Skip to contents

Creating a BDS Finding ADaM

Source: vignettes/bds_finding.Rmd
bds_finding.Rmd

Introduction

This article describes creating a BDS finding ADaM. Examples are currently presented and tested in the context of ADVS. However, the examples could be applied to other BDS Finding ADaMs such as ADEG, ADLB, etc. where a single result is captured in an SDTM Finding domain on a single date and/or time.

Note: All examples assume CDISC SDTM and/or ADaM format as input unless otherwise specified.

Programming Workflow

Read in Data

To start, all data frames needed for the creation of ADVS should be read into the environment. This will be a company specific process. Some of the data frames needed may be VS and ADSL.

For example purpose, the CDISC Pilot SDTM and ADaM datasets—which are included in {admiral.test}—are used.

library(admiral)
library(dplyr)
library(admiral.test)
library(lubridate)
library(stringr)
library(tibble)

data("admiral_adsl")
data("admiral_vs")

adsl <- admiral_adsl
vs <- admiral_vs

vs <- convert_blanks_to_na(vs)

At this step, it may be useful to join ADSL to your VS domain. Only the ADSL variables used for derivations are selected at this step. The rest of the relevant ADSL variables would be added later.


adsl_vars <- vars(TRTSDT, TRTEDT, TRT01A, TRT01P)

advs <- derive_vars_merged(
  vs,
  dataset_add = adsl,
  new_vars = adsl_vars,
  by_vars = vars(STUDYID, USUBJID)
)

Derive/Impute Numeric Date/Time and Analysis Day (ADT, ADTM, ADY, ADTF, ATMF)

The function derive_vars_dt() can be used to derive ADT. This function allows the user to impute the date as well.

Example calls:

advs <- derive_vars_dt(advs, new_vars_prefix = "A", dtc = VSDTC)

If imputation is needed and the date is to be imputed to the first of the month, the call would be:

advs <- derive_vars_dt(
  advs,
  new_vars_prefix = "A",
  dtc = VSDTC,
  highest_imputation = "M"
)

Similarly, ADTM may be created using the function derive_vars_dtm(). Imputation may be done on both the date and time components of ADTM.

# CDISC Pilot data does not contain times and the output of the derivation
# ADTM is not presented.
advs <- derive_vars_dtm(
  advs,
  new_vars_prefix = "A",
  dtc = VSDTC,
  highest_imputation = "M"
)

By default, the variable ADTF for derive_vars_dt() or ADTF and ATMF for derive_vars_dtm() will be created and populated with the controlled terminology outlined in the ADaM IG for date imputations.

See also Date and Time Imputation.

Once ADT is derived, the function derive_vars_dy() can be used to derive ADY. This example assumes both ADT and TRTSDT exist on the data frame.

advs <-
  derive_vars_dy(advs, reference_date = TRTSDT, source_vars = vars(ADT))

Assign PARAMCD, PARAM, PARAMN, PARCAT1

To assign parameter level values such as PARAMCD, PARAM, PARAMN, PARCAT1, etc., a lookup can be created to join to the source data.

For example, when creating ADVS, a lookup based on the SDTM --TESTCD value may be created:

VSTESTCD PARAMCD PARAM PARAMN PARCAT1 PARCAT1N
HEIGHT HEIGHT Height (cm) 1 Subject Characteristic 1
WEIGHT WEIGHT Weight (kg) 2 Subject Characteristic 1
DIABP DIABP Diastolic Blood Pressure (mmHg) 3 Vital Sign 2
MAP MAP Mean Arterial Pressure 4 Vital Sign 2
PULSE PULSE Pulse Rate (beats/min) 5 Vital Sign 2
SYSBP SYSBP Systolic Blood Pressure (mmHg) 6 Vital Sign 2
TEMP TEMP Temperature (C) 7 Vital Sign 2

This lookup may now be joined to the source data:

At this stage, only PARAMCD is required to perform the derivations. Additional derived parameters may be added, so only PARAMCD is joined to the datasets at this point. All other variables related to PARAMCD (e.g. PARAM, PARAMCAT1, …) will be added when all PARAMCD are derived.

advs <- derive_vars_merged(
  advs,
  dataset_add = param_lookup,
  new_vars = vars(PARAMCD),
  by_vars = vars(VSTESTCD)
)

Please note, it may be necessary to include other variables in the join. For example, perhaps the PARAMCD is based on VSTESTCD and VSPOS, it may be necessary to expand this lookup or create a separate look up for PARAMCD.

Derive Results (AVAL, AVALC)

The mapping of AVAL and AVALC is left to the ADaM programmer. An example mapping may be:

advs <- mutate(
  advs,
  AVAL = VSSTRESN,
  AVALC = VSSTRESC
)

Derive Additional Parameters (e.g. BSA, BMI or MAP for ADVS)

Optionally derive new parameters creating PARAMCD and AVAL. Note that only variables specified in the by_vars argument will be populated in the newly created records. This is relevant to the functions derive_param_map, derive_param_bsa, derive_param_bmi, and derive_param_qtc.

Below is an example of creating Mean Arterial Pressure for ADVS, see also Example 3 in section below Derive New Rows for alternative way of creating new parameters.

advs <- derive_param_map(
  advs,
  by_vars = vars(STUDYID, USUBJID, !!!adsl_vars, VISIT, VISITNUM, ADT, ADY, VSTPT, VSTPTNUM),
  set_values_to = vars(PARAMCD = "MAP"),
  get_unit_expr = VSSTRESU,
  filter = VSSTAT != "NOT DONE" | is.na(VSSTAT)
)

Likewise, function call below, to create parameter Body Surface Area and Body Mass Index for ADVS domain.

advs <- derive_param_bsa(
  advs,
  by_vars = vars(STUDYID, USUBJID, !!!adsl_vars, VISIT, VISITNUM, ADT, ADY, VSTPT, VSTPTNUM),
  method = "Mosteller",
  set_values_to = vars(PARAMCD = "BSA"),
  get_unit_expr = VSSTRESU,
  filter = VSSTAT != "NOT DONE" | is.na(VSSTAT)
)

advs <- derive_param_bmi(
  advs,
  by_vars = vars(STUDYID, USUBJID, !!!adsl_vars, VISIT, VISITNUM, ADT, ADY, VSTPT, VSTPTNUM),
  set_values_to = vars(PARAMCD = "BMI"),
  get_unit_expr = VSSTRESU,
  filter = VSSTAT != "NOT DONE" | is.na(VSSTAT)
)

Similarly, for ADEG, the parameters QTCBF QTCBS and QTCL can be created with a function call. See example below for PARAMCD = QTCF.


adeg <- tibble::tribble(
  ~USUBJID, ~EGSTRESU, ~PARAMCD, ~AVAL, ~VISIT,
  "P01", "msec", "QT", 350, "CYCLE 1 DAY 1",
  "P01", "msec", "QT", 370, "CYCLE 2 DAY 1",
  "P01", "msec", "RR", 842, "CYCLE 1 DAY 1",
  "P01", "msec", "RR", 710, "CYCLE 2 DAY 1"
)

adeg <- derive_param_qtc(
  adeg,
  by_vars = vars(USUBJID, VISIT),
  method = "Fridericia",
  set_values_to = vars(PARAMCD = "QTCFR"),
  get_unit_expr = EGSTRESU
)

Similarly, for ADLB, the function derive_param_wbc_abs() can be used to create new parameter for lab differentials converted to absolute values. See example below:

adlb <- tibble::tribble(
  ~USUBJID, ~PARAMCD, ~AVAL, ~PARAM, ~VISIT,
  "P01", "WBC", 33, "Leukocyte Count (10^9/L)", "CYCLE 1 DAY 1",
  "P01", "WBC", 38, "Leukocyte Count (10^9/L)", "CYCLE 2 DAY 1",
  "P01", "LYMLE", 0.90, "Lymphocytes (fraction of 1)", "CYCLE 1 DAY 1",
  "P01", "LYMLE", 0.70, "Lymphocytes (fraction of 1)", "CYCLE 2 DAY 1"
)

derive_param_wbc_abs(
  dataset = adlb,
  by_vars = vars(USUBJID, VISIT),
  set_values_to = vars(
    PARAMCD = "LYMPH",
    PARAM = "Lymphocytes Abs (10^9/L)",
    DTYPE = "CALCULATION"
  ),
  get_unit_expr = extract_unit(PARAM),
  wbc_code = "WBC",
  diff_code = "LYMLE",
  diff_type = "fraction"
)

When all PARAMCD have been derived and added to the dataset, the other information from the look-up table (PARAM, PARAMCAT1,…) should be added.


# Derive PARAM and PARAMN
advs <- derive_vars_merged(
  advs,
  dataset_add = select(param_lookup, -VSTESTCD),
  by_vars = vars(PARAMCD)
)

Derive Timing Variables (e.g. APHASE, AVISIT, APERIOD)

Categorical timing variables are protocol and analysis dependent. Below is a simple example.

advs <- advs %>%
  mutate(
    AVISIT = case_when(
      str_detect(VISIT, "SCREEN") ~ NA_character_,
      str_detect(VISIT, "UNSCHED") ~ NA_character_,
      str_detect(VISIT, "RETRIEVAL") ~ NA_character_,
      str_detect(VISIT, "AMBUL") ~ NA_character_,
      !is.na(VISIT) ~ str_to_title(VISIT)
    ),
    AVISITN = as.numeric(case_when(
      VISIT == "BASELINE" ~ "0",
      str_detect(VISIT, "WEEK") ~ str_trim(str_replace(VISIT, "WEEK", ""))
    )),
    ATPT = VSTPT,
    ATPTN = VSTPTNUM
  )


count(advs, VISITNUM, VISIT, AVISITN, AVISIT)
#> # A tibble: 15 x 5
#>    VISITNUM VISIT               AVISITN AVISIT       n
#>       <dbl> <chr>                 <dbl> <chr>    <int>
#>  1      1   SCREENING 1              NA <NA>       102
#>  2      2   SCREENING 2              NA <NA>        78
#>  3      3   BASELINE                  0 Baseline    84
#>  4      3.5 AMBUL ECG PLACEMENT      NA <NA>        65
#>  5      4   WEEK 2                    2 Week 2      84
#>  6      5   WEEK 4                    4 Week 4      70
#>  7      6   AMBUL ECG REMOVAL        NA <NA>        52
#>  8      7   WEEK 6                    6 Week 6      42
#>  9      8   WEEK 8                    8 Week 8      42
#> 10      9   WEEK 12                  12 Week 12     42
#> 11     10   WEEK 16                  16 Week 16     42
#> 12     11   WEEK 20                  20 Week 20     28
#> 13     12   WEEK 24                  24 Week 24     28
#> 14     13   WEEK 26                  26 Week 26     28
#> 15    201   RETRIEVAL                NA <NA>        26

count(advs, VSTPTNUM, VSTPT, ATPTN, ATPT)
#> # A tibble: 4 x 5
#>   VSTPTNUM VSTPT                        ATPTN ATPT                             n
#>      <dbl> <chr>                        <dbl> <chr>                        <int>
#> 1      815 AFTER LYING DOWN FOR 5 MINU…   815 AFTER LYING DOWN FOR 5 MINU…   232
#> 2      816 AFTER STANDING FOR 1 MINUTE    816 AFTER STANDING FOR 1 MINUTE    232
#> 3      817 AFTER STANDING FOR 3 MINUTES   817 AFTER STANDING FOR 3 MINUTES   232
#> 4       NA <NA>                            NA <NA>                           117

Timing Flag Variables (e.g. ONTRTFL)

In some analyses, it may be necessary to flag an observation as on-treatment. The admiral function derive_var_ontrtfl() can be used.

For example, if on-treatment is defined as any observation between treatment start and treatment end, the flag may be derived as:

advs <- derive_var_ontrtfl(
  advs,
  start_date = ADT,
  ref_start_date = TRTSDT,
  ref_end_date = TRTEDT
)

This function returns the original data frame with the column ONTRTFL added. Additionally, this function does have functionality to handle a window on the ref_end_date. For example, if on-treatment is defined as between treatment start and treatment end plus 60 days, the call would be:

advs <- derive_var_ontrtfl(
  advs,
  start_date = ADT,
  ref_start_date = TRTSDT,
  ref_end_date = TRTEDT,
  ref_end_window = 60
)

In addition, the function does allow you to filter out pre-treatment observations that occurred on the start date. For example, if observations with VSTPT == PRE should not be considered on-treatment when the observation date falls between the treatment start and end date, the user may specify this using the filter_pre_timepoint parameter:

advs <- derive_var_ontrtfl(
  advs,
  start_date = ADT,
  ref_start_date = TRTSDT,
  ref_end_date = TRTEDT,
  filter_pre_timepoint = ATPT == "AFTER LYING DOWN FOR 5 MINUTES"
)

Lastly, the function does allow you to create any on-treatment flag based on the analysis needs. For example, if variable ONTR01FL is needed, showing the on-treatment flag during Period 01, you need to set new var = ONTR01FL. In addition, for Period 01 Start Date and Period 01 End Date, you need ref_start_date = AP01SDT and ref_end_date = AP01EDT.

advs <- derive_var_ontrtfl(
  advs,
  new_var = ONTR01FL,
  start_date = ASTDT,
  end_date = AENDT,
  ref_start_date = AP01SDT,
  ref_end_date = AP01EDT,
  span_period = "Y"
)

Assign Reference Range Indicator (ANRIND)

The admiral function derive_var_anrind() may be used to derive the reference range indicator ANRIND.

This function requires the reference range boundaries to exist on the data frame (ANRLO, ANRHI) and also accommodates the additional boundaries A1LO and A1HI.

The function is called as:

advs <- derive_var_anrind(advs)

Derive Baseline (BASETYPE, ABLFL, BASE, BASEC, BNRIND)

The BASETYPE should be derived using the function derive_var_basetype(). The parameter basetypes of this function requires a named list of expression detailing how the BASETYPE should be assigned. Note, if a record falls into multiple expressions within the basetypes expression, a row will be produced for each BASETYPE.

advs <- derive_var_basetype(
  dataset = advs,
  basetypes = rlang::exprs(
    "LAST: AFTER LYING DOWN FOR 5 MINUTES" = ATPTN == 815,
    "LAST: AFTER STANDING FOR 1 MINUTE" = ATPTN == 816,
    "LAST: AFTER STANDING FOR 3 MINUTES" = ATPTN == 817,
    "LAST" = is.na(ATPTN)
  )
)

count(advs, ATPT, ATPTN, BASETYPE)
#> # A tibble: 4 x 4
#>   ATPT                           ATPTN BASETYPE                                n
#>   <chr>                          <dbl> <chr>                               <int>
#> 1 AFTER LYING DOWN FOR 5 MINUTES   815 LAST: AFTER LYING DOWN FOR 5 MINUT…   232
#> 2 AFTER STANDING FOR 1 MINUTE      816 LAST: AFTER STANDING FOR 1 MINUTE     232
#> 3 AFTER STANDING FOR 3 MINUTES     817 LAST: AFTER STANDING FOR 3 MINUTES    232
#> 4 <NA>                              NA LAST                                  117

It is important to derive BASETYPE first so that it can be utilized in subsequent derivations. This will be important if the data frame contains multiple values for BASETYPE.

Next, the analysis baseline flag ABLFL can be derived using the admiral function derive_var_extreme_flag(). For example, if baseline is defined as the last non-missing AVAL prior or on TRTSDT, the function call for ABLFL would be:

advs <- restrict_derivation(
  advs,
  derivation = derive_var_extreme_flag,
  args = params(
    by_vars = vars(STUDYID, USUBJID, BASETYPE, PARAMCD),
    order = vars(ADT, ATPTN, VISITNUM),
    new_var = ABLFL,
    mode = "last"
  ),
  filter = (!is.na(AVAL) & ADT <= TRTSDT & !is.na(BASETYPE))
)

Note: Additional examples of the derive_var_extreme_flag() function can be found above.

Lastly, the BASE, BASEC and BNRIND columns can be derived using the admiral function derive_var_base(). Example calls are:

advs <- derive_var_base(
  advs,
  by_vars = vars(STUDYID, USUBJID, PARAMCD, BASETYPE),
  source_var = AVAL,
  new_var = BASE
)

advs <- derive_var_base(
  advs,
  by_vars = vars(STUDYID, USUBJID, PARAMCD, BASETYPE),
  source_var = AVALC,
  new_var = BASEC
)

advs <- derive_var_base(
  advs,
  by_vars = vars(STUDYID, USUBJID, PARAMCD, BASETYPE),
  source_var = ANRIND,
  new_var = BNRIND
)

Derive Change from Baseline (CHG, PCHG)

Change and percent change from baseline can be derived using the admiral functions derive_var_chg() and derive_var_pchg(). These functions expect AVAL and BASE to exist in the data frame. The CHG is simply AVAL - BASE and the PCHG is (AVAL - BASE) / absolute value (BASE) * 100. Examples calls are:

advs <- derive_var_chg(advs)

advs <- derive_var_pchg(advs)

If the variables should not be derived for all records, e.g., for post-baseline records only, restrict_derivation() can be used.

Derive Shift (e.g. SHIFT1)

Shift variables can be derived using the admiral function derive_var_shift(). This function derives a character shift variable concatenating shift in values based on a user-defined pairing, e.g., shift from baseline reference range BNRIND to analysis reference range ANRIND. Examples calls are:

advs <- derive_var_shift(advs,
  new_var = SHIFT1,
  from_var = BNRIND,
  to_var = ANRIND
)

If the variables should not be derived for all records, e.g., for post-baseline records only, restrict_derivation() can be used.

Derive Analysis Ratio (R2BASE)

Analysis ratio variables can be derived using the admiral function derive_var_analysis_ratio(). This function derives a ratio variable based on user-specified pair. For example, Ratio to Baseline is calculated by AVAL / BASE and the function appends a new variable R2BASE to the dataset. Examples calls are:

advs <- derive_var_analysis_ratio(advs,
  numer_var = AVAL,
  denom_var = BASE
)

advs <- derive_var_analysis_ratio(advs,
  numer_var = AVAL,
  denom_var = ANRLO,
  new_var = R01ANRLO
)

If the variables should not be derived for all records, e.g., for post-baseline records only, restrict_derivation() can be used.

Derive Analysis Flags (e.g. ANL01FL)

In most finding ADaMs, an analysis flag is derived to identify the appropriate observation(s) to use for a particular analysis when a subject has multiple observations within a particular timing period.

In this situation, an analysis flag (e.g. ANLxxFL) may be used to choose the appropriate record for analysis.

This flag may be derived using the admiral function derive_var_extreme_flag(). For this example, we will assume we would like to choose the latest and highest value by USUBJID, PARAMCD, AVISIT, and ATPT.


advs <- restrict_derivation(
  advs,
  derivation = derive_var_extreme_flag,
  args = params(
    by_vars = vars(STUDYID, USUBJID, BASETYPE, PARAMCD, AVISIT),
    order = vars(ADT, ATPTN, AVAL),
    new_var = ANL01FL,
    mode = "last"
  ),
  filter = !is.na(AVISITN)
)

Another common example would be flagging the worst value for a subject, parameter, and visit. For this example, we will assume we have 3 PARAMCD values (SYSBP, DIABP, and RESP). We will also assume high is worst for SYSBP and DIABP and low is worst for RESP.


advs <- restrict_derivation(
  advs,
  derivation = derive_var_worst_flag,
  args = params(
    by_vars = vars(STUDYID, USUBJID, BASETYPE, PARAMCD, AVISIT),
    order = vars(ADT, ATPTN),
    new_var = WORSTFL,
    param_var = PARAMCD,
    analysis_var = AVAL,
    worst_high = c("SYSBP", "DIABP"),
    worst_low = "PULSE"
  ),
  filter = !is.na(AVISIT) & !is.na(AVAL)
)

Assign Treatment (TRTA, TRTP)

TRTA and TRTP must correlate to treatment TRTxxP and/or TRTxxA in ADSL. The derivation of TRTA and TRTP for a record are protocol and analysis specific.
admiral does not currently have functionality to assist with TRTA and TRTP assignment.

However, an example of a simple implementation could be:

advs <- mutate(advs, TRTP = TRT01P, TRTA = TRT01A)

count(advs, TRTP, TRTA, TRT01P, TRT01A)
#> # A tibble: 2 x 5
#>   TRTP               TRTA              TRT01P            TRT01A                n
#>   <chr>              <chr>             <chr>             <chr>             <int>
#> 1 Placebo            Placebo           Placebo           Placebo             588
#> 2 Xanomeline Low Do… Xanomeline Low D… Xanomeline Low D… Xanomeline Low D…   225

Assign ASEQ

The admiral function derive_var_obs_number() can be used to derive ASEQ. An example call is:

advs <- derive_var_obs_number(
  advs,
  new_var = ASEQ,
  by_vars = vars(STUDYID, USUBJID),
  order = vars(PARAMCD, ADT, AVISITN, VISITNUM, ATPTN),
  check_type = "error"
)