Read in Data

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

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

library(admiral)
library(admiralonco)
library(dplyr)
library(admiral.test)
library(lubridate)
library(stringr)
data("admiral_adsl")
data("admiral_rs")
data("admiral_tu")

adsl <- admiral_adsl
rs <- admiral_rs
tu <- admiral_tu

rs <- convert_blanks_to_na(rs)
tu <- convert_blanks_to_na(tu)

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

adsl_vars <- vars(RANDDT)
adrs <- derive_vars_merged(
  rs,
  dataset_add = adsl,
  new_vars = adsl_vars,
  by_vars = vars(STUDYID, USUBJID)
)

Pre-processing of Input Records

The first step involves company-specific pre-processing of records for the required input to the downstream parameter functions. Note that this could be needed multiple times (e.g. once for investigator and once for Independent Review Facility (IRF)/Blinded Independent Central Review (BICR) records). It could even involve merging input data from other sources besides RS, such as ADTR.

This step would include any required selection/derivation of ADT or applying any necessary partial date imputations, updating AVAL (e.g. this should be ordered from best to worst response), and setting analysis flag ANL01FL. Common options for ANL01FL would be to set null for invalid assessments or those occurring after new anti-cancer therapy, or to only flag assessments on or after after date of first treatment/randomization, or rules to cover the case when a patient has multiple observations per visit (e.g. by selecting worst value). Another consideration could be extra potential protocol-specific sources of Progressive Disease such as radiological assessments, which could be pre-processed here to create a PD record to feed downstream derivations.

For the derivation of the parameters it is expected that the subject identifier variables (usually STUDYID and USUBJID) and ADT are a unique key. This can be achieved by deriving an analysis flag (ANLzzFL). See Derive ANL01FL for an example.

The below shows an example of a possible company-specific implementation of this step.

adrs <- adrs %>%
  filter(RSEVAL == "INVESTIGATOR" & RSTESTCD == "OVRLRESP") %>%
  mutate(
    PARAMCD = "OVR",
    PARAM = "Overall Response by Investigator",
    PARCAT1 = "Tumor Response",
    PARCAT2 = "Investigator",
    PARCAT3 = "Recist 1.1"
  )

adrs <- adrs %>%
  derive_vars_dt(
    dtc = RSDTC,
    new_vars_prefix = "A",
    highest_imputation = "D",
    date_imputation = "last"
  ) %>%
  mutate(AVISIT = VISIT)

adrs <- adrs %>%
  mutate(
    AVALC = RSSTRESC,
    AVAL = aval_resp(AVALC)
  )

adrs <- adrs %>%
  restrict_derivation(
    derivation = derive_var_extreme_flag,
    args = params(
      by_vars = vars(STUDYID, USUBJID, ADT),
      order = vars(AVAL, RSSEQ),
      new_var = ANL01FL,
      mode = "last"
    ),
    filter = !is.na(AVAL) & ADT >= RANDDT
  )

adrs <- adrs %>%
  mutate(
    ANL01FL = case_when(
      !is.na(AVAL) & ADT >= RANDDT & ADT < NACTDT ~ "Y",
      TRUE ~ NA_character_
    )
  )

adrs <- adrs %>%
  derive_var_relative_flag(
    by_vars = vars(USUBJID),
    order = vars(ADT, AVISITN),
    new_var = ANL02FL,
    condition = AVALC == "PD",
    mode = "first",
    selection = "before",
    inclusive = TRUE
  )

Derive Progressive Disease Parameter

Now that we have the input records prepared above with any company-specific requirements, we can start to derive new parameter records. For the parameter derivations, all values except those overwritten by set_values_to argument are kept from the earliest occurring input record fulfilling the required criteria.

The function admiral::derive_param_extreme_event() can be used to find the date of first PD.

adrs <- adrs %>%
  derive_param_extreme_event(
    dataset_adsl = adsl,
    dataset_source = adrs,
    filter_source = PARAMCD == "OVR" & AVALC == "PD" & ANL01FL == "Y",
    order = vars(ADT, RSSEQ),
    set_values_to = vars(
      PARAMCD = "PD",
      PARAM = "Disease Progression by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

For progressive disease, response and death parameters shown in steps here and below, in our examples we show these as ADRS parameters, but they could equally be achieved via ADSL dates or ADEVENT parameters. If you prefer to store as an ADSL date, then the function admiral::derive_var_extreme_dt() could be used to find the date of first PD as a variable, rather than as a new parameter record. All the parameter derivation functions that use these dates are flexible to allow sourcing these from any input source using admiral::date_source(). See examples below.

Derive Response Parameter

The next required step is to define the source location for this newly derived PD date.

pd <- date_source(
  dataset_name = "adrs",
  date = ADT,
  filter = PARAMCD == "PD" & AVALC == "Y"
)

An equivalent example if using ADSL instead could be as follows (where PDDT would be pre-derived as first date of progressive disease).

pd <- date_source(
  dataset_name = "adsl",
  date = PDDT
)

The function derive_param_response() can then be used to find the date of first response. This differs from the admiral::derive_param_extreme_event() function in that it only looks for events occurring prior to first PD. In the below example, the response condition has been defined as CR or PR.

adrs <- adrs %>%
  derive_param_response(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & AVALC %in% c("CR", "PR") & ANL01FL == "Y",
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    set_values_to = vars(
      PARAMCD = "RSP",
      PARAM = "Response by Investigator (confirmation not required)",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Derive Clinical Benefit Parameter

Similarly, we now define the source location for this newly derived first response date.

resp <- date_source(
  dataset_name = "adrs",
  date = ADT,
  filter = PARAMCD == "RSP" & AVALC == "Y"
)

The function derive_param_clinbenefit() can then be used to derive the clinical benefit parameter, which we define as a patient having had a response or a sustained period of time before first PD. This could also be known as disease control. In this example the “sustained period” has been defined as 42 days after randomization date, using the ref_start_window argument.

adrs <- adrs %>%
  derive_param_clinbenefit(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & ANL01FL == "Y",
    source_resp = resp,
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    reference_date = RANDDT,
    ref_start_window = 42,
    set_values_to = vars(
      PARAMCD = "CB",
      PARAM = "Clinical Benefit by Investigator (confirmation for response not required)",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Derive Best Overall Response Parameter

The function derive_param_bor() can be used to derive the best overall response (without confirmation required) parameter. Similar to the above function you can optionally decide what period would you consider a SD or NON-CR/NON-PD as being eligible from. In this example, 42 days after randomization date has been used again.

adrs <- adrs %>%
  derive_param_bor(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & ANL01FL == "Y",
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    reference_date = RANDDT,
    ref_start_window = 42,
    set_values_to = vars(
      PARAMCD = "BOR",
      PARAM = "Best Overall Response by Investigator (confirmation not required)",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Note that the above gives pre-defined AVAL values of: "CR" ~ 1, "PR" ~ 2, "SD" ~ 3, "NON-CR/NON-PD" ~ 4, "PD" ~ 5, "NE" ~ 6, "MISSING" ~ 7.

If you’d like to provide your own company-specific ordering here you could do this as follows:

aval_resp_new <- function(arg) {
  case_when(
    arg == "CR" ~ 7,
    arg == "PR" ~ 6,
    arg == "SD" ~ 5,
    arg == "NON-CR/NON-PD" ~ 4,
    arg == "PD" ~ 3,
    arg == "NE" ~ 2,
    arg == "MISSING" ~ 1,
    TRUE ~ NA_real_
  )
}

Then add the additional argument aval_fun = aval_resp_new to the above derive_param_bor() call. Be aware that this will only impact the AVAL mapping, not the derivation of BOR in any way - as the function derivation relies only on AVALC here.

Derive Best Overall Response of CR/PR Parameter

The function admiral::derive_param_extreme_event() can be used to check if a patient had a response for BOR.

adrs <- adrs %>%
  derive_param_extreme_event(
    dataset_adsl = adsl,
    dataset_source = adrs,
    filter_source = PARAMCD == "BOR" & AVALC %in% c("CR", "PR") & ANL01FL == "Y",
    order = vars(ADT, RSSEQ),
    set_values_to = vars(
      PARAMCD = "BCP",
      PARAM = "Best Overall Response of CR/PR by Investigator (confirmation not required)",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Derive Response Parameters requiring Confirmation

Any of the above response parameters can be repeated for “confirmed” responses only. For these the functions derive_param_confirmed_resp() and derive_param_confirmed_bor() can be used. Some of the other functions from above can then be re-used passing in these confirmed response records. See the examples below of derived parameters requiring confirmation. The assessment and the confirmatory assessment here need to occur at least 28 days apart (without any +1 applied to this calculation of days between visits), using the ref_confirm argument.

adrs <- adrs %>%
  derive_param_confirmed_resp(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & AVALC %in% c("CR", "PR") & ANL01FL == "Y",
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    ref_confirm = 28,
    set_values_to = vars(
      PARAMCD = "CRSP",
      PARAM = "Confirmed Response by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

confirmed_resp <- date_source(
  dataset_name = "adrs",
  date = ADT,
  filter = PARAMCD == "CRSP" & AVALC == "Y"
)

adrs <- adrs %>%
  derive_param_clinbenefit(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & ANL01FL == "Y",
    source_resp = confirmed_resp,
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    reference_date = RANDDT,
    ref_start_window = 42,
    set_values_to = vars(
      PARAMCD = "CCB",
      PARAM = "Confirmed Clinical Benefit by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  ) %>%
  derive_param_confirmed_bor(
    dataset_adsl = adsl,
    filter_source = PARAMCD == "OVR" & ANL01FL == "Y",
    source_pd = pd,
    source_datasets = list(adrs = adrs),
    reference_date = RANDDT,
    ref_start_window = 42,
    ref_confirm = 28,
    set_values_to = vars(
      PARAMCD = "CBOR",
      PARAM = "Best Confirmed Overall Response by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  ) %>%
  derive_param_extreme_event(
    dataset_adsl = adsl,
    dataset_source = adrs,
    filter_source = PARAMCD == "CBOR" & AVALC %in% c("CR", "PR") & ANL01FL == "Y",
    order = vars(ADT, RSSEQ),
    set_values_to = vars(
      PARAMCD = "CBCP",
      PARAM = "Best Confirmed Overall Response of CR/PR by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Derive Parameters using Independent Review Facility (IRF)/Blinded Independent Central Review (BICR) responses

All of the above steps can be repeated for different sets of records, such as now using assessments from the IRF/BICR instead of investigator. For this you would just need to replace the first steps with selecting the required records, and then feed these as input to the downstream parameter functions.

Remember that a new progressive disease and response source object would be required for passing to source_pd and source_resp respectively.

adrsirf <- rs %>%
  filter(RSEVAL == "INDEPENDENT ASSESSOR" & RSEVALID == "RADIOLOGIST 1" & RSTESTCD == "OVRLRESP") %>%
  mutate(
    PARAMCD = "OVRR1",
    PARAM = "Overall Response by Radiologist 1",
    PARCAT1 = "Tumor Response",
    PARCAT2 = "Radiologist",
    PARCAT3 = "Recist 1.1"
  )

Then in all the calls to the parameter derivation functions you would replace the PARAMCD == "OVR" source with PARAMCD == "OVRR1".

Derive Death Parameter

The function admiral::derive_param_extreme_event() can be used to create a new death parameter using death date from ADSL. We need to restrict the columns from ADSL as we’ll merge all required variables later across all our ADRS records.

adsldth <- adsl %>%
  select(STUDYID, USUBJID, DTHDT, !!!adsl_vars)

adrs <- adrs %>%
  derive_param_extreme_event(
    dataset_adsl = adsldth,
    dataset_source = adsldth,
    filter_source = !is.na(DTHDT),
    set_values_to = vars(
      PARAMCD = "DEATH",
      PARAM = "Death",
      PARCAT1 = "Reference Event",
      ANL01FL = "Y",
      ADT = DTHDT
    )
  ) %>%
  select(-DTHDT)

Derive Last Disease Assessment Parameters

The function admiral::derive_param_extreme_event() can be used to create a parameter for last disease assessment. We need to set new_var to a dummy variable here, or otherwise the original AVALC value from the record would be overwritten with "Y".

adrs <- adrs %>%
  derive_param_extreme_event(
    dataset_adsl = adsl,
    dataset_source = adrs,
    filter_source = PARAMCD == "OVR" & ANL01FL == "Y",
    order = vars(ADT, RSSEQ),
    mode = "last",
    new_var = dummy,
    set_values_to = vars(
      PARAMCD = "LSTA",
      PARAM = "Last Disease Assessment by Investigator",
      PARCAT1 = "Tumor Response",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  ) %>%
  select(-dummy)

Derive Measurable Disease at Baseline Parameter

The function admiral::derive_param_exist_flag() can be used to check whether a patient has measurable disease at baseline, according to a company-specific condition. In this example we check TU for target lesions during the baseline visit. We need to restrict the columns from ADSL as we’ll merge all required variables later across all our ADRS records.

adslmdis <- adsl %>%
  select(STUDYID, USUBJID, !!!adsl_vars)

adrs <- adrs %>%
  derive_param_exist_flag(
    dataset_adsl = adslmdis,
    dataset_add = tu,
    condition = TUEVAL == "INVESTIGATOR" & TUSTRESC == "TARGET" & VISIT == "BASELINE",
    false_value = "N",
    missing_value = "N",
    set_values_to = vars(
      PARAMCD = "MDIS",
      PARAM = "Measurable Disease at Baseline by Investigator",
      PARCAT2 = "Investigator",
      PARCAT3 = "Recist 1.1",
      ANL01FL = "Y"
    )
  )

Derive AVAL for New Parameters

For cases where AVALC has been derived for new parameters above as "Y" or "N", we need to set AVAL to numeric versions such as 1/0.

adrs <- adrs %>%
  mutate(
    AVAL = case_when(
      AVALC == "Y" ~ 1,
      AVALC == "N" ~ 0,
      TRUE ~ AVAL
    )
  )

Assign ASEQ

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

adrs <- adrs %>%
  derive_var_obs_number(
    by_vars = vars(STUDYID, USUBJID),
    order = vars(PARAMCD, ADT, VISITNUM, RSSEQ),
    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.

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