1
0
mirror of https://github.com/msberends/AMR.git synced 2024-12-27 13:26:11 +01:00

icu_exclude in first_isolate(), fixes #110

This commit is contained in:
dr. M.S. (Matthijs) Berends 2023-05-17 22:12:10 +02:00
parent 5f9769a4f7
commit 3018fb87a9
3 changed files with 35 additions and 43 deletions

View File

@ -1,6 +1,6 @@
Package: AMR Package: AMR
Version: 2.0.0.9015 Version: 2.0.0.9018
Date: 2023-05-12 Date: 2023-05-17
Title: Antimicrobial Resistance Data Analysis Title: Antimicrobial Resistance Data Analysis
Description: Functions to simplify and standardise antimicrobial resistance (AMR) Description: Functions to simplify and standardise antimicrobial resistance (AMR)
data analysis and to work with microbial and antimicrobial properties by data analysis and to work with microbial and antimicrobial properties by

View File

@ -1,16 +1,17 @@
# AMR 2.0.0.9015 # AMR 2.0.0.9018
## Changed ## Changed
* Added oxygen tolerance to over 25,000 bacteria in the `microorganisms` data set * Added oxygen tolerance to over 25,000 bacteria in the `microorganisms` data set
* Added `mo_oxygen_tolerance()` to retrieve the values * Added `mo_oxygen_tolerance()` to retrieve the values
* Added `mo_is_anaerobic()` to determine which species are obligate anaerobic bacteria * Added `mo_is_anaerobic()` to determine which species are obligate anaerobic bacteria
* Added LPSN and GBIF identifiers, and oxygen tolerance to `mo_info()` * Added LPSN and GBIF identifiers, and oxygen tolerance to `mo_info()`
* formatting fix for `sir_interpretation_history()` * Added SAS Transport files (file extension `.xpt`) to [our download page](https://msberends.github.io/AMR/articles/datasets.html) to use in SAS software
* Fixed some WHONET codes for microorganisms and consequently a couple of entries in `clinical_breakpoints`
* Added microbial codes for Gram-negative/positive anaerobic bacteria * Added microbial codes for Gram-negative/positive anaerobic bacteria
* `mo_rank()` now returns `NA` for 'unknown' microorganisms (`B_ANAER`, `B_ANAER-NEG`, `B_ANAER-POS`, `B_GRAMN`, `B_GRAMP`, `F_FUNGUS`, `F_YEAST`, and `UNKNOWN`) * `mo_rank()` now returns `NA` for 'unknown' microorganisms (`B_ANAER`, `B_ANAER-NEG`, `B_ANAER-POS`, `B_GRAMN`, `B_GRAMP`, `F_FUNGUS`, `F_YEAST`, and `UNKNOWN`)
* Fixed formatting for `sir_interpretation_history()`
* Fixed some WHONET codes for microorganisms and consequently a couple of entries in `clinical_breakpoints`
* Fixed a bug for `as.mo()` that led to coercion of `NA` values when using custom microorganism codes * Fixed a bug for `as.mo()` that led to coercion of `NA` values when using custom microorganism codes
* Added SAS Transport files (file extension `.xpt`) to [our download page](https://msberends.github.io/AMR/articles/datasets.html) to use in SAS software * Fixed usage of `icu_exclude` in `first_isolates()`
# AMR 2.0.0 # AMR 2.0.0

View File

@ -191,13 +191,13 @@ first_isolate <- function(x = NULL,
} }
meet_criteria(col_specimen, allow_class = "character", has_length = 1, allow_NULL = TRUE, is_in = colnames(x)) meet_criteria(col_specimen, allow_class = "character", has_length = 1, allow_NULL = TRUE, is_in = colnames(x))
if (is.logical(col_icu)) { if (is.logical(col_icu)) {
meet_criteria(col_icu, allow_class = "logical", has_length = c(1, nrow(x)), allow_NULL = TRUE) meet_criteria(col_icu, allow_class = "logical", has_length = c(1, nrow(x)), allow_NA = TRUE)
if (length(col_icu) == 1) { x$newvar_is_icu <- col_icu
col_icu <- rep(col_icu, nrow(x)) } else if (!is.null(col_icu)) {
}
} else {
meet_criteria(col_icu, allow_class = c("character", "logical"), has_length = 1, allow_NULL = TRUE, is_in = colnames(x)) meet_criteria(col_icu, allow_class = c("character", "logical"), has_length = 1, allow_NULL = TRUE, is_in = colnames(x))
col_icu <- x[, col_icu, drop = TRUE] x$newvar_is_icu <- x[, col_icu, drop = TRUE]
} else {
x$newvar_is_icu <- NA_real_
} }
# method # method
method <- coerce_method(method) method <- coerce_method(method)
@ -251,14 +251,13 @@ first_isolate <- function(x = NULL,
"Determining first isolates ", "Determining first isolates ",
ifelse(method %in% c("episode-based", "phenotype-based"), ifelse(method %in% c("episode-based", "phenotype-based"),
ifelse(is.infinite(episode_days), ifelse(is.infinite(episode_days),
"without a specified episode length", paste(font_bold("without"), " a specified episode length"),
paste("using an episode length of", episode_days, "days") paste("using an episode length of", font_bold(paste(episode_days, "days")))
), ),
"" ""
) )
), ),
as_note = FALSE, add_fn = font_red
add_fn = font_black
) )
} }
@ -358,8 +357,7 @@ first_isolate <- function(x = NULL,
# remove testcodes # remove testcodes
if (!is.null(testcodes_exclude) && isTRUE(info) && message_not_thrown_before("first_isolate", "excludingtestcodes")) { if (!is.null(testcodes_exclude) && isTRUE(info) && message_not_thrown_before("first_isolate", "excludingtestcodes")) {
message_("Excluding test codes: ", vector_and(testcodes_exclude, quotes = TRUE), message_("Excluding test codes: ", vector_and(testcodes_exclude, quotes = TRUE),
add_fn = font_black, add_fn = font_red
as_note = FALSE
) )
} }
@ -372,8 +370,7 @@ first_isolate <- function(x = NULL,
check_columns_existance(col_specimen, x) check_columns_existance(col_specimen, x)
if (isTRUE(info) && message_not_thrown_before("first_isolate", "excludingspecimen")) { if (isTRUE(info) && message_not_thrown_before("first_isolate", "excludingspecimen")) {
message_("Excluding other than specimen group '", specimen_group, "'", message_("Excluding other than specimen group '", specimen_group, "'",
add_fn = font_black, add_fn = font_red
as_note = FALSE
) )
} }
} }
@ -455,15 +452,13 @@ first_isolate <- function(x = NULL,
message_("Basing inclusion on key antimicrobials, ", message_("Basing inclusion on key antimicrobials, ",
ifelse(ignore_I == FALSE, "not ", ""), ifelse(ignore_I == FALSE, "not ", ""),
"ignoring I", "ignoring I",
add_fn = font_black, add_fn = font_red
as_note = FALSE
) )
} }
if (type == "points") { if (type == "points") {
message_("Basing inclusion on all antimicrobial results, using a points threshold of ", message_("Basing inclusion on all antimicrobial results, using a points threshold of ",
points_threshold, points_threshold,
add_fn = font_black, add_fn = font_red
as_note = FALSE
) )
} }
} }
@ -506,33 +501,27 @@ first_isolate <- function(x = NULL,
(x$other_pat_or_mo | x$more_than_episode_ago) (x$other_pat_or_mo | x$more_than_episode_ago)
} }
decimal.mark <- getOption("OutDec")
big.mark <- ifelse(decimal.mark != ",", ",", " ")
# first one as TRUE # first one as TRUE
x[row.start, "newvar_first_isolate"] <- TRUE x[row.start, "newvar_first_isolate"] <- TRUE
# no tests that should be included, or ICU # no tests that should be included, or ICU
if (!is.null(col_testcode)) { if (!is.null(col_testcode)) {
x[which(x[, col_testcode] %in% tolower(testcodes_exclude)), "newvar_first_isolate"] <- FALSE x[which(x[, col_testcode] %in% tolower(testcodes_exclude)), "newvar_first_isolate"] <- FALSE
} }
if (any(!is.na(x$newvar_is_icu)) && any(x$newvar_is_icu == TRUE, na.rm = TRUE)) {
if (!is.null(col_icu)) {
if (icu_exclude == TRUE) { if (icu_exclude == TRUE) {
if (isTRUE(info)) { if (isTRUE(info)) {
message_("Excluding ", format(sum(col_icu, na.rm = TRUE), big.mark = " "), " isolates from ICU.", message_("Excluding ", format(sum(x$newvar_is_icu, na.rm = TRUE), decimal.mark = decimal.mark, big.mark = big.mark), " isolates from ICU.",
add_fn = font_black, add_fn = font_red)
as_note = FALSE
)
} }
x[which(col_icu), "newvar_first_isolate"] <- FALSE x[which(x$newvar_is_icu), "newvar_first_isolate"] <- FALSE
} else if (isTRUE(info)) { } else if (isTRUE(info)) {
message_("Including isolates from ICU.", message_("Including isolates from ICU.")
add_fn = font_black,
as_note = FALSE
)
} }
} }
decimal.mark <- getOption("OutDec")
big.mark <- ifelse(decimal.mark != ",", ",", " ")
if (isTRUE(info)) { if (isTRUE(info)) {
# print group name if used in dplyr::group_by() # print group name if used in dplyr::group_by()
cur_group <- import_fn("cur_group", "dplyr", error_on_fail = FALSE) cur_group <- import_fn("cur_group", "dplyr", error_on_fail = FALSE)
@ -560,11 +549,12 @@ first_isolate <- function(x = NULL,
# handle empty microorganisms # handle empty microorganisms
if (any(x$newvar_mo == "UNKNOWN", na.rm = TRUE) && isTRUE(info)) { if (any(x$newvar_mo == "UNKNOWN", na.rm = TRUE) && isTRUE(info)) {
message_( message_(
ifelse(include_unknown == TRUE, "Included ", "Excluded "), ifelse(include_unknown == TRUE, "Including ", "Excluding "),
format(sum(x$newvar_mo == "UNKNOWN", na.rm = TRUE), format(sum(x$newvar_mo == "UNKNOWN", na.rm = TRUE),
decimal.mark = decimal.mark, big.mark = big.mark decimal.mark = decimal.mark, big.mark = big.mark
), ),
" isolates with a microbial ID 'UNKNOWN' (in column '", font_bold(col_mo), "')" " isolates with a microbial ID 'UNKNOWN' (in column '", font_bold(col_mo), "')",
add_fn = font_red
) )
} }
x[which(x$newvar_mo == "UNKNOWN"), "newvar_first_isolate"] <- include_unknown x[which(x$newvar_mo == "UNKNOWN"), "newvar_first_isolate"] <- include_unknown
@ -572,10 +562,11 @@ first_isolate <- function(x = NULL,
# exclude all NAs # exclude all NAs
if (anyNA(x$newvar_mo) && isTRUE(info)) { if (anyNA(x$newvar_mo) && isTRUE(info)) {
message_( message_(
"Excluded ", format(sum(is.na(x$newvar_mo), na.rm = TRUE), "Excluding ", format(sum(is.na(x$newvar_mo), na.rm = TRUE),
decimal.mark = decimal.mark, big.mark = big.mark decimal.mark = decimal.mark, big.mark = big.mark
), ),
" isolates with a microbial ID 'NA' (in column '", font_bold(col_mo), "')" " isolates with a microbial ID `NA` (in column '", font_bold(col_mo), "')",
add_fn = font_red
) )
} }
x[which(is.na(x$newvar_mo)), "newvar_first_isolate"] <- FALSE x[which(is.na(x$newvar_mo)), "newvar_first_isolate"] <- FALSE