See https://github.com/openpharma/visR/runs/681093111?check_suite_focus=true

Will try and knock those down a bit in this issue

I was asked by @epijim if I could contribute in this project and that's why I would ask you about such possibility. @cschaerfe @diego-s @bailliem if these functions already initiated like for example: vr_plt_kaplan_meier.R and vr_sankey.R are already in progress / during implementation or not? If not and this assignments in comments are not finally set then I would be happy to work on this. Or maybe do you have any other kind of visualization-functions or other job to do which is still not initiated?

Issue description

I'm hoping to generate kaplan-meier plots/tables without stratifying my population. My equation looks like this:
surv_eq <- "survival::Surv(time, status)~1"

Steps to reproduce the issue

Please share code if possible.

surv_eq <- "survival::Surv(time, status)~1"

vr_kaplan_meier_summary(km_dt, surv_eq)

vr_kaplan_meier(km_dt, surv_eq)

What's the expected result?

I get back a summary table if I supply "ms_subtype" rather than "1" as the strata. For possibly unrelated reasons, I'm unable to generate a kaplan-meier plot even with strata supplied (see below).

What's the actual result?

vr_kaplan_meier_summary() generates this error if I do not supply strata:
Error: Column strata must be length 9 (the number of rows) or one, not 0

vr_kaplan_meier_summary() generates the expected tables if I do supply strata.

vr_kaplan_meier() generates this error if I do not supply strata:
Error in data.frame(time = survfit_summary$time, n.risk = survfit_summary$n.risk, : arguments imply differing number of rows: 43, 0

vr_kaplan_meier() generates this error if I do supply strata:
Error in summarize(., max_time = max(time)) : argument "by" is missing, with no default

Additional details / screenshot

We have multiple functions creating different kinds of summaries:

• vr_summarize and vr_summarize_tab1 are defined and used in the context of creating a summary table (in vr_utils.R)
• get_summary is defined to create a summary of a survfit object

I wonder if we should consolidate these methods or alternatively have more distinct names for them to avoid confusion?

Colors may be a second task to be tackled separately from the theming that is integral to styling the visualisation. There may be some requirements making this a non-trivial task to automate, though, so it'd be great to have a discussion on this prior to implementation.

Some requirements that came up before

• colors should be used sparingly and only where needed to convey message
• colors should be consistent across visualisations in the report (i.e., the same drug should always have the same color across a series of graphs)
• defaults needed for continuous and discrete variables
• ability to define specific colors for key variable values (e.g., drugs of particular interest, custom colors for factor values, ...)
• depending on use case: use corporate design colors as standard palette (parameterized by company or institution)
• colors should be discernible even if colorblind

To some extent, one could specify default color palettes at the beginning of the (rmarkdown) script using options for, e.g., ggplot2.continuous.colour and ggplot2.continuous.fill, but that would not account for most of the above customizations required for adhering to good graphics principles and our stakeholder requirements.

Does anyone have an idea how to best implement these requirements around colors and styling in the package? I guess we'd need to disentangle what could be meaningful defaults to be set at the document setup phase and what could be customized in a vr_style_plot function?

Can we chuck in a license? Something that says we have no liability or warranty, and make sure it's not copy left (e.g. you can use the code for anything).

Using terminology defined in a standard like CDISC ADaM has a lot of advantages. However, there are several aspects we need to consider, for example:

• general flexibility,
• the tradeoff between ease of use if data follows the standard vs. if it does not
• how easy it is to learn and understand

Generally, I would suggest not to make strong assumptions about column names. Then again, if we have to make assumptions it is a good idea to follow the standard. Also, I think we should have parameter names that are easy to read and to understand, especially for people familiar with survival analysis but not necessarily with the standard.

The easiest solution would probably be using the column names from the ADaM standard as default values and using simple parameter names that are meaningful in a functional way.

Suggestion: Change the function signature for vr_KM_est from
vr_KM_est(data = adtte, strata = "TRTP")
to either

1. vr_KM_est(data=NULL, stratifier=NULL, censor = "CNSR", time = "AVAL")
2. vr_KM_est(data=NULL, stratifier=NULL, event = "CNSR", time = "AVAL")

Greetings!

I'm hoping to use vr_table_one, but some of my values are missing/NA. I run into this error:

Error in quantile.default(x, probs = 0.25) :
missing values and NaN's not allowed if 'na.rm' is FALSE

If I change the NAs to "missing", as in the example in the doc, it then treats my continuous variable (age) as a categorical, and rather than get mean/median/etc, I get "number of unique values" in the table.

Is there a good workaround for this?

Thanks!
Kelly

We are currently not using a harmonized nomenclature for reporting sample size etc. Based on our scientist's feedback, this is something we should change soon, not only for the summary table (table one), but also for meta-data on the figures.

From Letizia: In general, capital letters refer to population attributes (i.e., parameters); and lower-case letters refer to sample attributes (i.e., statistics). For example, P refers to a population proportion; and p, to a sample proportion. @bailliem you mentioned there is also a more extensive guidance available that we could build upon?

With the switch to using equations in the KM functions, we need to fix the hardcoded reference to data$time in vr_est_km_risk_table. I guess it should be possible to extract the column name from the equation?

@bailliem This line has the columns study and row_id hard-coded(I guess from the crohn's dataset) which will likely break in all other cases. Do we need this function here or is the ordering sth. we should do separately?

Issue description

When the argument min_at_risk = 0, I am expecting the risk table to extend the entire x axis

Solution

I played around a bit and found an easy solution:

  if (min_at_risk > 0) {
  max_time <-
    tidy_object %>%
    dplyr::filter(n.risk >= min_at_risk) %>%
    dplyr::group_by(strata) %>%
    dplyr::summarize(max_time = max(time)) %>%
    dplyr::ungroup() %>%
    dplyr::summarize(min_time = min(max_time)) %>%
    dplyr::pull(min_time)
  } else {
    max_time = max(time_ticks)
  }

We started out implementing a few functions that are important in our daily work for visualising time to event analysis. However, this is not where the package should stop. Some ideas from discussions with colleagues and stakeholders for additional visualisations to be added are listed below. Please comment with other ideas or open an issue and start implementing if you wish to take one on.

Cohort Selection and Baseline Characteristics

• Table 1
• cohort attrition flowchart
• cohort attrition barplot
• tables of disease characteristics or prior therapies

Change from Baseline

• Scatterplots incl. diff options for overlapping data points like jitter
• Waterfall plots
• Cumulative distribution plots

Survival/Time to Event Analysis

• Kaplan-Meier plots (different defaults for descriptive vs. comparative questions)
• KM Summary tables
• Forest Plot (@bailliem is working on this)

Treatment/Testing Pattern Analysis

• Frequency Table of selected drugs by line
• Lineplot of year of start of 1st line vs % of patients by drug
• Stacked boxplot of types of line of therapy by year of diagnosis
• Donut plot showing distribution of cohort by max LOT with bars specifying most frequent drugs in each line
• Swimmer Plot showing patterns of treatment (or other activity) by patient

Treatment Sequences

• Flowchart specifying patient flow incl. Death, lost to follow up, and progression to next line
• Sankey plot (also showing numbers in non-interactive mode)
• Sunburst plot showing progression of drugs in lines

Post webinar set up a CI/CD workflow including

• unit tests
• dependencies
• styling and linting
• documentation build
• website build

This line is breaking the installation of the package with the message "Error in eval(parse(text = final_cond)) : object 'cancer_type' not found". Can someone (maybe @thanos-siadimas) have a look?

Issue description

We have bioconductor dependencies for the example data - this makes installation difficult I'm getting errors now on R 4.0

See here: #42

The guidance from the Kmunicate paper is to arrange n.risk, n.censor and n.events by stratum (see example of previos visR KM implementation)

With our new API, this ordering does not appear possible anymore:

I think it would be great to include the option to order the elements of the risk table following the KMunicate guidance, though, given that this was one of the motivators to actually write our own KM function. Maybe in addition to the more common ordering as implemented in ggsurvplot? What do you think?

Building the example analysis locally (either running each chunk in the notebook or knitting the full document) in RStudio works fine. Executing the following code results in the flowchart shown in the viewer or RStudio:

attrition_flow <- vr_attrition(
  N_array = cohort_attrition$`Remaining N`,
  descriptions = cohort_attrition$Criteria,
  complement_descriptions = complement_descriptions,
  display = F,
  output_path = attrition_chart_fn)

And embedding the exported svg in the report also works:

When building the vignette using pkgdown::build_articles(), the image included by knitr::include_graphics is broken:

Additionally, when knitting the vignette, it the svg (that is shown in the Viewer when working interactively) is not incorporated as an image, but as text:

And this svg returned by the call to attrition_flow does not appear at all when building the vignette.

One problem seems to be the file path used in knitr::include_graphics does not seem correct in the context of the fully built website. Not sure how to correct for this, though... Does anyone know have any ideas how to fix this problem and show the final flow chart in our vignette on the website?

Change the get_tableone function name to be more general. It calculates summary statistics for categorical and continuous variable types. Although it is used to derive table one, it could also be used with other wrapper functions / table types.

Update the graphical representation of the visR architecture. Maybe even add an example with concrete function (names or visualizations) to show how different functions map to the architecture concepts.

Issue description

Current CICD doesn't work... fix it.

Steps to reproduce the issue

Push a commit and Github actions doesnt work

What's the expected result?

• build and deploy docs
• README.rmd is built automatically
• check windows
• check ubuntu
• check osx
• calculate code coverage

Issue description

vr_render_table always collapses rows when using kable engine. This is not what we always need and should be an option or so.

Steps to reproduce the issue

Please share code if possible.

library(survival)
data(lung)


lung %>% 
  head(10) %>% 
  vr_render_table(title = "First 10 observations of `lung_cohort`.",
                  caption = "",
                  datasource = DATASET, 
                  engine = "kable")

What's the expected result?

• I want to see each row of the first ten samples

What's the actual result?

• Those consecutive row cells with same content are collapsed into one

Additional details / screenshot

The number of significant digits may change. At the moment, we mainly implement 2 or 3 digits after the decimal point, but this should probably be discussed further.
Do we want the number of digits to be sth. the user can specify in each function (risk: parameter creep) or enforce best practices? In the latter case, what would be the best practices we can agree upon with out scientists?

Several journals provide guidances, does anyone have any thoughts on which one's are preferable?

• Report to three significant digits (https://www.bmj.com/content/350/bmj.h1845). Here the size of the ratio is to use a rule based on significant digits. (adopting 2 decimals to all fractions makes the error larger for smaller ratios). Take the example—the ratio 1.23 with two decimal places and three significant digits. Dividing and multiplying it by 10 give 0.123 and 12.3, respectively, each still with three significant digits but with varying numbers of decimal places.
• 2 digits for p-values unless smaller than 0.01, in which case 3 digits should be shown (https://jamanetwork.com/DocumentLibrary/InstructionsForAuthors/InstructionsForTableCreation.pdf)

Issue description

pkgdown::build_site() not working due to issue in vr_alluvial_plot

Steps to reproduce the issue

Run vignettes/Alluvial.Rmd up to line 79 using R 4.0 and the latest version of tibble.

What's the actual result?

vr_alluvial_plot(dataset,
                 id = "PatientID",
                 linename = "LineName",
                 linenumber = "LineNumber",
                 data_source = "Simulatation")

Error: Must extract column with a single valid subscript.
x Subscript `tibble_vars[[i]]` has value 0 but must be a positive location.
Run `rlang::last_error()` to see where the error occurred. 

Additional details / screenshot

May be an R 4.0 issue?

Following the principle that every output should have a title and a reference to the underlying data source, we should add such a feature also to vr_attrition.

Ideally, this would be two parameters following terminology of other vr functions and be an integral part of the resulting image.

Currently the package supports a single tree and a single branch in a cohort attrition diagram. It would be great if the package can allow the users to add multiple branches (like having subset of samples).

When stratifying, it is sometimes useful to use ggplot's faceting (facet_grid and facet_wrap) However, this only works (if at all) only for the KM plot (e.g., when using survminer, but also visR), but it would be really useful to also have the risk table for each facet.
Maybe this is something we could implement in the future?

I would suggest to add the option to call the get_COX_HR method also on a data.frame. We should not make strong assumptions about the flow/hierarchy of different calls, if it is not necessary. Currently, the get_COX_HR method is only called on a survfit object but it would be very easy to add a function that calls coxph function on a data.frame.

This was an interesting idea from the Validation workshop of R in pharma:
They are using peer-reviewed literature for estimates of models to test again with testthat. I really like this idea for our wrapper functions for models and the summary measures.

Renamed the parameter "groupCols" in the functions to create and render a table_one with "group_cols". However, the html in the docs folder remain unchanged.

Any idea why?

Issue description

There seems to be an issue with the attrition code when using logical operators in the filtering strings. The table does not return correct N's

Steps to reproduce the issue

Please share code if possible.

data(lung)
lung_cohort <- lung %>% 
    dplyr::rename(ECOG = ph.ecog,
                  Karnofsky = ph.karno,
                  SITEID = inst,
                  CNSR = status,
                  AVAL = time,
                  WEIGHT = wt.loss
                 ) %>% 
    dplyr::mutate(USUBJID = sprintf("Pat%03d", row_number()),
                  SEX = factor(if_else(sex == 1, "male", "female")),
                  CNSR = if_else(CNSR == 2, 1, 0),
                  SUBGR1 = factor(case_when(ECOG == 0 ~ "0 asymptomatic",
                                            ECOG == 1 ~ "1 ambulatory",
                                            ECOG == 2 ~ "2 in bed less than 50% of day",
                                            ECOG == 3 ~ "3 in bed more than 50% of day",
                                            ECOG == 4 ~ "4 bedbound",
                                            ECOG == 5 ~ "5 dead")
                                 ),
                  AGEGR1 = factor(case_when(age < 30 ~ "< 30y",
                                            age >= 30 & age <= 50 ~ "30-50y",
                                            age > 50 & age <= 70 ~ "51-70y",
                                            age > 70 ~ "> 70y"),
                                  levels=c("< 30y", "30-50y", "51-70y", "> 70y")
                                 )
                 ) 
cohort_attrition <- vr_attrition_table(
  data = lung_cohort,
  criteria_descriptions = c("1. ECOG available", 
                            "2. Weight loss data available", 
                            "3. Non-missing censoring status",
                            "4. Positive follow up time",
                            "5. Meal Cal (missing or  > 1000"),
  criteria_conditions   = c("!is.na(SUBGR1)",
                            "!is.na(WEIGHT)",
                            "!is.na(CNSR)",
                            "AVAL >= 0",
                            "is.na(meal.cal) | meal.cal >= 1000"),
  subject_column_name   = 'USUBJID')

returns 132 patients after all filtering.

What's the expected result?

lung_cohort %>%
  dplyr::filter(!is.na(SUBGR1),
                !is.na(WEIGHT),
                !is.na(CNSR),
                 AVAL >= 0) %>% 
  dplyr::filter(is.na(meal.cal) | meal.cal >= 1000)

This returns a dataframe with 124 rows and not 132.

What's the actual result?

Additional details / screenshot

Add drop lines to show quantiles of the KM distribution.

Idea

What is your idea for a new plot/table?
Not only meta-analyses, but also cohort overviews or results of cox PH models and other regression models are often shown in ForestPlots. @bailliem had some initial ideas, but we haven't gotten around to fully implement this visualisation and help would be appreciated

Do you have any examples you can share of what it looks like?
Example for Cox PH Model (example from forestmodel)

Example for cohort overview from Morris et al.

What potential packages/functions could be used to make this plot?
There are the forestplot and forestmodel packages on CRAN as well as potentially some good ideas in the ggforest function of the survminer package. Ideally, this would be implemented using ggplot2 only.

Are you interested to implement this yourself?
@bailliem started, but not sure how it goes bandwidth wide.

It looks like we have several places where utility functions are defined: minimally R/vr_utils.R and R/utilities.R. Maybe it's good to agree on one single place for this? Happy to be convinced otherwise.

What are the core things that need to be decided in order to move towards to a simple but expandable way to structure the package?

I have tried a number of times to install vizR on Windows, but it keeps failing. The latest failure has to do with the files/packages (?) RTCGA and RTCGA.clinical as below. What am I doing wrong or what is the bug involved?
Thanks.
James Hunter

√  checking for file 'C:\Users\james\AppData\Local\Temp\Rtmp8McCI4\remotes77f838e42777\openpharma-visR-0c1b345/DESCRIPTION' (893ms)
-  preparing 'visR': (557ms)
√  checking DESCRIPTION meta-information ... 
-  excluding invalid files
   Subdirectory 'man' contains invalid file names:
     'visR architecture.png'
-  checking for LF line-endings in source and make files and shell scripts
-  checking for empty or unneeded directories
-  building 'visR_0.1.0.tar.gz'
   
Installing package into ‘C:/Users/james/OneDrive/Documents/R/win-library/3.6’
(as ‘lib’ is unspecified)
ERROR: dependencies 'RTCGA', 'RTCGA.clinical' are not available for package 'visR'
* removing 'C:/Users/james/OneDrive/Documents/R/win-library/3.6/visR'
Error: Failed to install 'visR' from GitHub:
  (converted from warning) installation of package ‘C:/Users/james/AppData/Local/Temp/Rtmp8McCI4/file77f8328a59f6/visR_0.1.0.tar.gz’ had non-zero exit status

The file "attrition_diagram.svg" is located in the "vignettes" folder but the codes searches it in the non-existing folder "/Example_analysis_files/figure-html".

I would fix this - any preferences how?

Following our current architecture, styling of plots should be separated out from the data wrangling and plotting task (see Wiki article on Architecture). We currently mainly rely on applying themes (or even color palettes) after the plotting task.

It may be nice to have convenience functions, though, that take care of multiple styling operations, similar to style_bbc() (from the BBC visuals cookbook).
To implement this, we would need some reasonable defaults to start with for font type/sizes, preferred grid styles, axis position etc. @bailliem do you happen to have some guidance from the graphical principles work

Hi @kawap,

here is a little writup on our requirements for Alluvial/Sankey plots which are often used for assessing patterns in patient flow, e.g., consecutively received treatments

There are some requirements for the plotting that one needs to take into account based on feedback we have received from scientists working with the output

• data wrangling and plotting should be 2 different functions
  • wrangling should result in a table with line of treatment, treatment name, pateint count and frequency (incl. those "lost" in each step)
  • visualisation then focus on showing that data as sankey plot
  • ideally, this is generic enough to not only focus on treatment, but any other kind of sequential data
• plotting functions should include annotation of plot with title (can be empty string), number of samples, and datasource in footnote
• has to work with the ggplot2 ecosystem (e.g., styling, adding titles etc.)
• shown data should include death/censoring, no later treatment (i.e., those patients lost to later observations)
• data needs to be interpretable in static form (containing information such as name of therapy and frequency)
• treatment groups should have meaningful order (or should be easily changeable)
• color scheme should match other analyses (i.e., fct. should have option to define custom color palette)
• optional: interactive version can hide some of the annotation and make visible through mouse-over etc.

There are a few sankey packages out there that may be useful in this context. Selecting which one is up to the discretion of whoever writes the visr function:

• https://github.com/corybrunson/ggalluvial
• https://ggforce.data-imaginist.com/reference/geom_parallel_sets.html
• https://erblast.github.io/parcats/
• https://github.com/Displayr/flipPlots (example here: https://www.r-graph-gallery.com/sankey-diagram.html)

@diego-s and @bailliem please add to this if I forgot anything important

When trying to install the package in our analytics environment, users often run into issues with resolving the V8 dependency:

Using PKG_CFLAGS=-I/usr/include/v8 -I/usr/include/v8-3.14
Using PKG_LIBS=-lv8 -lv8_libplatform
-----------------------------[ ANTICONF ]-------------------------------
Configuration failed to find the libv8 engine library. Try installing:
 * deb: libv8-dev or libnode-dev (Debian / Ubuntu)
 * rpm: v8-devel (Fedora, EPEL)
 * brew: v8 (OSX)
 * csw: libv8_dev (Solaris)
To use a custom libv8, set INCLUDE_DIR and LIB_DIR manually via:
R CMD INSTALL --configure-vars='INCLUDE_DIR=... LIB_DIR=...'
---------------------------[ ERROR MESSAGE ]----------------------------
<stdin>:1:10: fatal error: v8.h: No such file or directory
compilation terminated.
------------------------------------------------------------------------
ERROR: configuration failed for package 'V8'
* removing '/home/reyessia/R/x86_64-pc-linux-gnu-library/4.0/V8'
Error: Failed to install 'visR' from GitHub:
  (converted from warning) installation of package 'V8' had non-zero exit status

It can be resolved by following the instructions in the traceback, i.e., installing the corresponding libv8 library in the terminal (not in R, but into the system).
For ubuntu, apparently also the following worked
sudo apt install librsvg2-dev

If I remember correctly, we thought we had removed that dependency in develop a while ago, but it seems to be back both on master and develop. @bailliem, @SHAESEN2 do you know if we need this still?

The caption at the bottom of a DT-rendered table seems to replicate every time the user interacts with the table (filtering, sorting etc.).

MRE:

adtte %>%
  visR::estimate_KM("TRTP") %>%
  visR::get_pvalue() %>%
  visR::render(title = "P-values", 
               datasource = paste0("Analysis data - time to event"),
               engine = "DT")

Then interact with the table.

Likely caused by the caption being added to the DT object in using drawCallback = DT::JS(source_cap))

Issue description

"LCL", "UCL" and "CI" are allowed to be part of the statlist argument in the get_summary function, and by default all three are.
When CL or CI are a part if the statlist (l.55-61) ...
("conf.int" %in% names(survfit_object) &
survfit_object[["conf.type"]] != "none" &
((base::any(grepl(
"CL", statlist, fixed = TRUE
))) | (base::any(grepl(
"CI", statlist, fixed = TRUE
)))))
...they are changed to something like "0.95CI" or "0.95CL". And this happens also if only one of these stats is there (l.62-67):
CI <-
paste0(survfit_object[["conf.int"]], statlist[grepl("CL", statlist, fixed = TRUE)])
statlist[grepl("CL", statlist, fixed = TRUE)] <- CI
CI_Varname <-
paste0(survfit_object[["conf.int"]], statlist[grepl("CI", statlist, fixed = TRUE)])
statlist[grepl("CI", statlist, fixed = TRUE)] <- CI_Varname

In line 99 then CI_Varname is used independently of whether it was set or not (see condition l.55-61).

Steps to reproduce the issue

#works:
get_summary(survfit_object)

#breaks sometimes; cannot reproduce:
get_summary(survfit_object, statlist = c('median'))

#breaks always
get_summary(survfit_object, statlist = c('events', 'median', 'CI'))
--> Error: Problem with mutate() input 0.95CI.
x undefined columns selected
i Input 0.95CI is .CIpaste(.).

Currently we prefix all function calls, except for eg Survfit. One way to fix this is update the Call in estimate_KM, but this does not seem to be allowed. Another is to parse all calls downstream.

Converting this core package function into an S3 method would:

• make the package more extendable and accessible to more developers to add on different table engines that they use.
• make testing more compartmental and simpler to track since each engine would be self contained.
• create a simpler way to make knitr::knit_print methods to better connect with knitr and its methods.

Functions should be checked for consistent parameter names for a more intuitive behavior.

For example:
visR::add_annotation(..., lbl = NULL, ...)
vs
visR::vr_plot(..., y_label = NULL, ...)

Work through the list of variables referenced in visR globalvariable to set those variables to NULL within the function call, where possible. At the moment we are using this to suppress note in R CMD check "no visible binding". Ideally, we resolve this at the function level.

See here for more details on the alternative approach STAT545-UBC/Discussion#451

I think the API as shown in the kick-off by Steve is already great and I really like it a lot!
The key points are:

• Different classes model estimation / necessary calculations (implemented in the vr_KM_est.R) and plotting (vr_plot.R)
• Very generic plotting (vr_plot.R) and generic plotting-helpers (e.g. add_CI), with also consistent and clear naming.
• Use of the general tidyverse philosophy and especially the pipe operator.

A next step could be adding refactoring another plotting function to be integrated into the existing API, for example, the alluvial
The next steps would then be:

• Changing the referencing of column names. Right now they are renamed in the data.frame, I would use the parameter names instead (cf. issue 82).
• Creating an object for the result of the precalculations necessary to plot the alluvial (function vr_alluvial_wrangling).
• Separate alluvial base plot from addons (like adding the N) to have the same calling structure as in the survival example.
• Integrating the alluvial plotting into the vr_plot method.
  Each point would be a separate github issue but I don't want to clutter the issue list before we have decided if we do it like this

When following good principles for plotting KM curves, it is advised by Morris et al. to also show an extended risktable with N at risk, censored and events. In it's current implementation, this may lead to a rather small KM plot and the risk tables taking up a large part (I'd guess 50%) of the figure.
It would be nice to adjust this ratio (similarly to the tables.height parameter in ggsurvplot) or have some other options to reduce the plot real estate taken up by the table-part.

It could be a nice feature to add a print method to preview latex tables using texPreview (apologies for the shameless self plug). Many times it is a pain to tweak latex tables since they need to be rendered in a pdf doc to see them. texPreview quickly renders latex in the preview window of RStudio so users can iterate over tables much like they do in ggplot workflows.