mirror of https://github.com/msberends/AMR.git
398 lines
19 KiB
R
398 lines
19 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/ #
|
||
# ==================================================================== #
|
||
|
||
# This script runs in 20-30 minutes and renews all guidelines of CLSI and EUCAST!
|
||
# Run it with source("data-raw/reproduction_of_clinical_breakpoints.R")
|
||
|
||
library(dplyr)
|
||
library(readr)
|
||
library(tidyr)
|
||
devtools::load_all()
|
||
|
||
# Install the WHONET software on Windows (http://www.whonet.org/software.html),
|
||
# and copy the folder C:\WHONET\Resources to the data-raw/WHONET/ folder
|
||
# (for ASIARS-Net update, also copy C:\WHONET\Codes to the data-raw/WHONET/ folder)
|
||
|
||
# BE SURE TO RUN data-raw/reproduction_of_microorganisms.groups.R FIRST TO GET THE GROUPS!
|
||
|
||
# READ DATA ----
|
||
|
||
whonet_organisms <- read_tsv("data-raw/WHONET/Resources/Organisms.txt", na = c("", "NA", "-"), show_col_types = FALSE) %>%
|
||
# remove old taxonomic names
|
||
filter(TAXONOMIC_STATUS == "C") %>%
|
||
mutate(ORGANISM_CODE = toupper(WHONET_ORG_CODE))
|
||
|
||
whonet_breakpoints <- read_tsv("data-raw/WHONET/Resources/Breakpoints.txt", na = c("", "NA", "-"),
|
||
show_col_types = FALSE, guess_max = Inf) %>%
|
||
filter(GUIDELINES %in% c("CLSI", "EUCAST"))
|
||
|
||
whonet_antibiotics <- read_tsv("data-raw/WHONET/Resources/Antibiotics.txt", na = c("", "NA", "-"), show_col_types = FALSE) %>%
|
||
arrange(WHONET_ABX_CODE) %>%
|
||
distinct(WHONET_ABX_CODE, .keep_all = TRUE)
|
||
|
||
# MICROORGANISMS WHONET CODES ----
|
||
|
||
whonet_organisms <- whonet_organisms %>%
|
||
select(ORGANISM_CODE, ORGANISM, SPECIES_GROUP, GBIF_TAXON_ID) %>%
|
||
mutate(
|
||
# this one was called Issatchenkia orientalis, but it should be:
|
||
ORGANISM = if_else(ORGANISM_CODE == "ckr", "Candida krusei", ORGANISM)
|
||
) %>%
|
||
# try to match on GBIF identifier
|
||
left_join(microorganisms %>% distinct(mo, gbif, status) %>% filter(!is.na(gbif)), by = c("GBIF_TAXON_ID" = "gbif")) %>%
|
||
# remove duplicates
|
||
arrange(ORGANISM_CODE, GBIF_TAXON_ID, status) %>%
|
||
distinct(ORGANISM_CODE, .keep_all = TRUE) %>%
|
||
# add Enterobacterales, which is a subkingdom code in their data
|
||
bind_rows(data.frame(ORGANISM_CODE = "ebc", ORGANISM = "Enterobacterales", mo = as.mo("Enterobacterales"))) %>%
|
||
arrange(ORGANISM)
|
||
|
||
|
||
## Add new WHO codes to microorganisms.codes ----
|
||
|
||
matched <- whonet_organisms %>% filter(!is.na(mo))
|
||
unmatched <- whonet_organisms %>% filter(is.na(mo))
|
||
|
||
# generate the mo codes and add their names
|
||
message("Getting MO codes for WHONET input...")
|
||
unmatched <- unmatched %>%
|
||
mutate(mo = as.mo(gsub("(sero[a-z]*| nontypable| non[-][a-zA-Z]+|var[.]| not .*|sp[.],.*|, .*variant.*|, .*toxin.*|, microaer.*| beta-haem[.])", "", ORGANISM),
|
||
minimum_matching_score = 0.55,
|
||
keep_synonyms = TRUE,
|
||
language = "en"),
|
||
mo = case_when(ORGANISM %like% "Anaerobic" & ORGANISM %like% "negative" ~ as.mo("B_ANAER-NEG"),
|
||
ORGANISM %like% "Anaerobic" & ORGANISM %like% "positive" ~ as.mo("B_ANAER-POS"),
|
||
ORGANISM %like% "Anaerobic" ~ as.mo("B_ANAER"),
|
||
TRUE ~ mo),
|
||
mo_name = mo_name(mo,
|
||
keep_synonyms = TRUE,
|
||
language = "en"))
|
||
# check if coercion at least resembles the first part (genus)
|
||
unmatched <- unmatched %>%
|
||
mutate(
|
||
first_part = sapply(ORGANISM, function(x) strsplit(gsub("[^a-zA-Z _-]+", "", x), " ")[[1]][1], USE.NAMES = FALSE),
|
||
keep = mo_name %like_case% first_part | ORGANISM %like% "Gram " | ORGANISM == "Other" | ORGANISM %like% "anaerobic") %>%
|
||
arrange(keep)
|
||
unmatched %>%
|
||
View()
|
||
unmatched <- unmatched %>%
|
||
filter(keep == TRUE)
|
||
|
||
organisms <- matched %>% transmute(code = toupper(ORGANISM_CODE), group = SPECIES_GROUP, mo) %>%
|
||
bind_rows(unmatched %>% transmute(code = toupper(ORGANISM_CODE), group = SPECIES_GROUP, mo)) %>%
|
||
mutate(name = mo_name(mo, keep_synonyms = TRUE)) %>%
|
||
arrange(code)
|
||
|
||
# some subspecies exist, while their upper species do not, add them as the species level:
|
||
subspp <- organisms %>%
|
||
filter(mo_species(mo, keep_synonyms = TRUE) == mo_subspecies(mo, keep_synonyms = TRUE) &
|
||
mo_species(mo, keep_synonyms = TRUE) != "" &
|
||
mo_genus(mo, keep_synonyms = TRUE) != "Salmonella") %>%
|
||
mutate(mo = as.mo(paste(mo_genus(mo, keep_synonyms = TRUE),
|
||
mo_species(mo, keep_synonyms = TRUE)),
|
||
keep_synonyms = TRUE),
|
||
name = mo_name(mo, keep_synonyms = TRUE))
|
||
organisms <- organisms %>%
|
||
filter(!code %in% subspp$code) %>%
|
||
bind_rows(subspp) %>%
|
||
arrange(code)
|
||
|
||
# add the groups
|
||
organisms <- organisms %>%
|
||
bind_rows(tibble(code = organisms %>% filter(!is.na(group)) %>% pull(group) %>% unique(),
|
||
group = NA,
|
||
mo = organisms %>% filter(!is.na(group)) %>% pull(group) %>% unique() %>% as.mo(keep_synonyms = TRUE),
|
||
name = mo_name(mo, keep_synonyms = TRUE))) %>%
|
||
arrange(code, group) %>%
|
||
select(-group) %>%
|
||
distinct()
|
||
|
||
# 2023-07-08 SGM is also Strep gamma in WHONET, must only be Slowly-growing Mycobacterium
|
||
# 2024-06-14 still the case
|
||
organisms <- organisms %>%
|
||
filter(!(code == "SGM" & name %like% "Streptococcus"))
|
||
# this must be empty:
|
||
organisms$code[organisms$code %>% duplicated()]
|
||
|
||
saveRDS(organisms, "data-raw/organisms.rds", version = 2)
|
||
|
||
#---
|
||
# AT THIS POINT, `organisms` is clean and all entries have an mo code
|
||
#---
|
||
|
||
# update microorganisms.codes with the latest WHONET codes
|
||
microorganisms.codes2 <- microorganisms.codes %>%
|
||
# remove all old WHONET codes, whether we (in the end) keep them or not
|
||
filter(!toupper(code) %in% toupper(organisms$code)) %>%
|
||
# and add the new ones
|
||
bind_rows(organisms %>% select(code, mo)) %>%
|
||
arrange(code) %>%
|
||
distinct(code, .keep_all = TRUE)
|
||
# new codes:
|
||
microorganisms.codes2$code[which(!microorganisms.codes2$code %in% microorganisms.codes$code)]
|
||
mo_name(microorganisms.codes2$mo[which(!microorganisms.codes2$code %in% microorganisms.codes$code)], keep_synonyms = TRUE)
|
||
microorganisms.codes <- microorganisms.codes2
|
||
|
||
# Run this part to update ASIARS-Net:
|
||
# 2024-06-14: file not available anymore
|
||
# # start
|
||
# asiarsnet <- read_tsv("data-raw/WHONET/Codes/ASIARS_Net_Organisms_ForwardLookup.txt")
|
||
# asiarsnet <- asiarsnet %>%
|
||
# mutate(WHONET_Code = toupper(WHONET_Code)) %>%
|
||
# left_join(whonet_organisms %>% mutate(WHONET_Code = toupper(ORGANISM_CODE))) %>%
|
||
# mutate(
|
||
# mo1 = as.mo(ORGANISM_CODE),
|
||
# mo2 = as.mo(ORGANISM)
|
||
# ) %>%
|
||
# mutate(mo = if_else(mo2 == "UNKNOWN" | is.na(mo2), mo1, mo2)) %>%
|
||
# filter(!is.na(mo))
|
||
# insert1 <- asiarsnet %>% transmute(code = WHONET_Code, mo)
|
||
# insert2 <- asiarsnet %>% transmute(code = as.character(ASIARS_Net_Code), mo)
|
||
# # these will be updated
|
||
# bind_rows(insert1, insert2) %>%
|
||
# rename(mo_new = mo) %>%
|
||
# left_join(microorganisms.codes) %>%
|
||
# filter(mo != mo_new)
|
||
# microorganisms.codes <- microorganisms.codes %>%
|
||
# filter(!code %in% c(insert1$code, insert2$code)) %>%
|
||
# bind_rows(insert1, insert2) %>%
|
||
# arrange(code)
|
||
# # end
|
||
|
||
## Save to package ----
|
||
class(microorganisms.codes$mo) <- c("mo", "character")
|
||
usethis::use_data(microorganisms.codes, overwrite = TRUE, compress = "xz", version = 2)
|
||
rm(microorganisms.codes)
|
||
devtools::load_all()
|
||
|
||
|
||
# BREAKPOINTS ----
|
||
|
||
# now that we have the correct MO codes, get the breakpoints and convert them
|
||
|
||
whonet_breakpoints %>%
|
||
count(GUIDELINES, BREAKPOINT_TYPE) %>%
|
||
pivot_wider(names_from = BREAKPOINT_TYPE, values_from = n) %>%
|
||
janitor::adorn_totals(where = c("row", "col"))
|
||
# compared to current
|
||
AMR::clinical_breakpoints %>%
|
||
count(GUIDELINES = gsub("[^a-zA-Z]", "", guideline), type) %>%
|
||
arrange(tolower(type)) %>%
|
||
pivot_wider(names_from = type, values_from = n) %>%
|
||
as.data.frame() %>%
|
||
janitor::adorn_totals(where = c("row", "col"))
|
||
|
||
breakpoints <- whonet_breakpoints %>%
|
||
mutate(code = toupper(ORGANISM_CODE)) %>%
|
||
left_join(bind_rows(microorganisms.codes %>% filter(!code %in% c("ALL", "GEN")),
|
||
# GEN (Generic) and ALL (All) are PK/PD codes
|
||
data.frame(code = c("ALL", "GEN"),
|
||
mo = rep(as.mo("UNKNOWN"), 2))))
|
||
# these ones lack an MO name, they cannot be used:
|
||
unknown <- breakpoints %>%
|
||
filter(is.na(mo)) %>%
|
||
pull(code) %>%
|
||
unique()
|
||
breakpoints %>%
|
||
filter(code %in% unknown) %>%
|
||
count(GUIDELINES, YEAR, ORGANISM_CODE, BREAKPOINT_TYPE, sort = TRUE)
|
||
# 2024-06-14: these codes are currently: clu, kma, fso, tyi. No clue (are not in MO list of WHONET), and they are only ECOFFs, so remove them:
|
||
breakpoints <- breakpoints %>%
|
||
filter(!is.na(mo))
|
||
|
||
# and these ones have unknown antibiotics according to WHONET itself:
|
||
breakpoints %>%
|
||
filter(!WHONET_ABX_CODE %in% whonet_antibiotics$WHONET_ABX_CODE) %>%
|
||
count(YEAR, GUIDELINES, WHONET_ABX_CODE) %>%
|
||
arrange(desc(YEAR))
|
||
breakpoints %>%
|
||
filter(!WHONET_ABX_CODE %in% whonet_antibiotics$WHONET_ABX_CODE) %>%
|
||
pull(WHONET_ABX_CODE) %>%
|
||
unique()
|
||
# they are at the moment all old codes that have the right replacements in `antibiotics`, so we can use as.ab()
|
||
|
||
|
||
## Build new breakpoints table ----
|
||
|
||
breakpoints_new <- breakpoints %>%
|
||
filter(!is.na(WHONET_ABX_CODE)) %>%
|
||
transmute(
|
||
guideline = paste(GUIDELINES, YEAR),
|
||
type = ifelse(BREAKPOINT_TYPE == "ECOFF", "ECOFF", tolower(BREAKPOINT_TYPE)),
|
||
host = ifelse(BREAKPOINT_TYPE == "ECOFF", "ECOFF", tolower(HOST)),
|
||
method = TEST_METHOD,
|
||
site = SITE_OF_INFECTION,
|
||
mo,
|
||
rank_index = case_when(
|
||
is.na(mo_rank(mo, keep_synonyms = TRUE)) ~ 6, # for UNKNOWN, B_GRAMN, B_ANAER, B_ANAER-NEG, etc.
|
||
mo_rank(mo, keep_synonyms = TRUE) %like% "(infra|sub)" ~ 1,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "species" ~ 2,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "species group" ~ 2.5,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "genus" ~ 3,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "family" ~ 4,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "order" ~ 5,
|
||
TRUE ~ 6
|
||
),
|
||
ab = as.ab(WHONET_ABX_CODE),
|
||
ref_tbl = ifelse(type == "ECOFF" & is.na(REFERENCE_TABLE), "ECOFF", REFERENCE_TABLE),
|
||
disk_dose = POTENCY,
|
||
breakpoint_S = ifelse(type == "ECOFF" & is.na(S) & !is.na(ECV_ECOFF), ECV_ECOFF, S),
|
||
breakpoint_R = ifelse(type == "ECOFF" & is.na(R) & !is.na(ECV_ECOFF), ECV_ECOFF, R),
|
||
uti = ifelse(is.na(site), FALSE, gsub(".*(UTI|urinary|urine).*", "UTI", site) == "UTI"),
|
||
is_SDD = !is.na(SDD)
|
||
) %>%
|
||
# Greek symbols and EM dash symbols are not allowed by CRAN, so replace them with ASCII:
|
||
mutate(disk_dose = disk_dose %>%
|
||
gsub("μ", "u", ., fixed = TRUE) %>% # this is 'mu', \u03bc
|
||
gsub("µ", "u", ., fixed = TRUE) %>% # this is 'micro', u00b5 (yes, they look the same)
|
||
gsub("–", "-", ., fixed = TRUE)) %>%
|
||
arrange(desc(guideline), mo, ab, type, method) %>%
|
||
filter(!(is.na(breakpoint_S) & is.na(breakpoint_R)) & !is.na(mo) & !is.na(ab)) %>%
|
||
distinct(guideline, type, host, ab, mo, method, site, breakpoint_S, .keep_all = TRUE)
|
||
|
||
# fix reference table names
|
||
breakpoints_new %>% filter(guideline %like% "EUCAST", is.na(ref_tbl)) %>% View()
|
||
breakpoints_new <- breakpoints_new %>%
|
||
mutate(ref_tbl = case_when(is.na(ref_tbl) & guideline %like% "EUCAST 202" ~ lead(ref_tbl),
|
||
is.na(ref_tbl) ~ "Unknown",
|
||
TRUE ~ ref_tbl))
|
||
|
||
# clean disk zones
|
||
breakpoints_new[which(breakpoints_new$method == "DISK"), "breakpoint_S"] <- as.double(as.disk(breakpoints_new[which(breakpoints_new$method == "DISK"), "breakpoint_S", drop = TRUE]))
|
||
breakpoints_new[which(breakpoints_new$method == "DISK"), "breakpoint_R"] <- as.double(as.disk(breakpoints_new[which(breakpoints_new$method == "DISK"), "breakpoint_R", drop = TRUE]))
|
||
|
||
# regarding animal breakpoints, CLSI has adults and foals for horses, but only for amikacin - only keep adult horses
|
||
breakpoints_new %>%
|
||
filter(host %like% "foal") %>%
|
||
View()
|
||
breakpoints_new <- breakpoints_new %>%
|
||
filter(host %unlike% "foal") %>%
|
||
mutate(host = ifelse(host %like% "horse", "horse", host))
|
||
|
||
# FIXES FOR WHONET ERRORS ----
|
||
|
||
# WHONET has no >1024 but instead uses 1025, 513, etc, so as.mic() cannot be used to clean.
|
||
# instead, clean based on MIC factor levels
|
||
m <- unique(as.double(as.mic(levels(as.mic(1)))))
|
||
breakpoints_new[which(breakpoints_new$method == "MIC" &
|
||
is.na(breakpoints_new$breakpoint_S)), "breakpoint_S"] <- min(m)
|
||
breakpoints_new[which(breakpoints_new$method == "MIC" &
|
||
is.na(breakpoints_new$breakpoint_R)), "breakpoint_R"] <- max(m)
|
||
# raise these one higher valid MIC factor level:
|
||
breakpoints_new[which(breakpoints_new$breakpoint_R == 129), "breakpoint_R"] <- 128
|
||
breakpoints_new[which(breakpoints_new$breakpoint_R == 257), "breakpoint_R"] <- 256
|
||
breakpoints_new[which(breakpoints_new$breakpoint_R == 513), "breakpoint_R"] <- 512
|
||
breakpoints_new[which(breakpoints_new$breakpoint_R == 1025), "breakpoint_R"] <- 1024
|
||
|
||
# fix streptococci in WHONET table of EUCAST: Strep A, B, C and G must only include these groups and not all streptococci:
|
||
breakpoints_new$mo[breakpoints_new$mo == "B_STRPT" & breakpoints_new$ref_tbl %like% "^strep.* a.* b.*c.*g"] <- as.mo("B_STRPT_ABCG")
|
||
# Haemophilus same error (must only be H. influenzae)
|
||
breakpoints_new$mo[breakpoints_new$mo == "B_HMPHL" & breakpoints_new$ref_tbl %like% "^h.* influenzae"] <- as.mo("B_HMPHL_INFL")
|
||
# EUCAST says that for H. parainfluenzae the H. influenza rules can be used, so add them
|
||
breakpoints_new <- breakpoints_new %>%
|
||
bind_rows(
|
||
breakpoints_new %>%
|
||
filter(guideline %like% "EUCAST", mo == "B_HMPHL_INFL") %>%
|
||
mutate(mo = as.mo("B_HMPHL_PRNF"))
|
||
) %>%
|
||
arrange(desc(guideline), mo, ab, type, host, method)
|
||
# Achromobacter denitrificans is in WHONET included in their A. xylosoxidans table, must be removed
|
||
breakpoints_new <- breakpoints_new %>% filter(mo != as.mo("Achromobacter denitrificans"))
|
||
# WHONET contains gentamicin breakpoints for viridans streptocci, which are intrinsic R - they meant genta-high, which is ALSO in their table, so we just remove gentamicin in viridans streptococci
|
||
breakpoints_new <- breakpoints_new %>% filter(!(mo == as.mo("Streptococcus viridans") & ab == "GEN"))
|
||
# Nitrofurantoin in Staph (EUCAST) only applies to S. saprophyticus, while WHONET has the DISK correct but the MIC on genus level
|
||
breakpoints_new$mo[breakpoints_new$mo == "B_STPHY" & breakpoints_new$ab == "NIT" & breakpoints_new$guideline %like% "EUCAST"] <- as.mo("B_STPHY_SPRP")
|
||
# WHONET sets the 2023 breakpoints for SAM to MIC of 16/32 for Enterobacterales, should be MIC 8/32 like AMC (see issue #123 on github.com/msberends/AMR)
|
||
# UPDATE 2024-02-22: fixed now
|
||
|
||
# determine rank again now that some changes were made on taxonomic level (genus -> species)
|
||
breakpoints_new <- breakpoints_new %>%
|
||
mutate(rank_index = case_when(
|
||
is.na(mo_rank(mo, keep_synonyms = TRUE)) ~ 6, # for UNKNOWN, B_GRAMN, B_ANAER, B_ANAER-NEG, etc.
|
||
mo_rank(mo, keep_synonyms = TRUE) %like% "(infra|sub)" ~ 1,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "species" ~ 2,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "species group" ~ 2.5,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "genus" ~ 3,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "family" ~ 4,
|
||
mo_rank(mo, keep_synonyms = TRUE) == "order" ~ 5,
|
||
TRUE ~ 6
|
||
))
|
||
|
||
# WHONET adds one log2 level to the R breakpoint for their software, e.g. in AMC in Enterobacterales:
|
||
# EUCAST 2023 guideline: S <= 8 and R > 8
|
||
# WHONET file: S <= 8 and R >= 16
|
||
breakpoints_new %>% filter(guideline == "EUCAST 2023", ab == "AMC", mo == "B_[ORD]_ENTRBCTR", method == "MIC")
|
||
# but this will make an MIC of 12 I, which should be R according to EUCAST, so:
|
||
breakpoints_new <- breakpoints_new %>%
|
||
mutate(breakpoint_R = ifelse(guideline %like% "EUCAST" & method == "MIC" & log2(breakpoint_R) - log2(breakpoint_S) != 0,
|
||
pmax(breakpoint_S, breakpoint_R / 2),
|
||
breakpoint_R
|
||
))
|
||
# fix disks as well
|
||
breakpoints_new %>% filter(guideline == "EUCAST 2023", ab == "AMC", mo == "B_[ORD]_ENTRBCTR", method == "DISK")
|
||
breakpoints_new <- breakpoints_new %>%
|
||
mutate(breakpoint_R = ifelse(guideline %like% "EUCAST" & method == "DISK" & breakpoint_S - breakpoint_R != 0,
|
||
breakpoint_R + 1,
|
||
breakpoint_R
|
||
))
|
||
# fill missing R breakpoint where there is an S breakpoint
|
||
breakpoints_new[which(is.na(breakpoints_new$breakpoint_R)), "breakpoint_R"] <- breakpoints_new[which(is.na(breakpoints_new$breakpoint_R)), "breakpoint_S"]
|
||
|
||
|
||
# check the strange duplicates
|
||
breakpoints_new %>%
|
||
mutate(id = paste(guideline, type, host, method, site, mo, ab, uti)) %>%
|
||
filter(id %in% .$id[which(duplicated(id))]) %>%
|
||
arrange(desc(guideline))
|
||
# 2024-06-19 mostly ECOFFs, but there's no explanation in the whonet_breakpoints file, we have to remove duplicates
|
||
# remove duplicates
|
||
breakpoints_new <- breakpoints_new %>%
|
||
distinct(guideline, type, host, method, site, mo, ab, uti, .keep_all = TRUE)
|
||
|
||
|
||
# CHECKS AND SAVE TO PACKAGE ----
|
||
|
||
# check again
|
||
breakpoints_new %>% filter(guideline == "EUCAST 2024", ab == "AMC", mo == "B_[ORD]_ENTRBCTR", method == "MIC")
|
||
# compare with current version
|
||
clinical_breakpoints %>% filter(guideline == "EUCAST 2023", ab == "AMC", mo == "B_[ORD]_ENTRBCTR", method == "MIC")
|
||
|
||
# must have "human" and "ECOFF"
|
||
breakpoints_new %>% filter(mo == "B_STRPT_PNMN", ab == "AMP", guideline == "EUCAST 2020", method == "MIC")
|
||
|
||
# check dimensions
|
||
dim(breakpoints_new)
|
||
dim(clinical_breakpoints)
|
||
|
||
clinical_breakpoints <- breakpoints_new
|
||
clinical_breakpoints <- clinical_breakpoints %>% dataset_UTF8_to_ASCII()
|
||
usethis::use_data(clinical_breakpoints, overwrite = TRUE, compress = "xz", version = 2)
|
||
rm(clinical_breakpoints)
|
||
devtools::load_all(".")
|