Skip to content

Answers, Parts IV and V (chapters 15 to 19)

Each answer has a program that was run, and its output. The programs of chapters 15 and 16 read the ADaM data sets in programs/data/, so run the programs of chapters 9 to 12 first, in file order. Try the exercise yourself first. A different program that gives the same numbers is also a right answer.

programs/15_x1_trtdurd.R
# Answer 15-1: TRTDURD by a second way, compared with the reference
library(dplyr)
library(diffdf)
ours <- readRDS("data/adsl.rds") |>
transmute(USUBJID, TRTDURD = as.numeric(TRTEDT - TRTSDT) + 1) |>
haven::zap_label()
ref <- pharmaverseadam::adsl |>
filter(ARMCD != "Scrnfail") |>
select(USUBJID, TRTDURD) |>
haven::zap_label()
ours |>
filter(USUBJID == "01-701-1015")
#> # A tibble: 1 x 2
#> USUBJID TRTDURD
#> <chr> <dbl>
#> 1 01-701-1015 182
diffdf(ours, ref, keys = "USUBJID")
#> Warning message:
#> In diffdf(ours, ref, keys = "USUBJID") :
#> Not all Values Compared Equal
#> Differences found between the objects!
#>
#> Summary of BASE and COMPARE
#> ==================================================================
#> PROPERTY BASE COMP
#> ------------------------------------------------------------------
#> Name ours ref
#> Class "tbl_df, tbl, data.frame" "tbl_df, tbl, data.frame"
#> Rows(#) 254 254
#> Columns(#) 2 2
#> ------------------------------------------------------------------
#>
#>
#> Not all Values Compared Equal
#> =============================
#> Variable No of Differences
#> -----------------------------
#> TRTDURD 6
#> -----------------------------
#>
#>
#> ======================================
#> VARIABLE USUBJID BASE COMPARE
#> --------------------------------------
#> TRTDURD 01-704-1233 116 15
#> TRTDURD 01-705-1018 8 NA
#> TRTDURD 01-705-1031 166 22
#> TRTDURD 01-705-1303 169 15
#> TRTDURD 01-705-1377 63 22
#> TRTDURD 01-705-1382 1 NA
#> --------------------------------------
#>

For 01-701-1015, TRTDURD = 182 in both. Six subjects differ, and they are the six of TRTEDT in section 15.6. TRTDURD is computed from TRTEDT, so a rule of TRTEDT becomes a difference of TRTDURD. For 01-705-1018 and 01-705-1382 the reference has no TRTEDT, so it has no TRTDURD either. Write this as part of finding 15-01, not as a new finding: it has the same cause.

programs/15_x2_right.R
# Answer 15-2: cut() with right = FALSE moves the boundary to the other end
library(dplyr)
library(diffdf)
prod <- readRDS("data/adsl.rds") |>
select(USUBJID, AGE, AGEGR1) |>
haven::zap_label()
qc <- prod |>
mutate(AGEGR1 = as.character(cut(
AGE,
breaks = c(-Inf, 65, 80, Inf), labels = c("<65", "65-80", ">80"),
right = FALSE
)))
res <- diffdf(prod, qc, keys = "USUBJID")
#> Warning message:
#> In diffdf(prod, qc, keys = "USUBJID") :
#> Not all Values Compared Equal
res$NumDiff
#> # A tibble: 1 x 2
#> Variable `No of Differences`
#> * <chr> <int>
#> 1 AGEGR1 11
res$VarDiff_AGEGR1 |>
left_join(prod |> select(USUBJID, AGE), by = join_by(USUBJID)) |>
count(AGE, BASE, COMPARE)
#> # A tibble: 1 x 4
#> AGE BASE COMPARE n
#> <dbl> <chr> <chr> <int>
#> 1 80 65-80 >80 11

With right = FALSE, each interval is closed on the left: [65, 80) holds 65 to 79. Now the 4 subjects aged 65 are right, but 11 subjects aged 80 move to >80. The boundary moved from one end to the other. For whole years, the right break points are 64 and 80 with the default right = TRUE, as in program 15_1. A full compare finds either mistake.

programs/15_x3_ex_rows.R
# Answer 15-3: the EX rows of the six subjects behind the TRTEDT differences
library(dplyr)
ours <- readRDS("data/adsl.rds") |>
select(USUBJID, TRTEDT)
ref <- pharmaverseadam::adsl |>
select(USUBJID, TRTEDT_REF = TRTEDT)
six <- inner_join(ours, ref, by = join_by(USUBJID)) |>
filter(!TRTEDT %in% TRTEDT_REF | is.na(TRTEDT_REF)) |>
pull(USUBJID) |>
as.character()
six
#> [1] "01-704-1233" "01-705-1018" "01-705-1031" "01-705-1303" "01-705-1377"
#> [6] "01-705-1382"
pharmaversesdtm::ex |>
filter(USUBJID %in% six) |>
select(USUBJID, EXSEQ, EXSTDTC, EXENDTC) |>
print(n = 30)
#> # A tibble: 10 x 4
#> USUBJID EXSEQ EXSTDTC EXENDTC
#> <chr> <dbl> <chr> <chr>
#> 1 01-704-1233 1 2013-03-21 2013-04-04
#> 2 01-704-1233 2 2013-04-05 <NA>
#> 3 01-705-1018 1 2013-07-05 <NA>
#> 4 01-705-1031 1 2013-11-27 2013-12-18
#> 5 01-705-1031 2 2013-12-19 <NA>
#> 6 01-705-1303 1 2013-12-16 2013-12-30
#> 7 01-705-1303 2 2013-12-31 <NA>
#> 8 01-705-1377 1 2014-01-04 2014-01-25
#> 9 01-705-1377 2 2014-01-26 <NA>
#> 10 01-705-1382 1 2013-05-13 <NA>

All six subjects have an EX row without an end date (EXENDTC is NA), and it is always their last row. For 01-705-1018 and 01-705-1382 it is the only row. This is the “open dose row” of section 9.3. The list is the evidence for the finding note of section 15.7.

programs/15_x4_errors.R
# Answer 15-4: the log check also records an error, and then stops reading
library(dplyr)
check_log <- function(file) {
found <- tibble(kind = character(), text = character())
keep <- function(kind, cnd) {
text <- gsub("\\s+", " ", conditionMessage(cnd))
found <<- add_row(found, kind = kind, text = substr(text, 1, 60))
}
tryCatch(
withCallingHandlers(
source(file, local = new.env()),
warning = function(w) {
keep("WARNING", w)
invokeRestart("muffleWarning")
},
message = function(m) {
keep("MESSAGE", m)
invokeRestart("muffleMessage")
}
),
error = function(e) keep("ERROR", e)
)
found
}
prog <- tempfile(fileext = ".R")
writeLines(c(
"x <- mean(c('a', 'b'))",
"adsl <- dplyr::filter(pharmaverseadam::adsl, AEG > 80)",
"y <- 1"
), prog)
check_log(prog)
#> # A tibble: 2 x 2
#> kind text
#> <chr> <chr>
#> 1 WARNING argument is not numeric or logical: returning NA
#> 2 ERROR i In argument: `AEG > 80`. Caused by error: ! object 'AEG' n

tryCatch() around withCallingHandlers() catches the error, and keep() writes it as a row of kind ERROR. The test program has a warning (the mean of a text) and an error (AEG is not a column). The third line, y <- 1, never runs: R stops a program at its first error, so the check stops reading too. That is also what a reviewer must know: everything after the error is missing.

programs/15_x5_not_done.R
# Answer 15-5: the AVAL rule, written as the data needs it
library(dplyr)
advs <- readRDS("data/advs.rds") |>
left_join(
pharmaversesdtm::vs |> select(USUBJID, VSSEQ, VSSTAT),
by = join_by(USUBJID, VSSEQ)
)
# AVAL may be missing only when the test was not done, and then must be
advs |>
summarise(
missing_not_explained = sum(is.na(AVAL) & !VSSTAT %in% "NOT DONE"),
not_done_with_value = sum(!is.na(AVAL) & VSSTAT %in% "NOT DONE")
)
#> # A tibble: 1 x 2
#> missing_not_explained not_done_with_value
#> <int> <int>
#> 1 0 0

Both counts are 0: every missing AVAL is explained by VSSTAT = "NOT DONE", and no test that was not done has a value. ADVS passes. The rule now says what the data must be, not what is easy to count.

programs/16_x1_range.R
# Answer 16-1: check 9, weights outside a plausible range
library(dplyr)
weight <- pharmaversesdtm::vs |>
filter(VSTESTCD == "WEIGHT")
range(weight$VSSTRESN)
#> [1] 33.11 107.96
weight |>
filter(VSSTRESN < 30 | VSSTRESN > 200) |>
nrow()
#> [1] 0

The weights go from 33.11 kg to 107.96 kg, and 0 weights are outside 30 to 200 kg. The assumption holds. A plausibility check like this finds a weight in pounds or a typo with one digit too many.

programs/16_x2_strict.R
# Answer 16-2: a new decision D1: only a weight strictly before the dose
library(dplyr)
adsl <- readRDS("data/adsl.rds")
wt_010 <- pharmaversesdtm::vs |>
filter(VSTESTCD == "WEIGHT", !is.na(VSSTRESN)) |>
transmute(USUBJID, VSSEQ, VISIT, ADT = as.Date(VSDTC), AVAL = VSSTRESN) |>
inner_join(select(adsl, USUBJID, TRTSDT), by = join_by(USUBJID))
# Step 020 is the only step that changes: < instead of <=
wt_020 <- wt_010 |>
filter(ADT < TRTSDT) |>
arrange(USUBJID, ADT, VSSEQ) |>
slice_tail(n = 1, by = USUBJID)
adsl_wt <- adsl |>
left_join(
wt_020 |> select(USUBJID, WEIGHTBL = AVAL, WTBLVIS = VISIT),
by = join_by(USUBJID), relationship = "one-to-one"
)
count(adsl_wt, WTBLVIS)
#> # A tibble: 1 x 2
#> WTBLVIS n
#> <chr> <int>
#> 1 SCREENING 1 254
adsl_wt |>
filter(USUBJID == "01-701-1015") |>
select(USUBJID, WEIGHTBL, WTBLVIS)
#> # A tibble: 1 x 3
#> USUBJID WEIGHTBL WTBLVIS
#> <chr> <dbl> <chr>
#> 1 01-701-1015 54.0 SCREENING 1

Only step 020 changes: < instead of <=. Steps 010 and 030 stay the same, and so does the program header, except decision D1 and a new line in the change log. Now all 254 baselines come from SCREENING 1, because the BASELINE weight was always taken on the day of the first dose. 01-701-1015 gets 54.0 kg instead of 54.4 kg. One decision changes the baseline of every subject: that is why D1 must be written down and agreed.

programs/16_x3_height.R
# Answer 16-3: the check program for HEIGHTBL, before any derivation
library(dplyr)
adsl <- readRDS("data/adsl.rds") |>
select(USUBJID, TRTSDT)
height <- pharmaversesdtm::vs |>
filter(VSTESTCD == "HEIGHT") |>
inner_join(adsl, by = join_by(USUBJID))
count(height, VISIT, VSSTRESU)
#> # A tibble: 1 x 3
#> VISIT VSSTRESU n
#> <chr> <chr> <int>
#> 1 SCREENING 1 cm 254
no_height <- anti_join(adsl, height, by = join_by(USUBJID))
two_heights <- count(height, USUBJID) |> filter(n > 1)
after_dose <- filter(height, as.Date(VSDTC) > TRTSDT)
tribble(
~check, ~n,
"subjects without a height", nrow(no_height),
"subjects with more than one height", nrow(two_heights),
"heights after the first dose", nrow(after_dose),
"heights without a value", sum(is.na(height$VSSTRESN))
)
#> # A tibble: 4 x 2
#> check n
#> <chr> <int>
#> 1 subjects without a height 0
#> 2 subjects with more than one height 0
#> 3 heights after the first dose 0
#> 4 heights without a value 0

The height is measured once, at SCREENING 1, in cm, for all 254 subjects. All four checks give 0. So HEIGHTBL needs no baseline rule: it is the one height of the subject. Write that in the note too, so that a later data delivery with a second height is found by the check.

programs/16_x4_bmibl.R
# Answer 16-4: BMIBL from WEIGHTBL and the height, compared with the reference
library(dplyr)
library(diffdf)
adsl_wt <- readRDS("data/16_adsl_weightbl.rds")
# Step 040: the height of each subject (one per subject: exercise 16-3)
height <- pharmaversesdtm::vs |>
filter(VSTESTCD == "HEIGHT") |>
select(USUBJID, HEIGHTBL = VSSTRESN)
# Step 050: BMI = weight in kg / (height in m) squared
ours <- adsl_wt |>
left_join(height, by = join_by(USUBJID), relationship = "one-to-one") |>
transmute(USUBJID, BMIBL = WEIGHTBL / (HEIGHTBL / 100)^2) |>
haven::zap_label()
ours |>
filter(USUBJID == "01-701-1015")
#> # A tibble: 1 x 2
#> USUBJID BMIBL
#> <chr> <dbl>
#> 1 01-701-1015 25.1
ref <- pharmaverseadam::advs |>
filter(PARAMCD == "BMI", ABLFL == "Y") |>
select(USUBJID, BMIBL = AVAL) |>
haven::zap_label()
diffdf(ours, ref, keys = "USUBJID")
#> No issues were found!

BMIBL of 01-701-1015 is 25.1. The compare with the baseline BMI of the reference ADVS finds no issues for the 254 subjects. Steps 040 and 050 read the result of step 030: the numbered steps make the new variable one more step, not a new program.

programs/17_x1_lazy.R
# Answer 17-1: a default is evaluated when it is first used, not when the call starts
f <- function(a, b = a * 2) {
a <- 10
b
}
f(1)
#> [1] 20
g <- function(a, b = a * 2) {
force(b) # evaluate b now, while a is still 1
a <- 10
b
}
g(1)
#> [1] 2

f(1) gives 20, not 2. The default b = a * 2 is evaluated when the body first uses b, and by then the body has set a to 10. With force(b) as the first line, R evaluates b at once, while a is still 1: g(1) gives 2.

programs/17_x2_anrhi.R
# Answer 17-2: a variable with the name of a column. The column wins in filter()
library(dplyr)
alt <- pharmaverseadam::adlb |>
filter(PARAMCD == "ALT", !is.na(AVAL), is.na(DTYPE))
ANRHI <- 40
alt |> filter(AVAL > ANRHI) |> nrow() # the column ANRHI of each row
#> [1] 83
alt |> filter(AVAL > .env$ANRHI) |> nrow() # the value 40
#> [1] 58

filter(AVAL > ANRHI) keeps 83 rows: ANRHI is the column, the upper limit of each row. The variable of the program is hidden by the column. filter(AVAL > .env$ANRHI) uses the value 40 and keeps 58 rows. No error and no warning tell you which one you got. Use .env$ or another name.

programs/17_x3_n_over.R
# Answer 17-3: a column name made from a name and a value, with :=
library(dplyr)
adsl <- pharmaverseadam::adsl |>
filter(SAFFL == "Y")
n_over <- function(data, var, limit, by) {
data |>
summarise(
"n_{{ var }}_over_{limit}" := sum({{ var }} > limit, na.rm = TRUE),
.by = {{ by }}
)
}
n_over(adsl, AGE, 80, ARM)
#> # A tibble: 3 x 2
#> ARM n_AGE_over_80
#> <chr> <int>
#> 1 Placebo 30
#> 2 Xanomeline High Dose 18
#> 3 Xanomeline Low Dose 29
n_over(adsl, TRTDURD, 180, ARM)
#> # A tibble: 3 x 2
#> ARM n_TRTDURD_over_180
#> <chr> <int>
#> 1 Placebo 53
#> 2 Xanomeline High Dose 24
#> 3 Xanomeline Low Dose 24

"n_{{ var }}_over_{limit}" := builds the name from the argument and the value. 30 placebo subjects, 18 on the high dose and 29 on the low dose are older than 80. The same function gives n_TRTDURD_over_180 for the treatment duration.

programs/17_x4_text_name.R
# Answer 17-4: mean_of() for a name in a text, with .data[[ ]]
library(dplyr)
adsl <- pharmaverseadam::adsl |>
filter(SAFFL == "Y")
mean_of_text <- function(data, var) {
data |> summarise(mean = round(mean(.data[[var]]), 1))
}
mean_of_text(adsl, "AGE")
#> # A tibble: 1 x 1
#> mean
#> <dbl>
#> 1 75.1
# Or turn the text into a name first: rlang::sym(), then !!
mean_of <- function(data, var) {
data |> summarise(mean = round(mean({{ var }}), 1))
}
mean_of(adsl, !!rlang::sym("AGE"))
#> # A tibble: 1 x 1
#> mean
#> <dbl>
#> 1 75.1

Two ways give the same mean, 75.1:

  • .data[[var]] in the function: the caller gives a text.
  • !!rlang::sym("AGE") in the call: sym() turns the text into a name, and !! puts the name in. The function keeps its {{ }}.

The first way is the usual one. Use the second when you cannot change the function.

programs/18_x1_exclude.R
# Answer 18-1: leave some parameters out. Build the list without them
library(dplyr)
library(purrr)
exclude <- c("HEIGHT", "TEMP")
bl <- pharmaverseadam::advs |>
filter(ABLFL == "Y", is.na(ATPTN) | ATPTN == 815, !PARAMCD %in% exclude)
tables <- split(bl, bl$PARAMCD) |>
map(\(d) summarise(d, n = n(), mean = round(mean(AVAL), 1), .by = TRT01P))
length(tables)
#> [1] 7
names(tables)
#> [1] "BMI" "BSA" "DIABP" "MAP" "PULSE" "SYSBP" "WEIGHT"

7 tables are left. !PARAMCD %in% exclude removes the rows before split(), so the list never has the two parameters. A test inside the function would leave an empty element or a NULL in the list.

programs/18_x2_imap.R
# Answer 18-2: imap() gives each element and its name: a title per table
library(dplyr)
library(purrr)
bl <- pharmaverseadam::advs |>
filter(ABLFL == "Y", is.na(ATPTN) | ATPTN == 815)
by_param <- split(bl, bl$PARAMCD)
iwalk(by_param, \(d, code) {
cat(sprintf("%-7s n = %d, mean = %.1f\n", code, nrow(d), mean(d$AVAL)))
})
#> BMI n = 254, mean = 24.7
#> BSA n = 254, mean = 1.7
#> DIABP n = 254, mean = 76.4
#> HEIGHT n = 254, mean = 163.9
#> MAP n = 254, mean = 97.3
#> PULSE n = 254, mean = 69.7
#> SYSBP n = 254, mean = 139.2
#> TEMP n = 254, mean = 36.6
#> WEIGHT n = 254, mean = 66.6

iwalk() calls the function with two arguments: the element (d) and its name (code). Each parameter has 254 baseline rows in the lying-down position, one per subject. The mean baseline weight of all arms together is 66.6 kg.

programs/18_x3_switch_on.R
# Answer 18-3: switch DIABP on in a copy of the parameter file, run again
library(dplyr)
library(purrr)
settings <- "data/18_run_settings.csv" |>
readr::read_csv(show_col_types = FALSE) |>
mutate(include = if_else(paramcd == "DIABP", "Y", include))
file <- tempfile(fileext = ".csv")
readr::write_csv(settings, file)
bl <- pharmaverseadam::advs |>
filter(ABLFL == "Y", is.na(ATPTN) | ATPTN == 815)
readr::read_csv(file, show_col_types = FALSE) |>
filter(include == "Y", paramcd == "DIABP") |>
pwalk(\(paramcd, dec, title, ...) {
cat(title, "\n")
bl |>
filter(PARAMCD == paramcd) |>
summarise(
n = n(),
mean = formatC(mean(AVAL), format = "f", digits = dec + 1),
.by = TRT01P
) |>
arrange(TRT01P) |>
print()
})
#> Baseline diastolic blood pressure (mmHg)
#> # A tibble: 3 x 3
#> TRT01P n mean
#> <chr> <int> <chr>
#> 1 Placebo 86 75.7
#> 2 Xanomeline High Dose 84 77.2
#> 3 Xanomeline Low Dose 84 76.3

The mean baseline diastolic blood pressure on placebo is 75.7 mmHg, with dec = 0 in the file, so one decimal. The same numbers came from split() and map() in program 18_2. Only the copy of the settings changed, not the code.

programs/18_x4_missing_vars.R
# Answer 18-4: the variables in one data set only, both ways
library(dplyr)
spec <- tibble(name = names(pharmaverseadam::adsl))
ours <- tibble(name = names(readRDS("data/adsl.rds")))
only_spec <- anti_join(spec, ours, by = join_by(name))
only_ours <- anti_join(ours, spec, by = join_by(name))
nrow(only_spec)
#> [1] 37
head(only_spec$name, 5)
#> [1] "COUNTRY" "RFXSTDTC" "RFXENDTC" "RFPENDTC" "SCRFDT"
nrow(only_ours)
#> [1] 7
only_ours$name
#> [1] "TRT01PN" "TRT01AN" "ITTFL" "EFFFL" "COMP24FL" "DCSREAS" "RFENDT"

37 variables of the reference ADSL are not in your ADSL, such as COUNTRY and RFXSTDTC. 7 of your variables are not in the reference: the numeric treatments TRT01PN and TRT01AN, the flags ITTFL, EFFFL and COMP24FL, DCSREAS and RFENDT. In a study, the specification decides: a variable in the specification must be in the data, and a variable in the data needs a row in the specification.

programs/19_x1_tdd.R
# Answer 19-1: test first. A by column that does not exist: red, then green
library(dplyr)
copy <- file.path(tempdir(), "studylib_x1")
unlink(copy, recursive = TRUE)
dir.create(copy)
files <- list.files("studylib", full.names = TRUE)
invisible(file.copy(files, copy, recursive = TRUE))
# 1. The test, written before the code
test_file <- file.path(copy, "tests", "testthat", "test-by.R")
writeLines(c(
'test_that("count_subj() names a by column that does not exist", {',
' adae <- pharmaverseadam::adae',
' expect_error(count_subj(adae, ARMX), "column of .*by.* is not in")',
'})'
), test_file)
pkgload::load_all(copy, quiet = TRUE)
red <- testthat::test_file(test_file, reporter = "silent")
as_tibble(red) |> select(failed, error, passed)
#> # A tibble: 1 x 3
#> failed error passed
#> <int> <lgl> <int>
#> 1 0 TRUE 0
# 2. The code: a new check in step 010
code_file <- file.path(copy, "R", "count_subj.R")
code <- readLines(code_file)
step_020 <- grep("# Step 020", code)
check <- c(
" by_cols <- try(dplyr::select(data, {{ by }}), silent = TRUE)",
" if (inherits(by_cols, \"try-error\")) {",
" cli::cli_abort(\"A column of {.arg by} is not in {.arg data}.\")",
" }",
""
)
writeLines(append(code, check, after = step_020 - 1), code_file)
pkgload::load_all(copy, quiet = TRUE)
green <- testthat::test_file(test_file, reporter = "silent")
as_tibble(green) |> select(failed, error, passed)
#> # A tibble: 1 x 3
#> failed error passed
#> <int> <lgl> <int>
#> 1 0 FALSE 1
count_subj(pharmaverseadam::adae, ARMX) |> try()
#> Error in count_subj(pharmaverseadam::adae, ARMX) :
#> A column of `by` is not in `data`.

The program works on a copy of the library. First the test: error is TRUE and passed is 0. The dplyr error does not match the expected message, so expect_error() passes the error on, and the test fails with an error. This is the “red” step. Then the program adds the check to step 010 and loads the library again: passed is 1, the “green” step. The last call shows the new message.

try(dplyr::select(data, {{ by }})) tests the by columns: select() fails when a column does not exist, and inherits(..., "try-error") sees that failure.

programs/19_x2_pct.R
# Answer 19-2: subjects with a percent of the big N, built on the library
library(dplyr)
pkgload::load_all("studylib", quiet = TRUE)
count_subj_pct <- function(data, adsl, by = TRT01A) {
denom <- big_n(adsl, by = {{ by }})
count_subj(data, {{ by }}) |>
left_join(denom, by = join_by({{ by }})) |>
mutate(pct = round(100 * n_subj / N, 1))
}
teae <- pharmaverseadam::adae |>
filter(TRTEMFL == "Y")
res <- count_subj_pct(teae, pharmaverseadam::adsl)
res
#> # A tibble: 3 x 4
#> TRT01A n_subj N pct
#> <chr> <int> <int> <dbl>
#> 1 Placebo 65 86 75.6
#> 2 Xanomeline High Dose 68 72 94.4
#> 3 Xanomeline Low Dose 84 96 87.5
# The tests: silent when they pass
testthat::expect_equal(res$pct[res$TRT01A == "Placebo"], 75.6)
testthat::expect_true(all(res$n_subj <= res$N))

75.6% of the placebo subjects (65 of 86) had a TEAE. The new function is built on the two functions of the library, not copied from them, so a fix in count_subj() reaches it too. The two expectations are silent: they pass.

programs/19_x3_new_function.R
# Answer 19-3: a new function in the library: code, roxygen block, test
# A session can still hold the library from an earlier program: unload it, so the new copy is the only one
if ("studylib" %in% loadedNamespaces()) pkgload::unload("studylib")
copy <- file.path(tempdir(), "studylib_x3")
unlink(copy, recursive = TRUE)
dir.create(copy)
files <- list.files("studylib", full.names = TRUE)
invisible(file.copy(files, copy, recursive = TRUE))
writeLines(c(
"# n_pct.R: a count and its percent as one text, such as 65 (75.6)",
"#",
"# Change log:",
"# 0.2.0 2026-11-19 First version.",
"",
"#' A count with its percent",
"#'",
"#' @param n The counts.",
"#' @param total The totals, as many as `n` or one.",
"#' @param digits The decimals of the percent. Default 1.",
"#' @return A text vector, such as `\"65 (75.6)\"`.",
"#' @examples",
"#' n_pct(65, 86)",
"#' @export",
"n_pct <- function(n, total, digits = 1) {",
" # Step 010: check the arguments",
" if (!is.numeric(n) || !is.numeric(total) || any(total <= 0)) {",
" cli::cli_abort(\"{.arg n} and {.arg total}: numbers, total > 0.\")",
" }",
" # Step 020: the text",
" pct <- formatC(100 * n / total, format = \"f\", digits = digits)",
" paste0(n, \" (\", pct, \")\")",
"}"
), file.path(copy, "R", "n_pct.R"))
writeLines(c(
'test_that("n_pct() writes the count and the percent", {',
' expect_equal(n_pct(65, 86), "65 (75.6)")',
' expect_equal(n_pct(c(1, 2), 4, digits = 0), c("1 (25)", "2 (50)"))',
' expect_error(n_pct(5, 0), "total > 0")',
'})'
), file.path(copy, "tests", "testthat", "test-n_pct.R"))
roxygen2::roxygenise(copy)
#> i Loading studylib
#> Writing 'NAMESPACE'
#> Writing 'n_pct.Rd'
readLines(file.path(copy, "NAMESPACE"))
#> [1] "# Generated by roxygen2: do not edit by hand"
#> [2] ""
#> [3] "export(big_n)"
#> [4] "export(count_subj)"
#> [5] "export(n_pct)"
# test_local() runs the tests inside the package, so a function of the same name in your workspace
# (chapter 13 made an n_pct() of its own) cannot hide the package's one (chapter 17: where R finds a name)
res <- testthat::test_local(copy, reporter = "silent", stop_on_failure = FALSE)
sum(as.data.frame(res)$failed)
#> [1] 0
studylib::n_pct(65, 86)
#> [1] "65 (75.6)"

roxygen2 writes n_pct.Rd, and NAMESPACE gets the line export(n_pct). test_local() loads the package and runs its tests inside it. All tests of the library pass (0 failed), the three new expectations too. studylib::n_pct(65, 86) gives "65 (75.6)", the cell of a table.

Why test_local() and studylib::? If your session already has an n_pct() of its own (program 13 made one that writes 65 (75.6%)), a plain n_pct() call finds that one first: the workspace comes before every attached package (chapter 17). In the DataStep Studio for R session, where the objects of every run stay, this happened: the tests found the old function and failed. Inside the package, and with package::, there is no doubt which function runs.

programs/19_x4_debug.R
# Answer 19-4: stop in count_subj(), go to step 030, look at subj
commands <- c(
'pkgload::load_all("studylib", quiet = TRUE)',
'teae <- dplyr::filter(pharmaverseadam::adae, TRTEMFL == "Y")',
"debugonce(count_subj)",
"count_subj(teae, TRT01A)",
"n", "n", "n", "n", # the three checks of step 010
"n", # step 020 runs: now subj exists
"nrow(subj)",
"nrow(teae)",
"c" # continue to the end
)
out <- system2(
file.path(R.home("bin"), "R"), c("--vanilla", "--quiet"),
input = commands, stdout = TRUE, stderr = TRUE,
env = c(paste0("R_LIBS=", paste(.libPaths(), collapse = ":")), "LC_ALL=C")
)
out <- out[!grepl("built under R version|^Warning message:$", out)]
# Only the prompts and what they print, not the code of each step
out <- gsub(getwd(), ".", out, fixed = TRUE)
writeLines(grep("^(Browse|>)|^\\[1\\]|^debug at", out, value = TRUE))
#> > pkgload::load_all("studylib", quiet = TRUE)
#> > teae <- dplyr::filter(pharmaverseadam::adae, TRTEMFL == "Y")
#> > debugonce(count_subj)
#> > count_subj(teae, TRT01A)
#> debug at ./studylib/R/count_subj.R#21: {
#> Browse[1]> n
#> debug at ./studylib/R/count_subj.R#23: if (!is.data.frame(data)) {
#> Browse[1]> n
#> debug at ./studylib/R/count_subj.R#28: if (!rlang::is_string(id)) {
#> Browse[1]> n
#> debug at ./studylib/R/count_subj.R#31: if (!id %in% names(data)) {
#> Browse[1]> n
#> debug at ./studylib/R/count_subj.R#39: subj <- dplyr::distinct(data, dplyr::pick({
#> Browse[1]> n
#> debug at ./studylib/R/count_subj.R#43: dplyr::count(subj, dplyr::pick({
#> Browse[1]> nrow(subj)
#> [1] 217
#> Browse[1]> nrow(teae)
#> [1] 1122
#> Browse[1]> c
#> >

Five times n go past the three checks of step 010 and run step 020. At step 030, subj has 217 rows and teae has 1122. subj has one row per arm and subject, so 217 is the sum of the three counts (65, 68 and 84). The difference is the rule of chapter 1: 1122 records, 217 subjects.

Found an error or have a question? Write to datastep.studio@gmail.com· Every program runs in DataStep Studio for R, the R extension for VS Code and Positron.

Data: the CDISC pilot study (cdiscpilot01), from the R packagespharmaversesdtm andpharmaverseadam (Apache License 2.0).

R is a project of the R Foundation. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. This book is not affiliated with or endorsed by SAS Institute Inc. or the R Foundation. CDISC, SDTM and ADaM are trademarks of the Clinical Data Interchange Standards Consortium.

© 2026 DataStep Labs. All rights reserved.