{sdtmval} provides a set of tools to assist statistical programmers in validating Study Data Tabulation Model (SDTM) domain data sets.
Many data cleaning steps and SDTM processes are used repeatedly in different SDTM domain validation scripts. Functionalizing these repetitive tasks allows statistical programmers to focus on coding the unique aspects of a SDTM domain while standardize their code base across studies and domains. This should lead to fewer bugs and improved code readability too. {sdtmval} features include:
Automating the BLFL, DY, EPOCH, SEQ, and STAT methods to create new variables
Imputing and formatting full and partial dates: see vignette("Dates")
Applying specification data such as variable labels, lengths, code mapping, and sorting
Importing EDC and SDTM data from .csv and .sas7bdat files
Writing .xpt files (convenience wrapper for
haven::write_xpt())
Converting .Rmd files to .R scripts (convenience wrapper for
knitr::purl())
Logging R session information for reproducibility
Data formatting
You can install the release version of {sdtmval} from CRAN with:
install.packages("sdtmval")You can install the development version of {sdtmval} from GitHub with:
# install.packages("devtools")
devtools::install_github("skgithub14/sdtmval")In this example work flow, we will import a raw EDC table and transform it into a SDTM domain table. We will use the made-up domain ‘XX’ along with some example data included in {sdtmval}.
# set-up
library(sdtmval)
library(dplyr)
domain <- "XX"
# set working directory to location of sdtmval package example data
work_dir <- system.file("extdata", package = "sdtmval")The majority of the data needed is in the EDC form/table xx.csv.
There are also visit dates in the EDC table vd.csv and study start/end
dates in the SDTM table dm.sas7dbat. These can be imported using
read_edc_tbls() and read_sdtm_tbls().
# read in EDC tables from the forms XX and VD
edc_tbls <- c("xx", "vd")
edc_dat <- read_edc_tbls(edc_tbls, dir = work_dir)
# read in SDTM domain DM
sdtm_tbls <- c("dm")
sdtm_dat <- read_sdtm_tbls(sdtm_tbls, dir = work_dir)The raw data looks like this:
| STUDYID | USUBJID | VISIT | XXTESTCD | XXORRES |
|---|---|---|---|---|
| Study 1 | Subject 1 | Visit 1 | T1 | 1 |
| Study 1 | Subject 1 | Visit 2 | T1 | 0 |
| Study 1 | Subject 1 | Visit 3 | T1 | 2 |
| Study 1 | Subject 1 | Visit 3 | T2 | 100 |
| Study 1 | Subject 1 | Visit 4 | T3 | PASS |
| Study 1 | Subject 2 | Visit 1 | T1 | 1 |
| Study 1 | Subject 2 | Visit 2 | T1 | |
| Study 1 | Subject 2 | Visit 3 | T1 | 2 |
| Study 1 | Subject 2 | Visit 3 | T2 | 200 |
| Study 1 | Subject 2 | Visit 4 | T3 | FAIL |
The next thing we will do is get the relevant information from the SDTM specification for the study. The next set of functions assumes there is a .xlsx file which contains the sheets: ‘Datasets’, ‘XX’, and ‘Codelists’:
‘XX’ gives the variable information for the made-up XX domain.
get_data_spec() retrieves this entire tab.
‘Datasets’ contains the key variables by domain.
get_key_vars() retrieves the these for desired
domain.
‘Codelists’ provides a table of coded/decoded values by variable
for all domains. get_codelist() extracts a data frame of
coded/decoded values from this sheet just for the variables in desired
domain.
spec_fname <- "spec.xlsx"
spec <- get_data_spec(domain = domain, dir = work_dir, filename = spec_fname)
key_vars <- get_key_vars(domain = domain, dir = work_dir, filename = spec_fname)
codelists <- get_codelist(domain = domain, dir = work_dir, filename = spec_fname)
knitr::kable(spec)| Order | Dataset | Variable | Label | Data Type | Length |
|---|---|---|---|---|---|
| 1 | XX | STUDYID | Study Identifier | text | 200 |
| 2 | XX | DOMAIN | Domain Abbreviation | text | 200 |
| 3 | XX | USUBJID | Unique Subject Identifier | text | 200 |
| 4 | XX | XXSEQ | Sequence Number | integer | 8 |
| 5 | XX | XXTESTCD | XX Test Short Name | text | 8 |
| 6 | XX | XXTEST | XX Test Name | text | 40 |
| 7 | XX | XXORRES | Result or Finding in Original Units | text | 200 |
| 8 | XX | XXBLFL | Baseline Flag | text | 1 |
| 9 | XX | VISIT | Visit Name | text | 200 |
| 10 | XX | EPOCH | Epoch | text | 200 |
| 11 | XX | XXDTC | Date/Time of Measurements | datetime | 19 |
| 12 | XX | XXDY | Study Day of XX | integer | 8 |
knitr::kable(codelists)| ID | Term | Decoded Value |
|---|---|---|
| XXTESTCD | T1 | Test 1 |
| XXTESTCD | T2 | Test 2 |
| XXTESTCD | T3 | Test 3 |
key_vars
#> [1] "STUDYID" "USUBJID" "XXTESTCD" "VISIT"Now we will begin creating the SDTM XX domain using the EDC XX form as the basis.
First, it needs some pre-processing because there is extra white
space in some of the variables. We also want to turn all NA equivalent
values like "" and " " to NA for
the entire data set so we have consistent handling of missing values
during data processing. The function
trim_and_make_blanks_NA() does both of these tasks.
sdtm_xx1 <- trim_and_make_blanks_NA(edc_dat$xx)Next, using the codelist we retrieved earlier, we can create the
XXTEST variable.
# prepare the code list so it can be used by dplyr::recode()
xxtestcd_codelist <- codelists %>%
filter(ID == "XXTESTCD") %>%
select(Term, `Decoded Value`) %>%
tibble::deframe()
# create XXTEST variable
sdtm_xx2 <- mutate(sdtm_xx1, XXTEST = recode(XXTESTCD, !!!xxtestcd_codelist))
knitr::kable(sdtm_xx2)| STUDYID | USUBJID | VISIT | XXTESTCD | XXORRES | XXTEST |
|---|---|---|---|---|---|
| Study 1 | Subject 1 | Visit 1 | T1 | 1 | Test 1 |
| Study 1 | Subject 1 | Visit 2 | T1 | 0 | Test 1 |
| Study 1 | Subject 1 | Visit 3 | T1 | 2 | Test 1 |
| Study 1 | Subject 1 | Visit 3 | T2 | 100 | Test 2 |
| Study 1 | Subject 1 | Visit 4 | T3 | PASS | Test 3 |
| Study 1 | Subject 2 | Visit 1 | T1 | 1 | Test 1 |
| Study 1 | Subject 2 | Visit 2 | T1 | NA | Test 1 |
| Study 1 | Subject 2 | Visit 3 | T1 | 2 | Test 1 |
| Study 1 | Subject 2 | Visit 3 | T2 | 200 | Test 2 |
| Study 1 | Subject 2 | Visit 4 | T3 | FAIL | Test 3 |
In order to calculate the variables XXBLFL, EPOCH, and XXDY, we need the visit dates from the EDC VD table and the study start/end dates by subject from the SDTM DM table.
sdtm_xx3 <- sdtm_xx2 %>%
# get the VISITDTC column from the EDC VD form
left_join(edc_dat$vd, by = c("USUBJID", "VISIT")) %>%
# create the XXDTC variable
rename(XXDTC = VISITDTC) %>%
# get the study start/end dates by subject (RFSTDTC, RFXSTDTC, RFXENDTC)
left_join(sdtm_dat$dm, by = "USUBJID")Now, we can proceed with calculating those timing variables using the
create_BLFL(), create_EPOCH(), and
calc_DY() functions.
sdtm_xx4 <- sdtm_xx3 %>%
# XXBLFL
create_BLFL(sort_date = "XXDTC",
domain = domain,
grouping_vars = c("USUBJID", "XXTESTCD")) %>%
# EPOCH
create_EPOCH(date_col = "XXDTC") %>%
# XXDY
calc_DY(DY_col = "XXDY", DTC_col = "XXDTC")
# check the new variables and their related columns only
sdtm_xx4 %>%
select(USUBJID, XXTEST, XXORRES, XXDTC, XXBLFL,
EPOCH, XXDY, starts_with("RF")) %>%
knitr::kable()| USUBJID | XXTEST | XXORRES | XXDTC | XXBLFL | EPOCH | XXDY | RFSTDTC | RFXSTDTC | RFXENDTC |
|---|---|---|---|---|---|---|---|---|---|
| Subject 1 | Test 1 | 1 | 2023-08-01 | NA | SCREENING | -1 | 2023-08-02 | 2023-08-02 | 2023-08-03 |
| Subject 1 | Test 1 | 0 | 2023-08-02 | Y | TREATMENT | 1 | 2023-08-02 | 2023-08-02 | 2023-08-03 |
| Subject 1 | Test 1 | 2 | 2023-08-03 | NA | TREATMENT | 2 | 2023-08-02 | 2023-08-02 | 2023-08-03 |
| Subject 1 | Test 2 | 100 | 2023-08-03 | NA | TREATMENT | 2 | 2023-08-02 | 2023-08-02 | 2023-08-03 |
| Subject 1 | Test 3 | PASS | 2023-08-04 | NA | FOLLOW-UP | 3 | 2023-08-02 | 2023-08-02 | 2023-08-03 |
| Subject 2 | Test 1 | 1 | 2023-08-02 | Y | SCREENING | -1 | 2023-08-03 | 2023-08-03 | 2023-08-04 |
| Subject 2 | Test 1 | NA | 2023-08-03 | NA | TREATMENT | 1 | 2023-08-03 | 2023-08-03 | 2023-08-04 |
| Subject 2 | Test 1 | 2 | 2023-08-04 | NA | TREATMENT | 2 | 2023-08-03 | 2023-08-03 | 2023-08-04 |
| Subject 2 | Test 2 | 200 | 2023-08-04 | NA | TREATMENT | 2 | 2023-08-03 | 2023-08-03 | 2023-08-04 |
| Subject 2 | Test 3 | FAIL | 2023-08-05 | NA | FOLLOW-UP | 3 | 2023-08-03 | 2023-08-03 | 2023-08-04 |
Next, we will assign the sequence number using
assign_SEQ() (which also sorts your data frame).
sdtm_xx5 <- assign_SEQ(sdtm_xx4,
key_vars = c("USUBJID", "XXTESTCD", "VISIT"),
seq_prefix = domain)
# check the new variable
sdtm_xx5 %>%
select(USUBJID, XXTESTCD, VISIT, XXDTC, XXSEQ) %>%
knitr::kable()| USUBJID | XXTESTCD | VISIT | XXDTC | XXSEQ |
|---|---|---|---|---|
| Subject 1 | T1 | Visit 1 | 2023-08-01 | 1 |
| Subject 1 | T1 | Visit 2 | 2023-08-02 | 2 |
| Subject 1 | T1 | Visit 3 | 2023-08-03 | 3 |
| Subject 1 | T2 | Visit 3 | 2023-08-03 | 4 |
| Subject 1 | T3 | Visit 4 | 2023-08-04 | 5 |
| Subject 2 | T1 | Visit 1 | 2023-08-02 | 1 |
| Subject 2 | T1 | Visit 2 | 2023-08-03 | 2 |
| Subject 2 | T1 | Visit 3 | 2023-08-04 | 3 |
| Subject 2 | T2 | Visit 3 | 2023-08-04 | 4 |
| Subject 2 | T3 | Visit 4 | 2023-08-05 | 5 |
Now that the bulk of the data cleaning is complete, we will convert
all date columns to character columns and all NA values to
"" so that our validation table matches the production
table produced in SAS. To do this, we will use
format_chars_and_dates().
sdtm_xx6 <- format_chars_and_dates(sdtm_xx5)As a final step, we will assign the meta data from the spec to each
column using assign_meta_data(). The meta data includes the
labels for each column and their maximum allowed character lengths.
sdtm_xx7 <- sdtm_xx6 %>%
# only keep columns that are domain variables and order them per the spec
select(any_of(spec$Variable)) %>%
# assign variable lengths and labels
assign_meta_data(spec = spec)
# show the final SDTM domain
knitr::kable(sdtm_xx7)| STUDYID | USUBJID | XXSEQ | XXTESTCD | XXTEST | XXORRES | XXBLFL | VISIT | EPOCH | XXDTC | XXDY |
|---|---|---|---|---|---|---|---|---|---|---|
| Study 1 | Subject 1 | 1 | T1 | Test 1 | 1 | Visit 1 | SCREENING | 2023-08-01 | -1 | |
| Study 1 | Subject 1 | 2 | T1 | Test 1 | 0 | Y | Visit 2 | TREATMENT | 2023-08-02 | 1 |
| Study 1 | Subject 1 | 3 | T1 | Test 1 | 2 | Visit 3 | TREATMENT | 2023-08-03 | 2 | |
| Study 1 | Subject 1 | 4 | T2 | Test 2 | 100 | Visit 3 | TREATMENT | 2023-08-03 | 2 | |
| Study 1 | Subject 1 | 5 | T3 | Test 3 | PASS | Visit 4 | FOLLOW-UP | 2023-08-04 | 3 | |
| Study 1 | Subject 2 | 1 | T1 | Test 1 | 1 | Y | Visit 1 | SCREENING | 2023-08-02 | -1 |
| Study 1 | Subject 2 | 2 | T1 | Test 1 | Visit 2 | TREATMENT | 2023-08-03 | 1 | ||
| Study 1 | Subject 2 | 3 | T1 | Test 1 | 2 | Visit 3 | TREATMENT | 2023-08-04 | 2 | |
| Study 1 | Subject 2 | 4 | T2 | Test 2 | 200 | Visit 3 | TREATMENT | 2023-08-04 | 2 | |
| Study 1 | Subject 2 | 5 | T3 | Test 3 | FAIL | Visit 4 | FOLLOW-UP | 2023-08-05 | 3 |
# check the meta data was assigned
labels <- colnames(sdtm_xx7) %>%
purrr::map(~ attr(sdtm_xx7[[.]], "label")) %>%
unlist()
lengths <- colnames(sdtm_xx7) %>%
purrr::map(~ attr(sdtm_xx7[[.]], "width")) %>%
unlist()
data.frame(
column = colnames(sdtm_xx7),
labels = labels,
lengths = lengths
)
#> column labels lengths
#> 1 STUDYID Study Identifier 200
#> 2 USUBJID Unique Subject Identifier 200
#> 3 XXSEQ Sequence Number 8
#> 4 XXTESTCD XX Test Short Name 8
#> 5 XXTEST XX Test Name 40
#> 6 XXORRES Result or Finding in Original Units 200
#> 7 XXBLFL Baseline Flag 1
#> 8 VISIT Visit Name 200
#> 9 EPOCH Epoch 200
#> 10 XXDTC Date/Time of Measurements 19
#> 11 XXDY Study Day of XX 8Finally, we will write the SDTM XX domain validation table as a SAS
transport file using write_tbl_to_xpt().
write_tbl_to_xpt(sdtm_xx7, filename = domain, dir = work_dir)For each previous steps, we viewed the interim results to demonstrate the features of {sdtmval} however, {sdtmval} is designed to be used with pipe operators so that you can have one long, readable pipe. To demonstrate, we will reproduce the same results from above in one code chunk.
sdtm_xx <- edc_dat$xx %>%
# pre-processing
trim_and_make_blanks_NA() %>%
# XXTEST
dplyr::mutate(XXTEST = dplyr::recode(XXTESTCD, !!!xxtestcd_codelist)) %>%
# get the VISITDTC column from the EDC VD form
dplyr::left_join(edc_dat$vd, by = c("USUBJID", "VISIT")) %>%
# XXDTC
dplyr::rename(XXDTC = VISITDTC) %>%
# get the study start/end dates by subject (RFSTDTC, RFXSTDTC, RFXENDTC)
dplyr::left_join(sdtm_dat$dm, by = "USUBJID") %>%
# XXBLFL
create_BLFL(sort_date = "XXDTC",
domain = domain,
grouping_vars = c("USUBJID", "XXTESTCD")) %>%
# EPOCH
create_EPOCH(date_col = "XXDTC") %>%
# XXDY
calc_DY(DY_col = "XXDY", DTC_col = "XXDTC") %>%
# XXSEQ
assign_SEQ(key_vars = c("USUBJID", "XXTESTCD", "VISIT"),
seq_prefix = domain) %>%
# final formatting
format_chars_and_dates() %>%
dplyr::select(dplyr::any_of(spec$Variable)) %>%
assign_meta_data(spec = spec)
# check if the two data frames are identical
identical(sdtm_xx, sdtm_xx7)
#> [1] TRUE