generated from opensafely/research-template
-
Notifications
You must be signed in to change notification settings - Fork 0
/
validation_events.R
242 lines (199 loc) · 7.8 KB
/
validation_events.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
######################################
# This script compares ascertainment of covid-19 hospitalisations via SUS (APCS or ECDS) with that via ISARIC
######################################
# preliminaries ----
## Import libraries ----
library('tidyverse')
library('here')
library('glue')
library('fuzzyjoin')
library('data.table')
## Import custom user functions from lib
source(here("analysis", "lib", "utility.R"))
start_date = as.Date("2020-02-01")
end_date = as.Date("2022-11-30")
rounding_threshold <- 10
## output processed data to rds ----
output_dir <- here("output", "validation_events")
fs::dir_create(output_dir)
## import data ----
# import sus data
import_sus_data <- function(method){
dat <- read_rds(here("output", "admissions", glue("processed_sus_{method}.rds")))
dat <- transmute(
dat,
patient_id,
admission_number,
admission_date,
"admission_date_sus{method}" := admission_date,
"method_sus{method}" := TRUE
)
return(dat)
}
admissions_susA <- import_sus_data("A")
admissions_susB <- import_sus_data("B")
admissions_susC <- import_sus_data("C")
# import ISARIC data
admissions_isaric <- read_rds(here("output", "admissions", "processed_isaric.rds")) %>%
transmute(
patient_id,
admission_number,
admission_date,
admission_date_isaric=admission_date,
method_isaric=TRUE
)
# compare ascertainment of COVID-19 admissions between SUS and ISARIC ----
# Note, we assume ISARIC as a tool for identifying COVID-19 in-patient admissions is 100% specific, but not 100% sensitive.
# In other words, it may not pick up every COVID-19 admission, either because
# - lack of coverage in certain hospitals
# - some admissions were missed in participating hospitals
# - admission date was badly recorded
# but every ISARIC record corresponds to a "true" COVID-19 admission
# our goal is to see if national hospital data (provided via SUS) is capable of identifying those records picked up by SUS
## Number of first ISARIC admissions in TPP
nrow(admissions_isaric %>% filter(admission_number==1))
nrow(admissions_isaric %>% filter(admission_number==1, admission_date >= start_date, admission_date <= end_date))
## Number of first SUS admissions in TPP
nrow(admissions_susA %>% filter(admission_number ==1))
nrow(admissions_susB %>% filter(admission_number ==1))
nrow(admissions_susC %>% filter(admission_number ==1))
nrow(admissions_susA %>% filter(admission_number ==1, admission_date >= start_date, admission_date <= end_date))
nrow(admissions_susB %>% filter(admission_number ==1, admission_date >= start_date, admission_date <= end_date))
nrow(admissions_susC %>% filter(admission_number ==1, admission_date >= start_date, admission_date <= end_date))
# all dates of admission per patient up to <end_date>, and each ascertainment method where a match was found for that date
admissions_joined <-
reduce(
lst(
admissions_isaric %>% filter(admission_number == 1),
admissions_susA %>% select(-admission_number),
admissions_susB %>% select(-admission_number),
admissions_susC %>% select(-admission_number),
),
full_join,
by = c("patient_id", "admission_date")
) %>%
replace_na(
lst(
method_isaric=FALSE,
method_susA=FALSE,
method_susB=FALSE,
method_susC=FALSE,
)
) %>%
filter(
admission_date >= start_date,
admission_date <= end_date
)
# sensitivity of SUS for picking up admissions reported in ISARIC ----
## match records using exact date ----
ascertainment <-
admissions_joined %>%
# recover left-join on isaric admissions
filter(method_isaric) %>%
# remove multiple same-day matches from SUS
group_by(patient_id, admission_date) %>%
mutate(dup_id = row_number()) %>%
ungroup() %>%
filter(dup_id==1L) %>%
# summarise number of SUS admissions that match isaric admissions
summarise(
isaric_n = roundmid_any(n(), 10),
susA_n = roundmid_any(sum(method_susA),10),
susB_n = roundmid_any(sum(method_susB),10),
susC_n = roundmid_any(sum(method_susC),10),
susA_prop = susA_n/isaric_n,
susB_prop = susB_n/isaric_n,
susC_prop = susC_n/isaric_n,
) %>%
pivot_longer(
cols=everything(),
names_to = c("method", ".value"),
names_pattern = "(.*)_(.*)"
)
ascertainment
write_csv(ascertainment, fs::path(output_dir, "ascertainment.csv"))
remove(admissions_joined)
## match records using non-exact date ----
# we use fuzzy (=non-equi) matching in case admission dates are slightly different (and to pick up different spells/episodes within a single super spell, etc)
# all dates of admission via ISARIC per patient up to <end_date>, and each SUS ascertainment method where a match was found for that date +/- X days
function_fuzzy_join <- function(mx, my, x_days_before, x_days_after){
# this function is easy to read but is very very slow
# there is now an equivalent function to `fuzzy_left_join` in dplyr (and probably much faster), but this version of dplyr is not currently in the opensafely R image
joined <- fuzzy_left_join(
mx,
my,
by = c("patient_id", "admission_date"),
match_fun = list(
patient_id = `==`,
admission_date = function(x,y){x>=y-x_days_before & x<=y+x_days_after}
)
)
renamed <- rename(joined, patient_id=patient_id.x, admission_date=admission_date.x)
select(renamed, -patient_id.y, -admission_date.y)
}
function_nonequi_join <- function(mx, my, x_days_before, x_days_after){
# this function uses the data.table package. it's harder to read, but it is very very quick!
joined <- setDT(my)[
,
c("admission_date_pre", "admission_date_post") := list(admission_date-x_days_before, admission_date+x_days_after)
][
setDT(mx),
on = .(patient_id, admission_date_pre<=admission_date, admission_date_post>=admission_date)
]
joined <- select(joined, -admission_date_pre, -admission_date_post) %>% mutate(admission_date=admission_date_isaric)
joined
}
admissions_joined_nonequi <-
reduce(
lst(
admissions_isaric %>% filter(admission_number == 1),
admissions_susA %>% select(-admission_number),
admissions_susB %>% select(-admission_number),
admissions_susC %>% select(-admission_number),
),
~{function_nonequi_join(.x, .y, 2, 2)},
) %>%
mutate(
admission_date_diff_susA = admission_date_susA - admission_date_isaric,
admission_date_diff_susB = admission_date_susB - admission_date_isaric,
admission_date_diff_susC = admission_date_susC - admission_date_isaric
) %>%
replace_na(
lst(
method_isaric=FALSE,
method_susA=FALSE,
method_susB=FALSE,
method_susC=FALSE,
)
) %>%
filter(
admission_date >= start_date,
admission_date <= end_date
)
ascertainment_nonequi <-
admissions_joined_nonequi %>%
# remove multiple matches within period -- ignore thinking about how to select the most appropriate match for now as it doesn't matter in the summary
group_by(patient_id, admission_date) %>%
mutate(dup_id = row_number()) %>%
ungroup() %>%
filter(dup_id==1L) %>%
# summarise number of SUS admissions that closely match isaric admissions
summarise(
isaric_n = roundmid_any(n(), 10),
susA_n = roundmid_any(sum(method_susA),10),
susB_n = roundmid_any(sum(method_susB),10),
susC_n = roundmid_any(sum(method_susC),10),
susA_prop = susA_n/isaric_n,
susB_prop = susB_n/isaric_n,
susC_prop = susC_n/isaric_n,
susA_datediff = plyr::round_any(as.numeric(mean(admission_date_diff_susA, na.rm=TRUE)),1/24),
susB_datediff = plyr::round_any(as.numeric(mean(admission_date_diff_susB, na.rm=TRUE)),1/24),
susC_datediff = plyr::round_any(as.numeric(mean(admission_date_diff_susC, na.rm=TRUE)),1/24),
) %>%
pivot_longer(
cols=everything(),
names_to = c("method", ".value"),
names_pattern = "(.*)_(.*)"
)
ascertainment_nonequi
write_csv(ascertainment_nonequi, fs::path(output_dir, "ascertainment_nonequi.csv"))