Comparison with dplyr Tabulation

Introduction

In this vignette, we would like to discuss the similarities and differences between dplyr and rtable.

Much of the rtables framework focuses on tabulation/summarizing of data and then the visualization of the table. In this vignette, we focus on summarizing data using dplyr and contrast it to rtables. We won’t pay attention to the table visualization/markup and just derive the cell content.

Using dplyr to summarize data and gt to visualize the table is a good way if the tabulation is of a certain nature or complexity. However, there are tables such as the table created in the introduction vignette that take some effort to create with dplyr. Part of the effort is due to fact that when using dplyr the table data is stored in data.frames or tibbles which is not the most natural way to represent a table as we will show in this vignette.

If you know a more elegant way of deriving the table content with dplyr please let us know and we will update the vignette.

library(rtables)
library(dplyr)

Here is the table and data used in the introduction vignette:

n <- 400

set.seed(1)

df <- tibble(
arm = factor(sample(c("Arm A", "Arm B"), n, replace = TRUE), levels = c("Arm A", "Arm B")),
country = factor(sample(c("CAN", "USA"), n, replace = TRUE, prob = c(.55, .45)), levels = c("CAN", "USA")),
gender = factor(sample(c("Female", "Male"), n, replace = TRUE), levels = c("Female", "Male")),
handed = factor(sample(c("Left", "Right"), n, prob = c(.6, .4), replace = TRUE), levels = c("Left", "Right")),
age = rchisq(n, 30) + 10
) %>% mutate(
weight = 35 * rnorm(n, sd = .5) + ifelse(gender == "Female", 140, 180)
)

lyt <- basic_table(show_colcounts = TRUE) %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
summarize_row_groups() %>%
split_rows_by("handed") %>%
summarize_row_groups() %>%
analyze("age", afun = mean, format = "xx.x")

tbl <- build_table(lyt, df)
tbl
#                     Arm A                     Arm B
#              Female        Male        Female        Male
#              (N=96)      (N=105)       (N=92)      (N=107)
# ————————————————————————————————————————————————————————————
# CAN        45 (46.9%)   64 (61.0%)   46 (50.0%)   62 (57.9%)
#   Left     32 (33.3%)   42 (40.0%)   26 (28.3%)   37 (34.6%)
#     mean      38.9         40.4         40.3         37.7
#   Right    13 (13.5%)   22 (21.0%)   20 (21.7%)   25 (23.4%)
#     mean      36.6         40.2         40.2         40.6
# USA        51 (53.1%)   41 (39.0%)   46 (50.0%)   45 (42.1%)
#   Left     34 (35.4%)   19 (18.1%)   25 (27.2%)   25 (23.4%)
#     mean      40.4         39.7         39.2         40.1
#   Right    17 (17.7%)   22 (21.0%)   21 (22.8%)   20 (18.7%)
#     mean      36.9         39.8         38.5         39.0

Getting Started

We will start by deriving the first data cell on row 3 (note, row 1 and 2 have content cells, see the introduction vignette). Cell 3,1 contains the mean age for left handed & female Canadians in “Arm A”:

mean(df$age[df$country == "CAN" & df$arm == "Arm A" & df$gender == "Female" & df\$handed == "Left"])
# [1] 38.86979

or with dplyr:

df %>%
filter(country == "CAN", arm == "Arm A", gender == "Female", handed == "Left") %>%
summarise(mean_age = mean(age))
# # A tibble: 1 × 1
#   mean_age
#      <dbl>
# 1     38.9

Further, dplyr gives us other verbs to easily get the average age of left handed Canadians for each group defined by the 4 columns:

df %>%
group_by(arm, gender) %>%
filter(country == "CAN", handed == "Left") %>%
summarise(mean_age = mean(age))
# summarise() has grouped output by 'arm'. You can override using the .groups
# argument.
# # A tibble: 4 × 3
# # Groups:   arm [2]
#   arm   gender mean_age
#   <fct> <fct>     <dbl>
# 1 Arm A Female     38.9
# 2 Arm A Male       40.4
# 3 Arm B Female     40.3
# 4 Arm B Male       37.7

We can further get to all the average age cell values with:

average_age <- df %>%
group_by(arm, gender, country, handed) %>%
summarise(mean_age = mean(age))
# summarise() has grouped output by 'arm', 'gender', 'country'. You can
# override using the .groups argument.
average_age
# # A tibble: 16 × 5
# # Groups:   arm, gender, country [8]
#    arm   gender country handed mean_age
#    <fct> <fct>  <fct>   <fct>     <dbl>
#  1 Arm A Female CAN     Left       38.9
#  2 Arm A Female CAN     Right      36.6
#  3 Arm A Female USA     Left       40.4
#  4 Arm A Female USA     Right      36.9
#  5 Arm A Male   CAN     Left       40.4
#  6 Arm A Male   CAN     Right      40.2
#  7 Arm A Male   USA     Left       39.7
#  8 Arm A Male   USA     Right      39.8
#  9 Arm B Female CAN     Left       40.3
# 10 Arm B Female CAN     Right      40.2
# 11 Arm B Female USA     Left       39.2
# 12 Arm B Female USA     Right      38.5
# 13 Arm B Male   CAN     Left       37.7
# 14 Arm B Male   CAN     Right      40.6
# 15 Arm B Male   USA     Left       40.1
# 16 Arm B Male   USA     Right      39.0

In rtable syntax, we need the following code to get to the same content:

lyt <- basic_table() %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
split_rows_by("handed") %>%
analyze("age", afun = mean, format = "xx.x")

tbl <- build_table(lyt, df)
tbl
#                Arm A           Arm B
#            Female   Male   Female   Male
# ————————————————————————————————————————
# CAN
#   Left
#     mean    38.9    40.4    40.3    37.7
#   Right
#     mean    36.6    40.2    40.2    40.6
# USA
#   Left
#     mean    40.4    39.7    39.2    40.1
#   Right
#     mean    36.9    39.8    38.5    39.0

As mentioned in the introduction to this vignette, please ignore the difference in arranging and formatting the data: it’s possible to condense the rtable more and it is possible to make the tibble look more like the reference table using the gt R package.

In terms of tabulation for this example there was arguably not much added by rtables over dplyr.

Content Information

Unlike in rtables the different levels of summarization are discrete computations in dplyr which we will then need to combine

We first focus on the count and percentage information for handedness within each country (for each arm-gender pair), along with the analysis row mean values:

c_h_df <- df %>%
group_by(arm, gender, country, handed) %>%
summarize(mean = mean(age), c_h_count = n()) %>%
## we need the sum below to *not* be by country, so that we're dividing by the column counts
ungroup(country) %>%
# now the handed grouping has been removed, therefore we can calculate percent now:
mutate(n_col = sum(c_h_count), c_h_percent = c_h_count / n_col)
# summarise() has grouped output by 'arm', 'gender', 'country'. You can
# override using the .groups argument.
c_h_df
# # A tibble: 16 × 8
# # Groups:   arm, gender [4]
#    arm   gender country handed  mean c_h_count n_col c_h_percent
#    <fct> <fct>  <fct>   <fct>  <dbl>     <int> <int>       <dbl>
#  1 Arm A Female CAN     Left    38.9        32    96       0.333
#  2 Arm A Female CAN     Right   36.6        13    96       0.135
#  3 Arm A Female USA     Left    40.4        34    96       0.354
#  4 Arm A Female USA     Right   36.9        17    96       0.177
#  5 Arm A Male   CAN     Left    40.4        42   105       0.4
#  6 Arm A Male   CAN     Right   40.2        22   105       0.210
#  7 Arm A Male   USA     Left    39.7        19   105       0.181
#  8 Arm A Male   USA     Right   39.8        22   105       0.210
#  9 Arm B Female CAN     Left    40.3        26    92       0.283
# 10 Arm B Female CAN     Right   40.2        20    92       0.217
# 11 Arm B Female USA     Left    39.2        25    92       0.272
# 12 Arm B Female USA     Right   38.5        21    92       0.228
# 13 Arm B Male   CAN     Left    37.7        37   107       0.346
# 14 Arm B Male   CAN     Right   40.6        25   107       0.234
# 15 Arm B Male   USA     Left    40.1        25   107       0.234
# 16 Arm B Male   USA     Right   39.0        20   107       0.187

which has 16 rows (cells) like the average_age data frame defined above. Next, we will derive the group information for countries:

c_df <- df %>%
group_by(arm, gender, country) %>%
summarize(c_count = n()) %>%
# now the handed grouping has been removed, therefore we can calculate percent now:
mutate(n_col = sum(c_count), c_percent = c_count / n_col)
# summarise() has grouped output by 'arm', 'gender'. You can override using the
# .groups argument.
c_df
# # A tibble: 8 × 6
# # Groups:   arm, gender [4]
#   arm   gender country c_count n_col c_percent
#   <fct> <fct>  <fct>     <int> <int>     <dbl>
# 1 Arm A Female CAN          45    96     0.469
# 2 Arm A Female USA          51    96     0.531
# 3 Arm A Male   CAN          64   105     0.610
# 4 Arm A Male   USA          41   105     0.390
# 5 Arm B Female CAN          46    92     0.5
# 6 Arm B Female USA          46    92     0.5
# 7 Arm B Male   CAN          62   107     0.579
# 8 Arm B Male   USA          45   107     0.421

Finally, we left_join() the two levels of summary to get a data.frame containing the full set of values which make up the body of our table (note, however, they are not in the same order):

full_dplyr <- left_join(c_h_df, c_df) %>% ungroup()
# Joining with by = join_by(arm, gender, country, n_col)

Alternatively, we could calculate only the counts in c_h_df, and use mutate() after the left_join() to divide the counts by the n_col values which are more naturally calculated within c_df. This would simplify c_h_df’s creation somewhat by not requiring the explicit ungroup(), but it prevents each level of summarization from being a self-contained set of computations.

The rtables call in contrast is:

lyt <- basic_table(show_colcounts = TRUE) %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
summarize_row_groups() %>%
split_rows_by("handed") %>%
summarize_row_groups() %>%
analyze("age", afun = mean, format = "xx.x")

tbl <- build_table(lyt, df)
tbl
#                     Arm A                     Arm B
#              Female        Male        Female        Male
#              (N=96)      (N=105)       (N=92)      (N=107)
# ————————————————————————————————————————————————————————————
# CAN        45 (46.9%)   64 (61.0%)   46 (50.0%)   62 (57.9%)
#   Left     32 (33.3%)   42 (40.0%)   26 (28.3%)   37 (34.6%)
#     mean      38.9         40.4         40.3         37.7
#   Right    13 (13.5%)   22 (21.0%)   20 (21.7%)   25 (23.4%)
#     mean      36.6         40.2         40.2         40.6
# USA        51 (53.1%)   41 (39.0%)   46 (50.0%)   45 (42.1%)
#   Left     34 (35.4%)   19 (18.1%)   25 (27.2%)   25 (23.4%)
#     mean      40.4         39.7         39.2         40.1
#   Right    17 (17.7%)   22 (21.0%)   21 (22.8%)   20 (18.7%)
#     mean      36.9         39.8         38.5         39.0

We can now spot check that the values are the same

frm_rtables_h <- cell_values(
tbl,
rowpath = c("country", "CAN", "handed", "Right", "@content"),
colpath = c("arm", "Arm B", "gender", "Female")
)[[1]]
frm_rtables_h
# [1] 20.0000000  0.2173913
frm_dplyr_h <- full_dplyr %>%
filter(country == "CAN" & handed == "Right" & arm == "Arm B" & gender == "Female") %>%
select(c_h_count, c_h_percent)

frm_dplyr_h
# # A tibble: 1 × 2
#   c_h_count c_h_percent
#       <int>       <dbl>
# 1        20       0.217
frm_rtables_c <- cell_values(
tbl,
rowpath = c("country", "CAN", "@content"),
colpath = c("arm", "Arm A", "gender", "Male")
)[[1]]

frm_rtables_c
# [1] 64.0000000  0.6095238
frm_dplyr_c <- full_dplyr %>%
filter(country == "CAN" & arm == "Arm A" & gender == "Male") %>%
select(c_count, c_percent)

frm_dplyr_c
# # A tibble: 2 × 2
#   c_count c_percent
#     <int>     <dbl>
# 1      64     0.610
# 2      64     0.610

Further, the rtable syntax has hopefully also become a bit more straightforward to derive the cell values than with dplyr for this particular table.

Summary

In this vignette learned that:

• many tables are quite easily created with dplyr and data.frame or tibble as data structure
• dplyr keeps simple things simple
• if tables have group summaries then repeating of information is required
• rtables streamlines the construction of complex tables

We recommend that you continue reading the clinical_trials vignette where we create a number of more advanced tables using layouts.