\newpage
“If I had just five minutes to chop down a tree I would spend the first two and a half minutes sharpening my axe.”
r tufte::quote_footer('--- An anonymous woodsman')
\vspace{13pt}
This R package provides functions for handling data from the clinical data management system (CDMS) \textcolor{blue}{secuTrial}. The most important components are related to reading data exports from secuTrial into R. In brief, the package aims to enable swift execution of repetitive tasks in order to allow spending more time on the unique aspects of a dataset. It is developed and maintained by the Swiss Clinical Trial Organisation (\textcolor{blue}{SCTO}).
If you are still challenged by more basic operations in R we suggest reading \textcolor{blue}{Hands-On Programming with R}, which serves as an excellent introduction to the basic concepts of R.
This vignette will teach you how to use the secuTrialR
package and you will likely
learn quite a bit about secuTrial exports in general along the way.
Throughout the secuTrialR
package and within this vignette we refer to
patients, cases, subjects etc. enrolled in a secuTrial database as participants.
# needed so that the as.data.frame part of the vignette # does not need a restart of the session everytime the # vignette is built #rm(list = ls()) # removed this at the request of the CRAN submission knitr::opts_chunk$set( collapse = TRUE, comment = "#>" )
Please note that R versions >= 3.5
should be used to run secuTrialR
.
\vspace{5pt}
Stable release from CRAN \vspace{5pt}
install.packages("secuTrialR", dependencies = TRUE)
Development release from GitHub with devtools
For this you will need to have devtools
installed.
If you are working on Windows and would like to install with devtools
you
will likely need to install
\textcolor{blue}{Rtools}.
Installing everything, including the dependencies, from scratch may take a while (20-30 minutes).
\vspace{5pt}
# install devtools::install_github("SwissClinicalTrialOrganisation/secuTrialR")
# load silently suppressMessages(library(secuTrialR)) # show secuTrialR version packageVersion("secuTrialR")
Before we continue with the functionalities let's briefly talk about the test data which is
delivered as a part of the package. We refer to it as the CTU05
(clinical trial unit project 05)
data. This dataset has been fabricated for demonstration purposes only and is not
real clinical data. Principally it is made up of eight forms. These are called "surgeries",
"baseline", "outcome", "treatment", "allmedi", "studyterminat", "ae" and "sae".
You will see these names again later when the data has been read into R.
The project setup includes most data types implementable in secuTrial. It is, however,
not exhaustive. Since the data is delivered with the installation of the secuTrialR
package we can point to it via the system.file()
function.
\vspace{5pt}
ctu05_data_location <- system.file("extdata", "sT_exports", "exp_opt", "s_export_CSV-xls_CTU05_all_info.zip", package = "secuTrialR")
If you work on your own datasets you can specify a path as a regular character string without
using system.file()
.
Prior to reading your data into R you need to export it with the secuTrial ExportSearchTool. We suggest exporting non-rectangular, zipped, English data with reference values stored in a separate table including Add-IDs, centre information, structure information, form status, project setup, without duplicated meta data and without form data of hidden fields. Furthermore, it is important to use "CSV format"/"CSV format for MS Excel" and suggested to select UTF-8 encoding. Most of these options are truly optional and reading your data should work even with differences from the above specifications.
A description of how data can be exported from secuTrial can be found \textcolor{blue}{here}. This description includes screenshots of the export options configuration interface.
There is one principle function to read your data (i.e. read_secuTrial()
).
Below you can see it in action with the CTU05
dataset.
\vspace{5pt}
ctu05_data <- read_secuTrial(data_dir = ctu05_data_location)
If the "Read export successfully." message appears your data was correctly read. In this example you are also warned that hidden data fields are in the export which is a deviation from the suggested export option configuraion.
secuTrialdata
objectIf you inspect the class()
of ctu05_data
you will find that it is a secuTrialdata
object.
\vspace{5pt}
class(ctu05_data)
Really this is only a list
containing all the information from your secuTrial export.
\vspace{5pt}
typeof(ctu05_data)
\newpage
secuTrialdata
objectWe have implemented a custom variation of the print()
function for secuTrialdata
objects.
print(ctu05_data)
It shows you where the export archive of your secuTrialdata
object is located, tells you which
data tables (i.e. table
) it contains, what the source files (i.e. original_name
) are and
specifies each table's dimensions (i.e. ncol
, nrow
).
By now you have possibly realized that all the forms specified earlier (i.e. "surgeries", "baseline", "outcome", "treatment", "allmedi", "studyterminat", "ae" and "sae") are present, but also that there are many tables that do not correspond to the previously introduced forms.
The majority of the unfamiliar tables are tagged as TRUE
in the meta
column.
This means that they are metadata tables. Their names and data structures
are fixed in secuTrial exports. In the following we will briefly explain
which information the most relevant meta tables contain.
vp
- visitplan definitionvpfs
- visitplan form linkagefs
- forms informationqs
- questionsis
- items i.e. variable definitionsctr
- centre informationcn
- casenodes i.e. table of entered study participantscl
- information how the data in the variables is codedFurthermore, there is a set of tables whose names start with "at". These are
audit trail tables. They are only relevant if you need to investigate changes
in the data over time. For example certain values may be corrected (i.e. changed)
due to findings during monitoring visits at study centres.
Last but not least you may have also realized that the "surgeries" table is
called esurgeries
. This is because it is a so-called repetition form.
Repetition forms are labelled with a leading "e" and are implemented
as subforms in other forms. In this case, esurgeries
is a subform in
baseline
and the linkage is defined by the mnpdocid
column in both tables.
If this sounds cryptic to you we suggest you talk so someone who has implemented
a database in secuTrial and let them explain it with a specific example.
It is pretty straight forward when you look at a concrete implementation.
Since the secuTrialdata
object is a list
and the data tables within this list
are
data.frame
s you can simply access the tables using $
. Let's say you would like to have
a look at the placebo to verum ratio in your treatment
data or what types of other
medication were entered in allmedi
.
\vspace{5pt}
table(ctu05_data$treatment$rando_treatment) table(ctu05_data$allmedi$med_product)
During the loading process, coded categorical data is transformed. For example
the gender
variable in the baseline
form is categorical. The raw data is
accessible via gender
and the transformed version of the data is added during
the reading process and becomes accessible via gender.factor
. Thus, data is not overwritten
but added with the .factor
extension. If there are issues during factorization
a warning()
will inform you of this.
\vspace{5pt}
# raw gender data ctu05_data$baseline$gender # transformed gender data ctu05_data$baseline$gender.factor # raw more meds ctu05_data$allmedi$no_more_meds # transformed more meds ctu05_data$allmedi$no_more_meds.factor
Note that descriptive labels have also been automatically added to the data. \vspace{5pt}
label(ctu05_data$allmedi$no_more_meds.factor) label(ctu05_data$baseline$gender.factor) label(ctu05_data$esurgeries$surgery_organ.factor)
Datetime data is also transformed and similarly to the factorization
process the names are concatenated with
.date
or .datetime
.
\vspace{5pt}
# raw ctu05_data$baseline$visit_date # processed ctu05_data$baseline$visit_date.date # raw only head head(ctu05_data$baseline$hiv_date) # processed only head head(ctu05_data$baseline$hiv_date.datetime) # classes class(ctu05_data$baseline$visit_date) class(ctu05_data$baseline$visit_date.date) class(ctu05_data$baseline$hiv_date) class(ctu05_data$baseline$hiv_date.datetime)
Depending on the setup, incomplete dates can be valid entries in a secuTrial database.
Thus they will also occasionally appear in your exports. The datetime conversion
does not work in these cases and NA
s are created. If this happens, secuTrialR
will
warn you accordingly and you should have a closer look into the affected
datetime variables and whether you would like to perform so-called date imputation.
The secuTrialdata
object also contains information on the export options.
\vspace{5pt}
ctu05_data$export_options
export_options
itself is a list
. If you are interested in more information
than is printed you can also access it. Let's assume you would like to know
the project_name
and encoding
.
\vspace{5pt}
ctu05_data$export_options$project_name ctu05_data$export_options$encoding
Much more information is stored in the elements of export_options
.
The names of the elements should be descriptive enough to infer the contents.
\vspace{5pt}
names(ctu05_data$export_options)
secuTrialdata
objectsNow that you understand the secuTrialdata
object we will show you some generic functions
you can use on objects of this class.
First off you may be interested in a table of participants. \vspace{5pt}
get_participants(ctu05_data)
Please note that the mnpaid
column in this table corresponds to the pat_id
column in other tables.
\newpage
You can extract information about participant recruitment per centre and year by applying
annual_recruitment()
on a secuTrialdata
object.
\vspace{5pt}
annual_recruitment(ctu05_data)
Since the centre names often have a systematic addition (e.g. (RPACK)) we have
enabled the option to remove certain parts of the centre descriptions via
regular expressions (i.e. rm_regex
argument). In this case the regular expression
removes trailing parentheses and everything they enclose.
\vspace{5pt}
annual_recruitment(ctu05_data, rm_regex = "\\(.*\\)$")
It is also possible to plot the recruitment over time. \vspace{5pt}
plot_recruitment(ctu05_data, cex = 1.2, rm_regex = "\\(.*\\)$")
\newpage
secuTrialR
can provide a depiction of the visit structure, although only where
the visit plan is fixed. Black rectangles in the grid represent a form to be filled (x)
during one of the visits (y).
\vspace{5pt}
vs <- visit_structure(ctu05_data) plot(vs)
If you are not sure about how complete the data in your export is, it may be useful to get a quick overview of how well the forms have been filled. The below table shows both absolute and relative numbers for a few forms. \vspace{5pt}
fss <- form_status_summary(ctu05_data) tail(fss, n = 5)
Please note that a form is only complete if all required fields have been filled. Thus, a whole study may have 99% completeness on variable basis while showing 0% completeness on form basis. It is currently not technically possible to assess completeness on variable basis in a generic way. Hence, high completeness on form basis implies high completeness on variable basis but NOT vice versa.
If you would rather retrieve information on form completeness for each participant individually you can perform the following. \vspace{5pt}
fsc <- form_status_counts(ctu05_data) # show the top head(fsc)
Linkages amongst forms can be explored with the links_secuTrial()
function.
This relies on the igraph
package to create a network. It is possible to
interact with the network, e.g. move nodes around in order to read the labels better.
The R graphics device ID is returned to the console, but can be ignored. Forms are plotted in
deep yellow, variables in light blue.
\vspace{5pt}
links_secuTrial(ctu05_data)
The output can not be shown within this vignette but you should give it a try. Please note that the linkage plot is likely most useful without the audit trail data in the export.
\newpage
During study monitoring it is common practice to check random participants from a study database. These participants should be retrieved in a reproducible fashion, which can be achieved by setting a so-called seed. The below function allows reproducible retrieval for a loaded secuTrial data export. \vspace{5pt}
# randomly retrieve at least 25 percent of participants recorded after March 18th 2019 # from the centres "Inselspital Bern" and "Charité Berlin" return_random_participants(ctu05_data, percent = 0.25, seed = 1337, date = "2019-03-18", centres = c("Inselspital Bern (RPACK)", "Charité Berlin (RPACK)"))
Please note that earlier R versions may return different results because
there is a different rng_config
(i.e. RNGkind()
).
For this reason we have added the rng_config
to the output.
secuTrial allows implementing calculated variables (i.e. scores). Data is not directly entered into these variables but rather calculated automatically. Scores are defined by a set of rules and use the data in other variables as basis. For example the age of a study participant at data entry can be calculated as the difference between the participant's birthday and the day of data entry.
It is advisable to recalculate or validate score variable data before data analysis. A rule of thumb: The more complex a score is and the more data from different forms is necessary for its calculation the more likely its value should be recalculated. The below function will allow you to detect which variables this concerns. \vspace{5pt}
return_scores(ctu05_data)
Sometimes, during a study, certain fields may be hidden because data should no longer be entered into them. If this is the case and the data of these fields is part of your export is likely good to know about it. In this case nothing is hidden. \vspace{5pt}
return_hidden_items(ctu05_data)
In ongoing studies it is possible that changes to the secuTrial data entry
interface (i.e. the electronic case report forms) are made. Sometimes these
changes may call for adjustments in analysis code. It is considered good practice
to run diff_secuTrial()
on the last export and the current export of a project
to at least make yourself aware of potential changes in the setup. If there are
differences, the results of this function should be interpreted as a first indicator
since they may not cover all alterations. Information is returned on forms and
variables. A detailed list of changes can be produced in the secuTrial
FormBuilder with "Compare project setup".
For the below diff_secuTrial()
showcase we emulated a changed setup of CTU05
by copying the setup and importing it in the FormBuilder as a new secuTrial
project (CTU06). From this, we created a data export (v1) and then made a few
minor changes and exported again (v2). If this sounds confusing, never mind.
CTU06 v1 is simply a copy of CTU05. CTU06 v2 is a slighly altered version of CTU06 v1.
\vspace{5pt}
ctu06_v1 <- read_secuTrial(system.file("extdata", "sT_exports", "change_tracking", "s_export_CSV-xls_CTU06_version1.zip", package = "secuTrialR")) ctu06_v2 <- read_secuTrial(system.file("extdata", "sT_exports", "change_tracking", "s_export_CSV-xls_CTU06_version2.zip", package = "secuTrialR")) diff_secuTrial(ctu06_v1, ctu06_v2)
As you can see ctu06_v2
contains the two additional forms mnpctu06anewform
and
mnpctu06anothernewform
and the two additional variables new_item_in_fu
and
new_item_in_new_form
.
Given that you are working with R it is unlikely that you need such conversions for
yourself. However, collaborators may ask for data which is readily importable into SPSS,
STATA or SAS. For this you can use write_secuTrial()
.
Since this has not been heavily tested or used there may be issues and
you might prefer doing this manually with the haven
package. One particular
sticking point is the length of variable names - R is not restrictive in this
respect, but other software can be. secuTrialR
does not truncate names, prefering
to leave this to the user, which can cause write_secuTrial()
to fail with an error.
\vspace{5pt}
# retrieve path to a temporary directory tdir <- tempdir() # write spss write_secuTrial(ctu05_data, format = "sav", path = tdir)
secuTrialdata
In some cases it may be useful to subset your secuTrialdata
object.
For example if you have cohort data and would like to supply a subset of
the data for a retrospective study. We have implemented this
option with subset_secuTrial()
. It will truncate your secuTrialdata
object
and return a new secuTrialdata
object which is a subset of the original data.
It is possible to subset by including or excluding specific participant
ids or centres.
\vspace{5pt}
# initialize some subset identifiers participants <- c("RPACK-INS-011", "RPACK-INS-014", "RPACK-INS-015") centres <- c("Inselspital Bern (RPACK)", "Universitätsspital Basel (RPACK)") # exclude Bern and Basel ctu05_data_berlin <- subset_secuTrial(ctu05_data, centre = centres, exclude = TRUE) # exclude Berlin ctu05_data_bern_basel <- subset_secuTrial(ctu05_data, centre = centres) # keep only subset of participants ctu05_data_pids <- subset_secuTrial(ctu05_data, participant = participants) class(ctu05_data_berlin) class(ctu05_data_bern_basel) class(ctu05_data_pids)
If you subset based on centres all traces of deleted centres will be removed. If you remove based on participant ids all traces of deleted participants will be removed. \vspace{5pt}
# only Berlin remains ctu05_data_berlin$ctr # all centres remain even though all three participant ids are from Bern ctu05_data_pids$ctr
\newpage
Since the truncated object's class remains unchanged (i.e. secuTrialdata
) you can
still use the generic functions on it.
Let's say you would only like to look at the recruitment plot for Bern alone.
\vspace{5pt}
# keep only Bern ctu05_data_bern <- subset_secuTrial(ctu05_data, centre = "Inselspital Bern (RPACK)") # plot plot_recruitment(ctu05_data_bern)
... or Bern and Berlin. \vspace{5pt}
# keep only Bern and Berlin ctu05_data_bern_berlin <- subset_secuTrial(ctu05_data, centre = c("Inselspital Bern (RPACK)", "Charité Berlin (RPACK)")) # plot plot_recruitment(ctu05_data_bern_berlin)
If you are creating reports in which you would like to directly
link to specific pages of your secuTrial DataCapture you can
use build_secuTrial_url
. If you are no expert regarding the
secuTrial server architecture you would like to build links for, you should
talk to the server admin or consult the build_secuTrial_url
help page.
They will be able to guide you regarding the information for the server
,
instance
, customer
and project
parameters. The docid
, however, is
included in your export data in the non-meta data tables of the
secuTrialdata
object and can be found in the mnpdocid
columns.
head(ctu05_data$treatment$mnpdocid) head(ctu05_data$baseline$mnpdocid)
To demsonstrate build_secuTrial_url
we will use imaginary data
for the server
, instance
, customer
and project
parameters.
The real counterparts on your server will likely look structurally
similar.
server <- "server.secutrial.com" instance <- "ST21-setup-DataCapture" customer <- "TES" project <- "7036" # make three links with the first three baseline docids bl_docids <- head(ctu05_data$baseline$mnpdocid, n = 3) links <- build_secuTrial_url(server, instance, customer, project, bl_docids)
These are the links:
r links[1]
r links[2]
r links[3]
Of course they are dead ends but maybe you can use them to make out the arguments for your server.
as.data.frame
functionThis vignette has been working with the secuTrialdata
object, which is of type list
.
For some users, working with a list
can be tiresome so secuTrialR
provides an
as.data.frame()
method to save the data.frame
s in the secuTrialdata
object to
an environment of your choice.
As an example, we will create an environment called env
and check that it's empty
before running as.data.frame()
...
\vspace{5pt}
env <- new.env() ls(env)
# add files to env as.data.frame(ctu05_data, envir = env)
... and afterwards. \vspace{5pt}
ls(env)
Substituting env
with .GlobalEnv
instead would also be an option and would make the data.frames immediately accessible without having to refer to an environment.
Certain warning
messages can occur quite frequently when running read_secuTrial()
.
Some of them may call for deliberate action and thus it is important to understand them.
We briefly mentioned some of them earlier in this document but will now more closely
explain how they can be interpreted.
Please note that warning
messages may "pile up" depending on the export you are reading.
For example this may happen if there are many date variables with incomplete data. This is
no reason for concern. We suggest that you read them and interpret them based on the
explanations below. We use a_form_name
and a_variable_name
as place holders in the examples.
If in doubt you can always work with the raw data because it is never overwritten.
The below warning tells you that some data in a date variable could not be converted
during the process of date conversion (i.e. dates_secuTrial()
). This ususally occurs
if incomplete date entries are present. Since the raw data is not overwritten but rather a
variable_name.date
or variable_name.datetime
column are added to the dataset you can specifically see which
values could not be converted because the raw data will contain data while the corresponding .date
/.datetime
entires will be NA
. The warning
also indicates where to look. The dummy example below indicates to look
at the variable a_variable_name
in form a_form_name
.
# incomplete dates warning( "In dates_secuTrial.data.frame(tmp, datevars, timevars, dateformat, : Not all dates were converted for variable: 'a_variable_name' in form: 'a_form_name' This is likely due to incomplete date entries." )
In some cases secuTrial allows differently coded data to be decoded to the same target value for the
same variable. For instance this can happen if hierarchical lookuptables have been implemented
in the database. Because this interferes with the factorization (i.e. factorize_secuTrial()
)
we add the code to the duplicate decoded value and return the below message to make you aware.
If you run into this warning
message we suggest running the table()
function on the
variable in question. This will likely clarify the above explanation.
# duplicate factors warning( "In factorize_secuTrial.data.frame(curr_form_data, cl = object$cl, : Duplicate values found during factorization of a_variable_name")
Sometimes the labels of variables in a secuTrial database implementation may be changed
after release of the database. In these cases all labels (current and previous versions) are
added to the label
attribute during labelling (i.e. label_secuTrial()
) and the below warning
is triggered. It indicates which variables in which forms are affected.
# duplicate labels warning( "In label_secuTrial.secuTrialdata(d) : The labels attribute may be longer than 1 for the following variables and forms. Likely the label was changed from its original state in the secuTrial project setup. variables: a_variable_name forms: a_form_name" )
secuTrialdata
objectNaturally, you will sometimes need to merge/join some of the data from the individual form data stored
in your secuTrialdata
object. To achieve this you can use the base R merge()
function.
For our dataset we might be interested in merging the baseline
form data with that of the treatment
form. For this we can use the mnpcvpid
which uniquely identifies each participant visit. Since we
are only interested in the rando_treatment
variable we will shrink the data in the treatment
form prior to merging.
\vspace{5pt}
treatment_shrink <- ctu05_data$treatment[, c("mnpcvpid", "rando_treatment")]
Because we do not want to drop non-matching rows from baseline
we set all.x = TRUE
.
As you can see from the dim()
calls, one column has been added after the merge.
This corresponds to the rando_treatment
variable.
bl_treat <- merge(x = ctu05_data$baseline, y = treatment_shrink, by = "mnpcvpid", all.x = TRUE) # check dimensions dim(ctu05_data$baseline) dim(bl_treat)
Another common task may be to merge repetition form data to its parent form.
In our case esurgeries
can be naively merged with baseline
via the
mnpdocid
(from the secuTrial manual: "Each eCRF document record has a unique document identifier."):
\vspace{5pt}
bl_surg <- merge(x = ctu05_data$baseline, y = ctu05_data$esurgeries, by = "mnpdocid")
Please note, that such naive merging can cause duplication of data if the ids that the merge is directed by are not unique.
This also happened during the production of bl_surg
in the code above. Participant "RPACK-INS-012" exhibits the
mnpdocid
234 twice in the esurgeries
repetition form which causes a duplication of the baseline
data matching
mnpdocid
234.
\vspace{5pt}
table(ctu05_data$esurgeries$mnpdocid)
Lets briefly illustrate the consequences by looking at a table()
of the height
variable from the baseline
form before and after merging.
\vspace{5pt}
# before merge table(ctu05_data$baseline$height) # after merge table(bl_surg$height)
A closer look reveals that 180
now appears four times instead of three, which can be attributed to
the duplication. This is not a favourable outcome because it can cause confusion and misinterpretation.
A better approach is to change the structure of your repetition form before merging to make the ids you merge by unique.
For this you need to investigate which data you would like to merge and design an appropriate stategy.
In our example case we are interested in the surgery_organ
. Of course we also need to drag the mnpdocid
along
to perform the actual merge.
\vspace{5pt}
# write a temporary object surg <- ctu05_data$esurgeries[, c("mnpdocid", "surgery_organ.factor")] # only retain non NA rows surg <- surg[which(! is.na(surg$surgery_organ.factor)), ] # show it surg
\newpage
In order to prevent duplication we can restructure the data before merging.
\vspace{5pt}
library(tidyr) # pivot_wider # add a count surg$count <- 1 # show the data surg # make it wide surg_wide <- pivot_wider(surg, names_from = surgery_organ.factor, values_from = count) # show the wide data surg_wide
Checking the dimensions before and after merging reveals that the structure, especially the line count, remains
the same except for the data added from the esurgeries
repetition form (i.e. Stomach
and Other
). Also,
as expected, the table()
of the height variable returns the expected result (i.e. 180 is present three not four times).
\vspace{5pt}
# merge bl_surg_no_dup <- merge(x = ctu05_data$baseline, y = surg_wide, by = "mnpdocid", all.x = TRUE) # compare dimensions dim(bl_surg_no_dup) dim(ctu05_data$baseline) # check the height variable table(bl_surg_no_dup$height)
The above description only provides a very brief and simplified example. Merging
strategies need to be individually tailored and require a good understanding of the data at hand.
The links_secuTrial()
function may be helpful to understand which variables will allow you to
merge forms.
mnp*
variablesThere is a plethora of variables in the tables of secuTrial exports whose names start
with mnp
. These are metadata variables which are e.g. important to logically link the
different tables. Explaining them all is beyond the scope of this vignette.
For detailed explanations, please refer to the secuTrial "Export Formats" user manual.
\newpage
sessionInfo()
\vspace{185pt}
Disclaimer
The descriptions of the secuTrial exports used in this vignette and other secuTrialR
documentation correspond to our understanding of them and come with no warranty.
For in depth details please refer to the original secuTrial manuals.
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.