Pre-processing Delayed Data

Dawid Kałędkowski

15.05.2022

When creating apps which do not use DDL, once the datasets are created there is often some pre-processing required before initializing the teal app. Similarly, in the case of delayed data additional code instructions to pre-process data can be added to DDL objects which will be run after the data is loaded, which may happen after the launching of the shiny app or when the pull() method is called.

library(scda)
library(teal.data)
library(magrittr)

adsl_cf <- callable_function(function() synthetic_cdisc_data("latest")$adsl)
adsl <- cdisc_dataset_connector(
  dataname = "ADSL",
  pull_callable = adsl_cf,
  keys = get_cdisc_keys("ADSL")
) %>%
  mutate_dataset("ADSL$SEX <- as.factor(ADSL$SEX)")


adae_cf <- callable_function(function() synthetic_cdisc_data("latest")$adae)
adae <- cdisc_dataset_connector(
  dataname = "ADAE",
  pull_callable = adae_cf,
  keys = get_cdisc_keys("ADAE")
) %>%
  mutate_dataset("ADAE$X <- rep(ADSL$SEX[1])", vars = list(ADSL = adsl))

adsl$pull() %>%
  get_raw_data() %>%
  head(n = 3)
## # A tibble: 3 × 44
##   STUDYID USUBJID     SUBJID SITEID   AGE AGEU  SEX   RACE  ETHNIC COUNTRY DTHFL
##   <chr>   <chr>       <chr>  <chr>  <int> <fct> <fct> <fct> <fct>  <fct>   <fct>
## 1 AB12345 AB12345-CH… id-128 CHN-3     32 YEARS M     ASIAN NOT H… CHN     N    
## 2 AB12345 AB12345-CH… id-262 CHN-15    35 YEARS M     BLAC… NOT H… CHN     N    
## 3 AB12345 AB12345-RU… id-378 RUS-3     30 YEARS F     ASIAN NOT H… RUS     N    
## # … with 33 more variables: INVID <chr>, INVNAM <chr>, ARM <fct>, ARMCD <fct>,
## #   ACTARM <fct>, ACTARMCD <fct>, TRT01P <fct>, TRT01A <fct>, REGION1 <fct>,
## #   STRATA1 <fct>, STRATA2 <fct>, BMRKR1 <dbl>, BMRKR2 <fct>, ITTFL <fct>,
## #   SAFFL <fct>, BMEASIFL <fct>, BEP01FL <fct>, RANDDT <date>, TRTSDTM <dttm>,
## #   TRTEDTM <dttm>, EOSSTT <fct>, EOTSTT <fct>, EOSDT <date>, EOSDY <int>,
## #   DCSREAS <fct>, DTHDT <date>, DTHCAUS <fct>, DTHCAT <fct>, LDDTHELD <int>,
## #   LDDTHGR1 <fct>, LSTALVDT <date>, DTHADY <int>, study_duration_secs <dbl>
adae$pull() %>%
  get_raw_data() %>%
  head(n = 3)
## # A tibble: 3 × 75
##   STUDYID USUBJID     SUBJID SITEID   AGE AGEU  SEX   RACE  ETHNIC COUNTRY DTHFL
##   <chr>   <chr>       <chr>  <chr>  <int> <fct> <fct> <fct> <fct>  <fct>   <fct>
## 1 AB12345 AB12345-BR… id-134 BRA-1     47 YEARS M     WHITE NOT H… BRA     N    
## 2 AB12345 AB12345-BR… id-134 BRA-1     47 YEARS M     WHITE NOT H… BRA     N    
## 3 AB12345 AB12345-BR… id-134 BRA-1     47 YEARS M     WHITE NOT H… BRA     N    
## # … with 64 more variables: INVID <chr>, INVNAM <chr>, ARM <fct>, ARMCD <fct>,
## #   ACTARM <fct>, ACTARMCD <fct>, TRT01P <fct>, TRT01A <fct>, REGION1 <fct>,
## #   STRATA1 <fct>, STRATA2 <fct>, BMRKR1 <dbl>, BMRKR2 <fct>, ITTFL <fct>,
## #   SAFFL <fct>, BMEASIFL <fct>, BEP01FL <fct>, RANDDT <date>, TRTSDTM <dttm>,
## #   TRTEDTM <dttm>, EOSSTT <fct>, EOTSTT <fct>, EOSDT <date>, EOSDY <int>,
## #   DCSREAS <fct>, DTHDT <date>, DTHCAUS <fct>, DTHCAT <fct>, LDDTHELD <int>,
## #   LDDTHGR1 <fct>, LSTALVDT <date>, DTHADY <int>, study_duration_secs <dbl>, …
cdisc_data(adsl, adae, check = TRUE) %>%
  mutate_data("ADAE$x <- ADSL$SUBJID[1]")

The code is processed in the order the datasets are pulled so if there are dependencies between datasets it matters the order in which pre-processing code is added to the CDISCTealData object just as order matters when the arguments are inputted to the cdisc_data function to create the CDISCTealData object.

Finally, the code argument directly in teal_data and cdisc_data call does not need to be used for DDL because data loaded with DDL are reproducible by design. Because of this, it is recommended to set argument check = TRUE inside cdisc_data function when creating apps with DDL.

Processing dependencies

It may be required to generate a delayed data object that is dependent on some other delayed object or some constant value.

For this, when creating your delayed data object it’s possible to supply the additional variables that are to be accessed during the data loading (pull) using additional arguments through ...:

get_code(adsl)
## [1] "ADSL <- (function() synthetic_cdisc_data(\"latest\")$adsl)()\nADSL$SEX <- as.factor(ADSL$SEX)"
pull_fun_adae <- callable_function(
  function() {
    synthetic_cdisc_data("latest")$adae
  }
)
adae <- dataset_connector(
  dataname = "ADAE",
  pull_callable = pull_fun_adae,
  keys = get_cdisc_keys("ADAE")
)

get_code(adae)
## [1] "ADAE <- (function() {\n    synthetic_cdisc_data(\"latest\")$adae\n})()"

It’s also possible to supply these additional variables after creating your object using the mutate_dataset function.

last_run <- Sys.Date() # constant value stored as a variable in the current session

adsl_cf <- callable_function(function() synthetic_cdisc_data("latest")$adsl)
adsl <- cdisc_dataset_connector(
  dataname = "ADSL",
  pull_callable = adsl_cf,
  keys = get_cdisc_keys("ADSL")
) %>%
  mutate_dataset("ADSL$last_run <- last_run", vars = list(last_run = last_run))

cat(get_code(adsl))
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## last_run <- structure(19188, class = "Date")
## ADSL$last_run <- last_run
# compared to evaluating the variable at the time of loading
adsl_cf <- callable_function(function() synthetic_cdisc_data("latest")$adsl)
adsl <- cdisc_dataset_connector(
  dataname = "ADSL",
  pull_callable = adsl_cf,
  keys = get_cdisc_keys("ADSL")
) %>%
  mutate_dataset("last_run <- Sys.Date()\nADSL$last_run <- last_run")

adsl %>%
  get_code() %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## last_run <- Sys.Date()
## ADSL$last_run <- last_run

This is also required when creating the object depends on another delayed data object:

adsl <- synthetic_cdisc_data("latest")$adsl
adae_cf <- callable_function(function() synthetic_cdisc_data("latest")$adae)
adae <- cdisc_dataset_connector(
  dataname = "ADAE",
  pull_callable = adae_cf,
  keys = get_cdisc_keys("ADAE")
) %>%
  mutate_dataset("ADAE$n <- nrow(ADSL)")

cat(get_code(adae)) # the code returned by `adae` is not sufficient to reproduce `adae`
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
adsl_cf <- callable_function(function() synthetic_cdisc_data("latest")$adsl)
adsl <- cdisc_dataset_connector(
  dataname = "ADSL",
  pull_callable = adsl_cf,
  keys = get_cdisc_keys("ADSL")
)
adae_cf <- callable_function(function() synthetic_cdisc_data("latest")$adae)
adae <- cdisc_dataset_connector(
  dataname = "ADAE",
  pull_callable = adae_cf,
  keys = get_cdisc_keys("ADAE")
) %>%
  mutate_dataset("ADAE$n <- nrow(ADSL)", vars = list(ADSL = adsl))

cat(get_code(adae)) # this code can be run independently
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE$n <- nrow(ADSL)

Related to this idea, it is possible to provide the code on a Data level. However, this will always return all the code used to generate all the datasets in the object:

adsl_adae <- cdisc_data(
  adsl,
  adae
) %>% mutate_data("ADAE$avg_age <- mean(ADAE$AGE)")

# the output for all 3 are the same
adsl_adae %>%
  get_code() %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
## ADAE$avg_age <- mean(ADAE$AGE)
adsl_adae %>%
  get_code(dataname = "ADAE") %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
## ADAE$avg_age <- mean(ADAE$AGE)
adsl_adae %>%
  get_code(dataname = "ADSL") %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
## ADAE$avg_age <- mean(ADAE$AGE)

The better approach would be to supply the code on a Dataset level. This ensures that the code accessed on a dataset level only contains the snippets that pertains to itself:

adsl_adae <- cdisc_data(
  adsl,
  adae %>% mutate_dataset("ADAE$avg_age <- mean(ADAE$AGE)")
)

adsl_adae %>%
  get_code() %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
## ADAE$avg_age <- mean(ADAE$AGE)
adsl_adae %>%
  get_code("ADAE") %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()
## ADAE <- (function() synthetic_cdisc_data("latest")$adae)()
## ADAE$n <- nrow(ADSL)
## ADAE$avg_age <- mean(ADAE$AGE)
adsl_adae %>%
  get_code("ADSL") %>%
  cat()
## ADSL <- (function() synthetic_cdisc_data("latest")$adsl)()

Related to this idea, the delayed data object needs to be supplied with the code needed to reproduce the data. This can be provided at the Dataset level or the Data level.

Below is a comparison of these two approaches:

adsl <- synthetic_cdisc_data("latest")$adsl
cdisc_dataset("ADSL", adsl) %>% get_code() # no reproducible code
## [1] ""
# provide the code to reproduce the data:
cdisc_dataset("ADSL", adsl,
  code = "ADSL <- synthetic_cdisc_data(\"latest\")$adsl"
) %>%
  get_code()
## [1] "ADSL <- synthetic_cdisc_data(\"latest\")$adsl"
# it's possible to supply the code at the `Data` level:
adae <- synthetic_cdisc_data("latest")$adae
adsl_adae <- cdisc_data(
  cdisc_dataset("ADSL", adsl),
  cdisc_dataset("ADAE", adae),
  code = "ADSL <- synthetic_cdisc_data(\"latest\")$adsl\nADAE <- synthetic_cdisc_data(\"latest\")$adae"
)

adsl_adae %>%
  get_code() %>%
  cat()
## ADSL <- synthetic_cdisc_data("latest")$adsl
## ADAE <- synthetic_cdisc_data("latest")$adae
# but it's not possible then to access the code at a `Dataset` level:
adsl_adae %>%
  get_code("ADSL") %>%
  cat()
## ADSL <- synthetic_cdisc_data("latest")$adsl
## ADAE <- synthetic_cdisc_data("latest")$adae
# this can be avoided by storing the code like so:
adsl_adae <- cdisc_data(
  cdisc_dataset("ADSL", adsl, code = "ADSL <- synthetic_cdisc_data(\"latest\")$adsl"),
  cdisc_dataset("ADAE", adae, code = "ADAE <- synthetic_cdisc_data(\"latest\")$adsl")
)

adsl_adae %>%
  get_code("ADSL") %>%
  cat()
## ADSL <- synthetic_cdisc_data("latest")$adsl