Combining Data Extract with Data Merge

NEST CoreDev

teal.transform allows the app user to oversee transforming a relational set of data objects into the final dataset for analysis. User actions create a R expression that subsets and merges the input data.

In the following example we will create an analysis dataset ANL by:

1. Selecting the column AGE from ADSL
2. Selecting the column AVAL and filtering the rows where PARAMCD is OS from ADTTE
3. Merging the results from the above datasets using the primary keys.

Note that primary key columns are maintained when selecting columns from datasets.

Let’s see how to achieve this dynamic select, filter, and merge operations in a shiny app using teal.transform.

Step 1/5 - Preparing the Data

library(teal.transform)
library(teal.data)
#> Loading required package: teal.code
library(shiny)

# Define data.frame objects
ADSL <- teal.data::rADSL
ADTTE <- teal.data::rADTTE

# create a list of reactive data.frame objects
datasets <- list(
  ADSL = reactive(ADSL),
  ADTTE = reactive(ADTTE)
)

# create join_keys
join_keys <- join_keys(
  join_key("ADSL", "ADSL", c("STUDYID", "USUBJID")),
  join_key("ADSL", "ADTTE", c("STUDYID", "USUBJID")),
  join_key("ADTTE", "ADTTE", c("STUDYID", "USUBJID", "PARAMCD"))
)

Step 2/5 - Creating data extract specifications

In the following code block, we create a data_extract_spec object for each dataset, as illustrated above. It is created by the data_extract_spec() function which takes in four arguments:

1. dataname is the name of the dataset to be extracted.
2. select helps specify the columns from which we wish to allow the app user to select. It can be generated using the function select_spec(). In the case of ADSL, we restrict the selection to AGE, SEX, and BMRKR1, with AGE being the default selection.
3. filter helps specify the values of a variable we wish to filter during extraction. It can be generated using the function filter_spec(). In the case of ADTTE, we filter the variable PARAMCD by allowing users to choose from CRSD, EFS, OS, and PFS, with OS being the default filter.
4. reshape is a boolean which helps to specify if the data needs to be reshaped from long to wide format. By default it is set to FALSE.

adsl_extract <- data_extract_spec(
  dataname = "ADSL",
  select = select_spec(
    label = "Select variable:",
    choices = c("AGE", "SEX", "BMRKR1"),
    selected = "AGE",
    multiple = TRUE,
    fixed = FALSE
  )
)

adtte_extract <- data_extract_spec(
  dataname = "ADTTE",
  select = select_spec(
    choices = c("AVAL", "AVALC", "ASEQ"),
    selected = "AVAL",
    multiple = TRUE,
    fixed = FALSE
  ),
  filter = filter_spec(
    vars = "PARAMCD",
    choices = c("CRSD", "EFS", "OS", "PFS"),
    selected = "OS"
  )
)

data_extracts <- list(adsl_extract = adsl_extract, adtte_extract = adtte_extract)

Step 3/5 - Creating the UI

Here, we define the merge_ui function, which will be used to create the UI components for the shiny app.

Note that we take in the list of data_extract objects as input, and make use of the data_extract_ui function to create our UI.

merge_ui <- function(id, data_extracts) {
  ns <- NS(id)
  sidebarLayout(
    sidebarPanel(
      h3("Encoding"),
      tags$div(
        data_extract_ui(
          ns("adsl_extract"), # must correspond with data_extracts list names
          label = "ADSL extract",
          data_extracts[[1]]
        ),
        data_extract_ui(
          ns("adtte_extract"), # must correspond with data_extracts list names
          label = "ADTTE extract",
          data_extracts[[2]]
        )
      )
    ),
    mainPanel(
      h3("Output"),
      verbatimTextOutput(ns("expr")),
      dataTableOutput(ns("data"))
    )
  )
}

Step 4/5 - Creating the Server Logic

Here, we define the merge_srv function, which will be used to create the server logic for the shiny app.

This function takes as arguments the datasets (as a list of reactive data.frame), the data extract specifications created above (the data_extract list), and the join_keys object (read more about the join_keys in the Join Keys vignette of teal.data). We make use of the merge_expression_srv function to get a reactive list containing merge expression and information needed to perform the transformation - see more in merge_expression_srv documentation. We print this expression in the UI and also evaluate it to get the final ANL dataset which is also displayed as a table in the UI.

merge_srv <- function(id, datasets, data_extracts, join_keys) {
  moduleServer(id, function(input, output, session) {
    selector_list <- data_extract_multiple_srv(data_extracts, datasets, join_keys)
    merged_data <- merge_expression_srv(
      selector_list = selector_list,
      datasets = datasets,
      join_keys = join_keys,
      merge_function = "dplyr::left_join"
    )
    ANL <- reactive({
      data_list <- lapply(datasets, function(ds) ds())
      eval(envir = list2env(data_list), expr = as.expression(merged_data()$expr))
    })

    output$expr <- renderText(paste(merged_data()$expr, collapse = "\n"))
    output$data <- renderDataTable(ANL())
  })
}

Step 5/5 - Creating the shiny App

Finally, we include merge_ui and merge_srv in the UI and server components of the shinyApp, respectively, using the data_extracts defined in the first code block and the datasets object:

shinyApp(
  ui = fluidPage(merge_ui("data_merge", data_extracts)),
  server = function(input, output, session) {
    merge_srv("data_merge", datasets, data_extracts, join_keys)
  }
)