Required Packages

The examples of this vignette require the following packages.

library(admiral)
library(dplyr)
library(pharmaversesdtm)
library(pharmaverseadam)
library(lubridate)
library(stringr)
library(admiralonco)

Read in Data

To start, all data frames needed for the creation of ADTR should be read into the environment. This will be a company specific process. Some of the data frames needed may be ADSL, ADRS ,RS, TU, TR, SUPPTU SUPPTR.

For example purposes, the SDTM and ADaM datasets (based on CDISC Pilot test data)—which are included in {pharmaversesdtm} and {pharmaverseadam}—are used. Also, see Handling of Missing Values explains why we need to use the convert_blanks_to_na() function.

On the TR domain we filter on where tumor assessment short name TRTESTCD is "LDIAM" or "LPERP". Depending on your data collection standards you may need to update the code to filter on TRTESTCD == "DIAMETER". If this is the case then the template code where we derive parameters for lesions diameters should be updated accordingly.

data("adsl")
data("adrs_onco")
data("rs_onco_recist")
data("tu_onco_recist")
data("tr_onco_recist")
adrs <- adrs_onco
tu <- tu_onco_recist
tr <- tr_onco_recist
rs <- rs_onco_recist

tu <- convert_blanks_to_na(tu) %>%
  filter(TUEVAL == "INVESTIGATOR")
tr <- convert_blanks_to_na(tr) %>%
  filter(
    TREVAL == "INVESTIGATOR" & TRGRPID == "TARGET" & TRTESTCD %in% c("LDIAM", "LPERP")
  )
rs <- convert_blanks_to_na(rs)

At this step, it may be useful to join ADSL to your TR domain. For efficiency, only the ADSL variables (in this example, Randomization date (RANDDT) used for derivations are selected at this step. The rest of the relevant ADSL variables would be added later.

adsl_vars <- exprs(RANDDT)
tr <- derive_vars_merged(
  tr,
  dataset_add = adsl,
  new_vars = adsl_vars,
  by_vars = exprs(STUDYID, USUBJID)
)

Merge TR with TU and Derive New Variables

Depending on your company specifications you may want to merge TU with TR. At this point if you require any new variables these could be added in this step. As an example, here we are keeping the tumor location (TULOC), and deriving a new variable for the tumor group site (TULOCGR1).

tr <- derive_vars_merged(
  tr,
  dataset_add = tu,
  new_vars = exprs(TULOC),
  by_vars = exprs(STUDYID, USUBJID, TRLNKID = TULNKID)
) %>% mutate(
  TULOCGR1 = if_else(
    TULOC == "LYMPH NODE",
    "NODAL",
    "NON-NODAL"
  )
)

Furthermore, you could create additional new variables that are required downstream when deriving the required parameters. For example, here we include lesion ID expected (LSEXP) and lesion ID assessed (LSASS).

tr <- mutate(
  tr,
  LSEXP = TRLNKID,
  LSASS = if_else(!is.na(TRSTRESN), TRLNKID, NA_character_)
)

Pre-processing of Input Records

The next step involves company-specific pre-processing of records required downstream prior to the parameter derivations, to ensure that all the associated variable processing is done once before any new parameters are derived.

In order to calculate the sum of diameter of lesions, we need to ensure that lesions are grouped correctly, how this is done will be based on your company specifications but individual lesions sizes could be grouped either by using the analysis date ADT or visit AVISIT. This is to ensure that each lesion size is only counted once.

Below are some considerations depending on your analysis.

tr <- derive_vars_dt(
  tr,
  dtc = TRDTC,
  new_vars_prefix = "A",
  highest_imputation = "D",
  date_imputation = "first"
) %>%
  derive_vars_dy(
    reference_date = RANDDT,
    source_vars = exprs(ADT)
  )

tr <- mutate(
  tr,
  AVISIT = if_else(
    VISIT == "SCREENING",
    "BASELINE",
    VISIT
  ),
  AVISITN = if_else(
    AVISIT == "BASELINE",
    0,
    VISITNUM
  )
)

Derive Parameters for Lesion Diameters (LDIAMn & NLDIAMn)

Depending on your company specifications you may either create parameters for Target Lesion diameter for lymph nodes (LDIAMn) and non-lymph nodes (NLDIAMn) in ADTR or similar dataset.

The below examples show the set up of the parameter details and the analysis value & flag (AVAL/ANL01FL) variables.

tr <- mutate(tr, tmp_lesion_nr = str_sub(TRLNKID, 3))
adtr <- bind_rows(
  tr %>%
    filter(TRTESTCD == "LDIAM") %>%
    mutate(
      PARAMCD = paste0("LDIAM", tmp_lesion_nr),
      PARAM = paste("Target Lesion", tmp_lesion_nr, "Analysis Diameter")
    ),
  tr %>%
    filter(TRTESTCD == "LPERP") %>%
    mutate(
      PARAMCD = paste0("NLDIAM", tmp_lesion_nr),
      PARAM = paste("Target Lesion", tmp_lesion_nr, "Analysis Perpendicular")
    )
) %>%
  mutate(
    PARCAT1 = "Target Lesion(s)",
    PARCAT2 = "Investigator",
    PARCAT3 = "RECIST 1.1",
    AVAL = TRSTRESN,
    ANL01FL = if_else(!is.na(AVAL), "Y", NA_character_)
  ) %>%
  select(-tmp_lesion_nr)

Derive Parameter for Sum of Diameter and Analysis Flag ANL01FL

Sum of Target Lesions Diameters is the sum of the diameters of both lymph and non-lymph nodes (sum of analysis values AVAL from source observations LDIAMn & LNDIAMn).

In this vignette, the sum of the diameters is calculated across all the available lesion measurements, and we use the analysis flag ANL01FL to select the records that should be used as part of the analysis based on RECIST 1.1. Subsequently we then build on this analysis flag (see ANLzzFL) to demonstrate additional analysis flags for example purposes only. The analysis flags therefore should be setup according to your company specifications.

In the example below, the lesion sizes are counted and grouped according to the by variables specified, in this case we group by the visit (AVISIT and AVISITN) to ensure that the correct lesion measurements are summed together. If there are different analysis day/dates (ADY/ADT) associated with a particular visit, we take the minimum day/date.

adtr_sum <- derive_summary_records(
  dataset_add = adtr,
  by_vars = exprs(STUDYID, USUBJID, !!!adsl_vars, AVISIT, AVISITN),
  filter_add = (str_starts(PARAMCD, "LDIAM") & TULOCGR1 == "NON-NODAL") |
    (str_starts(PARAMCD, "NLDIAM") & TULOCGR1 == "NODAL"),
  set_values_to = exprs(
    AVAL = sum(AVAL, na.rm = TRUE),
    ADY = min(ADY, na.rm = TRUE),
    ADT = min(ADT, na.rm = TRUE),
    PARAMCD = "SDIAM",
    PARAM = "Target Lesions Sum of Diameters by Investigator",
    PARCAT1 = "Target Lesion(s)",
    PARCAT2 = "Investigator",
    PARCAT3 = "RECIST 1.1"
  )
)

The analysis flag ANL01FL flags the sum of diameters where the number of lesions assessed at baseline and at post-baseline match. To assess whether the number of lesions expected and assessed match, you could compare lesion identifiers (TR.TRLNKID) or use the previous variables set up earlier to compare lesion expected (LSEXP) vs lesion assessed (LSASS) or check for target lesion response (from RS) = NE. A target lesion response of NE means all lesions from baseline are not measured at that post-baseline visit.

Lesions can split or merge, if this is a consideration then this needs to be taken care of when deciding on which algorithm to use to check that all lesions are measured at post-baseline.

In the below example, we use derive_var_merged_summary() function multiple times in order to process additional variables for sum of diameter parameter, and calculate the analysis flag (ANL01FL).

In the first step we concatenate lesions IDs measured LSEXP at baseline, and merge in with the second step where we concatenate the lesions IDs assessed LSASS at that visit record. You can then compare LSEXP vs LSASS at each visit.

In the final step, the analysis flag is set ANL01FL = "Y" where lesions assessed at post-baseline match baseline.

adtr_sum <- adtr_sum %>%
  derive_var_merged_summary(
    dataset_add = adtr,
    by_vars = exprs(USUBJID),
    filter_add = AVISIT == "BASELINE" &
      ((str_starts(PARAMCD, "LDIAM") & TULOCGR1 == "NON-NODAL") |
        (str_starts(PARAMCD, "NLDIAM") & TULOCGR1 == "NODAL")),
    new_vars = exprs(LSEXP = paste(sort(TRLNKID), collapse = ", "))
  ) %>%
  derive_var_merged_summary(
    dataset_add = adtr,
    by_vars = exprs(USUBJID, AVISIT),
    filter_add = ((str_starts(PARAMCD, "LDIAM") & TULOCGR1 == "NON-NODAL") |
      (str_starts(PARAMCD, "NLDIAM") & TULOCGR1 == "NODAL")) & ANL01FL == "Y",
    new_vars = exprs(LSASS = paste(sort(TRLNKID), collapse = ", "))
  ) %>%
  mutate(
    ANL01FL = if_else(LSEXP == LSASS, "Y", NA_character_)
  )

Derive Baseline (ABLFL, BASE)

These functions are available in admiral, below examples show its usage for ADTR. Here we create baseline record flag ABLFL by selecting the last record prior to analysis day 1, and then derive the baseline value BASE. If in a particular study/or subject, baseline is defined differently, you need to make sure to update appropriate arguments.

adtr_sum <- adtr_sum %>%
  restrict_derivation(
    derivation = derive_var_extreme_flag,
    args = params(
      by_vars = exprs(USUBJID),
      order = exprs(ADY),
      new_var = ABLFL,
      mode = "last"
    ),
    filter = ADY <= 1
  ) %>%
  derive_var_base(
    by_vars = exprs(USUBJID)
  )

Derive NADIR

Once we have Sum of Target Lesions Diameters parameter (SDIAM), the NADIR is set to the lowest sum of diameters in Analysis Value (AVAL) before the current observation, where all lesions were assessed (i.e. ANL01F == "Y"). The first observations after baseline (depending on how you have defined baseline earlier) is set to the baseline value.

adtr_sum <- adtr_sum %>%
  derive_vars_joined(
    dataset_add = adtr_sum,
    by_vars = exprs(USUBJID),
    order = exprs(AVAL),
    new_vars = exprs(NADIR = AVAL),
    join_vars = exprs(ADY),
    join_type = "all",
    filter_add = ANL01FL == "Y",
    filter_join = ADY.join < ADY,
    mode = "first",
    check_type = "none"
  )

Derive Change from Baseline (CHG, PCHG)

These functions are available in admiral, below examples show its usage for ADTR. We calculate the change from baseline (CHG, CHGNAD) and percentage change from baseline (PCHG , PCHGNAD) for the parameter SDIAM (using AVAL - BASE) the variable NADIR (using AVAL - NADIR).

adtr_sum <- adtr_sum %>%
  derive_var_chg() %>%
  derive_var_pchg() %>%
  mutate(
    CHGNAD = AVAL - NADIR,
    PCHGNAD = if_else(NADIR == 0, NA_real_, 100 * CHGNAD / NADIR)
  )

Derive Additional Flag Variables

Depending on the analysis flags you require, you may want to create additional flag variables that will then aid setting up any additional analysis flag variables defined as per your company specifications.

adtr_sum <- adtr_sum %>%
  derive_var_merged_exist_flag(
    dataset_add = adrs,
    filter_add = PARAMCD == "PD",
    by_vars = exprs(USUBJID, ADT),
    new_var = PDFL,
    condition = AVALC == "Y"
  )

adtr_sum <- adtr_sum %>%
  derive_var_merged_exist_flag(
    dataset_add = derive_vars_dt(
      rs,
      dtc = RSDTC,
      new_vars_prefix = "A",
      highest_imputation = "D",
      flag_imputation = "none"
    ),
    filter_add = RSTESTCD == "OVRLRESP" & RSEVAL == "INVESTIGATOR",
    by_vars = exprs(USUBJID, ADT),
    new_var = PDFL,
    condition = RSSTRESC == "PD"
  )

adtr_sum <- adtr_sum %>% mutate(
  CRFL = if_else(AVAL == 0 & ANL01FL == "Y", "Y", NA_character_),
  CRNFL = if_else(NADIR == 0, "Y", NA_character_),
  PDFL = if_else(is.na(CRFL) & CRNFL == "Y", "Y", NA_character_)
)

adtr_sum <- adtr_sum %>% mutate(
  CRFL = if_else(AVAL == 0 & ANL01FL == "Y", "Y", NA_character_),
  CRNFL = if_else(NADIR == 0, "Y", NA_character_)
)

adtr_sum <- adtr_sum %>%
  mutate(
    PDFL = if_else(
      PCHGNAD >= 20 & CHGNAD >= 5 | is.na(CRFL) & CRNFL == "Y",
      "Y",
      NA_character_
    )
  )

Derive Analysis Flags ANLzzFL

Based on your specifications, additional analysis flags (ANLzzFL) could be set up in ADTR. Below are some examples which use the ANL01FL flag derived earlier for the sum of diameter parameter (SDIAM) as a basis.

Analysis 02 Flag (ANL02FL) only includes visits (where analysis date is e.g after randomization date (RANDDT)) with the lowest NADIR (lowest sum of diameter) so maximum tumor shrinkage by using the first non-missing lowest percentage change from baseline (PCHG).

adtr_sum <- adtr_sum %>%
  derive_var_relative_flag(
    by_vars = exprs(USUBJID),
    order = exprs(ADT),
    new_var = POSTRNDFL,
    condition = ADT > RANDDT,
    mode = "first",
    selection = "after",
    inclusive = TRUE,
    flag_no_ref_groups = FALSE
  ) %>%
  restrict_derivation(
    derivation = derive_var_extreme_flag,
    args = params(
      by_vars = exprs(USUBJID),
      new_var = ANL02FL,
      order = exprs(PCHG),
      mode = "first",
      check_type = "none"
    ),
    filter = ANL01FL == "Y" & POSTRNDFL == "Y"
  ) %>%
  select(-POSTRNDFL)

Analysis 03 flag (ANL03FL) includes all sum of diameters where all lesions were assessed (where ANL01FL == "Y") until the patient has PD, indicated by using the PDFL variable.

adtr_sum <- adtr_sum %>%
  restrict_derivation(
    derivation = derive_var_relative_flag,
    args = params(
      by_vars = exprs(USUBJID),
      new_var = ANL03FL,
      condition = PDFL == "Y",
      order = exprs(ADY),
      mode = "first",
      selection = "before",
      inclusive = FALSE
    ),
    filter = ANL01FL == "Y" | PDFL == "Y"
  )

Analysis 04 flag (ANL04FL) includes all sum of diameters where all lesions were assessed (where ANL01FL == "Y") and additionally those indicating PD using the PDFL variable.

adtr_sum <- adtr_sum %>%
  mutate(
    ANL04FL = if_else(ANL01FL == "Y" | PDFL == "Y", "Y", NA_character_)
  )

adtr <- bind_rows(adtr, adtr_sum)

Derive Analysis Sequence Number (ASEQ)

The {admiral} function admiral::derive_var_obs_number() can be used to derive ASEQ:

adtr <- adtr %>%
  derive_var_obs_number(
    by_vars = exprs(STUDYID, USUBJID),
    order = exprs(PARAMCD, AVISITN, TRSEQ),
    check_type = "error"
  )

Add ADSL Variables

If needed, the other ADSL variables can now be added. List of ADSL variables already merged held in vector adsl_vars.

adtr <- adtr %>%
  derive_vars_merged(
    dataset_add = select(adsl, !!!negate_vars(adsl_vars)),
    by_vars = exprs(STUDYID, USUBJID)
  )