Introduction to Variable Mapping

library(metalite)

Overview

We design an abstraction of variable mapping to enhance the robustness of the metadata information. The goal is to follow the dependency inversion principle in software design.

The design is inspired by ggplot2::aes().

For a typical analysis, we require developer to provide variable names for:

  • name: a reference name as a key to link other components.
  • Subject identifier (id): typically using USUBJID in ADaM data.
  • Treatment group (group): typically using TRTXX in ADaM data.
  • Other variables (var): additional variables required for analysis.
  • subset: a subset expression to define the analysis. Typically using analysis flag XXFL in ADaM data.
  • label: natural language to describe the purpose of this adam_mapping object.

Example 1: All Participants as Treated (apat)

In this example, we define an adam_mapping for APaT population. The example illustrate an use case to decouple variable name in ADSL based on data_mapping abstract layer.

x <- adam_mapping(
  name = "apat",
  id = "USUBJID",
  group = "TRT01A",
  subset = TRTFL == "Y",
  label = "All Participants as Treated"
)

x
#> ADaM mapping: 
#> * `name`   -> "apat"
#> * `id`     -> "USUBJID"
#> * `group`  -> "TRT01A"
#> * `var`    -> NULL
#> * `subset` -> TRTFL == "Y"
#> * `label`  -> "All Participants as Treated"

With the defined variable mapping, our development of other standard function can rely on the abstraction. As long as a study team provide proper data and the associate adam_mapping, our standard function can be used.

In R, the abstraction is a named list that can be accessed by .$subset and assigned by <-.

x$subset
#> TRTFL == "Y"
x$var <- "AGE"
x$subset <- quote(SAFFL == "Y") # using quote for an expression
x
#> ADaM mapping: 
#> * `name`   -> "apat"
#> * `id`     -> "USUBJID"
#> * `group`  -> "TRT01A"
#> * `var`    -> "AGE"
#> * `subset` -> SAFFL == "Y"
#> * `label`  -> "All Participants as Treated"

If our goal is to summarize var by group within the population defined in subset. We can write R scripts in this abstract layer using base R or tidy evaluation as below.

Base R

df <- r2rtf::r2rtf_adsl
ana <- df[eval(x$subset, df), ]

split(ana, ana[[x$group]]) |>
  sapply(function(y) mean(y[[x$var]]))
#>              Placebo Xanomeline High Dose  Xanomeline Low Dose 
#>             75.20930             74.38095             75.66667

Reference: eval and expression.

Tidy evaluation

library(dplyr)
df |>
  dplyr::filter(!!x$subset) |>
  dplyr::group_by(.data[[x$group]]) |>
  dplyr::summarise(mean = mean(.data[[x$var]]))

By using the adam_mapping abstract layer, it creates additional challenges to develop R program, yet we can reduce maintenance in the future.

Example 2: Serious adverse events (ser)

In this example, we define an adam_mapping for serious adverse events (AE). The example illustrate an use case to inherit default values defined in metalite.

An organization have conventions to define different analysis terms. We can define default values of those commonly used analysis terms.

For example, we define a default adam_mapping object as below. A real example can be found in metalite:::default_parameter_ae.

ser_default <- adam_mapping(
  name = "ser",
  label = "serious adverse events",
  subset = quote(AESER == "Y")
)

ser_default
#> ADaM mapping: 
#> * `name`   -> "ser"
#> * `id`     -> NULL
#> * `group`  -> NULL
#> * `var`    -> NULL
#> * `subset` -> AESER == "Y"
#> * `label`  -> "serious adverse events"

With default values, user can reduce input but still allow override default values if required.

Assuming user define an adam_mapping for a study, because the study require a footnote to explain the meaning of serious adverse events.

ser_user <- adam_mapping(
  name = "ser",
  id = "USUBJID",
  group = "TRT01A",
  label = "serious{^a} adverse events",
  footnote = "{^a} this is a footnote"
)
ser_user
#> ADaM mapping: 
#> * `name`     -> "ser"
#> * `id`       -> "USUBJID"
#> * `group`    -> "TRT01A"
#> * `var`      -> NULL
#> * `subset`   -> NULL
#> * `label`    -> "serious{^a} adverse events"
#> * `footnote` -> "{^a} this is a footnote"

After we merge the user defined and default adam_mapping objects. (always be left join) We can

  • keep all user defined variables.
  • add default subset variable, because it is not defined by user.
merge(ser_user, ser_default)
#> ADaM mapping: 
#> * `name`     -> "ser"
#> * `id`       -> "USUBJID"
#> * `group`    -> "TRT01A"
#> * `var`      -> NULL
#> * `subset`   -> AESER == "Y"
#> * `label`    -> "serious{^a} adverse events"
#> * `footnote` -> "{^a} this is a footnote"

Note: the adam_mapping object also allow user to define other variables. In this example, we added footnote variable.