Skip to contents

Long Data Summaries

Source: vignettes/articles/long_data.Rmd
long_data.Rmd

The goal of this article is to illustrate which {cards} functions are used to create long data summaries: think summaries from ADAE, ADLB, ADCM, and other similarly structured data sets.

Generally, the solution to long data summaries lies with ard_stack_hierarchical*(), ard_strata(), or even a call to a more basic function like ard_tabulate(). Herein, we will review these function and when each is needed.

Hierarchical or Nested Summaries

The ard_stack_hierarchical*() family of functions are useful when tabulating hierarchical or nested data and the tabulation needs to be repeated across more than one of the hierarchies. The most common example is the summary of adverse event (AE) data.

Primary System Organ Class
    Dictionary-Derived Term		 Placebo
N = 86		 Xanomeline High Dose
N = 84		 Xanomeline Low Dose
N = 84
Number Subjects with AE 69 (80%) 79 (94%) 77 (92%)
GASTROINTESTINAL DISORDERS 17 (20%) 21 (25%) 15 (18%)
    DIARRHOEA 9 (10%) 4 (4.8%) 5 (6.0%)
GENERAL DISORDERS AND ADMINISTRATION SITE CONDITIONS 21 (24%) 40 (48%) 47 (56%)
    APPLICATION SITE ERYTHEMA 3 (3.5%) 15 (18%) 12 (14%)
    APPLICATION SITE PRURITUS 6 (7.0%) 22 (26%) 22 (26%)
    APPLICATION SITE VESICLES 1 (1.2%) 6 (7.1%) 4 (4.8%)
SKIN AND SUBCUTANEOUS TISSUE DISORDERS 21 (24%) 42 (50%) 42 (50%)
    ERYTHEMA 9 (10%) 14 (17%) 15 (18%)
    PRURITUS GENERALISED 0 (0%) 1 (1.2%) 1 (1.2%)
Printing a few illustrative rows from the full table.

In the table above, the AE rates are reported for both the system organ class (SOC) and AE term. That is, each AE is counted once per subject and then rate for each AE calculated; this is repeated for SOC. A call to the ard_stack_hierarchical() function will return an ARD with the adverse event rates, the system organ class rates, the overall rates (row one from the example table), and the counts that appear in the header.

To create the ARD for this table, use the ard_stack_hierarchical() function.

ard_ae <-
  ADAE |>
  ard_stack_hierarchical(
    variables = c(AESOC, AEDECOD), # report rates for SOC and AE within SOC
    by = TRTA, # report all statistics by treatment
    id = USUBJID, # used to remove duplicate AEs within subject
    denominator = ADSL, # specified the denominator for rate calculations
    over_variables = TRUE # include summary statistics for Any AE
  )

The returned ARD contains four stacked sections: AE rates, SOC rates, Any AE rates, and Treatment counts. Let’s inspect each of these four sections.

Adverse Event Rates

To calculate the AE event counts, the ADAE data frame is subset to remove duplicate AEs reported from each subject. From there, the rates are calculated using the ADSL data frame passed in the denominator argument.

ard_ae |>
  filter(variable == "AEDECOD")
#> {cards} data frame: 2178 x 13
#>    group1 group1_level group2 group2_level variable variable_level stat_name stat_label  stat
#> 1    TRTA      Placebo  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         n          n     1
#> 2    TRTA      Placebo  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         N          N    86
#> 3    TRTA      Placebo  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         p          % 0.012
#> 4    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         n          n     3
#> 5    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         N          N    84
#> 6    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         p          % 0.036
#> 7    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         n          n     1
#> 8    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         N          N    84
#> 9    TRTA    Xanomeli…  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         p          % 0.012
#> 10   TRTA      Placebo  AESOC    CARDIAC …  AEDECOD      ATRIAL F…         n          n     0
#> ℹ 2168 more rows
#> ℹ Use `print(n = ...)` to see more rows
#> ℹ 4 more variables: context, fmt_fun, warning, error

System Organ Class Rates

The AE rate process is repeated for SOC.

ard_ae |>
  filter(variable == "AESOC") |>
  select(-all_missing_columns())
#> {cards} data frame: 207 x 9
#>    group1 group1_level variable variable_level stat_name stat_label  stat
#> 1    TRTA      Placebo    AESOC      CARDIAC …         n          n    13
#> 2    TRTA      Placebo    AESOC      CARDIAC …         N          N    86
#> 3    TRTA      Placebo    AESOC      CARDIAC …         p          % 0.151
#> 4    TRTA    Xanomeli…    AESOC      CARDIAC …         n          n    18
#> 5    TRTA    Xanomeli…    AESOC      CARDIAC …         N          N    84
#> 6    TRTA    Xanomeli…    AESOC      CARDIAC …         p          % 0.214
#> 7    TRTA    Xanomeli…    AESOC      CARDIAC …         n          n    13
#> 8    TRTA    Xanomeli…    AESOC      CARDIAC …         N          N    84
#> 9    TRTA    Xanomeli…    AESOC      CARDIAC …         p          % 0.155
#> 10   TRTA      Placebo    AESOC      CONGENIT…         n          n     0
#> ℹ 197 more rows
#> ℹ Use `print(n = ...)` to see more rows
#> ℹ 2 more variables: context, fmt_fun

Any AE Rates

The process is then repeated to calculate rates of any adverse event.

ard_ae |>
  filter(variable == "..ard_hierarchical_overall..") |>
  select(-all_missing_columns())
#> {cards} data frame: 9 x 9
#>   group1 group1_level                     variable variable_level stat_name stat_label  stat
#> 1   TRTA      Placebo ..ard_hierarchical_overall..           TRUE         n          n    69
#> 2   TRTA      Placebo ..ard_hierarchical_overall..           TRUE         N          N    86
#> 3   TRTA      Placebo ..ard_hierarchical_overall..           TRUE         p          % 0.802
#> 4   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         n          n    79
#> 5   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         N          N    84
#> 6   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         p          %  0.94
#> 7   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         n          n    77
#> 8   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         N          N    84
#> 9   TRTA    Xanomeli… ..ard_hierarchical_overall..           TRUE         p          % 0.917
#> ℹ 2 more variables: context, fmt_fun

Treatment Counts

Finally, a univariate tabulation of the TRTA column from ADSL is included.

ard_ae |>
  filter(variable == "TRTA") |>
  select(-all_missing_columns())
#> {cards} data frame: 9 x 7
#>   variable variable_level  context stat_name stat_label  stat
#> 1     TRTA        Placebo tabulate         n          n    86
#> 2     TRTA        Placebo tabulate         N          N   254
#> 3     TRTA        Placebo tabulate         p          % 0.339
#> 4     TRTA      Xanomeli… tabulate         n          n    84
#> 5     TRTA      Xanomeli… tabulate         N          N   254
#> 6     TRTA      Xanomeli… tabulate         p          % 0.331
#> 7     TRTA      Xanomeli… tabulate         n          n    84
#> 8     TRTA      Xanomeli… tabulate         N          N   254
#> 9     TRTA      Xanomeli… tabulate         p          % 0.331
#> ℹ 1 more variable: fmt_fun

The package exports a similar function for counting adverse event, rather than calculating rates: ard_stack_hierarchical_count().

Stratified Summaries

There are many types of stratified summaries that may be needed to report results from a trial. We will focus on a common lab summary where summary statistics are reported by lab type, visit and treatment.

Lab
     Visit
Placebo
N = 86
Xanomeline High Dose
N = 84
Xanomeline Low Dose
N = 84
Value at Visit Change from Baseline Value at Visit Change from Baseline Value at Visit Change from Baseline
Bilirubin (umol/L)





    Baseline





        n 7
7
6
        Mean (SD) 8.6 (3.0)
10.7 (5.1)
9.1 (3.0)
        Median 8.6
10.3
7.7
        Min - Max 5.1 - 12.0
5.1 - 18.8
6.8 - 13.7
    Week 24





        n 5 5 4 4 2 2
        Mean (SD) 7.87 (2.59) -0.34 (3.06) 10.26 (4.19) -2.99 (0.86) 9.41 (1.21) 0.00 (4.84)
        Median 6.84 0.00 9.41 -3.42 9.41 0.00
        Min - Max 5.13 - 11.97 -3.42 - 3.42 6.84 - 15.39 -3.42 - -1.71 8.55 - 10.26 -3.42 - 3.42
Creatinine (umol/L)





    Baseline





        n 7
7
6
        Mean (SD) 97 (18)
102 (15)
106 (22)
        Median 88
106
111
        Min - Max 80 - 124
80 - 124
80 - 133
    Week 24





        n 5 5 4 4 2 2
        Mean (SD) 99 (20) 5.3 (4.8) 106 (23) 2.2 (8.5) 102 (31) 0.0 (0.0)
        Median 97 8.8 106 4.4 102 0.0
        Min - Max 80 - 133 0.0 - 8.8 80 - 133 -8.8 - 8.8 80 - 124 0.0 - 0.0
Printing a few illustrative rows from the full table.

To build the ARD for this table, we use the ard_summary().

  • ard_summary(by="TRTA"): Use the by argument ensures each level of treatment has all associated summary statistics, even if there are combinations that are unobserved or all NA.

  • ard_summary(strata=c("PARAM", "AVISIT"): The strata argument will produce summary statistics for all observed combinations of 'PARAM' and 'AVISIT'. We opt to use strata because it is common a trial will not collect all labs at each visit, and we don’t want to report that bilirubin had no observations at week xx when it was never meant to be collected at that visit, for example.

  • ard_summary(variables=c("AVAL", "CHG")): These are the variables that will be summarized within 'TRTA', 'PARAM', and 'AVISIT'.

ADLB |>
  # subset on two labs and two study visits
  filter(
    PARAMCD %in% c("BILI", "CREAT"),
    AVISIT %in% c("Baseline", "Week 24")
  ) |>
  ard_summary(
    # calculate statistics by observed combinations of PARAM on AVISIT
    strata = c("PARAM", "AVISIT"),
    # `by='TRTA'` will provide results for each of the treatments, even if unobserved
    by = "TRTA",
    # provide summaries for the measurement and its change from baseline
    variables = c("AVAL", "CHG")
  )
#> {cards} data frame: 192 x 14
#>    group1 group1_level group2 group2_level group3 group3_level variable stat_name stat_label  stat
#> 1    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL         N          N     7
#> 2    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL      mean       Mean  8.55
#> 3    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL        sd         SD 2.962
#> 4    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL    median     Median  8.55
#> 5    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL       p25         Q1  5.13
#> 6    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL       p75         Q3 11.97
#> 7    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL       min        Min  5.13
#> 8    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline     AVAL       max        Max 11.97
#> 9    TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline      CHG         N          N     0
#> 10   TRTA      Placebo  PARAM    Bilirubi… AVISIT     Baseline      CHG      mean       Mean   NaN
#> ℹ 182 more rows
#> ℹ Use `print(n = ...)` to see more rows
#> ℹ 4 more variables: context, fmt_fun, warning, error

There are some cases, where slightly different behavior is needed within stratum. In these cases, use the ard_strata() function. For example, if we were tabulating character AVALC values, and the possible values are different depending on PARAM, the code may look something like this:

ard_strata(
  data = ADLB,
  .strata = "PARAM",
  .f = \(data_param) {
    # set factor depending on the PARAM value
    if (data_param$PARAM[1] == "XXX") data_param$AVALC <- factor(data_param$AVALC, levels = c("No", "Yes"))
    if (data_param$PARAM[1] == "YYY") data_param$AVALC <- factor(data_param$AVALC, levels = c("Low", "High"))

    ard_tabulate(
      data_param,
      strata = "AVISIT",
      by = "TRTA",
      variable = "AVALC"
    )
  }
)