Creating Questionnaire ADaMs

Introduction

This article describes creating questionnaire ADaMs. Although questionnaire data is collected in a single SDTM dataset (QS), usually it does not make sense to create a single ADQS dataset for all questionnaire analyses. For example, a univariate analysis of scores by visit requires different variables than a time-to-event analysis. Therefore this vignette does not provide a programming workflow for a complete dataset, but provides examples for deriving common types of questionnaire parameters.

At the moment, {admiral} does not provide functions or metadata for specific questionnaires nor functionality for handling the vast amount of questionnaires and related parameters, e.g. a metadata structure for storing parameter definitions and functions for reading such metadata. We plan to provide it in future releases.

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

Required Packages

The examples of this vignette require the following packages.

library(dplyr)
library(tidyr)
library(tibble)
library(admiral)

Example Data

In this vignette we use the example data from the CDISC ADaM Supplements (Generalized Anxiety Disorder 7-Item Version 2 (GAD-7), Geriatric Depression Scale Short Form (GDS-SF))1:

data("example_qs")
qs <- example_qs
STUDYID USUBJID QSSEQ QSTESTCD QSTEST VISIT VISITNUM QSDTC QSORRES QSCAT QSSTRESN
STUDYX P0001 1 GAD0201 GAD02-Feeling Nervous Anxious or On Edge VISIT 1 1 2012-11-16 More than half the days GAD-7 V2 2
STUDYX P0001 2 GAD0202 GAD02-Not Able to Stop/Control Worrying VISIT 1 1 2012-11-16 More than half the days GAD-7 V2 2
STUDYX P0001 3 GAD0203 GAD02-Worrying Too Much About Things VISIT 1 1 2012-11-16 More than half the days GAD-7 V2 2
STUDYX P0001 4 GAD0204 GAD02-Trouble Relaxing VISIT 1 1 2012-11-16 More than half the days GAD-7 V2 2
STUDYX P0001 5 GAD0205 GAD02-Being Restless Hard to Sit Still VISIT 1 1 2012-11-16 Nearly every day GAD-7 V2 3
STUDYX P0001 6 GAD0206 GAD02-Becoming Easily Annoyed/Irritable VISIT 1 1 2012-11-16 Nearly every day GAD-7 V2 3
STUDYX P0001 7 GAD0207 GAD02-Feel Afraid/Something Awful Happen VISIT 1 1 2012-11-16 Several days GAD-7 V2 1
STUDYX P0001 8 GAD0208 GAD02-Total Score VISIT 1 1 2012-11-16 15 GAD-7 V2 15
STUDYX P0001 9 GAD0201 GAD02-Feeling Nervous Anxious or On Edge UNSCHEDULED VISIT 5.01 501 2013-04-15 More than half the days GAD-7 V2 2
STUDYX P0001 10 GAD0202 GAD02-Not Able to Stop/Control Worrying UNSCHEDULED VISIT 5.01 501 2013-04-15 More than half the days GAD-7 V2 2
adsl <- tribble(
  ~STUDYID, ~USUBJID, ~SITEID, ~ITTFL, ~TRTSDT,                      ~DTHCAUS,
  "STUDYX",  "P0001",     13L,    "Y", lubridate::ymd("2012-11-16"), NA_character_,
  "STUDYX",  "P0002",     11L,    "Y", lubridate::ymd("2012-11-16"), "PROGRESSIVE DISEASE"
)
STUDYID USUBJID SITEID ITTFL TRTSDT DTHCAUS
STUDYX P0001 13 Y 2012-11-16 NA
STUDYX P0002 11 Y 2012-11-16 PROGRESSIVE DISEASE

Original Items

The original items, i.e. the answers to the questionnaire questions, can be handled in the same way as in a BDS finding ADaM. For example:

adqs <- qs %>%
  # Add ADSL variables
  derive_vars_merged(
    dataset_add = adsl,
    new_vars = exprs(TRTSDT, DTHCAUS),
    by_vars = exprs(STUDYID, USUBJID)
  ) %>%
  # Add analysis parameter variables
  mutate(
    PARAMCD = QSTESTCD,
    PARAM = QSTEST,
    PARCAT1 = QSCAT,
    AVALC = QSORRES,
    AVAL = QSSTRESN
  ) %>%
  # Add timing variables
  derive_vars_dt(new_vars_prefix = "A", dtc = QSDTC) %>%
  derive_vars_dy(reference_date = TRTSDT, source_vars = exprs(ADT)) %>%
  mutate(
    AVISIT = if_else(ADT <= TRTSDT, "BASELINE", VISIT),
    AVISITN = if_else(ADT <= TRTSDT, 0, VISITNUM)
  )
USUBJID PARAMCD PARAM PARCAT1 AVALC AVAL ADY AVISIT
P0001 GAD0201 GAD02-Feeling Nervous Anxious or On Edge GAD-7 V2 More than half the days 2 1 BASELINE
P0001 GAD0202 GAD02-Not Able to Stop/Control Worrying GAD-7 V2 More than half the days 2 1 BASELINE
P0001 GAD0203 GAD02-Worrying Too Much About Things GAD-7 V2 More than half the days 2 1 BASELINE
P0001 GAD0204 GAD02-Trouble Relaxing GAD-7 V2 More than half the days 2 1 BASELINE
P0001 GAD0205 GAD02-Being Restless Hard to Sit Still GAD-7 V2 Nearly every day 3 1 BASELINE
P0001 GAD0206 GAD02-Becoming Easily Annoyed/Irritable GAD-7 V2 Nearly every day 3 1 BASELINE
P0001 GAD0207 GAD02-Feel Afraid/Something Awful Happen GAD-7 V2 Several days 1 1 BASELINE
P0001 GAD0208 GAD02-Total Score GAD-7 V2 15 15 1 BASELINE
P0001 GAD0201 GAD02-Feeling Nervous Anxious or On Edge GAD-7 V2 More than half the days 2 151 UNSCHEDULED VISIT 5.01
P0001 GAD0202 GAD02-Not Able to Stop/Control Worrying GAD-7 V2 More than half the days 2 151 UNSCHEDULED VISIT 5.01

We handle unscheduled visits as normal visits. For deriving visits based on time-windows, see Visit and Period Variables. And for flagging values to be used for analysis, see derive_var_extreme_flag().

Please note that in the example data, the numeric values of the answers are mapped in SDTM (QSSTRESN) such that they can be used for deriving scores. Depending on the question, QSORRES == "YES" is mapped to QSSTRESN = 0 or QSSTRESN = 1. If the QSSTRESN values are not ready to be used for deriving scores and require transformation, it is recommended that QSSTRESN is kept in the ADaM dataset for traceability, and the transformed value is stored in AVAL, since that’s what will be used for the score calculation.

Scales and Scores

Scales and Scores are often derived as the sum or the average across a subset of the items. For the GAD-7 questionnaire, the total score is derived as the sum. The derive_summary_records() function with sum() can be used to derive it as a new parameter. For selecting the parameters to be summarized, regular expressions like in the example below may be helpful. In the example we derive a separate ADaM dataset for each questionnaire. Depending on the analysis needs, it is also possible that an ADaM contains more than one questionnaire or all questionnaires.

adgad7 <- adqs %>%
  # Select records to keep in the GAD-7 ADaM
  filter(PARCAT1 == "GAD-7 V2") %>%
  derive_summary_records(
    dataset = .,
    dataset_add = .,
    by_vars = exprs(STUDYID, USUBJID, AVISIT, ADT, ADY, TRTSDT, DTHCAUS),
    # Select records contributing to total score
    filter_add = str_detect(PARAMCD, "GAD020[1-7]"),
    set_values_to = exprs(
      AVAL = sum(AVAL, na.rm = TRUE),
      PARAMCD = "GAD02TS",
      PARAM = "GAD02-Total Score - Analysis"
    )
  )
USUBJID PARAMCD PARAM AVAL ADY AVISIT
P0001 GAD0201 GAD02-Feeling Nervous Anxious or On Edge 2 1 BASELINE
P0001 GAD0202 GAD02-Not Able to Stop/Control Worrying 2 1 BASELINE
P0001 GAD0203 GAD02-Worrying Too Much About Things 2 1 BASELINE
P0001 GAD0204 GAD02-Trouble Relaxing 2 1 BASELINE
P0001 GAD0205 GAD02-Being Restless Hard to Sit Still 3 1 BASELINE
P0001 GAD0206 GAD02-Becoming Easily Annoyed/Irritable 3 1 BASELINE
P0001 GAD0207 GAD02-Feel Afraid/Something Awful Happen 1 1 BASELINE
P0001 GAD0208 GAD02-Total Score 15 1 BASELINE
P0001 GAD02TS GAD02-Total Score - Analysis 15 1 BASELINE
P0001 GAD0201 GAD02-Feeling Nervous Anxious or On Edge 2 151 UNSCHEDULED VISIT 5.01

For the GDS-SF questionnaire, the total score is defined as the average of the item values transformed to the range [0, 15] and rounded up to the next integer. If more than five items are missing, the total score is considered as missing. This parameter can be derived by compute_scale() and derive_summary_records():

adgdssf <- adqs %>%
  # Select records to keep in the GDS-SF ADaM
  filter(PARCAT1 == "GDS SHORT FORM") %>%
  derive_summary_records(
    dataset = .,
    dataset_add = .,
    by_vars = exprs(STUDYID, USUBJID, AVISIT, ADT, ADY, TRTSDT, DTHCAUS),
    # Select records contributing to total score
    filter_add = str_detect(PARAMCD, "GDS02[01][0-9]"),
    set_values_to = exprs(
      AVAL = compute_scale(
        AVAL,
        source_range = c(0, 1),
        target_range = c(0, 15),
        min_n = 10
      ) %>%
        ceiling(),
      PARAMCD = "GDS02TS",
      PARAM = "GDS02- Total Score - Analysis"
    )
  )
USUBJID PARAMCD PARAM AVAL ADY AVISIT
P0001 GDS0201 GDS02-Satisfied With Life 0 1 BASELINE
P0001 GDS0202 GDS02-Dropped Activities and Interests 1 1 BASELINE
P0001 GDS0203 GDS02-Life Is Empty 0 1 BASELINE
P0001 GDS0204 GDS02-Bored Often 1 1 BASELINE
P0001 GDS0205 GDS02-Good Spirits Most of Time 1 1 BASELINE
P0001 GDS0206 GDS02-Afraid of Something Bad Happening 1 1 BASELINE
P0001 GDS0207 GDS02-Feel Happy Most of Time 1 1 BASELINE
P0001 GDS0208 GDS02-Often Feel Helpless 1 1 BASELINE
P0001 GDS0209 GDS02-Prefer to Stay Home 0 1 BASELINE
P0001 GDS0210 GDS02-Memory Problems 1 1 BASELINE

After deriving the scores by visit, the baseline and change from baseline variables can be derived:

adgdssf <- adgdssf %>%
  # Flag baseline records (last before treatement start)
  restrict_derivation(
    derivation = derive_var_extreme_flag,
    args = params(
      by_vars = exprs(STUDYID, USUBJID, PARAMCD),
      order = exprs(ADT),
      new_var = ABLFL,
      mode = "last"
    ),
    filter = !is.na(AVAL) & ADT <= TRTSDT
  ) %>%
  # Derive baseline and change from baseline variables
  derive_var_base(
    by_vars = exprs(STUDYID, USUBJID, PARAMCD),
    source_var = AVAL,
    new_var = BASE
  ) %>%
  derive_var_chg() %>%
  derive_var_pchg() %>%
  # Derive sequence number
  derive_var_obs_number(
    by_vars = exprs(STUDYID, USUBJID),
    order = exprs(PARAMCD, ADT),
    check_type = "error"
  )
USUBJID PARAMCD PARAM AVISIT AVAL BASE CHG PCHG
P0001 GDS0201 GDS02-Satisfied With Life BASELINE 0 0 0 NA
P0001 GDS0201 GDS02-Satisfied With Life VISIT 2 0 0 0 NA
P0001 GDS0201 GDS02-Satisfied With Life UNSCHEDULED 2.01 0 0 0 NA
P0001 GDS0201 GDS02-Satisfied With Life VISIT 3 NA 0 NA NA
P0001 GDS0201 GDS02-Satisfied With Life VISIT 4 0 0 0 NA
P0001 GDS0202 GDS02-Dropped Activities and Interests BASELINE 1 1 0 0
P0001 GDS0202 GDS02-Dropped Activities and Interests VISIT 2 0 1 -1 -100
P0001 GDS0202 GDS02-Dropped Activities and Interests UNSCHEDULED 2.01 0 1 -1 -100
P0001 GDS0202 GDS02-Dropped Activities and Interests VISIT 3 NA 1 NA NA
P0001 GDS0202 GDS02-Dropped Activities and Interests VISIT 4 0 1 -1 -100

Time to Deterioration/Improvement

As time to event parameters require specific variables like CNSR, STARTDT, and EVNTDESC, it makes sense to create a separate time to event dataset for them. However, it might be useful to create flags or categorization variables in ADQS. For example:

adgdssf <- adgdssf %>%
  mutate(
    AVALCAT1 = if_else(
      PARAMCD == "GDS02TS",
      case_when(
        AVAL <= 5 ~ "Normal",
        AVAL <= 10 ~ "Possible Depression",
        AVAL > 10 ~ "Likely Depression"
      ),
      NA_character_
    ),
    AVALCAT1N = if_else(
      PARAMCD == "GDS02TS",
      case_when(
        AVAL <= 5 ~ 0L,
        AVAL <= 10 ~ 1L,
        AVAL > 10 ~ 2L
      ),
      NA_integer_
    )
  ) %>%
  derive_var_base(
    by_vars = exprs(STUDYID, USUBJID, PARAMCD),
    source_var = AVALCAT1,
    new_var = BASECAT1
  ) %>%
  derive_var_base(
    by_vars = exprs(STUDYID, USUBJID, PARAMCD),
    source_var = AVALCAT1N,
    new_var = BASECA1N
  ) %>%
  mutate(
    CHGCAT1 = case_when(
      AVALCAT1N > BASECA1N ~ "WORSENED",
      AVALCAT1N == BASECA1N ~ "NO CHANGE",
      AVALCAT1N < BASECA1N ~ "IMPROVED",
    )
  )
USUBJID PARAMCD PARAM AVISIT AVAL AVALCAT1 CHGCAT1
P0001 GDS02TS GDS02- Total Score - Analysis BASELINE 10 Possible Depression NO CHANGE
P0001 GDS02TS GDS02- Total Score - Analysis VISIT 2 8 Possible Depression NO CHANGE
P0001 GDS02TS GDS02- Total Score - Analysis UNSCHEDULED 2.01 8 Possible Depression NO CHANGE
P0001 GDS02TS GDS02- Total Score - Analysis VISIT 3 7 Possible Depression NO CHANGE
P0001 GDS02TS GDS02- Total Score - Analysis VISIT 4 3 Normal IMPROVED
P0001 GDS0215 GDS02-Most People Better Off Than You BASELINE 0 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You VISIT 2 0 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You UNSCHEDULED 2.01 0 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You VISIT 3 0 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You VISIT 4 0 NA NA

Then a time to deterioration parameter can be derived by:

# Define event
deterioration_event <- event_source(
  dataset_name = "adqs",
  filter = PARAMCD == "GDS02TS" & CHGCAT1 == "WORSENED",
  date = ADT,
  set_values_to = exprs(
    EVNTDESC = "DEPRESSION WORSENED",
    SRCDOM = "ADQS",
    SRCVAR = "ADT",
    SRCSEQ = ASEQ
  )
)

# Define censoring at last assessment
last_valid_assessment <- censor_source(
  dataset_name = "adqs",
  filter = PARAMCD == "GDS02TS" & !is.na(CHGCAT1),
  date = ADT,
  set_values_to = exprs(
    EVNTDESC = "LAST ASSESSMENT",
    SRCDOM = "ADQS",
    SRCVAR = "ADT",
    SRCSEQ = ASEQ
  )
)

# Define censoring at treatment start (for subjects without assessment)
start <- censor_source(
  dataset_name = "adsl",
  date = TRTSDT,
  set_values_to = exprs(
    EVNTDESC = "TREATMENT START",
    SRCDOM = "ADSL",
    SRCVAR = "TRTSDT"
  )
)

adgdstte <- derive_param_tte(
  dataset_adsl = adsl,
  source_datasets = list(adsl = adsl, adqs = adgdssf),
  start_date = TRTSDT,
  event_conditions = list(deterioration_event),
  censor_conditions = list(last_valid_assessment, start),
  set_values_to = exprs(
    PARAMCD = "TTDEPR",
    PARAM = "Time to depression"
  )
) %>%
  derive_vars_duration(
    new_var = AVAL,
    start_date = STARTDT,
    end_date = ADT
  )
USUBJID PARAMCD PARAM AVAL CNSR EVNTDESC SRCDOM SRCVAR
P0001 TTDEPR Time to depression 90 1 LAST ASSESSMENT ADQS ADT
P0002 TTDEPR Time to depression 30 0 DEPRESSION WORSENED ADQS ADT

Time to Confirmed/Definitive Deterioration/Improvement

The derivation of confirmed/definitive deterioration/improvement parameters is very similar to the unconfirmed deterioration parameters except that the event is not based on CHGCATy, but on a confirmation flag variable. This confirmation flag can be derived by derive_var_joined_exist_flag(). For example, flagging deteriorations, which are confirmed by a second assessment at least seven days later:

adgdssf <- adgdssf %>%
  derive_var_joined_exist_flag(
    dataset_add = adgdssf,
    by_vars = exprs(USUBJID, PARAMCD),
    order = exprs(ADT),
    new_var = CDETFL,
    join_vars = exprs(CHGCAT1, ADY),
    join_type = "after",
    filter_join = CHGCAT1 == "WORSENED" &
      CHGCAT1.join == "WORSENED" &
      ADY.join >= ADY + 7
  )
USUBJID PARAMCD PARAM ADY CHGCAT1 CDETFL
P0001 GDS02TS GDS02- Total Score - Analysis 1 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 30 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 43 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 58 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 90 IMPROVED NA
P0001 GDS0215 GDS02-Most People Better Off Than You 1 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 30 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 43 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 58 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 90 NA NA

For flagging deteriorations at two consecutive assessments or considering death due to progression at the last assessment as confirmation, the tmp_obs_nr_var argument is helpful:

# Flagging deterioration at two consecutive assessments
adgdssf <- adgdssf %>%
  derive_var_joined_exist_flag(
    dataset_add = adgdssf,
    by_vars = exprs(USUBJID, PARAMCD),
    order = exprs(ADT),
    new_var = CONDETFL,
    join_vars = exprs(CHGCAT1),
    join_type = "after",
    tmp_obs_nr_var = tmp_obs_nr,
    filter_join = CHGCAT1 == "WORSENED" &
      CHGCAT1.join == "WORSENED" &
      tmp_obs_nr.join == tmp_obs_nr + 1
  ) %>%
  # Flagging deterioration confirmed by
  # - a second deterioration at least 7 days later or
  # - deterioration at the last assessment and death due to progression
  derive_var_joined_exist_flag(
    .,
    dataset_add = .,
    by_vars = exprs(USUBJID, PARAMCD),
    order = exprs(ADT),
    new_var = CDTDTHFL,
    join_vars = exprs(CHGCAT1, ADY),
    join_type = "all",
    tmp_obs_nr_var = tmp_obs_nr,
    filter_join = CHGCAT1 == "WORSENED" & (
      CHGCAT1.join == "WORSENED" & ADY.join >= ADY + 7 |
        tmp_obs_nr == max(tmp_obs_nr.join) & DTHCAUS == "PROGRESSIVE DISEASE")
  )
USUBJID PARAMCD PARAM ADY CHGCAT1 CONDETFL CDTDTHFL
P0001 GDS02TS GDS02- Total Score - Analysis 1 NO CHANGE NA NA
P0001 GDS02TS GDS02- Total Score - Analysis 30 NO CHANGE NA NA
P0001 GDS02TS GDS02- Total Score - Analysis 43 NO CHANGE NA NA
P0001 GDS02TS GDS02- Total Score - Analysis 58 NO CHANGE NA NA
P0001 GDS02TS GDS02- Total Score - Analysis 90 IMPROVED NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 1 NA NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 30 NA NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 43 NA NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 58 NA NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 90 NA NA NA

For definitive deterioration (deterioration at all following assessments), parameter summary functions like all() can be used in the filter condition:

adgdssf <- adgdssf %>%
  derive_var_joined_exist_flag(
    dataset_add = adgdssf,
    by_vars = exprs(USUBJID, PARAMCD),
    order = exprs(ADT),
    new_var = DEFDETFL,
    join_vars = exprs(CHGCAT1),
    join_type = "after",
    filter_join = CHGCAT1 == "WORSENED" & all(CHGCAT1.join == "WORSENED")
  )
USUBJID PARAMCD PARAM ADY CHGCAT1 DEFDETFL
P0001 GDS02TS GDS02- Total Score - Analysis 1 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 30 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 43 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 58 NO CHANGE NA
P0001 GDS02TS GDS02- Total Score - Analysis 90 IMPROVED NA
P0001 GDS0215 GDS02-Most People Better Off Than You 1 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 30 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 43 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 58 NA NA
P0001 GDS0215 GDS02-Most People Better Off Than You 90 NA NA

The time-to-event parameter can be derived in the same way as for the unconfirmed parameters (see Time to Deterioration/Improvement).

Worst/Best Answer

This class of parameters can be used when the worst answer of a set of yes/no answers should be selected. For example, if yes/no answers for “No sleep”, “Waking up more than three times”, “More than 30 minutes to fall asleep” are collected, a parameter for the worst sleeping problems could be derived. In the example, “no sleeping problems” is assumed if all questions were answered with “no”.

adsp <- adqs %>%
  filter(PARCAT1 == "SLEEPING PROBLEMS") %>%
  derive_extreme_event(
    by_vars = exprs(USUBJID, AVISIT),
    tmp_event_nr_var = event_nr,
    order = exprs(event_nr, ADY, QSSEQ),
    mode = "first",
    events = list(
      event(
        condition = PARAMCD == "SP0101" & AVALC == "YES",
        set_values_to = exprs(
          AVALC = "No sleep",
          AVAL = 1
        )
      ),
      event(
        condition = PARAMCD == "SP0102" & AVALC == "YES",
        set_values_to = exprs(
          AVALC = "Waking up more than three times",
          AVAL = 2
        )
      ),
      event(
        condition = PARAMCD == "SP0103" & AVALC == "YES",
        set_values_to = exprs(
          AVALC = "More than 30 mins to fall asleep",
          AVAL = 3
        )
      ),
      event(
        condition = all(AVALC == "NO"),
        set_values_to = exprs(
          AVALC = "No sleeping problems",
          AVAL = 4
        )
      ),
      event(
        condition = TRUE,
        set_values_to = exprs(
          AVALC = "Missing",
          AVAL = 99
        )
      )
    ),
    set_values_to = exprs(
      PARAMCD = "SP01WSP",
      PARAM = "Worst Sleeping Problems"
    )
  )
USUBJID PARAMCD PARAM AVISIT AVALC
P0001 SP0101 SP01-No sleep at all BASELINE NO
P0001 SP0102 SP01-Waking up more than three times BASELINE NO
P0001 SP0103 SP01-More than 30 mins to fall asleep BASELINE YES
P0001 SP01WSP Worst Sleeping Problems BASELINE More than 30 mins to fall asleep
P0001 SP0101 SP01-No sleep at all VISIT 2 NO
P0001 SP0102 SP01-Waking up more than three times VISIT 2 NO
P0001 SP0103 SP01-More than 30 mins to fall asleep VISIT 2 NO
P0001 SP01WSP Worst Sleeping Problems VISIT 2 No sleeping problems
P0001 SP0101 SP01-No sleep at all VISIT 4 NO
P0001 SP0102 SP01-Waking up more than three times VISIT 4 YES

Completion

Parameters for completion, like “at least 90% of the questions were answered”, can be derived by derive_summary_records().

adgdssf <- adgdssf %>%
  derive_summary_records(
    dataset_add = adgdssf,
    filter_add = str_detect(PARAMCD, "GDS02[01][0-9]"),
    by_vars = exprs(USUBJID, AVISIT),
    set_values_to = exprs(
      AVAL = sum(!is.na(AVAL)) / 15 >= 0.9,
      PARAMCD = "COMPL90P",
      PARAM = "Completed at least 90% of questions?",
      AVALC = if_else(AVAL == 1, "YES", "NO")
    )
  )
USUBJID PARAMCD PARAM AVISIT AVALC
P0001 COMPL90P Completed at least 90% of questions? BASELINE YES
P0001 COMPL90P Completed at least 90% of questions? UNSCHEDULED 2.01 YES
P0001 COMPL90P Completed at least 90% of questions? VISIT 2 YES
P0001 COMPL90P Completed at least 90% of questions? VISIT 3 NO
P0001 COMPL90P Completed at least 90% of questions? VISIT 4 YES
P0001 GDS0201 GDS02-Satisfied With Life BASELINE YES
P0001 GDS0201 GDS02-Satisfied With Life VISIT 2 YES
P0001 GDS0201 GDS02-Satisfied With Life UNSCHEDULED 2.01 YES
P0001 GDS0201 GDS02-Satisfied With Life VISIT 3 NA
P0001 GDS0201 GDS02-Satisfied With Life VISIT 4 YES

Please note that the denominator may depend on the answers of some of the questions. For example, a given questionnaire might direct someone to go from question #4 directly to question #8 based on their response to question #4, because questions #5, #6 and #7 would not apply in that case.

If missed visits need to be taken into account, the expected records can be added to the input dataset by calling derive_expected_records():

# Create dataset with expected visits and parameters (GDS0201 - GDS0215)
parm_visit_ref <- crossing(
  tribble(
    ~AVISIT,    ~AVISITN,
    "BASELINE",        0,
    "VISIT 2",         2,
    "VISIT 3",         3,
    "VISIT 4",         4,
    "VISIT 5",         5
  ),
  tibble(PARAMCD = sprintf("GDS02%02d", seq(1, 15)))
)

adgdssf <- adgdssf %>%
  derive_expected_records(
    dataset_ref = parm_visit_ref,
    by_vars = exprs(USUBJID),
    set_values_to = exprs(
      filled_in = 1
    )
  ) %>%
  derive_summary_records(
    dataset = .,
    dataset_add = .,
    filter_add = str_detect(PARAMCD, "GDS02[01][0-9]"),
    by_vars = exprs(USUBJID, AVISIT),
    set_values_to = exprs(
      AVAL = all(!is.na(AVAL)),
      PARAMCD = "COMPLALL",
      PARAM = "Completed all questions?",
      AVALC = if_else(AVAL == 1, "YES", "NO")
    )
  ) %>%
  filter(is.na(filled_in)) %>%
  select(-filled_in)
USUBJID PARAMCD PARAM AVISIT AVALC
P0001 COMPL90P Completed at least 90% of questions? BASELINE YES
P0001 COMPL90P Completed at least 90% of questions? UNSCHEDULED 2.01 YES
P0001 COMPL90P Completed at least 90% of questions? VISIT 2 YES
P0001 COMPL90P Completed at least 90% of questions? VISIT 3 NO
P0001 COMPL90P Completed at least 90% of questions? VISIT 4 YES
P0001 COMPLALL Completed all questions? BASELINE YES
P0001 COMPLALL Completed all questions? UNSCHEDULED 2.01 YES
P0001 COMPLALL Completed all questions? VISIT 2 YES
P0001 COMPLALL Completed all questions? VISIT 3 NO
P0001 COMPLALL Completed all questions? VISIT 4 YES

  1. The example QS data (example_qs) is included in the admiral package.↩︎