Get Started

Overview of TLG Programming

tidytlg provides a framework of creating TLG outputs for clinical study report. The TLG programming workflow includes the following steps:

• Prep environment: set up the R environment for the I/O paths.

• Process data: filter analysis data, perform data manipulation (e.g. convert character variable to factor), and define column variable.

• Generate results: create analysis rows of summary statistics (for tables) or plots (for graphs).

• Output results: output analysis results in designated format such as rtf or html.

We will illustrate the above steps by creating a demographic table first, and then follow by examples of creating listing and graph.

Prep environment

To set up the R environment, you can set the path objects of the input folder and output folder consistently for all TLG programs. The analysis datasets and other required inputs such as the titles file and column metadata file are placed in the input folder, while the output folder will be used to store the output files. The envsetup package can be used to set up the R environment for TLG programming.

# Prep Environment -------------------------------------------------------------------------------------
library(dplyr)
library(haven)
library(tidytlg)

# read adsl from PhUSE test data factory
testdata <- "https://github.com/phuse-org/TestDataFactory/raw/main/Updated/TDF_ADaM/"
adsl <- read_xpt(url(paste0(testdata,"adsl.xpt")))

Process data

Before generating analysis summary, the analysis data need to be processed first as shown in the code below.

# Process Data -----------------------------------------------------------------------------------------
adsl <- adsl %>%
  filter(ITTFL == "Y") %>%
  mutate(SEX = factor(SEX, levels = c("M", "F", "U"), labels = c("Male", "Female", "Unknown"))) %>% 
  tlgsetup(var = "TRT01PN",
           column_metadata_file = system.file("extdata/column_metadata.xlsx", package = "tidytlg"),
           tbltype = "type3")

The above code perform the tasks below:

• filtering analysis population

• convert the SEX variable from character type to factor type. So all the factor levels of SEX will be displayed in the analysis summary even there are no records for the factor level of “Unknown”.

• create the column variable, colnbr, through the tlgsetup function call: tlgsetup is using the numeric treatment variable (e.g. TRT01PN) to match with coldef in column metadata defined by column_metadata_file and tbltype to create the column variable, colnbr, in adsl for reflecting the column layout. Please see the vignette("tlgsetup") for more details. The column variable, colnbr, will be used in the subsequent analysis function calls for creating analysis results.

If you need multiple analysis datasets for creating TLG, tlgsetup will need to be applied to each dataset. Therefore, you will have a consistent column variable of colnbr for creating analysis summary.

Generate results

tidytlg provides 3 functions, univar, freq, and nested_freq, to generate analysis summary of descriptive statistics (univariate statistics and count (percentages)). For more details, please see the frequency analysis vignette("freq") and the univariate statistical analysis vignette("univar").

# Generate Results -------------------------------------------------------------------------------------

## Analysis set row
t1 <- adsl %>%
  freq(colvar = "colnbr",
       rowvar = "ITTFL",
       statlist = statlist("n"),
       subset = ITTFL == "Y",
       rowtext = "Analysis set: ITT") 

## Univariate summary for AGE
t2 <- adsl %>%
  univar(colvar = "colnbr",
         rowvar = "AGE",
         statlist = statlist(c("N", "MEANSD", "MEDIAN", "RANGE", "IQRANGE")),
         decimal = 0,
         row_header = "Age, years") 

## Count (percentages) for SEX
t3 <- adsl %>%
  freq(colvar = "colnbr",
       rowvar = "SEX",
       statlist = statlist(c("N","n (x.x%)")),
       row_header = "Gender")

The above function calls generate the requested analysis rows for the table output sequentially and store the results in individual objects (i.e. t1, t2, t3). The next step is to combine analysis results into a single tbl dataframe through the bind_table function call.

# Format Results ---------------------------------------------------------------------------------------

tbl <- bind_table(t1, t2, t3,
       column_metadata_file = system.file("extdata/column_metadata.xlsx", package = "tidytlg"),
       tbltype = "type3")

The above bind_table function call performs the following tasks:

• bind the analysis rows from t1, t2, t3 into tbl

• add formatting variables (indentme, newrows, newpage), which will be used in the gentlg function call below for creating the output.

• attach the column metadata specified by column_metadata_file and tbltype as an attribute of the tbl. So the column header and the spanning headers (i.e. decode, span1, span2, span3) defined in the column metadata can be used automatically in the gentlg function call.

Output results

The tbl data frame is the main input to the gentlg function for creating the RTF/HTML outputs.

The basic structure of tbl includes label, col1, col2, …, coln, where

• label: row text displayed on the 1st column of the table

• col1: statistic results displayed on the 2nd column of the table

• col2: statistic results displayed on the 3rd column of the table.

All other columns contain formatting instructions to create the RTF/HTML outputs. For tweaking the formatting variables to customize the table layout, please see the vignette("tbl_manipulation") for more details.

knitr::kable(tbl)

The gentlg function call below will create the rtf output using the tblid as the file name in the folder defined by the opath argument. Please ensure that the titles.xls file contains the records of titles and footnotes for the specified tblid.

tblid <- "Table01"

gentlg(huxme       = tbl,
       opath       = file.path(working_dir), 
       file        = tblid,
       orientation = "landscape",
       title_file = system.file("extdata/titles.xls", package = "tidytlg"))

To create the html output, users need to specify the format argument as “HTML” and print.hux argument as FALSE in the gentlg call.

gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = tblid,
       orientation = "landscape",
       title_file = system.file("extdata/titles.xls", package = "tidytlg"))

Users can also include superscripts, subscripts, or line breaks via unicode. Please see the vignette("symbols") for more details. Besides using univar, freq, and nested_freq functions to create the tbl dataframe, users can use other R packages to create analysis results and perform data wrangling to fit the tbl structure, which can be passed into the gentlg function call for generating the desired outputs.

Listing programming

The above workflow can also be used to create listings. Users need to prepare the data and assign it to tbl. In the gentlg function, users need to pay attention to:

• specify the tlf argument to Listing (i.e. tlf = "Listing")

• specify the idvars argument for identifying variables (such as treatment variable and USUBJID) where repeated values will be removed

• specify the colheader (column header) argument; if not specified, the column labels will be used as the column headers. For the below example, if colheader argument is not specified, some column headers will use variable names since these columns are newly created without labels.

• user has the option to control the column width by passing a vector of column width to the wcol argument. Please ensure that the length of the column width vector is the same as the number of columns in your data. For the example below, there are 8 columns in the data and users can specify customized column width such as c(0.15, 0.10, 0.05, 0.15, 0.20, 0.15, 0.05, 0.05) for the wcol argument to create the rtf output. However, for the html output shown here, the wcol argument can only take a single number as the column width and apply to every column.

# Prep Environment ---------------------------------------------------------------------------------------
library(dplyr)
library(haven)
library(tidytlg)

adsl <- cdisc_adsl
adae <- cdisc_adae

# Process Data --------------------------------------------------------------------------------------------
adsl <- adsl %>%
  filter(SAFFL == "Y") %>%
  select(USUBJID, SAFFL, TRT01AN, TRT01A) 

adae <- adae %>%
  filter(SAFFL == "Y" & TRTEMFL == "Y") %>%
  mutate(BSPT  = paste(AEBODSYS, "[", AEDECOD, "]"),
         SAEFL = if_else(AESER == "Y", "Yes", "No"),
         DTHFL = if_else(AEOUT == "FATAL", "Yes", "No")) %>%
   select(USUBJID, ASTDY, TRTA, BSPT, AETERM, SAEFL, DTHFL) 

tbl <- inner_join(adsl, adae, by = "USUBJID") %>%   
  arrange(TRT01AN, USUBJID, ASTDY) %>%   
  select(TRT01A, USUBJID, ASTDY, TRTA, BSPT, AETERM, SAEFL, DTHFL) %>% 
  filter(USUBJID %in% c("01-701-1015", "01-701-1023"))

# Output Results ------------------------------------------------------------------------------------------
gentlg(huxme       = tbl,
       tlf         = "l",
       format      = "HTML",
       print.hux = FALSE,
       orientation = "landscape",
       file        = "Listing01",
       title       = "Listing of Adverse Events",
       idvars      = c("TRT01A", "USUBJID"),
       wcol        = 0.15,
       colheader   = c("Treatment Group",
                       "Subject ID",
                       "Study Day of AE",
                       "Treatment Period",
                       "Body System [Preferred Term]",
                       "Verbatim Term",
                       "Serious",
                       "Fatal"))

Graph programming

To create the graph output, tidytlg provides a framework of integrating the png file with titles and footnotes for producing the rtf or html output.

In the gentlg function, users need to:

• specify the tlf argument to g for graph

• specify the plotnames argument with the full path of the png file

• define the plotwidth and plotheight: it’s advised here that users keep the aspect ratio of plot width and height approximately the same as the png image.

The code below will create the rtf output of the plot.

# Prep Environment ---------------------------------------------------------------------------------------
library(dplyr)
library(haven)
library(ggplot2)
#> 
#> Attaching package: 'ggplot2'
#> The following object is masked from 'package:huxtable':
#> 
#>     theme_grey
library(tidytlg)

# read adsl from PhUSE test data factory
testdata <- "https://github.com/phuse-org/TestDataFactory/raw/main/Updated/TDF_ADaM/"
adsl <- read_xpt(url(paste0(testdata,"adsl.xpt")))

tblid <- "Graph01"

# Process Data --------------------------------------------------------------------------------------------
adsl <- adsl  %>%
  filter(ITTFL == "Y") %>%
  select(USUBJID, ITTFL, TRT01PN, TRT01P, AGE, SEX, HEIGHTBL, WEIGHTBL) %>%
  mutate(SEX = factor(SEX, levels = c("M", "F"), labels = c("Male", "Female")))

# Generate Results ----------------------------------------------------------------------------------------
 
plot <- ggplot(data = adsl, aes(x = HEIGHTBL, y = WEIGHTBL)) +
  geom_point() +
  labs(x = "Baseline Height (cm)",
       y = "Baseline Weight (kg)") +
  facet_wrap(~SEX, nrow=1)

# create png file
png(file.path(working_dir, paste0(tblid,".png")), width=2800, height=1300, res=300, type = "cairo")

plot
#> Warning: Removed 1 rows containing missing values (`geom_point()`).

dev.off()
#> png 
#>   2

# Output Results ------------------------------------------------------------------------------------------

gentlg(tlf = "g",
       plotnames = file.path(system.file("extdata", package = "tidytlg"), paste0(tblid,".png")),
       plotwidth = 10,
       plotheight = 5,
       orientation = "landscape",
       opath       = file.path(working_dir),, 
       file = tblid,
       title_file = system.file("extdata/titles.xls", package = "tidytlg"))

Metadata method

Besides building the table section-by-section as shown above, we can use the table metadata approach as an efficient alternative for generating outputs. Table metadata is a data frame describing the data, functions and arguments needed to produce your table results. The table metadata shown below can be used to create the same table output as above. Each row in the table metadata describes how a tbl chunk will be created by the function defined in the func column. The rest of the columns defines the arguments (i.e. df, colvar, rowvar, statlist, rowtext, row_header) that will be passed into the function.

Once table metadata is defined, users just need to call the generate_results function with the column metadata define in the column_metadata_file and tbltype arguments to create the tbl dataframe. In the processing data step, users don’t need to call tlgsetp, since tlgsetup is embedded within the generate_results function. That’s why we need to specify the column metadata in the generate_results call.

library(dplyr)
library(haven)
library(tidytlg)

# read adsl from PhUSE test data factory
testdata <- "https://github.com/phuse-org/TestDataFactory/raw/main/Updated/TDF_ADaM/"
adsl <- read_xpt(url(paste0(testdata,"adsl.xpt")))

# Process data
adsl <- adsl %>%
  filter(ITTFL == "Y") %>%
  mutate(SEX = factor(SEX, levels = c("M", "F", "U"), labels = c("Male", "Female", "Unknown")))

# define table metadata
table_metadata <- tibble::tribble(
  ~func,     ~df,   ~rowvar, ~decimal, ~rowtext,     ~row_header, ~statlist,         ~subset,
  "freq",  "adsl",  "ITTFL",     NA, "Analysis set: ITT",     NA, statlist("n"),  "ITTFL == 'Y'",
  "univar", "adsl",   "AGE",      0,         NA, "Age (Years)",           NA,              NA,
  "freq",  "adsl",    "SEX",     NA,         NA, "Gender", statlist(c("N", "n (x.x%)")),   NA
) %>%
  mutate(colvar  = "TRT01PN")

# Generate results
tbl <- generate_results(table_metadata, 
                        column_metadata_file = system.file("extdata/column_metadata.xlsx", package = "tidytlg"),
                        tbltype = "type3") 

# Output results
tblid <- "Table01"

gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = tblid,
       orientation = "landscape",
       title_file = system.file("extdata/titles.xls", package = "tidytlg"))

By processing

There are two types of by-processing that tidytlg functions can provide:

• rowbyvar: split the summary statistics of rowvar by other variable(s) specified in rowbyvar. Please see the by processing section of frequency analysis vignette("freq") for further details. The rowbyvar argument can also be used in the univar function. A typical use case is to summarize the lab values by parameter and analysis visit, where we can call the univar function with rowvar = AVAL and rowbyvar = c("PARAM","AVISIT"). It is advised in this use case to turn on the .ord argument (i.e. .ord = TRUE) in the univar function call. So the numeric sorting columns associated with the by variables (PARAM_ord and AVISIT_ord) can be created and used for sorting the interleaved summary results of AVAL and CHG.

• tablebyvar: the argument of tablebyvar is designed to facilitate the sub-group analysis, which repeats a table summary by the sub-group variable. A typical use case is the summary of demographics table by country. For creating the sub-group analysis version of the same table, we just need to add the argument of tablebyvar with the sub-group variable in each function call.

The code below provides an example of summarizing age and race by gender using tablebyvar.

library(dplyr)
library(haven)
library(tidytlg)

# read adsl from PhUSE test data factory
testdata <- "https://github.com/phuse-org/TestDataFactory/raw/main/Updated/TDF_ADaM/"
adsl <- read_xpt(url(paste0(testdata,"adsl.xpt")))

# Process data
adsl <- adsl %>%
  filter(ITTFL == "Y") %>%
  mutate(SEX = factor(SEX, levels = c("M", "F"), labels = c("Male", "Female")))

# define table metadata
table_metadata <- tibble::tribble(
  ~func,     ~df,   ~rowvar, ~decimal, ~rowtext,     ~row_header, ~statlist,         ~subset, ~tablebyvar,
  "univar", "adsl",   "AGE",      0,         NA, "Age (Years)",           NA,              NA, "SEX",
  "freq",  "adsl",    "RACE",     NA,         NA, "Race", statlist(c("N", "n (x.x%)")),   NA, "SEX"
) %>%
  mutate(colvar  = "TRT01PN")

# Generate results
tbl <- generate_results(table_metadata, 
                        column_metadata_file = system.file("extdata/column_metadata.xlsx", package = "tidytlg"),
                        tbltype = "type3") 

# Output results
tblid <- "Table01"

gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = tblid,
       orientation = "landscape",
       title_file = system.file("extdata/titles.xls", package = "tidytlg"))

In summary, rowbyvar is used to create the by-variable summary for one rowvar in a single function call. To perform sub-group analysis, users need to specify tablebyvar in every function calls except the analysis population row.