-
Notifications
You must be signed in to change notification settings - Fork 0
/
summarizing_substance_use.Rmd
740 lines (634 loc) · 27.5 KB
/
summarizing_substance_use.Rmd
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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
---
title: "Summarizing Longitudinal Substance Use as Quinary Words"
author: "Gabriel Odom and Raymond Balise"
date: "`r Sys.Date()`"
output:
rmarkdown::html_vignette:
toc: true
toc_depth: 4
vignette: >
%\VignetteIndexEntry{Summarizing Substance Use}
%\VignetteEncoding{UTF-8}
%\VignetteEngine{knitr::rmarkdown}
editor_options:
chunk_output_type: console
---
```{r setup, echo=FALSE}
knitr::opts_chunk$set(
echo = TRUE,
message = FALSE,
warning = FALSE,
collapse = TRUE,
comment = "#>"
)
```
# Introduction
Here are the packages we need for this vignette:
```{r packages}
library(public.ctn0094data)
library(public.ctn0094extra)
library(dplyr)
library(purrr)
library(tibble)
library(tidyr)
library(stringr)
```
In this vignette, we will walk through the basic procedures to collapse participants' Urine Drug Screen (UDS) or Timeline Follow-Back (TLFB) data streams into [quinary word](https://doi.org/10.1371/journal.pone.0291248) summaries for downstream analysis. The basic steps are:
1. calculate the *induction* delay (difference, in days, between the day a participant was randomized to a trial arm and the day that they received the first dose of study drug). The `public.ctn0094data` database has a) the day of randomization and b) the days for which study drugs were administered, relative to Day 0 (the day study consent was signed).
2. partition the study days into weeks, then mark which weeks do not have a UDS collection event (these will be marked as "missing").
3. summarize the UDS data within each participant by week into symbols using [quinary word logic](https://doi.org/10.1371/journal.pone.0291248), then collapse the symbols into a participant-specific "word".
Rather than perform these calculations on all ~3600 participants included in `public.ctn0094data` and `public.ctn0094extra`, we will use the following 10 participants as examples:
```{r}
examplePeople_int <- c(1L, 163L, 210L, 242L, 4L, 17L, 13L, 1103L, 233L, 2089L)
```
-------------------------------------------------------------------------------
</br>
# Calculate Induction Delay
As we mentioned above, the *induction* delay is the difference, in days, between the day a participant was randomized to a trial arm and the day that they received the first dose of study drug. The reason we need this is because any UDS collected or TLFB recorded on or before the induction day must be considered *pre-treatment*, even if the participant has already been assigned to a treatment arm.
## Data Setup
The `randomization` day includes a first and second randomization day for all participants. For CTN-0027 and CTN-0051, these days are the same; for CTN-0030 these days are different. In order to find the induction delay, we only need the first randomization event
```{r trim-data-1}
data_ls <- loadRawData(c("randomization", "treatment"))
data_ls$randomization <-
data_ls$randomization %>%
select(who, when, treatment, randomized = which) %>%
# Remove second randomization events
filter(randomized != 2) %>%
# Retain example participants
filter(who %in% examplePeople_int)
data_ls$treatment <-
data_ls$treatment %>%
# Retain example participants
filter(who %in% examplePeople_int)
```
Let's inspect our results:
```{r view-data-1}
data_ls$randomization
data_ls$treatment
```
## Mark Study Days with Administered Treatment Drugs
Some participants were marked as receiving 0mg of the study drug to which they were assigned. We want to mark all the days where a participant actually got some of the assigned study drug.
```{r mark-study-drug-days}
treatTimeLong_df <-
# Collapse Data List
data_ls %>%
reduce(.f = full_join, by = c("who", "when")) %>%
as_tibble() %>%
arrange(who, when) %>%
# First pass: find any day with a dose of treatment drug
rowwise() %>%
mutate(
treated = if_else(
condition = !is.na(amount) & amount > 0,
true = TRUE,
false = FALSE,
missing = FALSE
)
)
# Inspect results
treatTimeLong_df
```
## Measure Difference from Randomization Day to Day of First Dose
Some participants received their first non-zero dose of their assigned study drug days after they were assigned to a treatment arm. If the subject supplies a UDS sample positive for substances of interest or records a substance use event in the TLFB after they were assigned to treatment but before they received treatment, these should not count against the efficacy of the study drug. **NOTE: under the "intent to treat" paradigm, this is not the case. For example, it is challenging for some patients to start treatment with Naltrexone, so evaluating treatment efficacy fairly often requires us to include this time after treatment assignment but before induction.**
```{r calculate-induction-delay}
inductDelay_df <-
treatTimeLong_df %>%
# Find the day of the first treatment
group_by(who) %>%
arrange(when) %>%
filter(treated) %>%
slice(1) %>%
mutate(treatStart = when) %>%
select(who, when, treatStart) %>%
# Add the first day back to the original data
left_join(treatTimeLong_df, ., by = c("who", "when")) %>%
group_by(who) %>%
fill(treatStart, .direction = "updown") %>%
# Calculate the delay
filter(randomized == 1) %>%
# This sets time to be missing if the induction was not observed
mutate(inductDelay = treatStart - when) %>%
select(who, treatment, inductDelay) %>%
ungroup()
# Inspect results
inductDelay_df
```
```{r echo=FALSE}
eg_df <-
inductDelay_df %>%
filter(inductDelay > 0)
```
We can see that participant `r eg_df$who` received their first dose of `r eg_df$treatment` `r eg_df$inductDelay` days after they were assigned to that treatment arm.
```{r echo=FALSE}
rm(eg_df)
```
Let's clean up our environment before we move on:
```{r clean-up-1}
rm(data_ls, treatTimeLong_df)
```
-------------------------------------------------------------------------------
</br>
# Partition Visit Weeks and Impute Missing Visits
Now that we have the study day for which each participant received their first dose of study drug (induction day), we can partition the study days into weeks before and after induction delay as pre-study / baseline period and study period, respectively.
## "Backbone" Timeline of Clinical Protocol
If we want to know which visits were missed, we first need to know which visits were required by the protocols of the three studies.
### CTN-0027 and CTN-0051 Protocols
This code will create a table with the subject ID, which trial they participated in, and a column of all possible trial contact days, from 30 days prior to consent to 24 weeks after. See the function documentation for `CreateProtocolHistory()` for more information.
```{r backbone-2751}
start_int <- c(`27` = -30L, `51` = -30L)
end_int <- c(`27` = 168L, `51` = 168L)
backbone2751_df <-
CreateProtocolHistory(
start_vec = start_int,
end_vec = end_int
) %>%
filter(who %in% examplePeople_int)
# Inspect
backbone2751_df
```
### CTN-0030 Protocol
Because the CTN-0030 protocol was adaptive, the amount of time requested to be spent in the study will change based on the participants' reactions to the first phase of treatment. See the function documentation for `CreateCTN30ProtocolHistory()` for more information.
```{r backbone-30}
backbone30_df <-
randomization %>%
full_join(everybody, by = "who") %>%
filter(project == "30") %>%
filter(who %in% examplePeople_int) %>%
CreateCTN30ProtocolHistory() %>%
mutate(project = "30") %>%
select(who, project, when)
# Inspect
backbone30_df
backbone30_df %>%
group_by(who) %>%
summarise(lastDay = max(when))
```
We can clearly see that the protocol length for some participants in CTN-0030 was much longer than others.
### Combined "Backbone"
We can now combine these two datasets and clean up our environment.
```{r backbone-first}
backbone_df <-
bind_rows(backbone2751_df, backbone30_df) %>%
arrange(who)
rm(backbone2751_df, backbone30_df, start_int, end_int)
```
## Mark Missing Visits
Now that we know the start of treatment and range of expected participation days per protocol, we can take the "difference" to find out when participants were supposed to visit the clinic, but did not.
### Data Setup
Let's set up the data for our example people:
```{r trim-data-2}
data_ls <- loadRawData(c("randomization", "visit"))
data_ls$randomization <-
data_ls$randomization %>%
select(who, when, treatment, randomized = which) %>%
# Remove second randomization events
filter(randomized != 2) %>%
# Retain example participants
filter(who %in% examplePeople_int)
data_ls$visit <-
data_ls$visit %>%
filter(who %in% examplePeople_int)
```
### Add First Randomization Day
```{r add-randomization-day}
timelineRand1_df <-
data_ls$randomization %>%
mutate(randomized = randomized == "1") %>%
# Join to backbone and arrange within subject by day
full_join(backbone_df, by = c("who", "when")) %>%
group_by(who) %>%
arrange(when, .by_group = TRUE) %>%
select(who, project, when, randomized)
# Inspect
timelineRand1_df
```
### Add on the Visit Days
```{r add-visit-days}
timelineVisit1_df <-
data_ls$visit %>%
select(who, when, visit, status = what) %>%
filter(status %in% c("visit", "final")) %>%
mutate(visit = TRUE) %>%
select(who, when, visit) %>%
left_join(timelineRand1_df, ., by = c("who", "when"))
# Inspect
timelineVisit1_df
```
Most (6 of 7 days, on average) of the data will be missing values (shown by `NA`) because participants were not expected to visit the clinic more than once per week.
### Impute the Missing Visits
This function will assign visit values to "missing" roughly every seven days that a visit hasn't been recorded. See the function documentation for `MarkMissing()` for more information. *Note: this function can take a while to run if used on thousands of participants over dozens of weeks each.*
```{r mark-missing-visits}
timelineMissing1_df <- MarkMissing(timelineVisit1_df)
```
### Clean up the Results
We aren't showing the results from the function above because they still need to be wrangled a bit.
```{r tidy-missing-visit-marks}
derived_visitImputed <-
timelineMissing1_df %>%
mutate(visit = as.character(visit)) %>%
replace_na(list(visit = "", visitYM = "")) %>%
mutate(visitImputed = paste0(visit, visitYM)) %>%
mutate(
visitImputed = str_replace(
visitImputed, pattern = "TRUETRUE", replacement = "Present"
)
) %>%
select(who, when, visitImputed) %>%
filter(visitImputed != "") %>%
ungroup()
# Inspect
derived_visitImputed
```
Let's now clean up our environment:
```{r clean-up-2}
rm(
backbone_df, data_ls, timelineMissing1_df, timelineRand1_df, timelineVisit1_df
)
```
-------------------------------------------------------------------------------
</br>
# Summarize Weekly Substance Use Results
This step is the most involved, so we will only show the steps necessary to complete this task for our 10 example participants. However, the full workflow is available at <LINK>. The general steps for randomized trial participants in this procedure are:
1. Determine which treatment arm to which each participant was assigned. Recall that buprenorphine can be used both as a standard treatment and an illicit substance. Therefore, having a TLFB record of or UDS positive for buprenorphine for a particular study day should not automatically be counted as "substance misuse".
2. For the substance(s) of interest (in our case, opioids), indicate their presence (or absence) in participants' data streams for each study visit day. This will result in a "long" data table with one row per participant per study visit and a logical value indicating if the participant used the substance(s) of interest on that day.
3. Create a "ticker" of study days, then partition these days into study weeks. For an "intent to treat" analysis, day 0 should be the randomization day. Otherwise, day 0 should be the induction day calculated above. *Note: we will use an "intent to treat" analysis from this point on.*
4. Summarize substance use within each study week using a [quinary word](https://doi.org/10.1371/journal.pone.0291248). For many participants, they have at least one week wherein more than one UDS sample was supplied. Our combination logic to summarize detected weekly substance use is as follows: if all UDS samples within that study week are positive for the substance(s) of interest, then the week is marked "+". If all UDS samples are negative for the substance(s) of interest, then the week is marked "-". If there is a mixture, the week is marked "\*". If the participant was supposed to supply a urine sample in that week but did not, then the week is marked "o". If the participant was not supposed to supply a urine sample in that week (and did not supply one), then the week is marked "\_".
5. Group these weekly summaries into trial phase (baseline / pre-randomization, phase 1, or phase 2 [CTN-0030 only]) and collapse into a single string.
You may have participants who consented to join the trial but were never randomized to a treatment arm. For these individuals, if a "complete study summary" is required (that is, you must include all participants who consented, even if they never were randomized), we recommend creating a use pattern word which is either all "o" or all "\_".
## Data Setup
First we build our data "backbone" as above (showing here to mirror the work done in the analysis script):
```{r backbone-again}
# CTN-0027 and CTN-0051
start_int <- c(`27` = -30L, `51` = -30L)
end_int <- c(`27` = 168L, `51` = 168L) # 24 weeks
backbone2751_df <- CreateProtocolHistory(
start_vec = start_int, end_vec = end_int
)
# CTN-0030
backbone30_df <-
randomization %>%
full_join(everybody, by = "who") %>%
filter(project == "30") %>%
CreateCTN30ProtocolHistory() %>%
mutate(project = "30") %>%
select(who, project, when)
# All Days
backbone_df <- bind_rows(
backbone2751_df, backbone30_df
) %>%
arrange(who) %>%
mutate(project = factor(project, levels = c("27", "30", "51"))) %>%
filter(who %in% examplePeople_int)
rm(backbone2751_df, backbone30_df, start_int, end_int)
```
Now we create a data set that combines the treatment arm with the imputed visits data we calculated above.
```{r}
randomized_df <-
randomization %>%
filter(who %in% examplePeople_int) %>%
mutate(randomized = as.integer(as.character(which))) %>%
select(who, when, randomized) %>%
left_join(everybody, by = "who") %>%
filter( !(randomized == 2 & project %in% c("27", "51")) ) %>%
select(-project)
udsUse2_df <-
backbone_df %>%
left_join(randomized_df, by = c("who", "when")) %>%
left_join(derived_visitImputed, by = c("who", "when")) %>%
left_join(uds, by = c("who", "when")) %>%
# So we can use MarkUse() with UDS data (instead of all_drugs)
mutate(source = "UDS")
```
This data set is an interesting one. It has one record per participant, per study day, *per substance reported*. So, because we are using UDS, this means that only rows which correspond to *study visit days* wherein a urine sample was supplied and positive for multiple substances will be duplicated. If we had used TLFB data instead, a data table in this form would be orders of magnitude longer, with many rows per participant per day. Here are visit days wherein substances were detected in the urine of participant 0017:
```{r eg-visit-day-substances}
udsUse2_df %>%
filter(visitImputed == "Present") %>%
filter(!is.na(what)) %>%
filter(who == 17) %>%
print(n = nrow(.))
```
## Participant's Assigned Treatment
Which substances are considered "approved" and which are "illicit"?
```{r good-drugs-bad-drugs}
nonStudyOpioids_ls <- list(
"Buprenorphine" = c("Opioid", "Methadone"),
"Methadone" = c("Opioid", "Buprenorphine"),
"Naltrexone" = c("Opioid", "Methadone", "Buprenorphine"),
"Not treated" = c("Opioid", "Methadone", "Buprenorphine")
)
```
Now we extract the treatment groups for each clinical trial, so that we can mark "illicit" buprenorphine and methadone appropriately.
```{r treatment-groups}
treatGroups_ls <-
public.ctn0094data::randomization %>%
filter(who %in% examplePeople_int) %>%
filter(which == 1) %>%
left_join(everybody, by = "who") %>%
select(who, treatment) %>%
mutate(
treat_drug = case_when(
str_detect(treatment, "BUP") ~ "Buprenorphine",
treatment == "Methadone" ~ "Methadone",
treatment == "Inpatient NR-NTX" ~ "Naltrexone"
)
) %>%
select(-treatment) %>%
split(f = .$treat_drug) %>%
map(.f = "who")
# Inspect
treatGroups_ls
```
This shows us which participants were assigned to buprenorphine, methadone, and naltrexone, respectively. We can now compare the substances present in urine against the substances the participants are supposed to have in their urine. For more information, see the documentation for the `MarkUse()` function.
```{r mark-non-study-drug-use, warning=TRUE}
opioidUse_df <-
udsUse2_df %>%
mutate(
treat_group = case_when(
who %in% treatGroups_ls$Buprenorphine ~ "Buprenorphine",
who %in% treatGroups_ls$Methadone ~ "Methadone",
who %in% treatGroups_ls$Naltrexone ~ "Naltrexone",
TRUE ~ "Not treated"
)
) %>%
split(f = .$treat_group) %>%
# List of data in alphabetical order, so the non-study drugs ls should match
map2(
.y = nonStudyOpioids_ls,
.f = ~{
# REQUIRES "source" COLUMN
MarkUse(
targetDrugs_char = .y,
drugs_df = .x,
# because we have participants with no recorded UDS; in practice DO NOT
# use this command
retainEmptyRows = TRUE
)
}
) %>%
bind_rows() %>%
mutate(
udsOpioid = case_when(
is.na(when) ~ NA,
!is.na(when) ~ TRUE
)
) %>%
select(who, when, udsOpioid)
```
All of the drugs marked above would still be individual rows, so we want to get back to "one row per person per day"
```{r timeline-to-one-row-per-day}
timelineUDS_df <-
udsUse2_df %>%
left_join(opioidUse_df, by = c("who", "when")) %>%
select(-what, -source) %>%
# 2,089 rows to 1,994
distinct()
rm(
derived_visitImputed, opioidUse_df, randomized_df, treatGroups_ls, udsUse2_df
)
```
## Counting Days Since Randomization
Do we state that by definition any person who wasn't randomized is an early treatment failure? In the sense of evaluating treatment efficacy, yes; in evaluating the subject, no. Regardless, no matter the treatment outcome definition, participants without a randomization date will be listed as treatment failures under the "intent to treat" paradigm.
```{r sets-of-randomized-people}
wasRandomized_int <-
timelineUDS_df %>%
group_by(who) %>%
summarise(randomized = any(randomized %in% 1:2)) %>%
filter(randomized) %>%
pull(who)
notRandomized_int <-
timelineUDS_df %>%
filter( !(who %in% wasRandomized_int) ) %>%
pull(who) %>%
unique()
# Was randomized:
wasRandomized_int
# Wasn't
notRandomized_int
```
Now we need a Study Day ticker (for randomized subjects only). Recall that CTN-0030 trial participants potentially have 2 randomization days.
```{r study-day-ticker-randomized}
timelineUDS2_df <-
timelineUDS_df %>%
filter(who %in% wasRandomized_int) %>%
group_by(who) %>%
filter(!is.na(randomized)) %>%
mutate(
whenRandomized1 = case_when(randomized == 1 ~ when),
whenRandomized2 = case_when(randomized == 2 ~ when)
) %>%
select(who, when, whenRandomized1, whenRandomized2) %>%
left_join(timelineUDS_df, ., by = c("who", "when")) %>%
filter(who %in% wasRandomized_int) %>%
# Add back in the groupings BEFORE the fill()
group_by(who) %>%
fill(whenRandomized1, .direction = "updown") %>%
fill(whenRandomized2, .direction = "updown") %>%
mutate(daysSinceRand1 = when - whenRandomized1) %>%
mutate(daysSinceRand2 = when - whenRandomized2) %>%
select(-whenRandomized1, -whenRandomized2)
```
## Symbolically Summarize Participant UDS by Study Week
We puts the first day of "study week 1" on the day after randomization, not on the day of consent, but consent could have been signed also on the same day as randomization.
```{r weekly-use-summary}
weeklyUse_df <-
timelineUDS2_df %>%
# The (daysSinceRand1 - 1) adjustment is to ensure that the first study week
# is a full 7 days, since "day 0" is the day before randomization. The "+1"
# at the end is to shift the study week label such that "week 0" is the
# week *BEFORE* treatment, rather than the first week of treatment. So, the
# randomization day is the last day of "week 0" (the pre-treatment period).
mutate(studyWeek1 = (daysSinceRand1 - 1) %/% 7 + 1) %>%
mutate(studyWeek2 = (daysSinceRand2 - 1) %/% 7 + 1) %>%
group_by(who, studyWeek1) %>%
# There are some study weeks with multiple UDS, so we count the number of
# positive and negative UDS per week.
summarise(
nPosUDS = sum(udsOpioid == 1, na.rm = TRUE),
nNegUDS = sum(visitImputed == "Present" & is.na(udsOpioid), na.rm = TRUE),
nMissing = sum(visitImputed == "Missing", na.rm = TRUE),
randWk1 = sum(randomized == 1, na.rm = TRUE) > 0,
randWk2 = sum(randomized == 2 & project == "30", na.rm = TRUE) > 0
) %>%
ungroup()
```
Now we assign a single symbol to represent the UDS results for each participant week. The symbols are (+) for only positive UDS, (--) only negative UDS, (\*) at least one of each positive and negative UDS, (o) UDS required but not collected, or (\_) UDS not required.
```{r single-week-symbols}
useByWeekRandomized_df <-
weeklyUse_df %>%
mutate(
udsStatus = case_when(
# If we see a positive UDS and no negative UDS, it's positive
nPosUDS > 0 & nNegUDS == 0 ~ "+",
# If we see a negative UDS and no positive UDS, it's negative
nPosUDS == 0 & nNegUDS > 0 ~ "-",
# If we see both positive and negative UDS in a single week, it's both
# (note that we can recode all "B"s to be "+" as necessary)
nPosUDS > 0 & nNegUDS > 0 ~ "*",
# If we don't have any UDS in a week after randomization, it's missing
# UPDATE 2022-03-08: I had this as a 0 originally, and I was using this
# in context of consent, not randomization. This was wrong.
nPosUDS == 0 & nNegUDS == 0 & studyWeek1 >= 1 ~ "o",
# If none of the above are true, but we still have a missing value as
# marked by the MarkMissing() function, then it's missing
nMissing > 0 ~ "o",
# If none of the above conditions are true (probably because it's a week
# before randomization but not during a baseline visit for consent),
# then leave it blank (pre-study)
TRUE ~ "_"
)
) %>%
group_by(who) %>%
# For CTN-0030, Phase II could have started on any day of the week, even in
# the middle of a treatment week. If we try to start counting Phase II
# weeks the day after treatment arms are switched, we can end up with the
# last "week" of Phase I not having 7 days. I'm going to leave the first
# week of Phase II as whatever week the switch happened in.
mutate(
rand1Active = studyWeek1 > 0,
# This returns 0 for any week before the Phase II randomization, and 1 for
# the Phase II randomization week and all subsequent weeks (because the
# randWk2 column is 1 only for the week of second randomization and 0
# all other rows).
rand2Active = cumsum(randWk2),
trialPhase = rand1Active + rand2Active
) %>%
select(
who, studyWeek = studyWeek1, randWk1, randWk2, udsStatus, trialPhase
)
```
We then split these rows by trial phase, and collapse the weekly symbols into a single string.
```{r use-patterns-by-phase}
weeklyOpioidPatterns_df <-
useByWeekRandomized_df %>%
mutate(
phase = case_when(
trialPhase == 0L ~ "Baseline",
trialPhase == 1L ~ "Phase_1",
trialPhase == 2L ~ "Phase_2"
)
) %>%
group_by(who, phase) %>%
summarise(
usePattern = paste0(udsStatus, collapse = "")
) %>%
pivot_wider(names_from = "phase", values_from = "usePattern")
weeklyOpioidPatterns_df
```
Notice that we have the opioid UDS results for each subject by week split into the various phases of the clinical trial. These summaries are in **quinary word** format, following the work in Odom et al. (2023) [<https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10490938/>].
Our last step for the randomized participants is to mark the study weeks when these phases start and end.
```{r use-pattern-phases-start-and-end}
derived_weeklyOpioidPatternRand <-
useByWeekRandomized_df %>%
mutate(
randWeek1 = randWk1 * studyWeek,
randWeek2 = randWk2 * studyWeek
) %>%
summarise(
startWeek = min(studyWeek),
randWeek1 = max(randWeek1),
randWeek2 = if_else( all(randWeek2 == 0), NA_real_, max(randWeek2) ),
endWeek = max(studyWeek)
) %>%
# View this smaller data set before joining. In practice, you can comment out
# this print() command.
print() %>%
left_join(weeklyOpioidPatterns_df, by = "who")
```
## Symbolically Summarize Non-Randomized Participants UDS by Study Week
There are some strange cases where non-randomized participants also submitted UDS samples during the clinical trial. We will also create quinary words for these participants, but note that most of the UDS results will be missing (symbolized by "o").
```{r mark-use-non-randomized}
nonRandUDS_df <-
backbone_df %>%
filter(who %in% notRandomized_int) %>%
left_join(uds) %>%
# MarkUse() requires a "source" column
mutate(source = "UDS") %>%
MarkUse(
targetDrugs_char = nonStudyOpioids_ls$`Not treated`,
drugs_df = .
) %>%
mutate(
udsOpioid = case_when(
!is.na(when) ~ TRUE,
is.na(when) ~ NA
)
) %>%
select(who, when, udsOpioid) %>%
# Uneccessary here, but some UDS records are duplicated
distinct()
```
Although not present in this small data example, there are positive UDS events among the non-randomized. We then construct a "study week" ticker to mark such weekly UDS results.
```{r non-randomized-study-week}
timelineNonRandUDS_df <-
backbone_df %>%
filter(who %in% notRandomized_int) %>%
left_join(nonRandUDS_df, by = c("who", "when")) %>%
# Because this week moves off of the consent date, there is no reason to add
# a `(week - 1)` adjustment
mutate(studyWeek = when %/% 7 + 1) %>%
group_by(who, studyWeek) %>%
summarise(
posUDS = sum(udsOpioid == 1, na.rm = TRUE) > 0
)
```
For these participants who were not randomized at all, they will only have "Phase I" missing values.
```{r single-week-symbols-non-randomized}
weeklyNonRandPatterns_df <-
timelineNonRandUDS_df %>%
mutate(
udsStatus = case_when(
# If they are positive, they are positive
posUDS ~ "+",
# If they aren't positive and it's after the consent week, they are
# missing (because they weren't randomized)
!posUDS & studyWeek >= 1 ~ "o",
# If they aren't positive and it's on or before the consent week, then
# leave it blank (pre-study)
!posUDS & studyWeek < 1 ~ "_"
)
) %>%
mutate(
phase = case_when(
studyWeek < 1 ~ "Baseline",
studyWeek >= 1 ~ "Phase_1"
)
) %>%
# Again, this print is unecessary, but here it make clear what we are doing
print() %>%
group_by(who, phase) %>%
summarise(
usePattern = paste0(udsStatus, collapse = "")
) %>%
pivot_wider(names_from = "phase", values_from = "usePattern") %>%
ungroup()
weeklyNonRandPatterns_df
```
Finally, we add in the start and end weeks for these two phases.
```{r non-randomized-start-and-end}
derived_weeklyOpioidPatternNonRand <-
timelineNonRandUDS_df %>%
group_by(who) %>%
summarise(
startWeek = min(studyWeek),
randWeek1 = NA_real_,
randWeek2 = NA_real_,
endWeek = max(studyWeek)
) %>%
left_join(weeklyNonRandPatterns_df, by = "who")
```
## Final Product
At the end of this very long process, we have this:
```{r final-product}
derived_weeklyOpioidPattern <-
derived_weeklyOpioidPatternRand %>%
bind_rows(derived_weeklyOpioidPatternNonRand) %>%
arrange(who) %>%
replace_na(list(Phase_2 = ""))
derived_weeklyOpioidPattern
```
-------------------------------------------------------------------------------
</br>
# Wrapping Up
```{r}
sessionInfo()
```