mirror of https://github.com/msberends/AMR.git
898 lines
57 KiB
R
898 lines
57 KiB
R
# ==================================================================== #
|
|
# TITLE: #
|
|
# AMR: An R Package for Working with Antimicrobial Resistance Data #
|
|
# #
|
|
# SOURCE CODE: #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# PLEASE CITE THIS SOFTWARE AS: #
|
|
# Berends MS, Luz CF, Friedrich AW, et al. (2022). #
|
|
# AMR: An R Package for Working with Antimicrobial Resistance Data. #
|
|
# Journal of Statistical Software, 104(3), 1-31. #
|
|
# https://doi.org/10.18637/jss.v104.i03 #
|
|
# #
|
|
# Developed at the University of Groningen and the University Medical #
|
|
# Center Groningen in The Netherlands, in collaboration with many #
|
|
# colleagues from around the world, see our website. #
|
|
# #
|
|
# 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/ #
|
|
# ==================================================================== #
|
|
|
|
library(dplyr)
|
|
|
|
# got EARS-Net codes (= ECDC/WHO codes) from here:
|
|
|
|
# Installed WHONET 2019 software on Windows (http://www.whonet.org/software.html),
|
|
# opened C:\WHONET\Codes\WHONETCodes.mdb in MS Access
|
|
# and exported table 'DRGLST' to MS Excel
|
|
library(readxl)
|
|
DRGLST <- read_excel("DRGLST.xlsx")
|
|
abx <- DRGLST %>%
|
|
select(
|
|
ab = WHON5_CODE,
|
|
name = ANTIBIOTIC
|
|
) %>%
|
|
# remove the ones without WHONET code
|
|
filter(!is.na(ab)) %>%
|
|
distinct(name, .keep_all = TRUE) %>%
|
|
# add the ones without WHONET code
|
|
bind_rows(
|
|
DRGLST %>%
|
|
select(
|
|
ab = WHON5_CODE,
|
|
name = ANTIBIOTIC
|
|
) %>%
|
|
filter(is.na(ab)) %>%
|
|
distinct(name, .keep_all = TRUE)
|
|
# add new ab code later
|
|
) %>%
|
|
arrange(name)
|
|
|
|
# add old ATC codes
|
|
ab_old <- AMR::antibiotics %>%
|
|
mutate(
|
|
official = gsub("( and |, )", "/", official),
|
|
abbr = tolower(paste(ifelse(is.na(abbr), "", abbr),
|
|
ifelse(is.na(certe), "", certe),
|
|
ifelse(is.na(umcg), "", umcg),
|
|
sep = "|"
|
|
))
|
|
)
|
|
for (i in 1:nrow(ab_old)) {
|
|
abbr <- ab_old[i, "abbr"]
|
|
abbr <- strsplit(abbr, "|", fixed = TRUE) %>%
|
|
unlist() %>%
|
|
unique()
|
|
abbr <- abbr[abbr != ""]
|
|
# print(abbr)
|
|
if (length(abbr) == 0) {
|
|
ab_old[i, "abbr"] <- NA_character_
|
|
} else {
|
|
ab_old[i, "abbr"] <- paste(abbr, collapse = "|")
|
|
}
|
|
}
|
|
|
|
# create reference data set: to be able to map ab to atc
|
|
abx_atc1 <- abx %>%
|
|
mutate(name_lower = tolower(name)) %>%
|
|
left_join(ab_old %>%
|
|
select(ears_net, atc), by = c(ab = "ears_net")) %>%
|
|
rename(atc1 = atc) %>%
|
|
left_join(ab_old %>%
|
|
mutate(official = gsub(", combinations", "", official, fixed = TRUE)) %>%
|
|
transmute(official = tolower(official), atc), by = c(name_lower = "official")) %>%
|
|
rename(atc2 = atc) %>%
|
|
left_join(ab_old %>%
|
|
mutate(official = gsub(", combinations", "", official, fixed = TRUE)) %>%
|
|
mutate(official = gsub("f", "ph", official)) %>%
|
|
transmute(official = tolower(official), atc), by = c(name_lower = "official")) %>%
|
|
rename(atc3 = atc) %>%
|
|
left_join(ab_old %>%
|
|
mutate(official = gsub(", combinations", "", official, fixed = TRUE)) %>%
|
|
mutate(official = gsub("t", "th", official)) %>%
|
|
transmute(official = tolower(official), atc), by = c(name_lower = "official")) %>%
|
|
rename(atc4 = atc) %>%
|
|
left_join(ab_old %>%
|
|
mutate(official = gsub(", combinations", "", official, fixed = TRUE)) %>%
|
|
mutate(official = gsub("f", "ph", official)) %>%
|
|
mutate(official = gsub("t", "th", official)) %>%
|
|
transmute(official = tolower(official), atc), by = c(name_lower = "official")) %>%
|
|
rename(atc5 = atc) %>%
|
|
left_join(ab_old %>%
|
|
mutate(official = gsub(", combinations", "", official, fixed = TRUE)) %>%
|
|
mutate(official = gsub("f", "ph", official)) %>%
|
|
mutate(official = gsub("t", "th", official)) %>%
|
|
mutate(official = gsub("ine$", "in", official)) %>%
|
|
transmute(official = tolower(official), atc), by = c(name_lower = "official")) %>%
|
|
rename(atc6 = atc) %>%
|
|
mutate(atc = case_when(
|
|
!is.na(atc1) ~ atc1,
|
|
!is.na(atc2) ~ atc2,
|
|
!is.na(atc3) ~ atc3,
|
|
!is.na(atc4) ~ atc4,
|
|
!is.na(atc4) ~ atc5,
|
|
TRUE ~ atc6
|
|
)) %>%
|
|
distinct(ab, name, .keep_all = TRUE) %>%
|
|
select(ab, atc, name)
|
|
|
|
abx_atc2 <- ab_old %>%
|
|
filter(
|
|
!atc %in% abx_atc1$atc,
|
|
is.na(ears_net),
|
|
!is.na(atc_group1),
|
|
atc_group1 %unlike% ("virus|vaccin|viral|immun"),
|
|
official %unlike% "(combinations| with )"
|
|
) %>%
|
|
mutate(ab = NA_character_) %>%
|
|
as.data.frame(stringsAsFactors = FALSE) %>%
|
|
select(ab, atc, name = official)
|
|
|
|
abx2 <- bind_rows(abx_atc1, abx_atc2)
|
|
|
|
rm(abx_atc1)
|
|
rm(abx_atc2)
|
|
|
|
abx2$ab[is.na(abx2$ab)] <- toupper(abbreviate(
|
|
gsub(
|
|
"[/0-9-]",
|
|
" ",
|
|
abx2$name[is.na(abx2$ab)]
|
|
),
|
|
minlength = 3,
|
|
method = "left.kept",
|
|
strict = TRUE
|
|
))
|
|
|
|
n_distinct(abx2$ab)
|
|
|
|
abx2 <- abx2 %>% arrange(ab)
|
|
seqnr <- 0
|
|
# add follow up nrs
|
|
for (i in 2:nrow(abx2)) {
|
|
if (abx2[i, "ab", drop = TRUE] == abx2[i - 1, "ab", drop = TRUE]) {
|
|
seqnr <- seqnr + 1
|
|
abx2[i, "seqnr"] <- seqnr
|
|
} else {
|
|
seqnr <- 0
|
|
}
|
|
}
|
|
for (i in 2:nrow(abx2)) {
|
|
if (!is.na(abx2[i, "seqnr"])) {
|
|
abx2[i, "ab"] <- paste0(abx2[i, "ab", drop = TRUE], abx2[i, "seqnr", drop = TRUE])
|
|
}
|
|
}
|
|
abx2 <- abx2 %>%
|
|
select(-seqnr) %>%
|
|
arrange(name)
|
|
|
|
# everything unique??
|
|
nrow(abx2) == n_distinct(abx2$ab)
|
|
|
|
# get ATC properties
|
|
abx2 <- abx2 %>%
|
|
left_join(ab_old %>%
|
|
select(
|
|
atc, abbr, atc_group1, atc_group2,
|
|
oral_ddd, oral_units, iv_ddd, iv_units
|
|
))
|
|
|
|
abx2$abbr <- lapply(as.list(abx2$abbr), function(x) unlist(strsplit(x, "|", fixed = TRUE)))
|
|
|
|
# Update Compound IDs and Synonyms ----
|
|
|
|
# vector with official names, returns vector with CIDs
|
|
get_CID <- function(ab) {
|
|
CID <- rep(NA_integer_, length(ab))
|
|
p <- AMR:::progress_ticker(n = length(ab), min_time = 0)
|
|
for (i in 1:length(ab)) {
|
|
p$tick()
|
|
|
|
CID[i] <- tryCatch(
|
|
data.table::fread(
|
|
paste0(
|
|
"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/name/",
|
|
URLencode(ab[i], reserved = TRUE),
|
|
"/cids/TXT?name_type=complete"
|
|
),
|
|
showProgress = FALSE
|
|
)[[1]][1],
|
|
error = function(e) NA_integer_
|
|
)
|
|
if (is.na(CID[i])) {
|
|
# try with removing the text in brackets
|
|
CID[i] <- tryCatch(
|
|
data.table::fread(
|
|
paste0(
|
|
"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/name/",
|
|
URLencode(trimws(gsub("[(].*[)]", "", ab[i])), reserved = TRUE),
|
|
"/cids/TXT?name_type=complete"
|
|
),
|
|
showProgress = FALSE
|
|
)[[1]][1],
|
|
error = function(e) NA_integer_
|
|
)
|
|
}
|
|
if (is.na(CID[i])) {
|
|
# try match on word and take the lowest CID value (sorted)
|
|
ab[i] <- gsub("[^a-z0-9]+", " ", ab[i], ignore.case = TRUE)
|
|
CID[i] <- tryCatch(
|
|
data.table::fread(
|
|
paste0(
|
|
"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/name/",
|
|
URLencode(ab[i], reserved = TRUE),
|
|
"/cids/TXT?name_type=word"
|
|
),
|
|
showProgress = FALSE
|
|
)[[1]][1],
|
|
error = function(e) NA_integer_
|
|
)
|
|
}
|
|
Sys.sleep(0.1)
|
|
}
|
|
CID
|
|
}
|
|
|
|
# get CIDs (4-5 min)
|
|
CIDs <- get_CID(antibiotics$name)
|
|
# take missing from previously found CIDs
|
|
CIDs[is.na(CIDs) & !is.na(antibiotics$cid)] <- antibiotics$cid[is.na(CIDs) & !is.na(antibiotics$cid)]
|
|
# These could not be found:
|
|
antibiotics[is.na(CIDs), ] %>% View()
|
|
|
|
# returns list with synonyms (brand names), with CIDs as names
|
|
get_synonyms <- function(CID, clean = TRUE) {
|
|
synonyms <- rep(NA_character_, length(CID))
|
|
p <- AMR:::progress_ticker(n = length(CID), min_time = 0)
|
|
|
|
for (i in 1:length(CID)) {
|
|
p$tick()
|
|
|
|
synonyms_txt <- ""
|
|
|
|
if (is.na(CID[i])) {
|
|
next
|
|
}
|
|
|
|
synonyms_txt <- tryCatch(
|
|
data.table::fread(
|
|
paste0(
|
|
"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/fastidentity/cid/",
|
|
CID[i],
|
|
"/synonyms/TXT"
|
|
),
|
|
sep = "\n",
|
|
showProgress = FALSE
|
|
)[[1]],
|
|
error = function(e) NA_character_
|
|
)
|
|
|
|
Sys.sleep(0.1)
|
|
|
|
if (clean == TRUE) {
|
|
# remove text between brackets
|
|
synonyms_txt <- trimws(gsub(
|
|
"[(].*[)]", "",
|
|
gsub(
|
|
"[[].*[]]", "",
|
|
gsub(
|
|
"[(].*[]]", "",
|
|
gsub("[[].*[)]", "", synonyms_txt)
|
|
)
|
|
)
|
|
))
|
|
synonyms_txt <- gsub("Co-", "Co", synonyms_txt, fixed = TRUE)
|
|
# only length 6 to 20 and no txt with reading marks or numbers and must start with capital letter (= brand)
|
|
synonyms_txt <- synonyms_txt[nchar(synonyms_txt) %in% c(6:20) &
|
|
!grepl("[-&{},_0-9/:]", synonyms_txt) &
|
|
grepl("^[A-Z]", synonyms_txt, ignore.case = FALSE)]
|
|
synonyms_txt <- unlist(strsplit(synonyms_txt, ";", fixed = TRUE))
|
|
}
|
|
synonyms_txt <- unique(trimws(synonyms_txt[tolower(synonyms_txt) %in% unique(tolower(synonyms_txt))]))
|
|
synonyms[i] <- list(sort(synonyms_txt))
|
|
}
|
|
names(synonyms) <- CID
|
|
synonyms
|
|
}
|
|
|
|
# get brand names from PubChem (3-4 min)
|
|
synonyms <- get_synonyms(CIDs)
|
|
synonyms.bak <- synonyms
|
|
|
|
# add existing ones (will be cleaned later)
|
|
for (i in seq_len(length(synonyms))) {
|
|
old <- antibiotics$synonyms[[i]]
|
|
if (old %like% "^c[()]") {
|
|
old <- eval(parse(text = old))
|
|
}
|
|
synonyms[[i]] <- c(synonyms[[i]], old)
|
|
}
|
|
|
|
antibiotics$synonyms <- synonyms
|
|
|
|
stop("remember to remove co-trimoxazole as synonyms from SXT (Sulfamethoxazole), so it only exists in SXT!")
|
|
sulfa <- antibiotics[which(antibiotics$ab == "SMX"), "synonyms", drop = TRUE][[1]]
|
|
cotrim <- antibiotics[which(antibiotics$ab == "SXT"), "synonyms", drop = TRUE][[1]]
|
|
sulfa <- sulfa[!sulfa %in% cotrim]
|
|
antibiotics[which(antibiotics$ab == "SMX"), "synonyms"][[1]][[1]] <- sulfa
|
|
|
|
|
|
# now go to end of this file
|
|
|
|
|
|
# -----
|
|
|
|
# add them to data set
|
|
antibiotics <- abx2 %>%
|
|
left_join(DRGLST %>%
|
|
select(ab = WHON5_CODE, CLASS, SUBCLASS) %>%
|
|
distinct(ab, .keep_all = TRUE), by = "ab") %>%
|
|
transmute(ab,
|
|
atc,
|
|
cid = CIDs,
|
|
# no capital after a slash: Ampicillin/Sulbactam -> Ampicillin/sulbactam
|
|
name = name %>%
|
|
gsub("([/-])([A-Z])", "\\1\\L\\2", ., perl = TRUE) %>%
|
|
gsub("edta", "EDTA", ., ignore.case = TRUE),
|
|
group = case_when(
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "am(ph|f)enicol" ~ "Amphenicols",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "aminoglycoside" ~ "Aminoglycosides",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "carbapenem" | name %like% "(imipenem|meropenem)" ~ "Carbapenems",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "First-generation cephalosporin" ~ "Cephalosporins (1st gen.)",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "Second-generation cephalosporin" ~ "Cephalosporins (2nd gen.)",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "Third-generation cephalosporin" ~ "Cephalosporins (3rd gen.)",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "Fourth-generation cephalosporin" ~ "Cephalosporins (4th gen.)",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "(tuberculosis|mycobacter)" ~ "Antimycobacterials",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "cephalosporin" ~ "Cephalosporins",
|
|
name %like% "^Ce" & is.na(atc_group1) & paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "beta-?lactam" ~ "Cephalosporins",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "(beta-?lactam|penicillin)" ~ "Beta-lactams/penicillins",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "quinolone" ~ "Quinolones",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "glycopeptide" ~ "Glycopeptides",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "macrolide" ~ "Macrolides/lincosamides",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "tetracycline" ~ "Tetracyclines",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "trimethoprim" ~ "Trimethoprims",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "polymyxin" ~ "Polymyxins",
|
|
paste(atc_group1, atc_group2, CLASS, SUBCLASS) %like% "(fungal|mycot)" ~ "Antifungals/antimycotics",
|
|
TRUE ~ "Other antibacterials"
|
|
),
|
|
atc_group1, atc_group2,
|
|
abbreviations = unname(abbr),
|
|
synonyms = unname(synonyms),
|
|
oral_ddd, oral_units,
|
|
iv_ddd, iv_units
|
|
) %>%
|
|
as.data.frame(stringsAsFactors = FALSE)
|
|
|
|
# some exceptions
|
|
antibiotics[which(antibiotics$ab == "DOX"), "abbreviations"][[1]] <- list(c("dox", "doxy"))
|
|
antibiotics[which(antibiotics$ab == "FLC"), "abbreviations"][[1]] <- list(c("clox"))
|
|
antibiotics[which(antibiotics$ab == "CEC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CEC"), "abbreviations"][[1]], "CFC")) # cefaclor old WHONET4 code
|
|
antibiotics[which(antibiotics$ab == "AMX"), "synonyms"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "AMX"), "synonyms"][[1]], "Amoxy")))
|
|
# 'Polymixin B' (POL) and 'Polymyxin B' (PLB) both exist, so:
|
|
antibiotics[which(antibiotics$ab == "PLB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PLB"), "abbreviations"][[1]], "POL", "Polymixin", "Polymixin B", "Poly B"))
|
|
antibiotics <- filter(antibiotics, ab != "POL")
|
|
# 'Latamoxef' (LTM) and 'Moxalactam (Latamoxef)' (MOX) both exist, so:
|
|
antibiotics[which(antibiotics$ab == "LTM"), "abbreviations"][[1]] <- list(c("MOX", "moxa"))
|
|
antibiotics <- filter(antibiotics, ab != "MOX")
|
|
# RFP and RFP1 (the J0 one) both mean 'rifapentine', although 'rifp' is not recognised, so:
|
|
antibiotics <- filter(antibiotics, ab != "RFP")
|
|
antibiotics[which(antibiotics$ab == "RFP1"), "ab"] <- "RFP"
|
|
antibiotics[which(antibiotics$ab == "RFP"), "abbreviations"][[1]] <- list(c("rifp"))
|
|
# Rifampicin is better known as a drug than Rifampin (Rifampin is still listed as a brand name), so:
|
|
antibiotics[which(antibiotics$ab == "RIF"), "name"] <- "Rifampicin"
|
|
# PME and PVM1 (the J0 one) both mean 'Pivmecillinam', so:
|
|
antibiotics <- filter(antibiotics, ab != "PME")
|
|
antibiotics[which(antibiotics$ab == "PVM1"), "ab"] <- "PME"
|
|
# Remove Sinecatechins
|
|
antibiotics <- filter(antibiotics, ab != "SNC")
|
|
# GLIMS codes
|
|
antibiotics[which(antibiotics$ab == as.ab("cefuroxim")), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == as.ab("cefuroxim")), "abbreviations"][[1]], "cfrx"))
|
|
antibiotics[which(antibiotics$ab == as.ab("cefotaxim")), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == as.ab("cefotaxim")), "abbreviations"][[1]], "cftx"))
|
|
antibiotics[which(antibiotics$ab == as.ab("ceftazidime")), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == as.ab("ceftazidime")), "abbreviations"][[1]], "cftz"))
|
|
antibiotics[which(antibiotics$ab == as.ab("cefepime")), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == as.ab("cefepime")), "abbreviations"][[1]], "cfpi"))
|
|
antibiotics[which(antibiotics$ab == as.ab("cefoxitin")), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == as.ab("cefoxitin")), "abbreviations"][[1]], "cfxt"))
|
|
# Add cefoxitin screening
|
|
class(antibiotics$ab) <- "character"
|
|
antibiotics <- rbind(antibiotics, data.frame(
|
|
ab = "FOX1", atc = NA, cid = NA,
|
|
name = "Cefoxitin screening",
|
|
group = "Cephalosporins (2nd gen.)", atc_group1 = NA, atc_group2 = NA,
|
|
abbreviations = "cfsc", synonyms = NA,
|
|
oral_ddd = NA, oral_units = NA, iv_ddd = NA, iv_units = NA,
|
|
loinc = NA,
|
|
stringsAsFactors = FALSE
|
|
))
|
|
# More GLIMS codes
|
|
antibiotics[which(antibiotics$ab == "AMB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMB"), "abbreviations"][[1]], "amf"))
|
|
antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]], "cftz"))
|
|
antibiotics[which(antibiotics$ab == "COL"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "COL"), "abbreviations"][[1]], "cst"))
|
|
antibiotics[which(antibiotics$ab == "CRO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CRO"), "abbreviations"][[1]], "cftr"))
|
|
antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]], "cftx"))
|
|
antibiotics[which(antibiotics$ab == "CXM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CXM"), "abbreviations"][[1]], "cfrx"))
|
|
antibiotics[which(antibiotics$ab == "CZO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CZO"), "abbreviations"][[1]], "cfzl"))
|
|
antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]], "fcu"))
|
|
antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]], "fluy"))
|
|
antibiotics[which(antibiotics$ab == "FLU"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FLU"), "abbreviations"][[1]], "flz"))
|
|
antibiotics[which(antibiotics$ab == "FOS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FOS"), "abbreviations"][[1]], "fof"))
|
|
antibiotics[which(antibiotics$ab == "FOX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FOX"), "abbreviations"][[1]], "cfxt"))
|
|
antibiotics[which(antibiotics$ab == "FUS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FUS"), "abbreviations"][[1]], "fa"))
|
|
antibiotics[which(antibiotics$ab == "GEH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "GEH"), "abbreviations"][[1]], "g_h"))
|
|
antibiotics[which(antibiotics$ab == "KAH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "KAH"), "abbreviations"][[1]], "k_h"))
|
|
antibiotics[which(antibiotics$ab == "KET"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "KET"), "abbreviations"][[1]], "ktc"))
|
|
antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]], "pipc"))
|
|
antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]], "PIPC"))
|
|
antibiotics[which(antibiotics$ab == "SPX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SPX"), "abbreviations"][[1]], "spa"))
|
|
antibiotics[which(antibiotics$ab == "STH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "STH"), "abbreviations"][[1]], "s_h"))
|
|
antibiotics[which(antibiotics$ab == "STR1"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "STR1"), "abbreviations"][[1]], "stm"))
|
|
antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]], "COTRIM"))
|
|
antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]], "trsx"))
|
|
antibiotics[which(antibiotics$ab == "TGC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TGC"), "abbreviations"][[1]], "tig"))
|
|
antibiotics[which(antibiotics$ab == "TMP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TMP"), "abbreviations"][[1]], "tri"))
|
|
antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]], "PIPTAZ"))
|
|
antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]], "pit"))
|
|
antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]], "pita"))
|
|
antibiotics[which(antibiotics$ab == "VOR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "VOR"), "abbreviations"][[1]], "vrc"))
|
|
|
|
# official RIVM codes (Dutch National Health Institute)
|
|
# https://www.rivm.nl/sites/default/files/2019-09/Bijlage_4_Lijst_antibiotica%202020%201.0.pdf
|
|
antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FCT"), "abbreviations"][[1]], "5flc"))
|
|
antibiotics[which(antibiotics$ab == "AMC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMC"), "abbreviations"][[1]], "amcl"))
|
|
antibiotics[which(antibiotics$ab == "AMB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMB"), "abbreviations"][[1]], "amfb"))
|
|
antibiotics[which(antibiotics$ab == "AMH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMH"), "abbreviations"][[1]], "amhl"))
|
|
antibiotics[which(antibiotics$ab == "AMK"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMK"), "abbreviations"][[1]], "amik"))
|
|
antibiotics[which(antibiotics$ab == "AMX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMX"), "abbreviations"][[1]], "amox"))
|
|
antibiotics[which(antibiotics$ab == "AMP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AMP"), "abbreviations"][[1]], "ampi"))
|
|
antibiotics[which(antibiotics$ab == "SAM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SAM"), "abbreviations"][[1]], "amsu"))
|
|
antibiotics[which(antibiotics$ab == "ANI"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ANI"), "abbreviations"][[1]], "anid"))
|
|
antibiotics[which(antibiotics$ab == "SAM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SAM"), "abbreviations"][[1]], "apsu"))
|
|
antibiotics[which(antibiotics$ab == "AZM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AZM"), "abbreviations"][[1]], "azit"))
|
|
antibiotics[which(antibiotics$ab == "AZL"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "AZL"), "abbreviations"][[1]], "azlo"))
|
|
antibiotics[which(antibiotics$ab == "ATM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ATM"), "abbreviations"][[1]], "aztr"))
|
|
antibiotics[which(antibiotics$ab == "PNV"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PNV"), "abbreviations"][[1]], "bepe"))
|
|
antibiotics[which(antibiotics$ab == "CAP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CAP"), "abbreviations"][[1]], "capr"))
|
|
antibiotics[which(antibiotics$ab == "CRB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CRB"), "abbreviations"][[1]], "carb"))
|
|
antibiotics[which(antibiotics$ab == "CAS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CAS"), "abbreviations"][[1]], "casp"))
|
|
antibiotics[which(antibiotics$ab == "CDC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CDC"), "abbreviations"][[1]], "cecl"))
|
|
antibiotics[which(antibiotics$ab == "CXA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CXA"), "abbreviations"][[1]], "cfax"))
|
|
antibiotics[which(antibiotics$ab == "CTB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTB"), "abbreviations"][[1]], "cfbu"))
|
|
antibiotics[which(antibiotics$ab == "CEC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CEC"), "abbreviations"][[1]], "cfcl"))
|
|
antibiotics[which(antibiotics$ab == "CFR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CFR"), "abbreviations"][[1]], "cfdx"))
|
|
antibiotics[which(antibiotics$ab == "CEP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CEP"), "abbreviations"][[1]], "cflt"))
|
|
antibiotics[which(antibiotics$ab == "LEX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LEX"), "abbreviations"][[1]], "cflx"))
|
|
antibiotics[which(antibiotics$ab == "MAN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MAN"), "abbreviations"][[1]], "cfmn"))
|
|
antibiotics[which(antibiotics$ab == "CPD"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CPD"), "abbreviations"][[1]], "cfpd"))
|
|
antibiotics[which(antibiotics$ab == "FEP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FEP"), "abbreviations"][[1]], "cfpi"))
|
|
antibiotics[which(antibiotics$ab == "CPO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CPO"), "abbreviations"][[1]], "cfpr"))
|
|
antibiotics[which(antibiotics$ab == "CFP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CFP"), "abbreviations"][[1]], "cfpz"))
|
|
antibiotics[which(antibiotics$ab == "CED"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CED"), "abbreviations"][[1]], "cfrd"))
|
|
antibiotics[which(antibiotics$ab == "CPT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CPT"), "abbreviations"][[1]], "cfro"))
|
|
antibiotics[which(antibiotics$ab == "CXM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CXM"), "abbreviations"][[1]], "cfrx"))
|
|
antibiotics[which(antibiotics$ab == "CFS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CFS"), "abbreviations"][[1]], "cfsl"))
|
|
antibiotics[which(antibiotics$ab == "CRO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CRO"), "abbreviations"][[1]], "cftr"))
|
|
antibiotics[which(antibiotics$ab == "CTT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTT"), "abbreviations"][[1]], "cftt"))
|
|
antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]], "cftx"))
|
|
antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]], "cftz"))
|
|
antibiotics[which(antibiotics$ab == "CFM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CFM"), "abbreviations"][[1]], "cfxm"))
|
|
antibiotics[which(antibiotics$ab == "FOX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FOX"), "abbreviations"][[1]], "cfxt"))
|
|
antibiotics[which(antibiotics$ab == "CZA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CZA"), "abbreviations"][[1]], "cfav"))
|
|
antibiotics[which(antibiotics$ab == "CZO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CZO"), "abbreviations"][[1]], "cfzl"))
|
|
antibiotics[which(antibiotics$ab == "CZX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CZX"), "abbreviations"][[1]], "cfzx"))
|
|
antibiotics[which(antibiotics$ab == "CHL"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CHL"), "abbreviations"][[1]], "chlo"))
|
|
antibiotics[which(antibiotics$ab == "CPC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CPC"), "abbreviations"][[1]], "cicl"))
|
|
antibiotics[which(antibiotics$ab == "CIN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CIN"), "abbreviations"][[1]], "cino"))
|
|
antibiotics[which(antibiotics$ab == "CIP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CIP"), "abbreviations"][[1]], "cipr"))
|
|
antibiotics[which(antibiotics$ab == "CIX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CIX"), "abbreviations"][[1]], "cipx"))
|
|
antibiotics[which(antibiotics$ab == "CLR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CLR"), "abbreviations"][[1]], "clar"))
|
|
antibiotics[which(antibiotics$ab == "CLI"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CLI"), "abbreviations"][[1]], "clin"))
|
|
antibiotics[which(antibiotics$ab == "CTR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTR"), "abbreviations"][[1]], "clot"))
|
|
antibiotics[which(antibiotics$ab == "CLO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CLO"), "abbreviations"][[1]], "clox"))
|
|
antibiotics[which(antibiotics$ab == "COL"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "COL"), "abbreviations"][[1]], "coli"))
|
|
antibiotics[which(antibiotics$ab == "CTC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTC"), "abbreviations"][[1]], "cxcl"))
|
|
antibiotics[which(antibiotics$ab == "CYC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CYC"), "abbreviations"][[1]], "cycl"))
|
|
antibiotics[which(antibiotics$ab == "CCV"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CCV"), "abbreviations"][[1]], "czcl"))
|
|
antibiotics[which(antibiotics$ab == "DAP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "DAP"), "abbreviations"][[1]], "dapt"))
|
|
antibiotics[which(antibiotics$ab == "DIC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "DIC"), "abbreviations"][[1]], "dicl"))
|
|
antibiotics[which(antibiotics$ab == "DOR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "DOR"), "abbreviations"][[1]], "dori"))
|
|
antibiotics[which(antibiotics$ab == "DOX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "DOX"), "abbreviations"][[1]], "doxy"))
|
|
antibiotics[which(antibiotics$ab == "ENX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ENX"), "abbreviations"][[1]], "enox"))
|
|
antibiotics[which(antibiotics$ab == "ETP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ETP"), "abbreviations"][[1]], "erta"))
|
|
antibiotics[which(antibiotics$ab == "ERY"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ERY"), "abbreviations"][[1]], "eryt"))
|
|
antibiotics[which(antibiotics$ab == "PHE"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PHE"), "abbreviations"][[1]], "fene"))
|
|
antibiotics[which(antibiotics$ab == "PHN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PHN"), "abbreviations"][[1]], "fepe"))
|
|
antibiotics[which(antibiotics$ab == "FLE"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FLE"), "abbreviations"][[1]], "fler"))
|
|
antibiotics[which(antibiotics$ab == "FLU"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FLU"), "abbreviations"][[1]], "fluc"))
|
|
antibiotics[which(antibiotics$ab == "FLC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FLC"), "abbreviations"][[1]], "flux"))
|
|
antibiotics[which(antibiotics$ab == "FOS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FOS"), "abbreviations"][[1]], "fosf"))
|
|
antibiotics[which(antibiotics$ab == "FRM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FRM"), "abbreviations"][[1]], "fram"))
|
|
antibiotics[which(antibiotics$ab == "FUS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FUS"), "abbreviations"][[1]], "fusi"))
|
|
antibiotics[which(antibiotics$ab == "GAT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "GAT"), "abbreviations"][[1]], "gati"))
|
|
antibiotics[which(antibiotics$ab == "GEH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "GEH"), "abbreviations"][[1]], "gehl"))
|
|
antibiotics[which(antibiotics$ab == "GEN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "GEN"), "abbreviations"][[1]], "gent"))
|
|
antibiotics[which(antibiotics$ab == "GRX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "GRX"), "abbreviations"][[1]], "grep"))
|
|
antibiotics[which(antibiotics$ab == "IPM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "IPM"), "abbreviations"][[1]], "imci"))
|
|
antibiotics[which(antibiotics$ab == "IPM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "IPM"), "abbreviations"][[1]], "imip"))
|
|
antibiotics[which(antibiotics$ab == "ISV"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ISV"), "abbreviations"][[1]], "isav"))
|
|
antibiotics[which(antibiotics$ab == "ITR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "ITR"), "abbreviations"][[1]], "itra"))
|
|
antibiotics[which(antibiotics$ab == "KAH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "KAH"), "abbreviations"][[1]], "kahl"))
|
|
antibiotics[which(antibiotics$ab == "KAN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "KAN"), "abbreviations"][[1]], "kana"))
|
|
antibiotics[which(antibiotics$ab == "KET"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "KET"), "abbreviations"][[1]], "keto"))
|
|
antibiotics[which(antibiotics$ab == "LVX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LVX"), "abbreviations"][[1]], "levo"))
|
|
antibiotics[which(antibiotics$ab == "LIN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LIN"), "abbreviations"][[1]], "linc"))
|
|
antibiotics[which(antibiotics$ab == "LNZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LNZ"), "abbreviations"][[1]], "line"))
|
|
antibiotics[which(antibiotics$ab == "LOR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LOR"), "abbreviations"][[1]], "lora"))
|
|
antibiotics[which(antibiotics$ab == "MEM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MEM"), "abbreviations"][[1]], "mero"))
|
|
antibiotics[which(antibiotics$ab == "MET"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MET"), "abbreviations"][[1]], "meti"))
|
|
antibiotics[which(antibiotics$ab == "MTR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MTR"), "abbreviations"][[1]], "metr"))
|
|
antibiotics[which(antibiotics$ab == "MEZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MEZ"), "abbreviations"][[1]], "mezl"))
|
|
antibiotics[which(antibiotics$ab == "MIF"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MIF"), "abbreviations"][[1]], "mica"))
|
|
antibiotics[which(antibiotics$ab == "MCZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MCZ"), "abbreviations"][[1]], "mico"))
|
|
antibiotics[which(antibiotics$ab == "MNO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MNO"), "abbreviations"][[1]], "mino"))
|
|
antibiotics[which(antibiotics$ab == "LTM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "LTM"), "abbreviations"][[1]], "moxa", "moxalactam"))
|
|
antibiotics[which(antibiotics$ab == "MFX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "MFX"), "abbreviations"][[1]], "moxi"))
|
|
antibiotics[which(antibiotics$ab == "NAL"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NAL"), "abbreviations"][[1]], "nali"))
|
|
antibiotics[which(antibiotics$ab == "NEO"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NEO"), "abbreviations"][[1]], "neom"))
|
|
antibiotics[which(antibiotics$ab == "NET"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NET"), "abbreviations"][[1]], "neti"))
|
|
antibiotics[which(antibiotics$ab == "NIT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NIT"), "abbreviations"][[1]], "nitr"))
|
|
antibiotics[which(antibiotics$ab == "NOR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NOR"), "abbreviations"][[1]], "norf"))
|
|
antibiotics[which(antibiotics$ab == "NYS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "NYS"), "abbreviations"][[1]], "nyst"))
|
|
antibiotics[which(antibiotics$ab == "OFX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "OFX"), "abbreviations"][[1]], "oflo"))
|
|
antibiotics[which(antibiotics$ab == "OXA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "OXA"), "abbreviations"][[1]], "oxal"))
|
|
antibiotics[which(antibiotics$ab == "PEF"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PEF"), "abbreviations"][[1]], "pefl"))
|
|
antibiotics[which(antibiotics$ab == "PEN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PEN"), "abbreviations"][[1]], "peni"))
|
|
antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PIP"), "abbreviations"][[1]], "pipc"))
|
|
antibiotics[which(antibiotics$ab == "PPA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PPA"), "abbreviations"][[1]], "pipz"))
|
|
antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]], "pita"))
|
|
antibiotics[which(antibiotics$ab == "PLB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PLB"), "abbreviations"][[1]], "polb"))
|
|
antibiotics[which(antibiotics$ab == "POS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "POS"), "abbreviations"][[1]], "posa"))
|
|
antibiotics[which(antibiotics$ab == "PRI"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "PRI"), "abbreviations"][[1]], "pris"))
|
|
antibiotics[which(antibiotics$ab == "QDA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "QDA"), "abbreviations"][[1]], "quda"))
|
|
antibiotics[which(antibiotics$ab == "RIF"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "RIF"), "abbreviations"][[1]], "rifa"))
|
|
antibiotics[which(antibiotics$ab == "RXT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "RXT"), "abbreviations"][[1]], "roxi"))
|
|
antibiotics[which(antibiotics$ab == "SMX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SMX"), "abbreviations"][[1]], "sfmx"))
|
|
antibiotics[which(antibiotics$ab == "SLF4"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SLF4"), "abbreviations"][[1]], "sfmz"))
|
|
antibiotics[which(antibiotics$ab == "SSS"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SSS"), "abbreviations"][[1]], "sfna"))
|
|
antibiotics[which(antibiotics$ab == "SLF"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SLF"), "abbreviations"][[1]], "sfsz"))
|
|
antibiotics[which(antibiotics$ab == "SPX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SPX"), "abbreviations"][[1]], "spar"))
|
|
antibiotics[which(antibiotics$ab == "SPT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SPT"), "abbreviations"][[1]], "spec"))
|
|
antibiotics[which(antibiotics$ab == "SPI"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SPI"), "abbreviations"][[1]], "spir"))
|
|
antibiotics[which(antibiotics$ab == "STH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "STH"), "abbreviations"][[1]], "sthl"))
|
|
antibiotics[which(antibiotics$ab == "STR1"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "STR1"), "abbreviations"][[1]], "stre"))
|
|
antibiotics[which(antibiotics$ab == "TAZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TAZ"), "abbreviations"][[1]], "tazo"))
|
|
antibiotics[which(antibiotics$ab == "TEC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TEC"), "abbreviations"][[1]], "teic"))
|
|
antibiotics[which(antibiotics$ab == "TLT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TLT"), "abbreviations"][[1]], "teli"))
|
|
antibiotics[which(antibiotics$ab == "TMX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TMX"), "abbreviations"][[1]], "tema"))
|
|
antibiotics[which(antibiotics$ab == "TEM"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TEM"), "abbreviations"][[1]], "temo"))
|
|
antibiotics[which(antibiotics$ab == "TRB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TRB"), "abbreviations"][[1]], "terb"))
|
|
antibiotics[which(antibiotics$ab == "TCY"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TCY"), "abbreviations"][[1]], "tetr"))
|
|
antibiotics[which(antibiotics$ab == "TIC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TIC"), "abbreviations"][[1]], "tica"))
|
|
antibiotics[which(antibiotics$ab == "TCC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TCC"), "abbreviations"][[1]], "ticl"))
|
|
antibiotics[which(antibiotics$ab == "TGC"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TGC"), "abbreviations"][[1]], "tige"))
|
|
antibiotics[which(antibiotics$ab == "TIN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TIN"), "abbreviations"][[1]], "tini"))
|
|
antibiotics[which(antibiotics$ab == "TOB"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TOB"), "abbreviations"][[1]], "tobr"))
|
|
antibiotics[which(antibiotics$ab == "TOH"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TOH"), "abbreviations"][[1]], "tohl"))
|
|
antibiotics[which(antibiotics$ab == "TMP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TMP"), "abbreviations"][[1]], "trim"))
|
|
antibiotics[which(antibiotics$ab == "TVA"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "TVA"), "abbreviations"][[1]], "trov"))
|
|
antibiotics[which(antibiotics$ab == "SLT4"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SLT4"), "abbreviations"][[1]], "trsm"))
|
|
antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "SXT"), "abbreviations"][[1]], "trsx"))
|
|
antibiotics[which(antibiotics$ab == "VAN"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "VAN"), "abbreviations"][[1]], "vanc"))
|
|
antibiotics[which(antibiotics$ab == "VOR"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "VOR"), "abbreviations"][[1]], "vori"))
|
|
|
|
antibiotics[which(antibiotics$ab == "FOS"), "synonyms"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "FOS"), "synonyms"][[1]], "Monuril")))
|
|
antibiotics[which(antibiotics$ab == "FOS"), "synonyms"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "FOS"), "synonyms"][[1]], "Monurol")))
|
|
|
|
antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "TZP"), "abbreviations"][[1]], "piptazo")))
|
|
|
|
antibiotics[which(antibiotics$ab == "RFP"), "abbreviations"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "RFP"), "abbreviations"][[1]], "RPT")))
|
|
antibiotics[which(antibiotics$ab == "RTP"), "abbreviations"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "RTP"), "abbreviations"][[1]], "RET")))
|
|
antibiotics[which(antibiotics$ab == "TYL1"), "abbreviations"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "TYL1"), "abbreviations"][[1]], "TVN")))
|
|
|
|
antibiotics <- antibiotics %>%
|
|
mutate(ab = as.character(ab)) %>%
|
|
rbind(antibiotics %>%
|
|
filter(ab == "GEH") %>%
|
|
mutate(
|
|
ab = "AMH",
|
|
name = "Amphotericin B-high",
|
|
abbreviations = list(c("amhl", "amfo b high", "ampho b high", "amphotericin high"))
|
|
)) %>%
|
|
rbind(antibiotics %>%
|
|
filter(ab == "GEH") %>%
|
|
mutate(
|
|
ab = "TOH",
|
|
name = "Tobramycin-high",
|
|
abbreviations = list(c("tohl", "tobra high", "tobramycin high"))
|
|
)) %>%
|
|
rbind(antibiotics %>%
|
|
filter(ab == "BUT") %>%
|
|
mutate(
|
|
ab = "CIX",
|
|
atc = "D01AE14",
|
|
name = "Ciclopirox",
|
|
group = "Antifungals/antimycotics",
|
|
atc_group1 = "Antifungals for topical use",
|
|
atc_group2 = "Other antifungals for topical use",
|
|
abbreviations = list(c("cipx"))
|
|
))
|
|
antibiotics[which(antibiotics$ab == "SSS"), "name"] <- "Sulfonamide"
|
|
# ESBL E-test codes:
|
|
antibiotics[which(antibiotics$ab == "CCV"), "abbreviations"][[1]] <- list(c("xtzl"))
|
|
antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CAZ"), "abbreviations"][[1]], "xtz", "cefta"))
|
|
antibiotics[which(antibiotics$ab == "CPC"), "abbreviations"][[1]] <- list(c("xpml"))
|
|
antibiotics[which(antibiotics$ab == "FEP"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "FEP"), "abbreviations"][[1]], "xpm"))
|
|
antibiotics[which(antibiotics$ab == "CTC"), "abbreviations"][[1]] <- list(c("xctl"))
|
|
antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]] <- list(c(antibiotics[which(antibiotics$ab == "CTX"), "abbreviations"][[1]], "xct"))
|
|
# High level Gentamcin and Streptomycin
|
|
antibiotics[which(antibiotics$ab == "GEH"), "abbreviations"][[1]] <- list(c("gehl", "gentamicin high", "genta high", "gehi"))
|
|
antibiotics[which(antibiotics$ab == "STH"), "abbreviations"][[1]] <- list(c("sthl", "streptomycin high", "strepto high", "sthi"))
|
|
# add imi and "imipenem/cilastatine" to imipenem
|
|
antibiotics[which(antibiotics$ab == "IPM"), "abbreviations"][[1]] <- list(c("imip", "imi", "imp"))
|
|
antibiotics[which(antibiotics$ab == "IPM"), "synonyms"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "IPM"), "synonyms"][[1]], "imipenem/cilastatin")))
|
|
# add synonyms of ones not found
|
|
antibiotics[which(antibiotics$ab == "TZP"), "synonyms"][[1]] <- list(sort(c(antibiotics[which(antibiotics$ab == "TZP"), "synonyms"][[1]], "Tazocel", "tazocillin", "Tazocin", "Zosyn")))
|
|
antibiotics[which(antibiotics$ab == "COL"), "synonyms"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "COL"), "synonyms"][[1]], "Colisticin", "Polymyxin E", "Colimycin", "Coly-Mycin", "Totazina", "Colistimethate", "Promixin", "Colistimethate Sodium"))))
|
|
# remove incorrect synonyms from rifampicin (RIF) and add them to the combination rifampicin/isoniazid (RFI)
|
|
old_sym <- antibiotics[which(antibiotics$ab == "RIF"), "synonyms"][[1]]
|
|
old_sym <- old_sym[!old_sym %in% c("Rifinah", "Rimactazid")]
|
|
antibiotics[which(antibiotics$ab == "RIF"), "synonyms"][[1]] <- list(old_sym)
|
|
antibiotics[which(antibiotics$ab == "RFI"), "synonyms"][[1]] <- list(sort(c("Rifinah", "Rimactazid")))
|
|
# remove incorrect synonyms from sulfamethoxazole (SMX) and add them to the combination trimethoprim/sulfamethoxazole (SXT)
|
|
old_sym <- antibiotics[which(antibiotics$ab == "SMX"), "synonyms"][[1]]
|
|
old_sym <- old_sym[!old_sym %in% c("Cotrimoxazole", "Bactrimel")]
|
|
antibiotics[which(antibiotics$ab == "SMX"), "synonyms"][[1]] <- list(old_sym)
|
|
antibiotics[which(antibiotics$ab == "SXT"), "synonyms"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "COL"), "synonyms"][[1]], "Cotrimoxazole", "Bactrimel", "Septra", "Bactrim", "Cotrimazole"))))
|
|
|
|
# Fix penicillins
|
|
antibiotics[which(antibiotics$ab == "PEN"), "abbreviations"][[1]] <- list(c("bepe", "pg", "pen", "peni", "peni g", "penicillin", "penicillin g"))
|
|
antibiotics[which(antibiotics$ab == "PEN"), "name"] <- "Benzylpenicillin"
|
|
antibiotics[which(antibiotics$ab == "PHN"), "abbreviations"][[1]] <- list(c("fepe", "peni v", "pv", "penicillin v", "PNV"))
|
|
antibiotics <- subset(antibiotics, antibiotics$ab != "PNV")
|
|
|
|
# New DDDs
|
|
antibiotics[which(antibiotics$ab == "PEN"), "iv_ddd"] <- 3.6
|
|
antibiotics[which(antibiotics$ab == "PEN"), "iv_units"] <- "g"
|
|
|
|
## new ATC codes
|
|
# ceftaroline
|
|
antibiotics[which(antibiotics$ab == "CPT"), "atc"] <- "J01DI02"
|
|
# faropenem
|
|
antibiotics[which(antibiotics$ab == "FAR"), "atc"] <- "J01DI03"
|
|
# ceftobiprole
|
|
antibiotics[which(antibiotics$ab == "BPR"), "atc"] <- "J01DI01"
|
|
# ceftazidime / avibactam
|
|
antibiotics[which(antibiotics$ab == "CZA"), "atc"] <- "J01DD52"
|
|
antibiotics[which(antibiotics$ab == "CZA"), "cid"] <- 90643431
|
|
antibiotics[which(antibiotics$ab == "CZA"), "atc_group1"] <- "Other beta-lactam antibacterials"
|
|
antibiotics[which(antibiotics$ab == "CZA"), "atc_group2"] <- "Third-generation cephalosporins"
|
|
antibiotics[which(antibiotics$ab == "CZA"), "iv_ddd"] <- 6
|
|
antibiotics[which(antibiotics$ab == "CZA"), "iv_units"] <- "g"
|
|
antibiotics[which(antibiotics$ab == "CZA"), "synonyms"] <- list(c("Avycaz", "Zavicefta"))
|
|
|
|
# typo
|
|
antibiotics[which(antibiotics$ab == "RXT"), "name"] <- "Roxithromycin"
|
|
antibiotics[which(antibiotics$ab == "PEN"), "atc"] <- "J01CE01"
|
|
|
|
# WHONET cleanup
|
|
antibiotics[which(antibiotics$ab == "BCZ"), "name"] <- "Bicyclomycin"
|
|
antibiotics[which(antibiotics$ab == "CCL"), "name"] <- "Cefetecol"
|
|
antibiotics[which(antibiotics$ab == "ENV"), "name"] <- "Enviomycin"
|
|
antibiotics[which(antibiotics$ab == "KIT"), "name"] <- "Kitasamycin"
|
|
antibiotics[which(antibiotics$ab == "LSP"), "name"] <- "Linco-spectin"
|
|
antibiotics[which(antibiotics$ab == "MEC"), "name"] <- "Mecillinam"
|
|
antibiotics[which(antibiotics$ab == "PMR"), "name"] <- "Pimaricin"
|
|
antibiotics[which(antibiotics$ab == "BCZ"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "BCZ"), "abbreviations"][[1]], "Bicozamycin"))))
|
|
antibiotics[which(antibiotics$ab == "CCL"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "CCL"), "abbreviations"][[1]], "Cefcatacol"))))
|
|
antibiotics[which(antibiotics$ab == "ENV"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "ENV"), "abbreviations"][[1]], "Tuberactinomycin"))))
|
|
antibiotics[which(antibiotics$ab == "KIT"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "KIT"), "abbreviations"][[1]], "Leucomycin"))))
|
|
antibiotics[which(antibiotics$ab == "LSP"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "LSP"), "abbreviations"][[1]], "lincomycin/spectinomycin"))))
|
|
antibiotics[which(antibiotics$ab == "MEC"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "MEC"), "abbreviations"][[1]], "Amdinocillin"))))
|
|
antibiotics[which(antibiotics$ab == "PMR"), "abbreviations"][[1]] <- list(sort(unique(c(antibiotics[which(antibiotics$ab == "PMR"), "abbreviations"][[1]], "Natamycin"))))
|
|
|
|
|
|
# set cephalosporins groups for the ones that could not be determined automatically:
|
|
antibiotics <- antibiotics %>%
|
|
mutate(group = case_when(
|
|
name == "Cefcapene" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefcapene pivoxil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefditoren pivoxil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefepime/clavulanic acid" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefepime/tazobactam" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefetamet pivoxil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefetecol (Cefcatacol)" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefetrizole" ~ "Cephalosporins (unclassified gen.)",
|
|
name == "Cefoselis" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefotaxime/clavulanic acid" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefotaxime/sulbactam" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefotiam hexetil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefovecin" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefozopran" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefpimizole" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefpodoxime proxetil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefpodoxime/clavulanic acid" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefquinome" ~ "Cephalosporins (4th gen.)",
|
|
name == "Cefsumide" ~ "Cephalosporins (unclassified gen.)",
|
|
name == "Ceftaroline" ~ "Cephalosporins (5th gen.)",
|
|
name == "Ceftaroline/avibactam" ~ "Cephalosporins (5th gen.)",
|
|
name == "Ceftazidime/avibactam" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefteram" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Cefteram pivoxil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Ceftiofur" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Ceftizoxime alapivoxil" ~ "Cephalosporins (3rd gen.)",
|
|
name == "Ceftobiprole" ~ "Cephalosporins (5th gen.)",
|
|
name == "Ceftobiprole medocaril" ~ "Cephalosporins (5th gen.)",
|
|
name == "Ceftolozane/enzyme inhibitor" ~ "Cephalosporins (5th gen.)",
|
|
name == "Ceftolozane/tazobactam" ~ "Cephalosporins (5th gen.)",
|
|
name == "Cefuroxime axetil" ~ "Cephalosporins (2nd gen.)",
|
|
TRUE ~ group
|
|
))
|
|
antibiotics[which(antibiotics$ab %in% c("CYC", "LNZ", "THA", "TZD")), "group"] <- "Oxazolidinones"
|
|
|
|
# add clindamycin inducible screening
|
|
clin <- antibiotics |>
|
|
filter(ab == "FOX1") |>
|
|
mutate(ab = as.character("CLI1"),
|
|
name = "Clindamycin inducible screening",
|
|
group = "Macrolides/lincosamides")
|
|
antibiotics <- antibiotics |>
|
|
mutate(ab = as.character(ab)) |>
|
|
bind_rows(clin)
|
|
class(antibiotics$ab) <- c("ab", "character")
|
|
antibiotics[which(antibiotics$ab == "CLI1"), "abbreviations"][[1]] <- list(c("clindamycin inducible", "clinda inducible", "clin inducible"))
|
|
|
|
# add pretomanid
|
|
antibiotics <- antibiotics %>%
|
|
mutate(ab = as.character(ab)) %>%
|
|
bind_rows(antibiotics %>%
|
|
mutate(ab = as.character(ab)) %>%
|
|
filter(ab == "SMF") %>%
|
|
mutate(
|
|
ab = "PMD",
|
|
atc = "J04AK08",
|
|
cid = 456199,
|
|
name = "Pretomanid",
|
|
abbreviations = list(""),
|
|
oral_ddd = NA_real_
|
|
))
|
|
|
|
|
|
|
|
# update ATC codes from WHOCC website -------------------------------------
|
|
|
|
# last time checked: 2024-02-22
|
|
|
|
library(rvest)
|
|
updated_atc <- as.list(antibiotics$atc)
|
|
|
|
get_atcs <- function(ab_name, type = "human") {
|
|
if (type == "human") {
|
|
url <- "https://atcddd.fhi.no/atc_ddd_index/"
|
|
} else if (type == "veterinary") {
|
|
url <- "https://atcddd.fhi.no/atcvet/atcvet_index/"
|
|
} else {
|
|
stop("invalid type")
|
|
}
|
|
|
|
ab_name <- gsub("/", " and ", tolower(ab_name), fixed = TRUE)
|
|
|
|
# we will do a search on their website, which means:
|
|
|
|
# go to the url
|
|
atc_tbl <- read_html(url) %>%
|
|
# get all forms
|
|
html_form() %>%
|
|
# get the second form (the first form is a global website form)
|
|
.[[2]] %>%
|
|
# set the name input box to our search parameter
|
|
html_form_set(name = ab_name) %>%
|
|
# hit Submit
|
|
html_form_submit() %>%
|
|
# read the resulting page
|
|
read_html() %>%
|
|
# retrieve the table on it
|
|
html_node("table") %>%
|
|
# transform it to an R data set
|
|
html_table(header = FALSE)
|
|
|
|
# and get the ATCs (first column) of only exact hits
|
|
unique(as.character(atc_tbl[which(tolower(atc_tbl[, 2, drop = TRUE]) == ab_name), 1, drop = TRUE]))
|
|
}
|
|
|
|
# this takes around 4 minutes (some are skipped and go faster)
|
|
for (i in seq_len(nrow(antibiotics))) {
|
|
message(percentage(i / nrow(antibiotics), digits = 1),
|
|
" - Downloading ", antibiotics$name[i],
|
|
appendLF = FALSE
|
|
)
|
|
atcs <- get_atcs(antibiotics$name[i], type = "human")
|
|
if (all(is.na(atcs))) {
|
|
atcs <- get_atcs(antibiotics$name[i], type = "veterinary")
|
|
}
|
|
if (length(atcs) > 0) {
|
|
updated_atc[[i]] <- atcs
|
|
message(" (", length(atcs), " results)")
|
|
# let the WHO server rest for a second - they might have a limitation on the queries per second
|
|
Sys.sleep(1)
|
|
} else {
|
|
message(" (skipping)")
|
|
}
|
|
}
|
|
|
|
antibiotics$atc <- updated_atc
|
|
|
|
# update DDDs from WHOCC website ------------------------------------------
|
|
|
|
# last time checked: 2024-02-22
|
|
ddd_oral <- rep(NA_real_, nrow(antibiotics))
|
|
ddd_oral_units <- rep(NA_character_, nrow(antibiotics))
|
|
ddd_iv <- rep(NA_real_, nrow(antibiotics))
|
|
ddd_iv_units <- rep(NA_character_, nrow(antibiotics))
|
|
progress <- progress_ticker(nrow(antibiotics))
|
|
for (i in seq_len(nrow(antibiotics))) {
|
|
on.exit(close(progress))
|
|
progress$tick()
|
|
atcs <- antibiotics$atc[[i]]
|
|
if (!all(is.na(atcs))) {
|
|
for (j in seq_len(length(atcs))) {
|
|
# oral
|
|
if (is.na(ddd_oral[i])) {
|
|
ddd_oral[i] <- atc_online_ddd(atcs[j], administration = "O")
|
|
if (!is.na(ddd_oral[i])) {
|
|
ddd_oral_units[i] <- atc_online_ddd_units(atcs[j], administration = "O")
|
|
}
|
|
}
|
|
# parenteral
|
|
if (is.na(ddd_iv[i])) {
|
|
ddd_iv[i] <- atc_online_ddd(atcs[j], administration = "P")
|
|
if (!is.na(ddd_iv[i])) {
|
|
ddd_iv_units[i] <- atc_online_ddd_units(atcs[j], administration = "P")
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (!is.na(ddd_oral[i]) | !is.na(ddd_iv[i])) {
|
|
# let the WHO server rest for 0.25 second - they might have a limitation on the queries per second
|
|
Sys.sleep(0.25)
|
|
}
|
|
}
|
|
|
|
antibiotics$oral_ddd <- ddd_oral
|
|
antibiotics$oral_units <- ddd_oral_units
|
|
antibiotics$iv_ddd <- ddd_iv
|
|
antibiotics$iv_units <- ddd_iv_units
|
|
|
|
# Wrap up -----------------------------------------------------------------
|
|
|
|
# set as data.frame again
|
|
antibiotics <- dataset_UTF8_to_ASCII(as.data.frame(antibiotics, stringsAsFactors = FALSE))
|
|
class(antibiotics$ab) <- c("ab", "character")
|
|
antibiotics <- dplyr::arrange(antibiotics, name)
|
|
|
|
# REFER TO data-raw/loinc.R FOR ADDING LOINC CODES
|
|
|
|
# make all abbreviations and synonyms lower case, unique and alphabetically sorted ----
|
|
for (i in 1:nrow(antibiotics)) {
|
|
abb <- as.character(sort(unique(tolower(antibiotics[i, "abbreviations", drop = TRUE][[1]]))))
|
|
abb <- abb[abb != "" & abb %unlike% ":"]
|
|
syn <- as.character(sort(unique(tolower(antibiotics[i, "synonyms", drop = TRUE][[1]]))))
|
|
syn <- syn[syn != "" & syn %unlike% ":"]
|
|
antibiotics[i, "abbreviations"][[1]] <- ifelse(length(abb) == 0, list(""), list(abb))
|
|
antibiotics[i, "synonyms"][[1]] <- ifelse(length(syn) == 0, list(""), list(syn))
|
|
if ("loinc" %in% colnames(antibiotics)) {
|
|
loinc <- as.character(sort(unique(tolower(antibiotics[i, "loinc", drop = TRUE][[1]]))))
|
|
loinc <- loinc[loinc != ""]
|
|
antibiotics[i, "loinc"][[1]] <- ifelse(length(loinc) == 0, list(""), list(loinc))
|
|
}
|
|
}
|
|
|
|
|
|
usethis::use_data(antibiotics, overwrite = TRUE, version = 2, compress = "xz")
|
|
rm(antibiotics)
|