4.2.1.1 Data transformation

The only necessary elements that must be provided are the dataset to be transformed and the corresponding dictionary. If the project is longitudinal, as in the case of covican, also the event-form dataset should be specified. These elements can be specified directly using the output of the redcap_data function or separately in different arguments.

#Option A: list object 
covican_transformed <- rd_transform(covican)

#Option B: separately with different arguments
covican_transformed <- rd_transform(data = covican$data, 
                                    dic = covican$dictionary, 
                                    event_form = covican$event_form)

#Print the results of the transformation
covican_transformed$results

1. Recalculating calculated fields and saving them as '[field_name]_recalc'

| Total calculated fields | Non-transcribed fields | Recalculated different fields |
|:-----------------------:|:----------------------:|:-----------------------------:|
|            2            |         0 (0%)         |            1 (50%)            |


|     field_name      | Transcribed? | Is equal? |
|:-------------------:|:------------:|:---------:|
|         age         |     Yes      |   FALSE   |
| screening_fail_crit |     Yes      |   TRUE    |

2. Transforming checkboxes: changing their values to No/Yes and changing their names to the names of its options. For checkboxes that have a branching logic, when the logic is missing their values will be set to missing

Table: Checkbox variables advisable to be reviewed

| Variables without any branching logic |
|:-------------------------------------:|
|        type_underlying_disease        |

3. Replacing original variables for their factor version

4. Deleting variables that contain some patterns

This function will return a list with the transformed dataset, dictionary and the output of the results of the transformation.

As we can see, there are 4 steps in the transformation:

1. Recalculation of REDCap calculated fields: it finds all the calculated fields and recalculates them using the REDCap logic specified in the calculation field translated into R. The recalculated variable is saved as the original name adding ’_recalc’ at the end. It can happen that the logic found contains some specific smart-variables or other complex structures which the function is not able to transcribe. With the summary found in results we can see how many calculated fields have been found, if they have been transcribed and, if that is the case, if the recalculated variable is equal to the original one.

Note: If the REDCap project is longitudinal and the event-form is not specified, this step will not be executed.

In the example, we can see how there are two REDCap calculated fields, both have been transcribed successfully and the recalculation of the age does not match the original calculated variable from REDCap.

2. Checkbox transformation: by default, it changes the names of the checkboxes to the name of its corresponding option and the name of their labels to ‘No/Yes’. If we want to specify another pair of label names we can specify them using the checkbox_labels argument as we will see. Furthermore, if the checkbox contains a branching logic, when this logic evaluated returns a missing value (some of the variables specified in it are missing) the values of the checkbox will be set to missing.

Note: If the REDCap project is longitudinal and the event-form is not specified, the evaluations of the branching logic will not be done.

For example, let’s explain the transformation that undergo the variables corresponding to the checkbox field of the type of underlying disease. The variables were named originally as _type_underlying_disease__0 and _type_underlying_disease__1 although the name of the options are ‘Haematological cancer’ and ‘Solid tumour’. Thus, in the transformed dataset, the names are converted to type_underlying_disease_haematological_cancer and type_underlying_disease_solid_tumour. Then, since this checkbox variable does not have a branching logic, the variable is advised to be reviewed by the user in the results, as seen above. When reviewed we could use an additional function rd_insert_na to insert the necessary missing values into this variable, as we will explain later. If a branching logic was found for this variable, rd_transform will insert automatically the missing values when the logic is not satisfied and no further transformation will be needed.

3. Replacement of the original variable by its factor version: REDCap creates two versions of the variables in the dataset for multiple-choice fields: a numerical one with the number that corresponds to each category and a factor one containing the labels of each category. In this step, we will replace the original variables with their factor versions, except for redcap_event_name and redcap_data_access_group, for which we will keep both versions. We can specify other variables that we do not want to transform to factor using the argument exclude_to_factor which we will see later.
4. Elimination of variables containing some pattern: by default, the pattern that the function looks for is ’_complete’, but we can specify any other pattern using the argument delete_vars, as explained later.

In this case, we do not have any variable with the pattern ’_complete’ since the built-in dataset only contains a sample of the variables of the project. All REDCap projects, when downloaded, contain one variable with the pattern ’_complete’ for each form indicating if the form has been marked as incomplete/unverified/completed. In general, we do not need this information so these variables are removed by default.

4.2.1.2 Data transformation and classification by event

Additionally, we can change the final structure of the transformed dataset. For example, we can split it by each event. This can be done by specifying in the final_format argument that we want our data to be split by event.

dataset <- rd_transform(covican,
                        final_format = "by_event")

#To print the results
dataset$results

1. Recalculating calculated fields and saving them as '[field_name]_recalc'

| Total calculated fields | Non-transcribed fields | Recalculated different fields |
|:-----------------------:|:----------------------:|:-----------------------------:|
|            2            |         0 (0%)         |            1 (50%)            |


|     field_name      | Transcribed? | Is equal? |
|:-------------------:|:------------:|:---------:|
|         age         |     Yes      |   FALSE   |
| screening_fail_crit |     Yes      |   TRUE    |

2. Transforming checkboxes: changing their values to No/Yes and changing their names to the names of its options. For checkboxes that have a branching logic, when the logic is missing their values will be set to missing

Table: Checkbox variables advisable to be reviewed

| Variables without any branching logic |
|:-------------------------------------:|
|        type_underlying_disease        |

3. Replacing original variables for their factor version

4. Deleting variables that contain some patterns

5. Final arrangment of the data by event

Now, a final step in the transformation has been added, which consists in splitting the data according to the events in the study. So, now the transformed dataset found in the output of the function is a tibble object with as many data frames as events there are in the REDCap project:

dataset$data

# A tibble: 2 × 3
  events                   vars       df             
  <chr>                    <list>     <list>         
1 baseline_visit_arm_1     <chr [36]> <df [190 × 36]>
2 follow_up_visit_da_arm_1 <chr [8]>  <df [152 × 8]> 

The column df of the nested dataframe is a list containing the data corresponding to each event. Also the variables of the forms that are found in each event are reported in the column vars.

Note: If the REDCap project is longitudinal and the event-form is not specified, this transformation is not posible.

4.2.1.3 Data transformation and classification by form

Another option is to split the data by the forms found in the REDCap project. We will use also the final_format argument to specify that we want to split data by form:

dataset <- rd_transform(covican,
                        final_format = "by_form")

#To print the results
dataset$results

1. Recalculating calculated fields and saving them as '[field_name]_recalc'

| Total calculated fields | Non-transcribed fields | Recalculated different fields |
|:-----------------------:|:----------------------:|:-----------------------------:|
|            2            |         0 (0%)         |            1 (50%)            |


|     field_name      | Transcribed? | Is equal? |
|:-------------------:|:------------:|:---------:|
|         age         |     Yes      |   FALSE   |
| screening_fail_crit |     Yes      |   TRUE    |

2. Transforming checkboxes: changing their values to No/Yes and changing their names to the names of its options. For checkboxes that have a branching logic, when the logic is missing their values will be set to missing

Table: Checkbox variables advisable to be reviewed

| Variables without any branching logic |
|:-------------------------------------:|
|        type_underlying_disease        |

3. Replacing original variables for their factor version

4. Deleting variables that contain some patterns
5. Final arrangment of the data by form

As before, a final step in the transformation has been added, which is to split the data according to the forms in the study. Thus, the transformed dataset will now be a tibble object with as many data frames as forms there are in the REDCap project:

dataset$data

# A tibble: 7 × 4
  form                        events    vars       df             
  <chr>                       <list>    <list>     <list>         
1 inclusionexclusion_criteria <chr [1]> <chr [11]> <df [190 × 11]>
2 demographics                <chr [1]> <chr [9]>  <df [190 × 9]> 
3 comorbidities               <chr [1]> <chr [10]> <df [190 × 10]>
4 cancer                      <chr [1]> <chr [16]> <df [190 × 16]>
5 vital_signs                 <chr [2]> <chr [7]>  <df [177 × 7]> 
6 laboratory_findings         <chr [2]> <chr [7]>  <df [177 × 7]> 
7 microbiological_studies     <chr [1]> <chr [6]>  <df [190 × 6]> 

Note: If the REDCap project is longitudinal and the event-form is not specified, this transformation is not posible.

4.2.1.4 Additional arguments

There are other arguments which can be used to customize some of the transformation steps that the function performs by default:

checkbox_labels: specifies the name of the categories for the checkbox variables. Default is ‘No/Yes’, but we can change it to ‘N/Y’:

dataset <- rd_transform(covican,
                        checkbox_labels = c("N", "Y"))

checkbox_na: logical argument involved in the transformation of checkbox variables. For checkbox variables that have a branching logic specified, when the logic is missing the values of the checkbox will be always converted to missing. Additionally, if this argument is true then also when the branching logic isn’t satisfied their values will be converted to missing.

dataset <- rd_transform(covican,
                        checkbox_na = TRUE)

exclude_to_factor: specifies the name of the variables that we do not want to transform into a factor. For example, if we want the variable dm to keep its original numeric version:

dataset <- rd_transform(covican,
                        exclude_to_factor = "dm")

keep_labels: logical argument, retains the labeling of the dataset columns from REDCap. By default, the function will remove the labels of the dataset and the labels can be found in the dictionary:

dataset <- rd_transform(covican,
                        keep_labels = TRUE)

str(dataset$data[,1:2])

'data.frame':   342 obs. of  2 variables:
 $ record_id        : 'labelled' chr  "100-13" "100-13" "100-16" "100-16" ...
  ..- attr(*, "label")= Named chr "Record ID"
  .. ..- attr(*, "names")= chr "record_id"
 $ redcap_event_name: 'labelled' chr  "baseline_visit_arm_1" "follow_up_visit_da_arm_1" "baseline_visit_arm_1" "follow_up_visit_da_arm_1" ...
  ..- attr(*, "label")= Named chr "Event Name"
  .. ..- attr(*, "names")= chr "redcap_event_name"

delete_vars: every variable containing the strings specified in this argument will be removed from the dataset. By default, the value of delete_vars is ‘_complete’. For example, we can change the argument to remove the inclusion and exclusion criteria variables from the dataset (variables that contain ‘inc_’ and ‘exc_’ in their names):

dataset <- rd_transform(covican,
                        delete_vars = c("inc_", "exc_"))

which_event: in the transformation by event explained earlier, we can specify whether we want to keep only one out of all the events in the dataset. For example, if we only want to keep the baseline visit:

dataset <- rd_transform(covican,
                        final_format = "by_event",
                        which_event = "baseline_visit_arm_1")

which_form: in the transformation by form explained earlier, we can specify whether we want to keep only one of the forms. For example, if we only want to keep the demographic form:

dataset <- rd_transform(covican,
                        final_format = "by_form",
                        which_form = "demographics")

data <- dataset$data

names(data)

[1] "record_id"                       "redcap_event_name"              
[3] "redcap_data_access_group"        "redcap_event_name.factor"       
[5] "redcap_data_access_group.factor" "d_admission"                    
[7] "d_birth"                         "age"                            
[9] "age_recalc"                     

wide: in the transformation by form, we can specify that we want each of the split datasets to be in a wide format. This is useful if the form appears in more than one event (or in a repeated event). Then, we will only have one row per patient and all the variables of the form will be in columns repeated by each event in the order that the events appear in REDCap. For example, if we want to keep only the laboratory findings in a wide format we can do:

dataset <- rd_transform(covican,
                        final_format = "by_form",
                        which_form = "laboratory_findings",
                        wide = TRUE)

head(dataset$data)

# A tibble: 6 × 5
  record_id available_analytics_1 available_analytics_2 potassium_1 potassium_2
  <chr>     <fct>                 <fct>                       <dbl>       <dbl>
1 100-13    Yes                   Yes                          3.66        4.1 
2 100-16    Yes                   No                           4.04       NA   
3 100-31    Yes                   <NA>                         4.58       NA   
4 100-34    Yes                   No                           3.48       NA   
5 100-36    Yes                   No                           4.09       NA   
6 100-52    Yes                   Yes                          3.7         7.15

4.3.1.1 Output

First, we will examine the output of this function. When the rd_query function is executed, it returns a list that includes a data frame with all the queries identified and a second element with a summary of the number of generated queries in each specified variable for each expression applied:

The data frame is designed to aid users in locating each query in their REDCap project. It includes information such as the record identifier, the Data Access Group (DAG), the event in which each query can be found, along with the name and the description of the analyzed variable and a brief description of the query.

Let’s see some examples of the usability of the function in generating different types of queries.

4.3.1.2 Missings

If we want to identify missing values in the variables copd and age in the raw transformed data, a list of required arguments needs to be supplied. We must use the variables argument to specify the variables from the database that will be examined and the expression argument to describe the expression that will be applied to those variables, in this case ‘%in%NA’ to detect missing values. Additionaly, we must use the data and dic arguments to indicate the R objects containing the REDCap data and dictionary, respectively. If the REDCap project presents a longitudinal design, we should also specify the event in which the described variables are present through the use of the event argument:

In this case, we can observe that there are 6 missing values in the copd variable and 5 missing values in age.

4.3.1.3 Missings of variables with a branching logic

Another example is when we try to identify missing values in variables where a branching logic is employed. In this scenario, when the conditions of the branching logic are not satisfied, by definition, all of the values should be missing and thus queries for this specific missing values (conditions not met) should not be reported. To adress this, if a variable presents a branching logic, the function will issue a warning with a message to check the results element of our output:

example <- rd_query(covican_transformed,
                    variables = c("age", "copd", "potassium"),
                    expression = c("%in%NA", "%in%NA", "%in%NA"),
                    event = "baseline_visit_arm_1")

Warning: Some of the variables that were checked for missings present a branching logic. 
Check the results tab of output for more details (...$results).

# Printing results
example$results

As we can see, in addition to the missing values of the age and copd variables already identified, there are 31 missing values in the potassium variable. The variable potassium has the following branching logic [available_analytics]=‘1’, which means that we should only identify the missing values when available_analytics has the value ‘1’. To accomplish this, we can use the filter argument to ensure that the condition in this branching logic is fulfilled. Recall that, in the transformed dataset, the value ‘1’ was changed to ‘Yes’.

example <- rd_query(covican_transformed,
                    variables = c("potassium"),
                    expression = c("%in%NA"),
                    event = "baseline_visit_arm_1",
                    filter = c("available_analytics=='Yes'"))

Warning: Some of the variables that were checked for missings present a branching logic. 
Check the results tab of output for more details (...$results).

# Printing results
example$results

The total number of missing values changes when we use the filter argument, the variable potassium now presents 21 missing values instead of the previous 31 cases identified. This means that we were identifying 10 missing values in which available_analytics did not have the value 1 and, therefore, should not be considered as missing values.

4.3.1.4 Expressions

Up until this point, we have illustrated examples where the expression applied is for detecting missing values. But, as previously mentioned, the rd_query function is also able to identify outliers or observations that fulfill a specific condition. Hence, to identify, for example, all the observations where age is greater than 70, we should use the expression argument again specifying ‘>70’:

example <- rd_query(variables="age",
                    expression=">70",
                    event="baseline_visit_arm_1",
                    dic=covican_transformed$dictionary,
                    data=covican_transformed$data)

# Printing results
example$results

We can add other variables with other specific expressions in the same function because it is designed to treat the arguments variables and expression as vectors, so that the element at position ‘n’ of expression is applied to the element at position ‘n’ of variables.

For example, if we wanted to identify all the observations where age is greater than 70 and all the observations where copd is ‘Yes’ we should use:

example <- rd_query(covican_transformed,
                    variables=c("age", "copd"),
                    expression=c(">70", "=='Yes'"),
                    event="baseline_visit_arm_1")

# Printing results
example$results

In a more complex scenario, for example, to identify all the observations where age is greater than 70, less than 80, or is a missing value we should use the following expression:

example <- rd_query(covican_transformed,
                    variables="age",
                    expression="(>70 & <80) | %in%NA",
                    event="baseline_visit_arm_1")

# Printing results
example$results

4.3.1.5 Special cases

Same expression for all variables

In order to evaluate the same expression for all variables, the user should supply just a single element for expression:

example <- rd_query(covican_transformed,
                    variables = c("copd","age","dm"),
                    expression = "%in%NA",
                    event = "baseline_visit_arm_1")

Warning: There are more variables than expressions, so the same expression was
applied to all variables

# Printing results
example$results

The function issues a warning every time the same expression is applied to all variables to ensure that the user did not make a mistake when providing the information for each argument.

Not defining an event in a dataset with multiple events

Another special case is when the data analysed corresponds to a REDCap longitudinal project, but the event argument of the function is not defined:

example <- rd_query(covican_transformed,
                    variables = "copd",
                    expression = "%in%NA")

Warning: event = NA, but the dataset presents a variable that indicates the
presence of events, please specify the event.

# Printing results
example$results

As we can see, there are 152 new missing values in the follow-up visit because the variable copd it is only present in the baseline visit. Thus, it might result in an overestimation of the number of missing values, as the function considers all the events of the study if no event is specified.

The function will issue a warning if it detects that the REDCap project contains multiple events and the event argument is not specified.

4.3.1.6 Additional arguments

variable_names, query_name, instrument

These arguments allow users to customize the data frame returned by the function. We can change the variables names using the variables_names argument, alter the description of the query using the query_name argument or even change the name of the instrument using the instrument argument:

example<- rd_query(covican_transformed,
                   variables = c("copd"),
                   variables_names = c("Chronic obstructive pulmonary disease (Yes/No)"),
                   expression = c("%in%NA"),
                   query_name = c("COPD is a missing value."),
                   instrument = c("Admission"),
                   event = "baseline_visit_arm_1")

Output:

negate

This argument can be used to negate the expression applied to the variables. For example, if we want to identify all the non missing values of the variable copd, we can apply the expression ‘%in%NA’ which normally would report the missing values and add negate = TRUE, so the result will be the number of non missing values in copd:

example <- rd_query(covican_transformed,
                    variables = "copd",
                    expression = "%in%NA",
                    negate = TRUE,
                    event = "baseline_visit_arm_1")

# Printing results
example$results

There are 184 non missing values in the variable copd.

addTo

In order to keep all queries in the same R object, we can use the addTo argument to specify the output of another query dataset.

example2 <- rd_query(covican_transformed,
                     variables = "age",
                     expression = "%in%NA",
                     event = "baseline_visit_arm_1",
                     addTo = example)

# Printing results
example2$results

We have joined our former output of 184 non missing values in the variable copd with the new query dataset composed by the 5 missing values of the variable age.

report_title

To customize the title of the summary of queries, we can use the report_title argument:

example <- rd_query(covican_transformed,
                    variables = c("copd", "age"),
                    expression = c("%in%NA", "<20"),
                    event = "baseline_visit_arm_1",
                    report_title = "Missing COPD values in the baseline event")

# Printing results
example$results

The default title of the summary is “Report of queries” but we have changed it to “Missing COPD values in the baseline event”.

report_zeros

By default, the function will only report, in the summary of queries, variables with at least one query and will omit those with zero queries. To include these omitted variables in the summary, we can use the report_zeros argument:

example <- rd_query(covican_transformed,
                    variables = c("copd", "age"),
                    expression = c("%in%NA", "<20"),
                    event = "baseline_visit_arm_1",
                    report_zeros = TRUE)

# Printing results
example$results

The variable age is reported in the summary in spite of not having any queries identified.

by_dag

If the REDCap project has Data Access Groups (DAGs), it might be of our interest to report the summary by each one of the DAGs. To do that, we can use the by_dag argument:

example <- rd_query(covican_transformed,
                    variables = c("copd", "age"),
                    expression = c("%in%NA", ">60"),
                    event = "baseline_visit_arm_1",
                    by_dag = TRUE)

Now we can choose to see the report of a specific DAG, for example, the summary of the generated queries for the Hospital 2:

# Printing results
example$results$`Hospital 2`

For this DAG, there are 3 values of age bigger than 60 and 2 missing values in the variable copd.