doudpackage

doudpackage builds the “table one” of a bio-medical paper: one row per variable, one column per sub-group, the right statistical test picked for you, and an HTML or LaTeX table ready to paste into the manuscript.

You give it a data.frame and the name of the grouping variable. It works out which variables are quantitative and which are qualitative, describes each of them, compares them across the groups, and renders the result with kableExtra.

Installation

install.packages("doudpackage")

Development version:

# install.packages("devtools")
devtools::install_github("tiago972/doudpackage")

Getting started

Two functions do the work: descTab() computes the table, parseClassFun() renders it.

library(doudpackage)

data(iris)
iris$treated <- factor(sample(c("no", "yes"), 150, replace = TRUE))
iris$dose <- runif(150, min = 0, max = 100)
iris$dose[sample(1:150, 5)] <- NA

tab <- descTab(iris, group = "Species")
tab["table"]
#>                      var      setosa  versicolor virginica       Total pvalue
#> 1         dose mean (SD) 56.3 (24.6) 58.4 (27.9) 51.5 (31) 55.4 (27.9)  0.464
#> 2 Petal.Length mean (SD)   1.5 (0.2)   4.3 (0.5) 5.6 (0.6)   3.8 (1.8)  0.000
#> 3  Petal.Width mean (SD)   0.2 (0.1)   1.3 (0.2)   2 (0.3)   1.2 (0.8)  0.000
#> 4 Sepal.Length mean (SD)     5 (0.4)   5.9 (0.5) 6.6 (0.6)   5.8 (0.8)  0.000
#> 5  Sepal.Width mean (SD)   3.4 (0.4)   2.8 (0.3)   3 (0.3)   3.1 (0.4)  0.000
#> 6            treated, no     23 (46)     26 (52)   28 (56)   77 (51.3)  0.602
#> 7           treated, yes     27 (54)     24 (48)   22 (44)   73 (48.7)  0.602

descTab() returns an S4 object; the raw data frame is under ["table"] and parseClassFun() turns it into the formatted table:

parseClassFun(tab)

What goes into a row

Variable	Described as	Compared with
numeric / integer, normal	mean (SD)	t test (2 groups), ANOVA (3+)
numeric / integer, non normal	median (IQR)	Wilcoxon (2 groups), Kruskal-Wallis (3+)
factor / ordered factor	n (%) per level	Chi-squared, Fisher when the approximation is unreliable

Variables of any other type (character, Date, logical) are ignored with a warning. Set normality to choose how quantitative variables are handled:

descTab(iris, group = "Species", normality = "normal")      # mean (SD) everywhere
descTab(iris, group = "Species", normality = "non normal")  # median (IQR) everywhere
descTab(iris, group = "Species", normality = "assess")      # Shapiro-Wilk, variable by variable

Shaping the output

tab <- descTab(iris, group = "Species", na.print = TRUE)

parseClassFun(
  tab,
  # only one line for a binary variable, and choose which one
  levels_to_keep = list("treated" = "yes"),
  # regroup rows under labels, in this order
  group_rows_labels = list("Petal" = c("Petal.Length", "Petal.Width"),
                           "Sepal" = c("Sepal.Length", "Sepal.Width")),
  # column order
  col.order = c("Total", "setosa", "versicolor", "virginica"),
  font = "Times New Roman"
)

Useful descTab() arguments:

Argument	Effect
`group`	the factor defining the columns; omit it for a single Total column
`na.print`	add a “Missing values” row under each variable
`pvalue`	drop the p value column
`quanti`, `quali`	restrict the table to one kind of variable
`digits.p`, `digits.qt`, `digits.ql`	rounding for p values, mean/SD, proportions
`parallel`, `mc.cores`	compute the analyses with `parallel::mclapply()`

Notes on missing and degenerate data

Rows with a missing value in group are dropped, with a warning.
Unused levels of group (typically left behind by a subset) are dropped, with a warning, rather than producing an empty column.
When a test cannot be computed — a constant variable, an empty level, too few observations — the p value is NA and a warning names the variable. The rest of the table is still produced.

Bugs

Please report them at https://github.com/tiago972/doudpackage/issues.