mirror of https://github.com/msberends/AMR.git
306 lines
11 KiB
R
306 lines
11 KiB
R
# ==================================================================== #
|
|
# TITLE #
|
|
# AMR: An R Package for Working with Antimicrobial Resistance Data #
|
|
# #
|
|
# SOURCE #
|
|
# https://github.com/msberends/AMR #
|
|
# #
|
|
# CITE AS #
|
|
# Berends MS, Luz CF, Friedrich AW, Sinha BNM, Albers CJ, Glasner C #
|
|
# (2022). AMR: An R Package for Working with Antimicrobial Resistance #
|
|
# Data. Journal of Statistical Software, 104(3), 1-31. #
|
|
# doi: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(openxlsx)
|
|
library(dplyr)
|
|
library(tidyr)
|
|
library(cleaner)
|
|
library(AMR)
|
|
|
|
# USE THIS FUNCTION TO READ THE EUCAST EXCEL FILE THAT CONTAINS THE BREAKPOINT TABLES
|
|
|
|
read_EUCAST <- function(sheet, file, guideline_name) {
|
|
message("\nGetting sheet: ", sheet)
|
|
sheet.bak <- sheet
|
|
|
|
uncertainties <- NULL
|
|
add_uncertainties <- function(old, new) {
|
|
if (is.null(old)) {
|
|
new
|
|
} else {
|
|
bind_rows(old, new)
|
|
}
|
|
}
|
|
|
|
raw_data <- read.xlsx(
|
|
xlsxFile = file,
|
|
sheet = sheet,
|
|
colNames = FALSE,
|
|
skipEmptyRows = FALSE,
|
|
skipEmptyCols = FALSE,
|
|
fillMergedCells = TRUE,
|
|
na.strings = c("", "-", "NA", "IE", "IP")
|
|
)
|
|
probable_rows <- suppressWarnings(raw_data %>% mutate_all(as.double) %>% summarise_all(~ sum(!is.na(.))) %>% unlist() %>% max())
|
|
if (probable_rows == 0) {
|
|
message("NO ROWS FOUND")
|
|
message("------------------------")
|
|
return(NULL)
|
|
}
|
|
|
|
# in the info header in the Excel file, EUCAST mentions which genera are targeted
|
|
if (sheet %like% "anaerob.*Gram.*posi") {
|
|
sheet <- paste0(c(
|
|
"Actinomyces", "Bifidobacterium", "Clostridioides",
|
|
"Clostridium", "Cutibacterium", "Eggerthella",
|
|
"Eubacterium", "Lactobacillus", "Propionibacterium",
|
|
"Staphylococcus saccharolyticus"
|
|
),
|
|
collapse = "_"
|
|
)
|
|
} else if (sheet %like% "anaerob.*Gram.*nega") {
|
|
sheet <- paste0(c(
|
|
"Bacteroides",
|
|
"Bilophila",
|
|
"Fusobacterium",
|
|
"Mobiluncus",
|
|
"Parabacteroides",
|
|
"Porphyromonas",
|
|
"Prevotella"
|
|
),
|
|
collapse = "_"
|
|
)
|
|
} else if (sheet == "Streptococcus A,B,C,G") {
|
|
sheet <- paste0(microorganisms %>%
|
|
filter(genus == "Streptococcus") %>%
|
|
mutate(lancefield = mo_name(mo, Lancefield = TRUE)) %>%
|
|
filter(lancefield %like% "^Streptococcus group") %>%
|
|
pull(fullname),
|
|
collapse = "_"
|
|
)
|
|
} else if (sheet %like% "PK.*PD") {
|
|
sheet <- "UNKNOWN"
|
|
}
|
|
mo_sheet <- paste0(suppressMessages(as.mo(unlist(strsplit(sheet, "_")))), collapse = "|")
|
|
if (!is.null(mo_uncertainties())) uncertainties <- add_uncertainties(uncertainties, mo_uncertainties())
|
|
|
|
set_columns_names <- function(x, cols) {
|
|
colnames(x) <- cols[1:length(colnames(x))]
|
|
x
|
|
}
|
|
|
|
get_mo <- function(x) {
|
|
for (i in seq_len(length(x))) {
|
|
y <- trimws2(unlist(strsplit(x[i], "(,|and)")))
|
|
y <- trimws2(gsub("[(].*[)]", "", y))
|
|
y <- suppressWarnings(suppressMessages(as.mo(y)))
|
|
if (!is.null(mo_uncertainties())) uncertainties <<- add_uncertainties(uncertainties, mo_uncertainties())
|
|
y <- y[!is.na(y) & y != "UNKNOWN"]
|
|
x[i] <- paste(y, collapse = "|")
|
|
}
|
|
x
|
|
}
|
|
|
|
MICs_with_trailing_superscript <- c(
|
|
seq(from = 0.0011, to = 0.0019, by = 0.0001),
|
|
seq(from = 0.031, to = 0.039, by = 0.001),
|
|
seq(from = 0.061, to = 0.069, by = 0.001),
|
|
seq(from = 0.1251, to = 0.1259, by = 0.0001),
|
|
seq(from = 0.251, to = 0.259, by = 0.001),
|
|
seq(from = 0.51, to = 0.59, by = 0.01),
|
|
seq(from = 11, to = 19, by = 1),
|
|
seq(from = 161, to = 169, by = 01),
|
|
seq(from = 21, to = 29, by = 1),
|
|
seq(from = 321, to = 329, by = 1),
|
|
seq(from = 41, to = 49, by = 1),
|
|
seq(from = 81, to = 89, by = 1)
|
|
)
|
|
has_superscript <- function(x) {
|
|
# because due to floating point error, 0.1252 is not in:
|
|
# seq(from = 0.1251, to = 0.1259, by = 0.0001)
|
|
sapply(x, function(x) any(near(x, MICs_with_trailing_superscript)))
|
|
}
|
|
|
|
has_zone_diameters <- rep(any(unlist(raw_data) %like% "zone diameter"), nrow(raw_data))
|
|
|
|
cleaned <- raw_data %>%
|
|
as_tibble() %>%
|
|
set_columns_names(LETTERS) %>%
|
|
transmute(
|
|
drug = A,
|
|
MIC_S = B,
|
|
MIC_R = C,
|
|
disk_dose = ifelse(has_zone_diameters, E, NA_character_),
|
|
disk_S = ifelse(has_zone_diameters, `F`, NA_character_),
|
|
disk_R = ifelse(has_zone_diameters, G, NA_character_)
|
|
) %>%
|
|
filter(
|
|
!is.na(drug),
|
|
!(is.na(MIC_S) & is.na(MIC_R) & is.na(disk_S) & is.na(disk_R)),
|
|
MIC_S %unlike% "(MIC|S ≤|note)",
|
|
MIC_S %unlike% "^[-]",
|
|
drug != MIC_S,
|
|
) %>%
|
|
mutate(
|
|
administration = case_when(
|
|
drug %like% "[( ]oral" ~ "oral",
|
|
drug %like% "[( ]iv" ~ "iv",
|
|
TRUE ~ NA_character_
|
|
),
|
|
uti = ifelse(drug %like% "(UTI|urinary|urine)", TRUE, FALSE),
|
|
systemic = ifelse(drug %like% "(systemic|septic)", TRUE, FALSE),
|
|
mo = ifelse(drug %like% "([.]|spp)", get_mo(drug), mo_sheet)
|
|
) %>%
|
|
# clean disk doses
|
|
mutate(disk_dose = clean_character(disk_dose, remove = "[^0-9.-]")) %>%
|
|
# clean MIC and disk values
|
|
mutate(
|
|
MIC_S = gsub(".,.", "", MIC_S), # remove superscript notes with comma, like 0.5^2,3
|
|
MIC_R = gsub(".,.", "", MIC_R),
|
|
disk_S = gsub(".,.", "", disk_S),
|
|
disk_R = gsub(".,.", "", disk_R),
|
|
MIC_S = clean_double(MIC_S), # make them valid numeric values
|
|
MIC_R = clean_double(MIC_R),
|
|
disk_S = clean_integer(disk_S),
|
|
disk_R = clean_integer(disk_R),
|
|
# invalid MIC values have a superscript text, delete those
|
|
MIC_S = ifelse(has_superscript(MIC_S),
|
|
substr(MIC_S, 1, nchar(MIC_S) - 1),
|
|
MIC_S
|
|
),
|
|
MIC_R = ifelse(has_superscript(MIC_R),
|
|
substr(MIC_R, 1, nchar(MIC_R) - 1),
|
|
MIC_R
|
|
),
|
|
# and some are just awful
|
|
MIC_S = ifelse(MIC_S == 43.4, 4, MIC_S),
|
|
MIC_R = ifelse(MIC_R == 43.4, 4, MIC_R),
|
|
) %>%
|
|
# clean drug names
|
|
mutate(
|
|
drug = gsub(" ?[(, ].*$", "", drug),
|
|
drug = gsub("[1-9]+$", "", drug),
|
|
ab = as.ab(drug)
|
|
) %>%
|
|
select(ab, mo, everything(), -drug) %>%
|
|
as.data.frame(stringsAsFactors = FALSE)
|
|
|
|
# new row for every different MO mentioned
|
|
for (i in 1:nrow(cleaned)) {
|
|
mo <- cleaned[i, "mo", drop = TRUE]
|
|
if (grepl(pattern = "|", mo, fixed = TRUE)) {
|
|
mo_vect <- unlist(strsplit(mo, "|", fixed = TRUE))
|
|
cleaned[i, "mo"] <- mo_vect[1]
|
|
for (j in seq_len(length(mo_vect))) {
|
|
cleaned <- bind_rows(cleaned, cleaned[i, , drop = FALSE])
|
|
cleaned[nrow(cleaned), "mo"] <- mo_vect[j]
|
|
}
|
|
}
|
|
}
|
|
|
|
cleaned <- cleaned %>%
|
|
distinct(ab, mo, administration, uti, systemic, .keep_all = TRUE) %>%
|
|
arrange(ab, mo) %>%
|
|
mutate_at(c("MIC_S", "MIC_R", "disk_S", "disk_R"), as.double) %>%
|
|
pivot_longer(c("MIC_S", "MIC_R", "disk_S", "disk_R"), "type") %>%
|
|
mutate(
|
|
method = ifelse(type %like% "MIC", "MIC", "DISK"),
|
|
type = gsub("^.*_", "breakpoint_", type)
|
|
) %>%
|
|
pivot_wider(names_from = type, values_from = value) %>%
|
|
mutate(
|
|
guideline = guideline_name,
|
|
disk_dose = ifelse(method == "DISK", disk_dose, NA_character_),
|
|
mo = ifelse(mo == "", mo_sheet, mo)
|
|
) %>%
|
|
filter(!(is.na(breakpoint_S) & is.na(breakpoint_R))) %>%
|
|
# comply with rsi_translation for now
|
|
transmute(guideline,
|
|
method,
|
|
site = case_when(
|
|
uti ~ "UTI",
|
|
systemic ~ "Systemic",
|
|
TRUE ~ administration
|
|
),
|
|
mo, ab,
|
|
ref_tbl = sheet.bak,
|
|
disk_dose = ifelse(!is.na(disk_dose), paste0(disk_dose, "ug"), NA_character_),
|
|
breakpoint_S,
|
|
breakpoint_R
|
|
) %>%
|
|
as.data.frame(stringsAsFactors = FALSE)
|
|
|
|
if (!is.null(uncertainties)) {
|
|
print(uncertainties %>% distinct(input, mo, .keep_all = TRUE))
|
|
}
|
|
|
|
message("Estimated: ", probable_rows, ", gained: ", cleaned %>% count(ab) %>% nrow())
|
|
message("------------------------")
|
|
cleaned
|
|
}
|
|
|
|
|
|
# Actual import -----------------------------------------------------------
|
|
|
|
file <- "data-raw/v_11.0_Breakpoint_Tables.xlsx"
|
|
sheets <- readxl::excel_sheets(file)
|
|
guideline_name <- "EUCAST 2021"
|
|
|
|
sheets_to_analyse <- sheets[!sheets %in% c("Content", "Changes", "Notes", "Guidance", "Dosages", "Technical uncertainty", "Topical agents")]
|
|
|
|
# takes the longest time:
|
|
new_EUCAST <- read_EUCAST(
|
|
sheet = sheets_to_analyse[1],
|
|
file = file,
|
|
guideline_name = guideline_name
|
|
)
|
|
for (i in 2:length(sheets_to_analyse)) {
|
|
tryCatch(
|
|
new_EUCAST <<- bind_rows(
|
|
new_EUCAST,
|
|
read_EUCAST(
|
|
sheet = sheets_to_analyse[i],
|
|
file = file,
|
|
guideline_name = guideline_name
|
|
)
|
|
),
|
|
error = function(e) message(e$message)
|
|
)
|
|
}
|
|
|
|
# 2021-07-12 fix for Morganellaceae (check other lines too next time)
|
|
morg <- rsi_translation %>%
|
|
as_tibble() %>%
|
|
filter(
|
|
ab == "IPM",
|
|
guideline == "EUCAST 2021",
|
|
mo == as.mo("Enterobacterales")
|
|
) %>%
|
|
mutate(mo = as.mo("Morganellaceae"))
|
|
morg[which(morg$method == "MIC"), "breakpoint_S"] <- 0.001
|
|
morg[which(morg$method == "MIC"), "breakpoint_R"] <- 4
|
|
morg[which(morg$method == "DISK"), "breakpoint_S"] <- 50
|
|
morg[which(morg$method == "DISK"), "breakpoint_R"] <- 19
|
|
|
|
rsi_translation <- rsi_translation %>%
|
|
bind_rows(morg) %>%
|
|
bind_rows(morg %>%
|
|
mutate(guideline = "EUCAST 2020")) %>%
|
|
arrange(desc(guideline), ab, mo, method)
|