Summary type in `tbl_roche_summary()` is a bit stochastic....should it be?

[ddsjoberg]

In gtsummary's tbl_summary() (and subsequently in crane's tbl_roche_summary()), the summary type of a variable is initially determined by inspecting the class of the column and it's spread. For example, a factor/character will default to categorical summaries. Most numeric columns will default to continuous, but if that numeric column very few levels, it will then default to categorical.

Should we make this more deterministic for tbl_roche_summary()? For example, all numeric results default to continuous (which can of course be changed). This way, there will be no surprises from a study team as their data accumulates and perhaps a default summary type changes.

The default precision of summary statistics is similar in that it is determined by the spread of the data. It could occur that at the beginning of a study, the algorithm chooses to round to the nearest integer, but as more data is added the default changes to 1 decimal place.

[ddsjoberg]

@Melkiades can you investigate this?

We can begin by adding list("assign_summary_type-arg:cat_threshold" = 0L) to our theme so all numeric columns will default to continuous (the default bahviour is for numeric variables with fewer than 10 unique levels default categorical. But it could be the case that in the beginning of a trial, a variable defaults to categorical but as more data accrues it changes to continuous (which could be confusing for users).

library(gtsummary)
packageVersion("gtsummary")
#> [1] '2.4.0.9001'

tbl_summary(mtcars, include = cyl) |> 
  as_kable()

Characteristic	N = 32
cyl
4	11 (34%)
6	7 (22%)
8	14 (44%)

with_gtsummary_theme(
  x = list("assign_summary_type-arg:cat_threshold" = 0L),
  expr = tbl_summary(mtcars, include = cyl)
) |> 
  as_kable()

Characteristic	N = 32
cyl	6.00 (4.00, 8.00)

^{Created on 2025-09-10 with reprex v2.1.1}