rtables is an R package developed by Gabriel Becker & Adrian Waddell that is sponsored by the Roche Group and is available open source at github.com/Roche/rtables. The rtables package was initially designed by Adrian Waddell as a proof of concept and was used within Roche for tabulating clinical trials data. Starting with 2019, Gabriel Becker joined the rtables project and togehter we redesigned the package with focus on a more involved data structure and a more powerful tabulation framework. The redesign allows more general accessor on modifiers using pathing, better pagination, and a future feature allowing tiles, footnotes, and a cell referencing system. Further, the redesign improved the tabulation speed.
This document includes two vignettes, introduction and clinical_trials from the rtables package on the commit 83f653080cd5b21bc87b74c5701f664d474c1d74. These vignettes are supposed to give a good overview of the capability of rtables. There are other vignettes available in the package if one wants to get a deeper understanding of the rtables framework. You can install the version used for this document with:
devtools::install_github("Roche/rtables", ref = "83f653080cd5b21bc87b74c5701f664d474c1d74")The rtables package is currently not published on CRAN as we are in the process of refining some desgin details (mainly around pathing and visualizing the tree data structure) before submitting it to CRAN.
rtables outputs the tables to ASCII with the toString function and HTML with the as_html function. Note that it is currently not possible to add row gaps (empty rows or white spaces) when outputting the table. Row gaps are a feature that are neither essential for tabulation nor for designing the table data structure. Instead, most often the row gaps can be determined from the underlying table data structure by the outputting algorithm. However, specifying row gaps is a feature that is on our roadmap.
The rtables R package provides a framework to create, tabulate and output tables in R. Most of the design requirements for rtables have their origin in studying tables that are commonly used to report analyses from clinical trials; however, we were careful to keep rtables a general purpose toolkit.
There are a number of other table frameworks available in R such as gt from RStudio, xtable, tableone, and tables to name a few. There is a number of reasons to implement rtables (yet another tables R package):
rtables has two powerful tabulation frameworks: rtabulate and the layouting based tabulation frameworkIn the remainder of this section, we give a short introduction into rtables and tabulating a table. The content is based on the useR 2020 presentation from Gabriel Becker.
The packages used for this section are rtables and dplyr:
library(rtables)
library(dplyr)The data used in this section is a made up using random number generators. The data content is relatively simple: one row per imaginary person and one column per measurement: study arm, the country of origin, gender, handedness, age, and weight.
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)
)
head(df)# A tibble: 6 x 6
arm country gender handed age weight
<fct> <fct> <fct> <fct> <dbl> <dbl>
1 Arm A USA Female Left 31.3 139.
2 Arm B CAN Female Right 50.5 116.
3 Arm A USA Male Right 32.4 186.
4 Arm A USA Male Right 34.6 169.
5 Arm B USA Female Right 43.0 160.
6 Arm A USA Female Right 43.2 126.Note that we use factors variables so that the level order is represented in the row or column order when we tabulate the information of df below.
The aim of this section is to build the following table step by step:
Arm A Arm B
Female Male Female Male
(N=96) (N=105) (N=92) (N=107)
------------------------------------------------------------
CAN 45 (46.9%) 64 (61%) 46 (50%) 62 (57.9%)
Left 32 (33.3%) 42 (40%) 26 (28.3%) 37 (34.6%)
mean 38.9 40.4 40.3 37.7
Right 13 (13.5%) 22 (21%) 20 (21.7%) 25 (23.4%)
mean 36.6 40.2 40.2 40.6
USA 51 (53.1%) 41 (39%) 46 (50%) 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%) 21 (22.8%) 20 (18.7%)
mean 36.9 39.8 38.5 39 In rtables a basic table is defined to have 0 rows and one column representing all data. Analyzing a variable is one way of adding a row:
l <- basic_table() %>%
analyze("age", mean, format = "xx.x")
build_table(l, df) all obs
--------------
mean 39.4 In the code above we first described the table and assigned that description to a variable l. We then built the table using the actual data with build_table. The description of a table is called a table layout. basic_table is the start of every table layout and contains the information that we have one column representing all data. The analyze instruction adds to the layout that the age variable should be analyzed with the mean analysis function and the result should be rounded to 1 decimal place.
Hence, a layout is “pre-data”, that is, it’s a description of how to build a table once we get data. We can look at the layout isolated:
lA Pre-data Table Layout
Column-Split Structure:
( () -> () -> ) ()
Row-Split Structure:
age (** analysis **) The general layouting instructions are summarized below:
basic_table is a layout representing a table with zero rows and one columnsplit_rows_by, split_rows_by_multivar, split_rows_by_cuts, split_rows_by_cutfun, split_rows_by_quartilessplit_cols_by, split_cols_by_cuts, split_cols_by_cutfun, split_cols_by_quartilessummarize_row_groupsanalyze, analyze_against_baseline, analyze_colvars, analyze_row_groupsusing those functions it is possible to create a wide variety of tables as we will show in this document.
We will now add more structure to the columns by adding a column split based on the factor variable arm:
l <- basic_table() %>%
split_cols_by("arm") %>%
analyze("age", afun = mean, format = "xx.x")
build_table(l, df) Arm A Arm B
--------------------
mean 39.5 39.4 The resulting table has one column per factor level of arm. So the data represented by the first column is df[df$arm == "ARM A", ]. Hence, the split_cols_by partitions the data among the columns by default.
Column splitting can be done in a recursive/nested manner by adding sequential split_cols_by layout instruction. It’s also possible to add a non-nested split. Here we splitting each arm further by the gender:
l <- basic_table() %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
analyze("age", afun = mean, format = "xx.x")
build_table(l, df) Arm A Arm B
Female Male Female Male
------------------------------------
mean 38.8 40.1 39.6 39.2The first column represents the data in df where df$arm == "A" & df$gender == "Female" and the second column the data in df where df$arm == "A" & df$gender == "Male", and so on.
So far, we have created layouts with analysis and column splitting instructions, i.e. analyze and split_cols_by, respectively. This resulted with a table with multiple columns and one data row. We will add more row structure by stratifying the mean analysis by country (i.e. adding a split in the row space):
l <- basic_table() %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
analyze("age", afun = mean, format = "xx.x")
build_table(l, df) Arm A Arm B
Female Male Female Male
--------------------------------------
CAN
mean 38.2 40.3 40.3 38.9
USA
mean 39.2 39.7 38.9 39.6In this table the data used to derive the first data cell (average of age of female canadians in Arm A) is where df$country == "CAN" & df$arm == "Arm A" & df$gender == "Female". This cell value can also be calculated manually:
mean(df$age[df$country == "CAN" & df$arm == "Arm A" & df$gender == "Female"])[1] 38.22447When adding row splits we get by default label rows for each split level, for example CAN and USA in the table above. Besides the column space subsetting, we have now further subsetted the data for each cell. It is often useful when defining a row splitting to display information about each row group. In rtables this is referred to as content information, i.e. mean on row 2 is a descendant of CAN (visible via the indenting, though the table has an underlying tree structure that is not of importance for this section). In order to add content information and turn the CAN label row into a content row the summarize_row_groups function is required. By default, the count (nrows) and percentage of data relative to the column associated data is calculated:
l <- basic_table() %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
summarize_row_groups() %>%
analyze("age", afun = mean, format = "xx.x")
build_table(l, df) Arm A Arm B
Female Male Female Male
------------------------------------------------------
CAN 45 (46.9%) 64 (61%) 46 (50%) 62 (57.9%)
mean 38.2 40.3 40.3 38.9
USA 51 (53.1%) 41 (39%) 46 (50%) 45 (42.1%)
mean 39.2 39.7 38.9 39.6 The relative percentage for average age of female Canadians is calculated as follows:
df_cell <- subset(df, df$country == "CAN" & df$arm == "Arm A" & df$gender == "Female")
df_col_1 <- subset(df, df$arm == "Arm A" & df$gender == "Female")
c(count = nrow(df_cell), percentage = nrow(df_cell)/nrow(df_col_1)) count percentage
45.00000 0.46875 so the group percentages per row split sum up to 1 for each column.
We can further split the row space by dividing each country by handedness:
l <- basic_table() %>%
split_cols_by("arm") %>%
split_cols_by("gender") %>%
split_rows_by("country") %>%
summarize_row_groups() %>%
split_rows_by("handed") %>%
analyze("age", afun = mean, format = "xx.x")
build_table(l, df) Arm A Arm B
Female Male Female Male
--------------------------------------------------------
CAN 45 (46.9%) 64 (61%) 46 (50%) 62 (57.9%)
Left
mean 38.9 40.4 40.3 37.7
Right
mean 36.6 40.2 40.2 40.6
USA 51 (53.1%) 41 (39%) 46 (50%) 45 (42.1%)
Left
mean 40.4 39.7 39.2 40.1
Right
mean 36.9 39.8 38.5 39 Next, we further add a count and percentage summary for handedness within each country:
l <- basic_table() %>%
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")
build_table(l, df) Arm A Arm B
Female Male Female Male
------------------------------------------------------------
CAN 45 (46.9%) 64 (61%) 46 (50%) 62 (57.9%)
Left 32 (33.3%) 42 (40%) 26 (28.3%) 37 (34.6%)
mean 38.9 40.4 40.3 37.7
Right 13 (13.5%) 22 (21%) 20 (21.7%) 25 (23.4%)
mean 36.6 40.2 40.2 40.6
USA 51 (53.1%) 41 (39%) 46 (50%) 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%) 21 (22.8%) 20 (18.7%)
mean 36.9 39.8 38.5 39 In this section we create a
using the rtables layout facility. That is, we demonstrate how the layout based tabulation framework can specify the structure and relations that are commonly found when analyzing clinical trials data.
Note that all the data is created using random number generators. All ex_* data which is currently attached to the rtables package were created using random.cdisc.data another R package that we intend to release as open source soon.
The packages used in this section are:
library(rtables)
library(tibble)
library(dplyr)Demographic tables summarize the variables content for different population subsets (encoded in the columns).
One feature of analyze that we have not introduced in the previous section is that the analysis function afun can specify multiple rows with the in_rows function:
ADSL <- ex_adsl # Example ADSL dataset
basic_table() %>%
split_cols_by("ARM") %>%
analyze(vars = "AGE", afun = function(x) {
in_rows(
"Mean (sd)" = rcell(c(mean(x), sd(x)), format = "xx.xx (xx.xx)"),
"Range" = rcell(range(x), format = "xx.xx - xx.xx")
)
}) %>%
build_table(ADSL) A: Drug X B: Placebo C: Combination
-------------------------------------------------------
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
Range 21 - 50 21 - 62 20 - 69 Multiple variables can be analyzed in one analyze call:
basic_table() %>%
split_cols_by("ARM") %>%
analyze(vars = c("AGE", "BMRKR1"), afun = function(x) {
in_rows(
"Mean (sd)" = rcell(c(mean(x), sd(x)), format = "xx.xx (xx.xx)"),
"Range" = rcell(range(x), format = "xx.xx - xx.xx")
)
}) %>%
build_table(ADSL) A: Drug X B: Placebo C: Combination
----------------------------------------------------------
AGE
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
Range 21 - 50 21 - 62 20 - 69
BMRKR1
Mean (sd) 5.97 (3.55) 5.7 (3.31) 5.62 (3.49)
Range 0.41 - 17.67 0.65 - 14.24 0.17 - 21.39 Hence, if afun can process different data vector types (i.e. variables selected from the data) then we are fairly close to a standard demographic table. Here is a function that either creates a count table or some number summary if the argument x is a factor or numeric, respectively:
s_summary <- function(x) {
if (is.numeric(x)) {
in_rows(
"n" = rcell(sum(!is.na(x)), format = "xx"),
"Mean (sd)" = rcell(c(mean(x, na.rm = TRUE), sd(x, na.rm = TRUE)), format = "xx.xx (xx.xx)"),
"IQR" = rcell(IQR(x, na.rm = TRUE), format = "xx.xx"),
"min - max" = rcell(range(x, na.rm = TRUE), format = "xx.xx - xx.xx")
)
} else if (is.factor(x)) {
vs <- as.list(table(x))
do.call(in_rows, lapply(vs, rcell, format = "xx"))
} else (
stop("type not supported")
)
}Note we use rcells to wrap the results in order to add formatting instructions for rtables. We can use s_summary outside the context of tabulation:
s_summary(ADSL$AGE)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod row_label
1 n 400 0 n
2 Mean (sd) 34.88 (7.44) 0 Mean (sd)
3 IQR 10 0 IQR
4 min - max 20 - 69 0 min - maxand
s_summary(ADSL$SEX)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod row_label
1 F 222 0 F
2 M 166 0 M
3 U 9 0 U
4 UNDIFFERENTIATED 3 0 UNDIFFERENTIATEDWe can now create a commonly used variant of the demographic table:
lyt <- basic_table() %>%
split_cols_by(var = "ARM") %>%
analyze(c("AGE", "SEX"), afun = s_summary)
tbl <- build_table(lyt, ADSL)
tbl A: Drug X B: Placebo C: Combination
----------------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
SEX
F 79 77 66
M 51 55 60
U 3 2 4
UNDIFFERENTIATED 1 0 2 Note that analyze can also be called multiple times in sequence:
tbl2 <- basic_table() %>%
split_cols_by(var = "ARM") %>%
analyze("AGE", s_summary) %>%
analyze("SEX", s_summary) %>%
build_table(ADSL)
tbl2 A: Drug X B: Placebo C: Combination
----------------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
SEX
F 79 77 66
M 51 55 60
U 3 2 4
UNDIFFERENTIATED 1 0 2 which leads to the identical table as tbl:
identical(tbl, tbl2)[1] TRUEIn clinical trials analyses the number of patients per column is often referred to as N (rather than the overall population which outside of clinical trials is commonly referred to as N). Column Ns are added using the add_colcounts function:
basic_table() %>%
split_cols_by(var = "ARMCD") %>%
add_colcounts() %>%
analyze(c("AGE", "SEX"), s_summary) %>%
build_table(ADSL) ARM A ARM B ARM C
(N=134) (N=134) (N=132)
--------------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
SEX
F 79 77 66
M 51 55 60
U 3 2 4
UNDIFFERENTIATED 1 0 2 We will now show a couple of variations of the demographic table that we developed above. These variations are in structure and not in analysis, hence they don’t require a modification to the s_summary function.
We will start with a standard table analyzing the variables AGE and BMRKR2 variables:
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
analyze(c("AGE", "BMRKR2"), s_summary) %>%
build_table(ADSL) A: Drug X B: Placebo C: Combination
(N=134) (N=134) (N=132)
---------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
BMRKR2
LOW 50 45 40
MEDIUM 37 56 42
HIGH 47 33 50 Assume we would like to have this analysis carried out per gender encoded in the row space:
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX") %>%
analyze(c("AGE", "BMRKR2"), s_summary) %>%
build_table(ADSL) A: Drug X B: Placebo C: Combination
(N=134) (N=134) (N=132)
---------------------------------------------------------------
F
AGE
n 79 77 66
Mean (sd) 32.76 (6.09) 34.12 (7.06) 35.2 (7.43)
IQR 9 8 6.75
min - max 21 - 47 23 - 58 21 - 64
BMRKR2
LOW 26 21 26
MEDIUM 21 38 17
HIGH 32 18 23
M
AGE
n 51 55 60
Mean (sd) 35.57 (7.08) 37.44 (8.69) 35.38 (8.24)
IQR 11 9 11
min - max 23 - 50 21 - 62 20 - 69
BMRKR2
LOW 21 23 11
MEDIUM 15 18 23
HIGH 15 14 26
U
AGE
n 3 2 4
Mean (sd) 31.67 (3.21) 31 (5.66) 35.25 (3.1)
IQR 3 4 3.25
min - max 28 - 34 27 - 35 31 - 38
BMRKR2
LOW 2 1 1
MEDIUM 1 0 2
HIGH 0 1 1
UNDIFFERENTIATED
AGE
n 1 0 2
Mean (sd) 28 (NA) NaN (NA) 45 (1.41)
IQR 0 NA 1
min - max 28 - 28 Inf - -Inf 44 - 46
BMRKR2
LOW 1 0 2
MEDIUM 0 0 0
HIGH 0 0 0 We will now subset ADSL to include only males and females in the analysis in order to reduces the number of rows in the table:
ADSL_M_F <- filter(ADSL, SEX %in% c("M", "F"))
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX") %>%
analyze(c("AGE", "BMRKR2"), s_summary) %>%
build_table(ADSL_M_F) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
---------------------------------------------------------------
F
AGE
n 79 77 66
Mean (sd) 32.76 (6.09) 34.12 (7.06) 35.2 (7.43)
IQR 9 8 6.75
min - max 21 - 47 23 - 58 21 - 64
BMRKR2
LOW 26 21 26
MEDIUM 21 38 17
HIGH 32 18 23
M
AGE
n 51 55 60
Mean (sd) 35.57 (7.08) 37.44 (8.69) 35.38 (8.24)
IQR 11 9 11
min - max 23 - 50 21 - 62 20 - 69
BMRKR2
LOW 21 23 11
MEDIUM 15 18 23
HIGH 15 14 26
U
AGE
n 0 0 0
Mean (sd) NaN (NA) NaN (NA) NaN (NA)
IQR NA NA NA
min - max Inf - -Inf Inf - -Inf Inf - -Inf
BMRKR2
LOW 0 0 0
MEDIUM 0 0 0
HIGH 0 0 0
UNDIFFERENTIATED
AGE
n 0 0 0
Mean (sd) NaN (NA) NaN (NA) NaN (NA)
IQR NA NA NA
min - max Inf - -Inf Inf - -Inf Inf - -Inf
BMRKR2
LOW 0 0 0
MEDIUM 0 0 0
HIGH 0 0 0 Note that the UNDIFFERENTIATED and U levels still show up in the table. This is because tabulation respects the factor levels and level order, exactly as the split and table function do. If empty levels should be dropped then rtables needs to know that at splitting time via the split_fun argument in split_rows_by. There are a number of predefined functions. For this example drop_split_levels is required to drop the empty levels at splitting time. Splitting is a big topic and will be eventually addressed in a specific package vignette.
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", split_fun = drop_split_levels, child_labels = "visible") %>%
analyze(c("AGE", "BMRKR2"), s_summary) %>%
build_table(ADSL_M_F) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
------------------------------------------------------------
F
AGE
n 79 77 66
Mean (sd) 32.76 (6.09) 34.12 (7.06) 35.2 (7.43)
IQR 9 8 6.75
min - max 21 - 47 23 - 58 21 - 64
BMRKR2
LOW 26 21 26
MEDIUM 21 38 17
HIGH 32 18 23
M
AGE
n 51 55 60
Mean (sd) 35.57 (7.08) 37.44 (8.69) 35.38 (8.24)
IQR 11 9 11
min - max 23 - 50 21 - 62 20 - 69
BMRKR2
LOW 21 23 11
MEDIUM 15 18 23
HIGH 15 14 26 In the table above the labels M and F are not very descriptive. You can add the full label as follows:
ADSL_M_F_l <- ADSL_M_F %>%
mutate(lbl_sex = case_when(
SEX == "M" ~ "Male",
SEX == "F" ~ "Female",
SEX == "U" ~ "Unknown",
SEX == "UNDIFFERENTIATED" ~ "Undifferentiated"
))
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels, child_labels = "visible") %>%
analyze(c("AGE", "BMRKR2"), s_summary) %>%
build_table(ADSL_M_F_l) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
------------------------------------------------------------
Female
AGE
n 79 77 66
Mean (sd) 32.76 (6.09) 34.12 (7.06) 35.2 (7.43)
IQR 9 8 6.75
min - max 21 - 47 23 - 58 21 - 64
BMRKR2
LOW 26 21 26
MEDIUM 21 38 17
HIGH 32 18 23
Male
AGE
n 51 55 60
Mean (sd) 35.57 (7.08) 37.44 (8.69) 35.38 (8.24)
IQR 11 9 11
min - max 23 - 50 21 - 62 20 - 69
BMRKR2
LOW 21 23 11
MEDIUM 15 18 23
HIGH 15 14 26 For the next table variation we only stratify by gender for the AGE analysis. To do this the nested argument has to be set to FALSE in analyze call:
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels, child_labels = "visible") %>%
analyze("AGE", s_summary, show_labels = "visible") %>%
analyze("BMRKR2", s_summary, nested = FALSE, show_labels = "visible") %>%
build_table(ADSL_M_F_l) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
------------------------------------------------------------
Female
AGE
n 79 77 66
Mean (sd) 32.76 (6.09) 34.12 (7.06) 35.2 (7.43)
IQR 9 8 6.75
min - max 21 - 47 23 - 58 21 - 64
Male
AGE
n 51 55 60
Mean (sd) 35.57 (7.08) 37.44 (8.69) 35.38 (8.24)
IQR 11 9 11
min - max 23 - 50 21 - 62 20 - 69
BMRKR2
LOW 47 44 37
MEDIUM 36 56 40
HIGH 47 32 49 Once we split the rows into groups (Male and Female here) one might want to summarize groups: usually by showing count and column percentages. This is especially important if we have missing data. For example if we create the above table but add missing data to the AGE variable:
insert_NAs <- function(x) {
x[sample(c(TRUE, FALSE), length(x), TRUE, prob = c(0.2, 0.8))] <- NA
x
}
set.seed(1)
ADSL_NA <- ADSL_M_F_l %>%
mutate(AGE = insert_NAs(AGE))
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels, child_labels = "visible") %>%
analyze("AGE", s_summary) %>%
analyze("BMRKR2", s_summary, nested = FALSE, show_labels = "visible") %>%
build_table(filter(ADSL_NA, SEX %in% c("M", "F"))) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
----------------------------------------------------------
Female
n 65 61 54
Mean (sd) 32.71 (6.07) 34.33 (7.31) 34.61 (6.78)
IQR 9 10 6.75
min - max 21 - 47 23 - 58 21 - 54
Male
n 44 44 50
Mean (sd) 35.66 (6.78) 36.93 (8.18) 35.64 (8.42)
IQR 10.5 8.25 10.75
min - max 24 - 48 21 - 58 20 - 69
BMRKR2
LOW 47 44 37
MEDIUM 36 56 40
HIGH 47 32 49 Here it is not easy to see how many females and males there are in each arm as n represents the number of non-missing data elements in the variables. Groups within rows that are defined by splitting can be summarized with summarize_row_groups, for example:
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels) %>%
summarize_row_groups() %>%
analyze("AGE", s_summary) %>%
analyze("BMRKR2", afun = s_summary, nested = FALSE, show_labels = "visible") %>%
build_table(filter(ADSL_NA, SEX %in% c("M", "F"))) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
----------------------------------------------------------
Female 79 (60.8%) 77 (58.3%) 66 (52.4%)
n 65 61 54
Mean (sd) 32.71 (6.07) 34.33 (7.31) 34.61 (6.78)
IQR 9 10 6.75
min - max 21 - 47 23 - 58 21 - 54
Male 51 (39.2%) 55 (41.7%) 60 (47.6%)
n 44 44 50
Mean (sd) 35.66 (6.78) 36.93 (8.18) 35.64 (8.42)
IQR 10.5 8.25 10.75
min - max 24 - 48 21 - 58 20 - 69
BMRKR2
LOW 47 44 37
MEDIUM 36 56 40
HIGH 47 32 49 There are a couple of things to note here.
summarize_row_groups).We can recreate this default behavior (count percentage) by defining a cfun for illustrative purposes here as it results in the same table as above:
basic_table() %>%
split_cols_by(var = "ARM") %>%
add_colcounts() %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels) %>%
summarize_row_groups(cfun = function(df, labelstr, .N_col, ...) {
in_rows(
rcell(nrow(df) * c(1, 1/.N_col), format = "xx (xx.xx%)"),
.labels = labelstr
)
}) %>%
analyze("AGE", s_summary) %>%
analyze("BEP01FL", afun = s_summary, nested = FALSE, show_labels = "visible") %>%
build_table(filter(ADSL_NA, SEX %in% c("M", "F"))) A: Drug X B: Placebo C: Combination
(N=130) (N=132) (N=126)
----------------------------------------------------------
Female 79 (60.77%) 77 (58.33%) 66 (52.38%)
n 65 61 54
Mean (sd) 32.71 (6.07) 34.33 (7.31) 34.61 (6.78)
IQR 9 10 6.75
min - max 21 - 47 23 - 58 21 - 54
Male 51 (39.23%) 55 (41.67%) 60 (47.62%)
n 44 44 50
Mean (sd) 35.66 (6.78) 36.93 (8.18) 35.64 (8.42)
IQR 10.5 8.25 10.75
min - max 24 - 48 21 - 58 20 - 69
BEP01FL
Y 67 63 65
N 63 69 61 Note that cfun differs from afun (which is used in analyze) in that cfun does not operate on variables but rather on data.frames or tibbles which are passed via the df argument (afun can optionally request df too). Further, cfun gives the default group label (factor level from splitting) as an argument to labelstr and hence it could be modified:
basic_table() %>%
split_cols_by(var = "ARM") %>%
split_rows_by("SEX", labels_var = "lbl_sex", split_fun = drop_split_levels, child_labels = "hidden") %>%
summarize_row_groups(cfun = function(df, labelstr, .N_col, ...) {
in_rows(
rcell(nrow(df) * c(1, 1/.N_col), format = "xx (xx.xx%)"),
.labels = paste0(labelstr, ": count (perc.)")
)
}) %>%
analyze("AGE", s_summary) %>%
analyze("BEP01FL", s_summary, nested = FALSE, show_labels = "visible") %>%
build_table(filter(ADSL_NA, SEX %in% c("M", "F"))) A: Drug X B: Placebo C: Combination
--------------------------------------------------------------------
Female: count (perc.) 79 (60.77%) 77 (58.33%) 66 (52.38%)
n 65 61 54
Mean (sd) 32.71 (6.07) 34.33 (7.31) 34.61 (6.78)
IQR 9 10 6.75
min - max 21 - 47 23 - 58 21 - 54
Male: count (perc.) 51 (39.23%) 55 (41.67%) 60 (47.62%)
n 44 44 50
Mean (sd) 35.66 (6.78) 36.93 (8.18) 35.64 (8.42)
IQR 10.5 8.25 10.75
min - max 24 - 48 21 - 58 20 - 69
BEP01FL
Y 67 63 65
N 63 69 61 Layouts have a couple of advantages over tabulating the tables directly:
Here is an example that demonstrates the reusability of layouts:
lyt <- NULL %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze(c("AGE", "SEX"), afun = s_summary)
lytA Pre-data Table Layout
Column-Split Structure:
ARM (lvls)
Row-Split Structure:
( (** multivar analysis **) -> AGE, SEX (** multivar analysis **) -> ) (** multivar analysis **) We can now build a table for ADSL
build_table(lyt, ADSL) A: Drug X B: Placebo C: Combination
(N=134) (N=134) (N=132)
----------------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
SEX
F 79 77 66
M 51 55 60
U 3 2 4
UNDIFFERENTIATED 1 0 2 or for all patients that are older than 18:
build_table(lyt, ADSL %>% filter(AGE > 18)) A: Drug X B: Placebo C: Combination
(N=134) (N=134) (N=132)
----------------------------------------------------------------
AGE
n 134 134 132
Mean (sd) 33.77 (6.55) 35.43 (7.9) 35.43 (7.72)
IQR 11 10 10
min - max 21 - 50 21 - 62 20 - 69
SEX
F 79 77 66
M 51 55 60
U 3 2 4
UNDIFFERENTIATED 1 0 2 There are a number of different adverse event tables. We will now present two tables that show adverse events by id and then by grade and by id.
This time we won’t use the ADAE dataset from random.cdisc.data but rather generate a dataset on the fly (see Adrian’s 2016 Phuse paper):
set.seed(1)
lookup <- tribble(
~AEDECOD, ~AEBODSYS, ~AETOXGR,
'HEADACHE', "NERVOUS SYSTEM DISORDERS", "5",
'BACK PAIN', "MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS", "2",
'GINGIVAL BLEEDING', "GASTROINTESTINAL DISORDERS", "1",
'HYPOTENSION', "VASCULAR DISORDERS", "3",
'FAECES SOFT', "GASTROINTESTINAL DISORDERS", "2",
'ABDOMINAL DISCOMFORT', "GASTROINTESTINAL DISORDERS", "1",
'DIARRHEA', "GASTROINTESTINAL DISORDERS", "1",
'ABDOMINAL FULLNESS DUE TO GAS', "GASTROINTESTINAL DISORDERS", "1",
'NAUSEA (INTERMITTENT)', "GASTROINTESTINAL DISORDERS", "2",
'WEAKNESS', "MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS", "3",
'ORTHOSTATIC HYPOTENSION', "VASCULAR DISORDERS", "4"
)
normalize <- function(x) x/sum(x)
weightsA <- normalize(c(0.1, dlnorm(seq(0, 5, length.out = 25), meanlog = 3)))
weightsB <- normalize(c(0.2, dlnorm(seq(0, 5, length.out = 25))))
N_pop <- 300
ADSL2 <- data.frame(
USUBJID = seq(1, N_pop, by = 1),
ARM = sample(c('ARM A', 'ARM B'), N_pop, TRUE),
SEX = sample(c('F', 'M'), N_pop, TRUE),
AGE = 20 + rbinom(N_pop, size=40, prob=0.7)
)
l.adae <- mapply(ADSL2$USUBJID, ADSL2$ARM, ADSL2$SEX, ADSL2$AGE, FUN = function(id, arm, sex, age) {
n_ae <- sample(0:25, 1, prob = if (arm == "ARM A") weightsA else weightsB)
i <- sample(1:nrow(lookup), size = n_ae, replace = TRUE, prob = c(6, rep(1, 10))/16)
lookup[i, ] %>%
mutate(
AESEQ = seq_len(n()),
USUBJID = id, ARM = arm, SEX = sex, AGE = age
)
}, SIMPLIFY = FALSE)
ADAE2 <- do.call(rbind, l.adae)
ADAE2 <- ADAE2 %>%
mutate(
ARM = factor(ARM, levels = c("ARM A", "ARM B")),
AEDECOD = as.factor(AEDECOD),
AEBODSYS = as.factor(AEBODSYS),
AETOXGR = factor(AETOXGR, levels = as.character(1:5))
) %>%
select(USUBJID, ARM, AGE, SEX, AESEQ, AEDECOD, AEBODSYS, AETOXGR)
ADAE2# A tibble: 3,118 x 8
USUBJID ARM AGE SEX AESEQ AEDECOD AEBODSYS AETOXGR
<dbl> <fct> <dbl> <chr> <int> <fct> <fct> <fct>
1 1 ARM A 45 F 1 NAUSEA (INTERMIT… GASTROINTESTINAL D… 2
2 1 ARM A 45 F 2 HEADACHE NERVOUS SYSTEM DIS… 5
3 1 ARM A 45 F 3 HEADACHE NERVOUS SYSTEM DIS… 5
4 1 ARM A 45 F 4 HEADACHE NERVOUS SYSTEM DIS… 5
5 1 ARM A 45 F 5 HEADACHE NERVOUS SYSTEM DIS… 5
6 1 ARM A 45 F 6 HEADACHE NERVOUS SYSTEM DIS… 5
7 1 ARM A 45 F 7 HEADACHE NERVOUS SYSTEM DIS… 5
8 1 ARM A 45 F 8 HEADACHE NERVOUS SYSTEM DIS… 5
9 1 ARM A 45 F 9 HEADACHE NERVOUS SYSTEM DIS… 5
10 1 ARM A 45 F 10 FAECES SOFT GASTROINTESTINAL D… 2
# … with 3,108 more rowsWe start by defining an events summary function:
s_events_patients <- function(x, labelstr, .N_col) {
in_rows(
"Total number of patients with at least one event" =
rcell(length(unique(x)) * c(1, 1/.N_col), format = "xx (xx.xx%)"),
"Total number of events" = rcell(length(x), format = "xx")
)
}So, for a population of 5 patients where
we would get the following summary:
s_events_patients(x = c("id 1", "id 1", "id 2"), .N_col = 5)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod
1 Total number of patients with at least one event 2 (40%) 0
2 Total number of events 3 0
row_label
1 Total number of patients with at least one event
2 Total number of eventsThe .N_col argument is a special keyword argument which build_table passes the population size for each respective column. For a list of keyword arguments for the functions passed to afun in analyze refer to the documentation with ?analyze.
We now use the s_events_patients summary function in a tabulation:
basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze("USUBJID", s_events_patients) %>%
build_table(ADAE2) ARM A ARM B
(N=2060) (N=1058)
-----------------------------------------------------------------------------
Total number of patients with at least one event 114 (5.53%) 150 (14.18%)
Total number of events 2060 1058 Note that the column N’s are wrong as by default they are set to the number of rows per group (i.e. number of AEs per arm here). This also affects the percentages. For this table we are interested in the number of patients per column/arm which is usually taken from ADSL (variable ADSL2 here):
N_per_arm <- table(ADSL2$ARM)
N_per_arm
ARM A ARM B
146 154 Since this information is not “pre-data” it needs to go to the table creation function build_table:
basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze("USUBJID", s_events_patients) %>%
build_table(ADAE2, col_counts = N_per_arm) ARM A ARM B
(N=146) (N=154)
-----------------------------------------------------------------------------
Total number of patients with at least one event 114 (78.08%) 150 (97.4%)
Total number of events 2060 1058 We next calculate this information per system organ class:
l <- basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze("USUBJID", s_events_patients) %>%
split_rows_by("AEBODSYS", child_labels = "visible", nested = FALSE) %>%
summarize_row_groups("USUBJID", cfun = s_events_patients)
build_table(l, ADAE2, col_counts = N_per_arm) ARM A ARM B
(N=146) (N=154)
--------------------------------------------------------------------------------
Total number of patients with at least one event 114 (78.08%) 150 (97.4%)
Total number of events 2060 1058
GASTROINTESTINAL DISORDERS
Total number of patients with at least one event 114 (78.08%) 130 (84.42%)
Total number of events 760 374
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
Total number of patients with at least one event 98 (67.12%) 81 (52.6%)
Total number of events 273 142
NERVOUS SYSTEM DISORDERS
Total number of patients with at least one event 113 (77.4%) 133 (86.36%)
Total number of events 787 420
VASCULAR DISORDERS
Total number of patients with at least one event 93 (63.7%) 75 (48.7%)
Total number of events 240 122 We now have to the add a count table of AEDECOD for each AEBODSYS. The default analyze behavior for a factor is to create the count table per level (using rtab_inner):
tbl1 <- basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
split_rows_by("AEBODSYS", child_labels = "visible", indent_mod = 1) %>%
summarize_row_groups("USUBJID", cfun = s_events_patients) %>%
analyze("AEDECOD", indent_mod = -1) %>%
build_table(ADAE2, col_counts = N_per_arm)
tbl1 ARM A ARM B
(N=146) (N=154)
--------------------------------------------------------------------------------
GASTROINTESTINAL DISORDERS
Total number of patients with at least one event 114 (78.08%) 130 (84.42%)
Total number of events 760 374
ABDOMINAL DISCOMFORT 113 65
ABDOMINAL FULLNESS DUE TO GAS 119 65
BACK PAIN 0 0
DIARRHEA 107 53
FAECES SOFT 122 58
GINGIVAL BLEEDING 147 71
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 152 62
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
Total number of patients with at least one event 98 (67.12%) 81 (52.6%)
Total number of events 273 142
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 135 75
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 138 67
NERVOUS SYSTEM DISORDERS
Total number of patients with at least one event 113 (77.4%) 133 (86.36%)
Total number of events 787 420
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 787 420
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
VASCULAR DISORDERS
Total number of patients with at least one event 93 (63.7%) 75 (48.7%)
Total number of events 240 122
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 104 58
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 136 64
WEAKNESS 0 0 The indent_mod argument enables relative indenting changes if the tree structure of the table does not result in the desired indentation by default.
This table so far is however not the usual adverse event table as it counts the total number of events and not the number of subjects one or more events for a particular term. To get the correct table we need to write a custom analysis function:
table_count_once_per_id <- function(df, termvar = "AEDECOD", idvar = "USUBJID") {
x <- df[[termvar]]
id <- df[[idvar]]
counts <- table(x[!duplicated(id)])
in_rows(
.list = as.vector(counts),
.labels = names(counts)
)
}
table_count_once_per_id(ADAE2)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod
1 ABDOMINAL DISCOMFORT 23 0
2 ABDOMINAL FULLNESS DUE TO GAS 21 0
3 BACK PAIN 20 0
4 DIARRHEA 7 0
5 FAECES SOFT 11 0
6 GINGIVAL BLEEDING 15 0
7 HEADACHE 100 0
8 HYPOTENSION 16 0
9 NAUSEA (INTERMITTENT) 21 0
10 ORTHOSTATIC HYPOTENSION 14 0
11 WEAKNESS 16 0
row_label
1 ABDOMINAL DISCOMFORT
2 ABDOMINAL FULLNESS DUE TO GAS
3 BACK PAIN
4 DIARRHEA
5 FAECES SOFT
6 GINGIVAL BLEEDING
7 HEADACHE
8 HYPOTENSION
9 NAUSEA (INTERMITTENT)
10 ORTHOSTATIC HYPOTENSION
11 WEAKNESSSo the desired AE table is:
basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
split_rows_by("AEBODSYS", child_labels = "visible", indent_mod = 1) %>%
summarize_row_groups("USUBJID", cfun = s_events_patients) %>%
analyze("AEDECOD", afun = table_count_once_per_id, show_labels = "hidden", indent_mod = -1) %>%
build_table(ADAE2, col_counts = N_per_arm) ARM A ARM B
(N=146) (N=154)
--------------------------------------------------------------------------------
GASTROINTESTINAL DISORDERS
Total number of patients with at least one event 114 (78.08%) 130 (84.42%)
Total number of events 760 374
ABDOMINAL DISCOMFORT 24 28
ABDOMINAL FULLNESS DUE TO GAS 18 26
BACK PAIN 0 0
DIARRHEA 17 17
FAECES SOFT 17 14
GINGIVAL BLEEDING 18 25
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 20 20
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
Total number of patients with at least one event 98 (67.12%) 81 (52.6%)
Total number of events 273 142
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 58 45
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 40 36
NERVOUS SYSTEM DISORDERS
Total number of patients with at least one event 113 (77.4%) 133 (86.36%)
Total number of events 787 420
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 113 133
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
VASCULAR DISORDERS
Total number of patients with at least one event 93 (63.7%) 75 (48.7%)
Total number of events 240 122
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 44 31
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 49 44
WEAKNESS 0 0 Note that we are missing the overall summary in the first two rows. This can be added with another analyze call and then setting nested to FALSE in the subsequent summarize_row_groups call:
tbl <- basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze("USUBJID", afun = s_events_patients) %>%
split_rows_by("AEBODSYS", child_labels = "visible", nested = FALSE, indent_mod = 1) %>%
summarize_row_groups("USUBJID", cfun = s_events_patients) %>%
analyze("AEDECOD", table_count_once_per_id, show_labels = "hidden", indent_mod = -1) %>%
build_table(ADAE2, col_counts = N_per_arm)
tbl ARM A ARM B
(N=146) (N=154)
--------------------------------------------------------------------------------
Total number of patients with at least one event 114 (78.08%) 150 (97.4%)
Total number of events 2060 1058
GASTROINTESTINAL DISORDERS
Total number of patients with at least one event 114 (78.08%) 130 (84.42%)
Total number of events 760 374
ABDOMINAL DISCOMFORT 24 28
ABDOMINAL FULLNESS DUE TO GAS 18 26
BACK PAIN 0 0
DIARRHEA 17 17
FAECES SOFT 17 14
GINGIVAL BLEEDING 18 25
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 20 20
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
Total number of patients with at least one event 98 (67.12%) 81 (52.6%)
Total number of events 273 142
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 58 45
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 40 36
NERVOUS SYSTEM DISORDERS
Total number of patients with at least one event 113 (77.4%) 133 (86.36%)
Total number of events 787 420
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 113 133
HYPOTENSION 0 0
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 0 0
WEAKNESS 0 0
VASCULAR DISORDERS
Total number of patients with at least one event 93 (63.7%) 75 (48.7%)
Total number of events 240 122
ABDOMINAL DISCOMFORT 0 0
ABDOMINAL FULLNESS DUE TO GAS 0 0
BACK PAIN 0 0
DIARRHEA 0 0
FAECES SOFT 0 0
GINGIVAL BLEEDING 0 0
HEADACHE 0 0
HYPOTENSION 44 31
NAUSEA (INTERMITTENT) 0 0
ORTHOSTATIC HYPOTENSION 49 44
WEAKNESS 0 0 Finally, if we wanted to prune the 0 counts row we can do that with the trim_rows function:
trim_rows(tbl) ARM A ARM B
(N=146) (N=154)
--------------------------------------------------------------------------------
Total number of patients with at least one event 114 (78.08%) 150 (97.4%)
Total number of events 2060 1058
GASTROINTESTINAL DISORDERS
Total number of patients with at least one event 114 (78.08%) 130 (84.42%)
Total number of events 760 374
ABDOMINAL DISCOMFORT 24 28
ABDOMINAL FULLNESS DUE TO GAS 18 26
DIARRHEA 17 17
FAECES SOFT 17 14
GINGIVAL BLEEDING 18 25
NAUSEA (INTERMITTENT) 20 20
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
Total number of patients with at least one event 98 (67.12%) 81 (52.6%)
Total number of events 273 142
BACK PAIN 58 45
WEAKNESS 40 36
NERVOUS SYSTEM DISORDERS
Total number of patients with at least one event 113 (77.4%) 133 (86.36%)
Total number of events 787 420
HEADACHE 113 133
VASCULAR DISORDERS
Total number of patients with at least one event 93 (63.7%) 75 (48.7%)
Total number of events 240 122
HYPOTENSION 44 31
ORTHOSTATIC HYPOTENSION 49 44 Pruning is a larger topic with a separate rtables package vignette.
The adverse events table by ID and by grade shows how many patients had at least one adverse event per grade for different subsets of the data (e.g. defined by system organ class).
For this table we do not show the zero count grades. Note that we add the “overall” groups with a custom split function.
table_count_grade_once_per_id <- function(df, labelstr = "", gradevar = "AETOXGR", idvar = "USUBJID", grade_levels = NULL) {
id <- df[[idvar]]
grade <- df[[gradevar]]
if (!is.null(grade_levels)) {
stopifnot(all(grade %in% grade_levels))
grade <- factor(grade, levels = grade_levels)
}
id_sel <- !duplicated(id)
in_rows(
"--Any Grade--" = sum(id_sel),
.list = as.list(table(grade[id_sel]))
)
}
table_count_grade_once_per_id(ex_adae, grade_levels = 1:5)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod row_label
1 --Any Grade-- 365 0 --Any Grade--
2 1 131 0 1
3 2 70 0 2
4 3 74 0 3
5 4 25 0 4
6 5 65 0 5All of the layouting concepts needed to create this table have already been introduced so far:
basic_table() %>%
split_cols_by("ARM") %>%
add_colcounts() %>%
analyze("AETOXGR",
afun = table_count_grade_once_per_id,
extra_args = list(grade_levels = 1:5),
var_labels = "- Any adverse events -", show_labels = "visible") %>%
split_rows_by("AEBODSYS", child_labels = "visible", nested = FALSE, indent_mod = 1) %>%
summarize_row_groups(cfun = table_count_grade_once_per_id, format = "xx", indent_mod = 1) %>%
split_rows_by("AEDECOD", child_labels = "visible", indent_mod = -2) %>%
analyze("AETOXGR",
afun = table_count_grade_once_per_id,
extra_args = list(grade_levels = 1:5), show_labels = "hidden") %>%
build_table(ADAE2, col_counts = N_per_arm) ARM A ARM B
(N=146) (N=154)
-------------------------------------------------------------------
- Any adverse events -
--Any Grade-- 114 150
1 32 34
2 22 30
3 11 21
4 8 6
5 41 59
GASTROINTESTINAL DISORDERS
--Any Grade-- 114 130
1 77 96
2 37 34
3 0 0
4 0 0
5 0 0
ABDOMINAL DISCOMFORT
--Any Grade-- 68 49
1 68 49
2 0 0
3 0 0
4 0 0
5 0 0
ABDOMINAL FULLNESS DUE TO GAS
--Any Grade-- 73 51
1 73 51
2 0 0
3 0 0
4 0 0
5 0 0
BACK PAIN
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
DIARRHEA
--Any Grade-- 68 40
1 68 40
2 0 0
3 0 0
4 0 0
5 0 0
FAECES SOFT
--Any Grade-- 76 44
1 0 0
2 76 44
3 0 0
4 0 0
5 0 0
GINGIVAL BLEEDING
--Any Grade-- 80 52
1 80 52
2 0 0
3 0 0
4 0 0
5 0 0
HEADACHE
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
NAUSEA (INTERMITTENT)
--Any Grade-- 83 50
1 0 0
2 83 50
3 0 0
4 0 0
5 0 0
ORTHOSTATIC HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
WEAKNESS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
MUSCULOSKELETAL AND CONNECTIVE TISSUE DISORDERS
--Any Grade-- 98 81
1 0 0
2 58 45
3 40 36
4 0 0
5 0 0
ABDOMINAL DISCOMFORT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ABDOMINAL FULLNESS DUE TO GAS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
BACK PAIN
--Any Grade-- 79 62
1 0 0
2 79 62
3 0 0
4 0 0
5 0 0
DIARRHEA
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
FAECES SOFT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
GINGIVAL BLEEDING
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HEADACHE
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
NAUSEA (INTERMITTENT)
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ORTHOSTATIC HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
WEAKNESS
--Any Grade-- 73 43
1 0 0
2 0 0
3 73 43
4 0 0
5 0 0
NERVOUS SYSTEM DISORDERS
--Any Grade-- 113 133
1 0 0
2 0 0
3 0 0
4 0 0
5 113 133
ABDOMINAL DISCOMFORT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ABDOMINAL FULLNESS DUE TO GAS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
BACK PAIN
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
DIARRHEA
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
FAECES SOFT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
GINGIVAL BLEEDING
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HEADACHE
--Any Grade-- 113 133
1 0 0
2 0 0
3 0 0
4 0 0
5 113 133
HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
NAUSEA (INTERMITTENT)
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ORTHOSTATIC HYPOTENSION
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
WEAKNESS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
VASCULAR DISORDERS
--Any Grade-- 93 75
1 0 0
2 0 0
3 44 31
4 49 44
5 0 0
ABDOMINAL DISCOMFORT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ABDOMINAL FULLNESS DUE TO GAS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
BACK PAIN
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
DIARRHEA
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
FAECES SOFT
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
GINGIVAL BLEEDING
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HEADACHE
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
HYPOTENSION
--Any Grade-- 66 43
1 0 0
2 0 0
3 66 43
4 0 0
5 0 0
NAUSEA (INTERMITTENT)
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
ORTHOSTATIC HYPOTENSION
--Any Grade-- 70 54
1 0 0
2 0 0
3 0 0
4 70 54
5 0 0
WEAKNESS
--Any Grade-- 0 0
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0 The response table that we will create here is composed of 3 parts:
Let’s start with the first part which is fairly simple to derive:
ADRS_BESRSPI <- ex_adrs %>%
filter(PARAMCD == "BESRSPI") %>%
mutate(
rsp = factor(AVALC %in% c("CR", "PR"), levels = c(TRUE, FALSE), labels = c("Responders", "Non-Responders")),
is_rsp = (rsp == "Responders")
)
s_proportion <- function(x, .N_col) {
in_rows(.list = lapply(as.list(table(x)), function(xi) rcell(xi * c(1, 1/.N_col), format = "xx.xx (xx.xx%)")))
}
basic_table() %>%
split_cols_by("ARMCD", ref_group = "ARM A") %>%
add_colcounts() %>%
analyze("rsp", s_proportion, show_labels = "hidden") %>%
build_table(ADRS_BESRSPI) ARM A ARM B ARM C
(N=134) (N=134) (N=132)
----------------------------------------------------------
Responders 114 (85.07%) 90 (67.16%) 120 (90.91%)
Non-Responders 20 (14.93%) 44 (32.84%) 12 (9.09%) Note that we did set the ref_group argument in split_cols_by which for the current table had no effect as we only use the cell data for the responder and non-responder counting. The ref_group argument is needed for the part 2. and 3. of the table.
We will now look the implementation of part “2. Unstratified analysis comparison vs. control group.” Let’s start with the analysis function:
s_unstratified_response_analysis <- function(x, .ref_group, .in_ref_col) {
if (.in_ref_col) {
return(in_rows(
"Difference in Response Rates (%)" = rcell(numeric(0)),
"95% CI (Wald, with correction)" = rcell(numeric(0)),
"p-value (Chi-Squared Test)" = rcell(numeric(0)),
"Odds Ratio (95% CI)" = rcell(numeric(0))
))
}
fit <- stats::prop.test(
x = c(sum(x), sum(.ref_group)),
n = c(length(x), length(.ref_group)),
correct = FALSE
)
fit_glm <- stats::glm(
formula = rsp ~ group,
data = data.frame(
rsp = c(.ref_group, x),
group = factor(rep(c("ref", "x"), times = c(length(.ref_group), length(x))), levels = c("ref", "x"))
),
family = binomial(link = "logit")
)
in_rows(
"Difference in Response Rates (%)" = non_ref_rcell((mean(x) - mean(.ref_group)) * 100,
.in_ref_col, format = "xx.xx") ,
"95% CI (Wald, with correction)" = non_ref_rcell(fit$conf.int * 100,
.in_ref_col, format = "(xx.xx, xx.xx)"),
"p-value (Chi-Squared Test)" = non_ref_rcell(fit$p.value,
.in_ref_col, format = "x.xxxx | (<0.0001)"),
"Odds Ratio (95% CI)" = non_ref_rcell(c(
exp(stats::coef(fit_glm)[-1]),
exp(stats::confint.default(fit_glm, level = .95)[-1, , drop = FALSE])
),
.in_ref_col, format = "xx.xx (xx.xx - xx.xx)")
)
}
s_unstratified_response_analysis(
x = ADRS_BESRSPI %>% filter(ARM == "A: Drug X") %>% pull(is_rsp),
.ref_group = ADRS_BESRSPI %>% filter(ARM == "B: Placebo") %>% pull(is_rsp),
.in_ref_col = FALSE
)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod
1 Difference in Response Rates (%) 17.91 0
2 95% CI (Wald, with correction) (7.93, 27.89) 0
3 p-value (Chi-Squared Test) 0.0006 0
4 Odds Ratio (95% CI) 2.79 (1.53 - 5.06) 0
row_label
1 Difference in Response Rates (%)
2 95% CI (Wald, with correction)
3 p-value (Chi-Squared Test)
4 Odds Ratio (95% CI)Hence we can now add the next section to the table:
basic_table() %>%
split_cols_by("ARMCD", ref_group = "ARM A") %>%
add_colcounts() %>%
analyze("rsp", s_proportion, show_labels = "hidden") %>%
analyze("is_rsp", s_unstratified_response_analysis, show_labels = "visible", var_labels = "Unstratified Response Analysis") %>%
build_table(ADRS_BESRSPI) ARM A ARM B ARM C
(N=134) (N=134) (N=132)
------------------------------------------------------------------------------------------
Responders 114 (85.07%) 90 (67.16%) 120 (90.91%)
Non-Responders 20 (14.93%) 44 (32.84%) 12 (9.09%)
Unstratified Response Analysis
Difference in Response Rates (%) -17.91 5.83
95% CI (Wald, with correction) (-27.89, -7.93) (-1.94, 13.61)
p-value (Chi-Squared Test) 0.0006 0.1436
Odds Ratio (95% CI) 0.36 (0.2 - 0.65) 1.75 (0.82 - 3.75)Next we will add part 3. the “multinomial response table”. To do so, we are adding a row-split by response level, and then doing the same thing as we did for the binary response table above.
s_prop <- function(df, .N_col) {
in_rows(
"95% CI (Wald, with correction)" = rcell(binom.test(nrow(df), .N_col)$conf.int * 100, format = "(xx.xx, xx.xx)")
)
}
s_prop(
df = ADRS_BESRSPI %>% filter(ARM == "A: Drug X", AVALC == "CR"),
.N_col = sum(ADRS_BESRSPI$ARM == "A: Drug X")
)in_rows object print method:
----------------------------
row_name formatted_cell indent_mod
1 95% CI (Wald, with correction) (49.38, 66.67) 0
row_label
1 95% CI (Wald, with correction)We can now create the final response table with all three parts:
basic_table() %>%
split_cols_by("ARMCD", ref_group = "ARM A") %>%
add_colcounts() %>%
analyze("rsp", s_proportion, show_labels = "hidden") %>%
analyze("is_rsp", s_unstratified_response_analysis,
show_labels = "visible", var_labels = "Unstratified Response Analysis") %>%
split_rows_by(
var = "AVALC",
split_fun = reorder_split_levels(neworder = c("CR", "PR", "SD", "NON CR/PD", "PD", "NE"), drlevels = TRUE),
nested = FALSE
) %>%
summarize_row_groups() %>%
analyze("AVALC", afun = s_prop) %>%
build_table(ADRS_BESRSPI) ARM A ARM B ARM C
(N=134) (N=134) (N=132)
--------------------------------------------------------------------------------------------
Responders 114 (85.07%) 90 (67.16%) 120 (90.91%)
Non-Responders 20 (14.93%) 44 (32.84%) 12 (9.09%)
Unstratified Response Analysis
Difference in Response Rates (%) -17.91 5.83
95% CI (Wald, with correction) (-27.89, -7.93) (-1.94, 13.61)
p-value (Chi-Squared Test) 0.0006 0.1436
Odds Ratio (95% CI) 0.36 (0.2 - 0.65) 1.75 (0.82 - 3.75)
CR 78 (58.2%) 55 (41%) 97 (73.5%)
95% CI (Wald, with correction) (49.38, 66.67) (32.63, 49.87) (65.1, 80.79)
PR 36 (26.9%) 35 (26.1%) 23 (17.4%)
95% CI (Wald, with correction) (19.58, 35.2) (18.92, 34.41) (11.38, 24.99)
SD 20 (14.9%) 44 (32.8%) 12 (9.1%)
95% CI (Wald, with correction) (9.36, 22.11) (24.97, 41.47) (4.79, 15.34) In case the we wanted to rename the levels of AVALC and remove the CI for NE we could do that as follows:
rsp_label <- function(x) {
rsp_full_label <- c(
CR = "Complete Response (CR)",
PR = "Partial Response (PR)",
SD = "Stable Disease (SD)",
`NON CR/PD` = "Non-CR or Non-PD (NON CR/PD)",
PD = "Progressive Disease (PD)",
NE = "Not Evaluable (NE)",
Missing = "Missing",
`NE/Missing` = "Missing or unevaluable"
)
stopifnot(all(x %in% names(rsp_full_label)))
rsp_full_label[x]
}
tbl <- basic_table() %>%
split_cols_by("ARMCD", ref_group = "ARM A") %>%
add_colcounts() %>%
analyze("rsp", s_proportion, show_labels = "hidden") %>%
analyze("is_rsp", s_unstratified_response_analysis,
show_labels = "visible", var_labels = "Unstratified Response Analysis") %>%
split_rows_by(
var = "AVALC",
split_fun = keep_split_levels(c("CR", "PR", "SD", "NON CR/PD", "PD"), reorder = TRUE),
nested = FALSE
) %>%
summarize_row_groups(cfun = function(df, labelstr, .N_col) {
in_rows(nrow(df) * c(1, 1/.N_col), .formats = "xx (xx.xx%)", .labels = rsp_label(labelstr))
}) %>%
analyze("AVALC", afun = s_prop) %>%
analyze("AVALC", afun = function(x, .N_col) {
in_rows(rcell(sum(x == "NE") * c(1, 1/.N_col), format = "xx.xx (xx.xx%)"), .labels = rsp_label("NE"))
}, nested = FALSE) %>%
build_table(ADRS_BESRSPI)
tbl ARM A ARM B ARM C
(N=134) (N=134) (N=132)
--------------------------------------------------------------------------------------------
Responders 114 (85.07%) 90 (67.16%) 120 (90.91%)
Non-Responders 20 (14.93%) 44 (32.84%) 12 (9.09%)
Unstratified Response Analysis
Difference in Response Rates (%) -17.91 5.83
95% CI (Wald, with correction) (-27.89, -7.93) (-1.94, 13.61)
p-value (Chi-Squared Test) 0.0006 0.1436
Odds Ratio (95% CI) 0.36 (0.2 - 0.65) 1.75 (0.82 - 3.75)
Complete Response (CR) 78 (58.21%) 55 (41.04%) 97 (73.48%)
95% CI (Wald, with correction) (49.38, 66.67) (32.63, 49.87) (65.1, 80.79)
Partial Response (PR) 36 (26.87%) 35 (26.12%) 23 (17.42%)
95% CI (Wald, with correction) (19.58, 35.2) (18.92, 34.41) (11.38, 24.99)
Stable Disease (SD) 20 (14.93%) 44 (32.84%) 12 (9.09%)
95% CI (Wald, with correction) (9.36, 22.11) (24.97, 41.47) (4.79, 15.34)
Progressive Disease (PD) 0 (0%) 0 (0%) 0 (0%)
95% CI (Wald, with correction) (0, 2.72) (0, 2.72) (0, 2.76)
Not Evaluable (NE) 0 (0%) 0 (0%) 0 (0%) Note that the table is missing the rows gaps to make it more readable. The row spacing feature is on the rtables roadmap and will be implemented in future.
The table topic poses a rich set of problems on its own right including but not only: table data structures, tabulation, outputting, formatting, and table processing. We are still actively working on rtables and expect that in the next year the rtables framework keeps evolving to meet all requirements for submitting clinical trials data analyses in a regulatory context and we also hope that our framework proves to be useful for other industries that rely on visualizing the data with tables.
We would like to thank Roche for financing the rtables project and allowing to be developed open source. Further, we would also like to thank the NEST project (at Roche) team members for their valuable feedback and involvement in the refinement of rtables. That is, many thanks go to Tadeusz Lewandowski who is the NEST business lead, and to the subject matter expert team members: Nick Paszty, Jana Stoilova, Heng Wang, Francois Collin, Daniel Sabanés Bové, and Nina Qi.