|
# devtools::use_data(microorganisms.oldDT, overwrite = TRUE)
|
|
"microorganismsDT"
|
|
|
|
#' @rdname supplementary_data
|
|
"microorganisms.prevDT"
|
|
|
|
#' @rdname supplementary_data
|
|
"microorganisms.unprevDT"
|
|
|
|
#' @rdname supplementary_data
|
|
"microorganisms.oldDT"
|