问题
I have a dataset of patients getting treatments in various hospitals (in-patient only) wherein some analysis has revealed several inconsistencies. One of these was that - software was allowing patients to get admission without closure of their previously open case_id.
In order to understand it better, let us consider the sample dataset
sample data
dput(df)
df <- structure(list(case_id = 1:22, patient_id = c(1L, 1L, 1L, 1L,
1L, 1L, 1L, 2L, 2L, 2L, 1L, 3L, 3L, 3L, 4L, 4L, 5L, 5L, 6L, 7L,
8L, 8L), pack_id = c(12L, 62L, 59L, 68L, 77L, 86L, 20L, 55L,
86L, 72L, 7L, 54L, 75L, 26L, 21L, 12L, 49L, 35L, 51L, 31L, 10L,
54L), hosp_id = c(1L, 1L, 2L, 2L, 1L, 1L, 2L, 3L, 3L, 4L, 2L,
3L, 3L, 3L, 4L, 5L, 6L, 6L, 7L, 7L, 8L, 8L), admn_date = structure(c(18262,
18264, 18265, 18266, 18277, 18279, 18283, 18262, 18264, 18277,
18287, 18275, 18301, 18291, 18366, 18374, 18309, 18319, 18364,
18303, 18328, 18341), class = "Date"), discharge_date = structure(c(18275,
18276, 18271, 18275, 18288, 18280, 18286, 18275, 18276, 18288,
18291, 18283, 18309, 18297, 18375, 18381, 18347, 18328, 18367,
18309, 18341, 18344), class = "Date")), row.names = c(NA, -22L
), class = "data.frame")
> df
case_id patient_id pack_id hosp_id admn_date discharge_date
1 1 1 12 1 2020-01-01 2020-01-14
2 2 1 62 1 2020-01-03 2020-01-15
3 3 1 59 2 2020-01-04 2020-01-10
4 4 1 68 2 2020-01-05 2020-01-14
5 5 1 77 1 2020-01-16 2020-01-27
6 6 1 86 1 2020-01-18 2020-01-19
7 7 1 20 2 2020-01-22 2020-01-25
8 8 2 55 3 2020-01-01 2020-01-14
9 9 2 86 3 2020-01-03 2020-01-15
10 10 2 72 4 2020-01-16 2020-01-27
11 11 1 7 2 2020-01-26 2020-01-30
12 12 3 54 3 2020-01-14 2020-01-22
13 13 3 75 3 2020-02-09 2020-02-17
14 14 3 26 3 2020-01-30 2020-02-05
15 15 4 21 4 2020-04-14 2020-04-23
16 16 4 12 5 2020-04-22 2020-04-29
17 17 5 49 6 2020-02-17 2020-03-26
18 18 5 35 6 2020-02-27 2020-03-07
19 19 6 51 7 2020-04-12 2020-04-15
20 20 7 31 7 2020-02-11 2020-02-17
21 21 8 10 8 2020-03-07 2020-03-20
22 22 8 54 8 2020-03-20 2020-03-23
If we see in the data above, patient with id 1 got admission in hospital_1 (row-1) on 1 January and took a discharge on 14 January. Before this discharge, the patient took admission in same hospital again (row-2) ; and in hospital_2 again two times (rows 3 & 4) before finally getting all these four records closed on 15 January (row-2).
I have already filtered such records where the patient/s were admitted in multiple hospitals/same hospital multiple times; by the following code
Code tried
df_2 <- df %>% arrange(patient_id, admn_date, discharge_date) %>%
mutate(sort_key = row_number()) %>%
pivot_longer(c(admn_date, discharge_date), names_to ="activity",
values_to ="date", names_pattern = "(.*)_date") %>%
mutate(activity = factor(activity, ordered = T,
levels = c("admn", "discharge")),
admitted = ifelse(activity == "admn", 1, -1)) %>%
group_by(patient_id) %>%
arrange(date, sort_key, activity, .by_group = TRUE) %>%
mutate (admitted = cumsum(admitted)) %>%
ungroup()
> df_2
# A tibble: 44 x 8
case_id patient_id pack_id hosp_id sort_key activity date admitted
<int> <int> <int> <int> <int> <ord> <date> <dbl>
1 1 1 12 1 1 admn 2020-01-01 1
2 2 1 62 1 2 admn 2020-01-03 2
3 3 1 59 2 3 admn 2020-01-04 3
4 4 1 68 2 4 admn 2020-01-05 4
5 3 1 59 2 3 discharge 2020-01-10 3
6 1 1 12 1 1 discharge 2020-01-14 2
7 4 1 68 2 4 discharge 2020-01-14 1
8 2 1 62 1 2 discharge 2020-01-15 0
9 5 1 77 1 5 admn 2020-01-16 1
10 6 1 86 1 6 admn 2020-01-18 2
# ... with 34 more rows
With this code df_2 %>% filter(admitted >1 & activity == "admn") I can filter out the inconsistent records at once.
However, I want to include/generate one list column where-ever a new record/case_id has been opened without closing of any previous one, where the hsopital_ids get accumulated whenever activity == 'admn' and hospital_id is removed from existing entries whenever activity == 'discharge'. So basically my desired output for df_2 be something like
Desired OUTPUT
# A tibble: 44 x 8
case_id patient_id pack_id hosp_id sort_key activity date admitted open_records
<int> <int> <int> <int> <int> <ord> <date> <dbl> <list>
1 1 1 12 1 1 admn 2020-01-01 1 1
2 2 1 62 1 2 admn 2020-01-03 2 1, 1
3 3 1 59 2 3 admn 2020-01-04 3 1, 1, 2
4 4 1 68 2 4 admn 2020-01-05 4 1, 1, 2, 2
5 3 1 59 2 3 discharge 2020-01-10 3 1, 1, 2
6 1 1 12 1 1 discharge 2020-01-14 2 1, 2
7 4 1 68 2 4 discharge 2020-01-14 1 1,
8 2 1 62 1 2 discharge 2020-01-15 0 <NULL>
9 5 1 77 1 5 admn 2020-01-16 1 1
10 6 1 86 1 6 admn 2020-01-18 2 1, 1
# ... with 34 more rows
NOTE I am aware that list column won't be displayed in the tibble/data.frame like the one I have shown for explanation purpose only. However, if there is any method by which that can be printed I would like to know about that for sure.
MOREOVER If there is any better strategy to store the hospital ids in the column instead of generating list column, I would also like to know about that for sure.
回答1:
If you don't mind using a loop
library(stringi)
df3 <- df2
df3$open_records <- NA
df3$hosp_id <- as.character(df3$hosp_id) #makes pasting easier
for(i in 1:nrow(df3)){
#if re-admn
if(df3$activity[i] == "admn"){
df3$open_records[i] <- paste(lag(df3$open_records, default = "")[i],
df3$hosp_id[i],
sep = ",")
#we'll handle pretty commas later
}
#if discharge
if(df3$activity[i] == "discharge"){
df3$open_records[i] <- sub(df3$hosp_id[i], "",
stri_reverse(df3$open_records[i-1]))
#sub out one hospital if discharge
#we reverse the string before removing to get the last hosp_id
}
#if admitted == 0
if(df3$admitted[i] == 0) df3$open_records[i] <- NA
#if just starting the group
if(df3$activity[i] == "admn" & df3$admitted[i] == 1){
df3$open_records[i] <- df3$hosp_id[i]
}
}
#comma clean
df3$open_records <- gsub("^,*|(?<=,),|,*$", "", df3$open_records, perl=T)
df3$open_records <- gsub(",", ", ", df3$open_records)
If your dataset is really large this might not be optimal. It might be worth it to add next() commands to each if statement, too (if you do this this, I think it makes sense to move the starting group if statement to the top of the loop).
(comma clean source: Removing multiple commas and trailing commas using gsub)
EDIT, based on need to not use loop
library(tidyverse)
paste3 <- function(out, input, activity, sep = ",") {
if (activity == "admn") {
paste(out, input, sep = sep)
} else
if (activity == "discharge") {
sub(input, "", out)
}
}
df4 <- df2 %>%
mutate(temp_act = lead(activity)) %>%
mutate(open_records = accumulate2(hosp_id, head(temp_act, -1), paste3)
) %>%
select(-temp_act)
df4$open_records <- gsub("^,*|(?<=,),|,*$", "", df4$open_records, perl=T)
df4$open_records <- gsub(",", ", ", df4$open_records)
I noticed that patients can be admitted to the same hospital more than once concurrently. One thing you might want to consider is concatenating the case_id and the hosp_id so instead of removing the first matching hosp_id when a discharge occurs, you can remove the one that corresponds to the correct case_id. (Replace hosp_id in the code with your new variable.)
This doesn't show up in your sample code, but if someone has open_records of 2, 1, 2, 1, 2 and is discharged from their 3rd admittance, my code will return 1, 2, 1, 2 when you probably want 2, 1, 1, 2.
来源：https://stackoverflow.com/questions/65388891/r-how-to-accumulate-values-in-a-list-column-based-on-multiple-criteria