These functions determine which items in a vector can be considered (the start of) a new episode. This can be used to determine clinical episodes for any epidemiological analysis. The get_episode() function returns the index number of the episode per group, while the is_new_episode() function returns TRUE for every new get_episode() index. Both absolute and relative episode determination are supported.
Usage
get_episode(x, episode_days = NULL, case_free_days = NULL, ...)
is_new_episode(x, episode_days = NULL, case_free_days = NULL, ...)Arguments
- x
vector of dates (class
DateorPOSIXt), will be sorted internally to determine episodes- episode_days
episode length in days to specify the time period after which a new episode begins, can also be less than a day or
Inf, see Details- case_free_days
(inter-epidemic) interval length in days after which a new episode will start, can also be less than a day or
Inf, see Details- ...
ignored, only in place to allow future extensions
Details
Episodes can be determined in two ways: absolute and relative.
Absolute
This method uses
episode_daysto define an episode length in days, after which a new episode will start. A common use case in AMR data analysis is microbial epidemiology: episodes of S. aureus bacteraemia in ICU patients for example. The episode length could then be 30 days, so that new S. aureus isolates after an ICU episode of 30 days will be considered a different (or new) episode.Thus, this method counts since the start of the previous episode.
Relative
This method uses
case_free_daysto quantify the duration of case-free days (the inter-epidemic interval), after which a new episode will start. A common use case is infectious disease epidemiology: episodes of norovirus outbreaks in a hospital for example. The case-free period could then be 14 days, so that new norovirus cases after that time will be considered a different (or new) episode.Thus, this methods counts since the last case in the previous episode.
In a table:
| Date | Using episode_days = 7 | Using case_free_days = 7 |
| 2023-01-01 | 1 | 1 |
| 2023-01-02 | 1 | 1 |
| 2023-01-05 | 1 | 1 |
| 2023-01-08 | 2** | 1 |
| 2023-02-21 | 3 | 2*** |
| 2023-02-22 | 3 | 2 |
| 2023-02-23 | 3 | 2 |
| 2023-02-24 | 3 | 2 |
| 2023-03-01 | 4 | 2 |
** This marks the start of a new episode, because 8 January 2023 is more than 7 days since the start of the previous episode (1 January 2023).
*** This marks the start of a new episode, because 21 January 2023 is more than 7 days since the last case in the previous episode (8 January 2023).
Either episode_days or case_free_days must be provided in the function.
Difference between get_episode() and is_new_episode()
The get_episode() function returns the index number of the episode, so all cases/patients/isolates in the first episode will have the number 1, all cases/patients/isolates in the second episode will have the number 2, etc.
The is_new_episode() function returns TRUE for every new get_episode() index, and is thus equal to !duplicated(get_episode(...)).
To specify, when setting episode_days = 365 (using method 1 as explained above), this is how the two functions differ:
| patient | date | get_episode() | is_new_episode() |
| A | 2019-01-01 | 1 | TRUE |
| A | 2019-03-01 | 1 | FALSE |
| A | 2021-01-01 | 2 | TRUE |
| B | 2008-01-01 | 1 | TRUE |
| B | 2008-01-01 | 1 | FALSE |
| C | 2020-01-01 | 1 | TRUE |
Other
The first_isolate() function is a wrapper around the is_new_episode() function, but is more efficient for data sets containing microorganism codes or names and allows for different isolate selection methods.
The dplyr package is not required for these functions to work, but these episode functions do support variable grouping and work conveniently inside dplyr verbs such as filter(), mutate() and summarise().
Examples
# difference between absolute and relative determination of episodes:
x <- data.frame(dates = as.Date(c(
"2021-01-01",
"2021-01-02",
"2021-01-05",
"2021-01-08",
"2021-02-21",
"2021-02-22",
"2021-02-23",
"2021-02-24",
"2021-03-01",
"2021-03-01"
)))
x$absolute <- get_episode(x$dates, episode_days = 7)
x$relative <- get_episode(x$dates, case_free_days = 7)
x
#> dates absolute relative
#> 1 2021-01-01 1 1
#> 2 2021-01-02 1 1
#> 3 2021-01-05 1 1
#> 4 2021-01-08 2 1
#> 5 2021-02-21 3 2
#> 6 2021-02-22 3 2
#> 7 2021-02-23 3 2
#> 8 2021-02-24 3 2
#> 9 2021-03-01 4 2
#> 10 2021-03-01 4 2
# `example_isolates` is a data set available in the AMR package.
# See ?example_isolates
df <- example_isolates[sample(seq_len(2000), size = 100), ]
get_episode(df$date, episode_days = 60) # indices
#> [1] 11 38 25 37 31 42 44 9 13 8 9 7 48 47 3 6 46 41 11 16 43 9 41 29 44
#> [26] 42 30 28 36 1 30 39 40 32 35 6 3 27 24 1 2 35 17 20 9 36 34 48 23 22
#> [51] 7 21 12 43 21 17 50 31 48 26 5 26 14 15 10 2 35 36 47 8 46 16 27 13 45
#> [76] 40 28 25 21 4 50 36 45 17 24 6 18 6 3 5 46 19 4 37 49 48 13 33 10 41
is_new_episode(df$date, episode_days = 60) # TRUE/FALSE
#> [1] TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE FALSE TRUE
#> [13] TRUE TRUE TRUE TRUE TRUE TRUE FALSE TRUE TRUE FALSE FALSE TRUE
#> [25] FALSE FALSE TRUE TRUE TRUE TRUE FALSE TRUE TRUE TRUE TRUE FALSE
#> [37] FALSE TRUE TRUE FALSE TRUE FALSE TRUE TRUE FALSE FALSE TRUE FALSE
#> [49] TRUE TRUE FALSE TRUE TRUE FALSE FALSE FALSE TRUE FALSE FALSE TRUE
#> [61] TRUE FALSE TRUE TRUE TRUE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
#> [73] FALSE FALSE TRUE FALSE FALSE FALSE FALSE TRUE FALSE FALSE FALSE FALSE
#> [85] FALSE FALSE TRUE FALSE FALSE FALSE FALSE TRUE FALSE FALSE TRUE FALSE
#> [97] FALSE TRUE FALSE FALSE
# filter on results from the third 60-day episode only, using base R
df[which(get_episode(df$date, 60) == 3), ]
#> # A tibble: 3 × 46
#> date patient age gender ward mo PEN OXA FLC AMX
#> <date> <chr> <dbl> <chr> <chr> <mo> <sir> <sir> <sir> <sir>
#> 1 2002-10-22 718963 82 M ICU B_STPHY_EPDR R NA R R
#> 2 2002-10-20 F35553 51 M ICU B_STPHY_AURS S NA S NA
#> 3 2002-09-08 B8CB09 60 F Outpatie… B_STPHY_CONS S NA S NA
#> # … with 36 more variables: AMC <sir>, AMP <sir>, TZP <sir>, CZO <sir>,
#> # FEP <sir>, CXM <sir>, FOX <sir>, CTX <sir>, CAZ <sir>, CRO <sir>,
#> # GEN <sir>, TOB <sir>, AMK <sir>, KAN <sir>, TMP <sir>, SXT <sir>,
#> # NIT <sir>, FOS <sir>, LNZ <sir>, CIP <sir>, MFX <sir>, VAN <sir>,
#> # TEC <sir>, TCY <sir>, TGC <sir>, DOX <sir>, ERY <sir>, CLI <sir>,
#> # AZM <sir>, IPM <sir>, MEM <sir>, MTR <sir>, CHL <sir>, COL <sir>,
#> # MUP <sir>, RIF <sir>
# the functions also work for less than a day, e.g. to include one per hour:
get_episode(
c(
Sys.time(),
Sys.time() + 60 * 60
),
episode_days = 1 / 24
)
#> [1] 1 2
# \donttest{
if (require("dplyr")) {
# is_new_episode() can also be used in dplyr verbs to determine patient
# episodes based on any (combination of) grouping variables:
df %>%
mutate(condition = sample(
x = c("A", "B", "C"),
size = 100,
replace = TRUE
)) %>%
group_by(patient, condition) %>%
mutate(new_episode = is_new_episode(date, 365)) %>%
select(patient, date, condition, new_episode) %>%
arrange(patient, condition, date)
}
#> # A tibble: 100 × 4
#> # Groups: patient, condition [98]
#> patient date condition new_episode
#> <chr> <date> <chr> <lgl>
#> 1 032343 2003-06-09 B TRUE
#> 2 059414 2006-07-21 A TRUE
#> 3 063988 2006-12-26 A TRUE
#> 4 065133 2011-12-19 C TRUE
#> 5 067927 2002-02-05 A TRUE
#> 6 069276 2015-06-18 B TRUE
#> 7 083080 2012-04-16 B TRUE
#> 8 092034 2016-08-07 B TRUE
#> 9 097186 2015-10-28 B TRUE
#> 10 125492 2017-09-15 B TRUE
#> # … with 90 more rows
if (require("dplyr")) {
df %>%
group_by(ward, patient) %>%
transmute(date,
patient,
new_index = get_episode(date, 60),
new_logical = is_new_episode(date, 60)
) %>%
arrange(patient, ward, date)
}
#> # A tibble: 100 × 5
#> # Groups: ward, patient [92]
#> ward date patient new_index new_logical
#> <chr> <date> <chr> <int> <lgl>
#> 1 Clinical 2003-06-09 032343 1 TRUE
#> 2 Clinical 2006-07-21 059414 1 TRUE
#> 3 Clinical 2006-12-26 063988 1 TRUE
#> 4 Clinical 2011-12-19 065133 1 TRUE
#> 5 ICU 2002-02-05 067927 1 TRUE
#> 6 ICU 2015-06-18 069276 1 TRUE
#> 7 Clinical 2012-04-16 083080 1 TRUE
#> 8 Clinical 2016-08-07 092034 1 TRUE
#> 9 Clinical 2015-10-28 097186 1 TRUE
#> 10 Clinical 2017-09-15 125492 1 TRUE
#> # … with 90 more rows
if (require("dplyr")) {
df %>%
group_by(ward) %>%
summarise(
n_patients = n_distinct(patient),
n_episodes_365 = sum(is_new_episode(date, episode_days = 365)),
n_episodes_60 = sum(is_new_episode(date, episode_days = 60)),
n_episodes_30 = sum(is_new_episode(date, episode_days = 30))
)
}
#> # A tibble: 3 × 5
#> ward n_patients n_episodes_365 n_episodes_60 n_episodes_30
#> <chr> <int> <int> <int> <int>
#> 1 Clinical 64 15 44 52
#> 2 ICU 21 9 20 20
#> 3 Outpatient 7 6 7 7
# grouping on patients and microorganisms leads to the same
# results as first_isolate() when using 'episode-based':
if (require("dplyr")) {
x <- df %>%
filter_first_isolate(
include_unknown = TRUE,
method = "episode-based"
)
y <- df %>%
group_by(patient, mo) %>%
filter(is_new_episode(date, 365)) %>%
ungroup()
identical(x, y)
}
#> [1] TRUE
# but is_new_episode() has a lot more flexibility than first_isolate(),
# since you can now group on anything that seems relevant:
if (require("dplyr")) {
df %>%
group_by(patient, mo, ward) %>%
mutate(flag_episode = is_new_episode(date, 365)) %>%
select(group_vars(.), flag_episode)
}
#> # A tibble: 100 × 4
#> # Groups: patient, mo, ward [95]
#> patient mo ward flag_episode
#> <chr> <mo> <chr> <lgl>
#> 1 219024 B_ESCHR_COLI Clinical TRUE
#> 2 960698 B_ESCHR_COLI Clinical TRUE
#> 3 696587 B_BCTRD_FRGL Clinical TRUE
#> 4 850429 B_STPHY_EPDR Clinical TRUE
#> 5 8CBCF2 B_STPHY_CONS Clinical TRUE
#> 6 559068 B_STPHY_CPTS Clinical TRUE
#> 7 E52521 B_KLBSL_AERG Clinical TRUE
#> 8 304508 B_STRPT_PNMN Clinical TRUE
#> 9 990867 B_STPHY_AURS Clinical TRUE
#> 10 2F0932 B_STPHY_CONS Clinical TRUE
#> # … with 90 more rows
# }