Split Forest Plots

Split forest plots create separate plots for each value of a categorical variable, with an interactive sidebar for navigation. This is ideal for subgroup analyses where you want to compare results across different populations.

When to Use Split vs Group

Feature	Use When
`split_by`	Many subgroups (5+), each deserves focused attention
`group`	Few subgroups (2-4), want to see all at once

Split creates a sidebar with one plot per subgroup. Group creates collapsible sections in a single plot.

Basic Split

Use split_by to create separate plots for each unique value:

Code

set.seed(42)
trial_data <- data.frame(
  study = paste0("Study ", 1:30),
  region = sample(c("North America", "Europe", "Asia Pacific"), 30, replace = TRUE),
  or = exp(rnorm(30, log(0.8), 0.3)),
  lower = NA, upper = NA
)
trial_data$lower <- trial_data$or * exp(-1.96 * 0.25)
trial_data$upper <- trial_data$or * exp(1.96 * 0.25)

forest_plot(trial_data,
  point = "or", lower = "lower", upper = "upper",
  label = "study",
  split_by = "region",
  scale = "log", null_value = 1,
  axis_label = "Odds Ratio (95% CI)",
  title = "Treatment Effect by Region"
)

The floating sidebar shows each region. Click to navigate between subgroups.

Hierarchical Splits

Split by multiple variables for nested navigation:

Code

set.seed(123)
n <- 50
subgroup_data <- data.frame(
  study = paste0("Study ", sprintf("%02d", 1:n)),
  sex = sample(c("Male", "Female"), n, replace = TRUE),
  age_group = sample(c("18-40", "41-65", "65+"), n, replace = TRUE),
  or = exp(rnorm(n, log(0.75), 0.25)),
  lower = NA, upper = NA
)
subgroup_data$lower <- subgroup_data$or * exp(-1.96 * 0.2)
subgroup_data$upper <- subgroup_data$or * exp(1.96 * 0.2)

forest_plot(subgroup_data,
  point = "or", lower = "lower", upper = "upper",
  label = "study",
  split_by = c("sex", "age_group"),  # Hierarchical: Sex > Age Group
  scale = "log", null_value = 1,
  theme = web_theme_modern(),
  axis_label = "Odds Ratio"
)

The sidebar shows “Sex” as the first level header, with “Age Group” nested under each sex.

Using split_forest() Directly

For more control, use the split_forest() function with the pipe syntax:

Code

subgroup_data |>
  web_spec(
    point = "or", lower = "lower", upper = "upper",
    label = "study",
    scale = "log", null_value = 1,
    columns = list(
      col_text("sex", header = "Sex"),
      col_text("age_group", header = "Age"),
      col_interval("OR (95% CI)")
    )
  ) |>
  split_forest(by = c("sex", "age_group")) |>
  forest_plot()

Shared Axis Range

By default, each sub-plot auto-scales its axis. Use shared_axis = TRUE for consistent comparison:

Code

forest_plot(trial_data,
  point = "or", lower = "lower", upper = "upper",
  label = "study",
  split_by = "region",
  shared_axis = TRUE,  # Same axis range across all regions
  scale = "log", null_value = 1,
  theme = web_theme_modern()
)

This ensures the same visual scale across all subgroups.

Exporting Split Forests

Export all sub-plots to a directory structure:

split_result <- data |>
  web_spec(point = "or", lower = "lower", upper = "upper") |>
  split_forest(by = c("sex", "age_group"))

# Exports to: output/Male/Male_Young.svg, etc.
save_split_forest(split_result, "output", format = "svg")

The directory structure mirrors the split hierarchy.

When to Use Split vs. Group

Feature	`group =`	`split_by =`
Display	Single plot with collapsible rows	Separate plots with sidebar
Navigation	Click group headers	Click sidebar items
Comparison	All visible at once	One subgroup at a time
Best for	Small number of groups	Many subgroups, detailed analysis

Use group when you want to see all data in one view with collapsible sections.

Use split_by when each subgroup deserves its own focused view, or when you have many subgroups that would clutter a single plot.

--- title: "Split Forest Plots" --- ```{r} #| include: false library(webforest) library(dplyr) ``` Split forest plots create **separate plots for each value** of a categorical variable, with an interactive sidebar for navigation. This is ideal for subgroup analyses where you want to compare results across different populations. ::: {.callout-tip} ## When to Use Split vs Group | Feature | Use When | |---------|----------| | **`split_by`** | Many subgroups (5+), each deserves focused attention | | **`group`** | Few subgroups (2-4), want to see all at once | Split creates a sidebar with one plot per subgroup. Group creates collapsible sections in a single plot. ::: ## Basic Split Use `split_by` to create separate plots for each unique value: ```{r} set.seed(42) trial_data <- data.frame( study = paste0("Study ", 1:30), region = sample(c("North America", "Europe", "Asia Pacific"), 30, replace = TRUE), or = exp(rnorm(30, log(0.8), 0.3)), lower = NA, upper = NA ) trial_data$lower <- trial_data$or * exp(-1.96 * 0.25) trial_data$upper <- trial_data$or * exp(1.96 * 0.25) forest_plot(trial_data, point = "or", lower = "lower", upper = "upper", label = "study", split_by = "region", scale = "log", null_value = 1, axis_label = "Odds Ratio (95% CI)", title = "Treatment Effect by Region" ) ``` The floating sidebar shows each region. Click to navigate between subgroups. ## Hierarchical Splits Split by multiple variables for nested navigation: ```{r} set.seed(123) n <- 50 subgroup_data <- data.frame( study = paste0("Study ", sprintf("%02d", 1:n)), sex = sample(c("Male", "Female"), n, replace = TRUE), age_group = sample(c("18-40", "41-65", "65+"), n, replace = TRUE), or = exp(rnorm(n, log(0.75), 0.25)), lower = NA, upper = NA ) subgroup_data$lower <- subgroup_data$or * exp(-1.96 * 0.2) subgroup_data$upper <- subgroup_data$or * exp(1.96 * 0.2) forest_plot(subgroup_data, point = "or", lower = "lower", upper = "upper", label = "study", split_by = c("sex", "age_group"), # Hierarchical: Sex > Age Group scale = "log", null_value = 1, theme = web_theme_modern(), axis_label = "Odds Ratio" ) ``` The sidebar shows "Sex" as the first level header, with "Age Group" nested under each sex. ## Using split_forest() Directly For more control, use the `split_forest()` function with the pipe syntax: ```{r} subgroup_data |> web_spec( point = "or", lower = "lower", upper = "upper", label = "study", scale = "log", null_value = 1, columns = list( col_text("sex", header = "Sex"), col_text("age_group", header = "Age"), col_interval("OR (95% CI)") ) ) |> split_forest(by = c("sex", "age_group")) |> forest_plot() ``` ## Shared Axis Range By default, each sub-plot auto-scales its axis. Use `shared_axis = TRUE` for consistent comparison: ```{r} forest_plot(trial_data, point = "or", lower = "lower", upper = "upper", label = "study", split_by = "region", shared_axis = TRUE, # Same axis range across all regions scale = "log", null_value = 1, theme = web_theme_modern() ) ``` This ensures the same visual scale across all subgroups. ## Sidebar Features The sidebar includes: - **Section headers** - Split variable names displayed as uppercase labels - **Collapsible** - Click the chevron to collapse the sidebar - **Search** - Filter subgroups by typing in the search box - **Keyboard navigation** - Use arrow keys to move between subgroups - **Bold active** - Currently selected subgroup is highlighted ## Exporting Split Forests Export all sub-plots to a directory structure: ```r split_result <- data |> web_spec(point = "or", lower = "lower", upper = "upper") |> split_forest(by = c("sex", "age_group")) # Exports to: output/Male/Male_Young.svg, etc. save_split_forest(split_result, "output", format = "svg") ``` The directory structure mirrors the split hierarchy. ## When to Use Split vs. Group | Feature | `group =` | `split_by =` | |---------|-----------|--------------| | Display | Single plot with collapsible rows | Separate plots with sidebar | | Navigation | Click group headers | Click sidebar items | | Comparison | All visible at once | One subgroup at a time | | Best for | Small number of groups | Many subgroups, detailed analysis | **Use `group`** when you want to see all data in one view with collapsible sections. **Use `split_by`** when each subgroup deserves its own focused view, or when you have many subgroups that would clutter a single plot.