This post explains how to use the
gtsummary package for creating table summary, especially
with descriptive statistics, regression models, medical values or
demographics data.
This post showcases the key
features of gtsummary and provides a set of
table examples using the package.
{gtsummary}
The gtsummary package in R is made for creating tables
that summarize information, statistics or more in a given dataset. You
can use it in combination of the pipe %>% symbol for
easy-to-read code and publication-ready tables!
The main function is tbl_summary() that becomes very
powerful when combined with other functions available. If you’re working
a regression problem, you have the tbl_regression()
function. If you need to merge some tables, you have
tbl_merge(). Those are just examples of things you can
do!
✍️ author → Daniel D. Sjoberg
📘 documentation → github
⭐️ more than 800 stars on github
| Characteristic | Drug, N = 571 | Placebo, N = 431 | Difference2 | 95% CI2,3 | p-value2 |
|---|---|---|---|---|---|
| age | -0.57 | -4.4, 3.2 | 0.8 | ||
| Median (IQR) | 51 (45, 56) | 51 (45, 59) | |||
| sex | 0.17 | -0.23, 0.57 | |||
| Female | 30 (53%) | 19 (44%) | |||
| Male | 27 (47%) | 24 (56%) | |||
| bmi | 0.75 | -1.1, 2.6 | 0.4 | ||
| Median (IQR) | 25.3 (21.9, 28.8) | 23.6 (21.4, 26.3) | |||
| 1 n (%) | |||||
| 2 Welch Two Sample t-test; Standardized Mean Difference | |||||
| 3 CI = Confidence Interval | |||||
To get started with gtsummary, you can install it
directly from CRAN using the install.packages function:
The gtsummary package lets you automatically
summarize information about your dataset. In the
following case, we use the tbl_summary() function to obtain
the main information on the iris dataset. The package detects
the variable type and generates the appropriate summary
type.
| Characteristic | N = 1501 |
|---|---|
| Sepal.Length | 5.80 (5.10, 6.40) |
| Sepal.Width | 3.00 (2.80, 3.30) |
| Petal.Length | 4.35 (1.60, 5.10) |
| Petal.Width | 1.30 (0.30, 1.80) |
| Species | |
| setosa | 50 (33%) |
| versicolor | 50 (33%) |
| virginica | 50 (33%) |
| 1 Median (IQR); n (%) | |
With the tbl_regression() function, we can super easily
display the statistical results of a regression model.
Example with a logistic regression on the Titanic dataset:
# load dataset
data(Titanic)
df = as.data.frame(Titanic)
# load library
library(gtsummary)
# create the model
model = glm(Survived ~ Age + Class + Sex + Freq, family=binomial, data=df)
# generate table
model %>%
tbl_regression() %>% # regression summary function
add_global_p() %>% # add p-values
bold_labels() %>% # make label in bold
italicize_levels() # make categories in label in italic| Characteristic | log(OR)1 | 95% CI1 | p-value |
|---|---|---|---|
| Age | 0.5 | ||
| Child | — | — | |
| Adult | 0.62 | -1.0, 2.4 | |
| Class | >0.9 | ||
| 1st | — | — | |
| 2nd | -0.03 | -2.0, 2.0 | |
| 3rd | 0.25 | -1.8, 2.4 | |
| Crew | 0.27 | -1.8, 2.4 | |
| Sex | 0.6 | ||
| Male | — | — | |
| Female | -0.37 | -1.9, 1.1 | |
| Freq | -0.01 | -0.02, 0.00 | 0.2 |
| 1 OR = Odds Ratio, CI = Confidence Interval | |||
As its name suggests it, the gtsummary package makes
very easy to generate summary of your dataset. In practice, it uses the
tbl_summary() function to compute descriptive
statistics for every column in your dataset depending to the
type of variable.
What’s even better is that you can add inferential statistics (like p-values) to these tables to make them even more informative!
Example:
# load dataset and filter to keep just a few columns
data(mtcars)
mtcars = mtcars %>%
select(vs, mpg, drat, hp, gear)
# load package
library(gtsummary)
# create summary table
mtcars %>%
tbl_summary(
by=vs, # group by the `vs` variable (dichotomous: 0 or 1)
statistic = list(
all_continuous() ~ "{mean} ({sd})", # will display: mean (standard deviation)
all_categorical() ~ "{n} / {N} ({p}%)" # will display: n / N (percentage)
)
) %>%
add_overall() %>% # statistics for all observations
add_p() %>% # add p-values
bold_labels() %>% # make label in bold
italicize_levels() # make categories in label in italic| Characteristic | Overall, N = 321 | 0, N = 181 | 1, N = 141 | p-value2 |
|---|---|---|---|---|
| mpg | 20.1 (6.0) | 16.6 (3.9) | 24.6 (5.4) | <0.001 |
| drat | 3.60 (0.53) | 3.39 (0.47) | 3.86 (0.51) | 0.013 |
| hp | 147 (69) | 190 (60) | 91 (24) | <0.001 |
| gear | 0.001 | |||
| 3 | 15 / 32 (47%) | 12 / 18 (67%) | 3 / 14 (21%) | |
| 4 | 12 / 32 (38%) | 2 / 18 (11%) | 10 / 14 (71%) | |
| 5 | 5 / 32 (16%) | 4 / 18 (22%) | 1 / 14 (7.1%) | |
| 1 Mean (SD); n / N (%) | ||||
| 2 Wilcoxon rank sum test; Fisher’s exact test | ||||
The package has a whole set of functions that can be used to custom what your table looks like. You can even call functions from others packages such as gt
Example:
data(iris)
library(gtsummary)
library(gt)
iris %>%
tbl_summary(by=Species) %>%
add_overall() %>% # info ignoring the `by` argument
add_n() %>% # number of observations
modify_header(label ~ "**Variables from the dataset**") %>% # title of the variables
modify_spanning_header(c("stat_0", "stat_1", "stat_2", "stat_3") ~ "*Descriptive statistics of the iris flowers*, grouped by Species") %>%
as_gt() %>%
gt::tab_source_note(gt::md("*The iris dataset is probably the **most famous** dataset in the world*"))| Variables from the dataset | N | Descriptive statistics of the iris flowers, grouped by Species | |||
|---|---|---|---|---|---|
| Overall, N = 1501 | setosa, N = 501 | versicolor, N = 501 | virginica, N = 501 | ||
| Sepal.Length | 150 | 5.80 (5.10, 6.40) | 5.00 (4.80, 5.20) | 5.90 (5.60, 6.30) | 6.50 (6.23, 6.90) |
| Sepal.Width | 150 | 3.00 (2.80, 3.30) | 3.40 (3.20, 3.68) | 2.80 (2.53, 3.00) | 3.00 (2.80, 3.18) |
| Petal.Length | 150 | 4.35 (1.60, 5.10) | 1.50 (1.40, 1.58) | 4.35 (4.00, 4.60) | 5.55 (5.10, 5.88) |
| Petal.Width | 150 | 1.30 (0.30, 1.80) | 0.20 (0.20, 0.30) | 1.30 (1.20, 1.50) | 2.00 (1.80, 2.30) |
| The iris dataset is probably the most famous dataset in the world | |||||
| 1 Median (IQR) | |||||
