如何在gtsummary/bstfun的as_forest_plot函数中实现分类变量层级缩进？

使用gtsummary包的实验性函数as_forest_plot生成模型系数森林图时，遇到两个问题：

• 分类变量的层级缩进无法实现，尝试modify_column_indent函数未成功
• 无法显示分类变量的参考水平

原代码如下：

library(titanic)
library(tidyverse, warn.conflicts = FALSE)
library(gtsummary, warn.conflicts = FALSE)
library(bstfun, warn.conflicts = FALSE)
library(magrittr, warn.conflicts = FALSE)
library(janitor, warn.conflicts = FALSE)


titanic::titanic_train %>% 
  clean_names %>% 
  select(-c(name, parch, ticket, cabin, embarked)) %>% 
  mutate(pclass = factor(pclass)) %>% 
  glm(survived ~ . - passenger_id, data = ., family = "binomial") %>%
  tbl_regression(exponentiate = TRUE) %>% 
  modify_column_indent(columns = label, rows = (header_row == TRUE)) %>%
  modify_cols_merge(
  pattern = "{estimate} ({ci})",
  rows = !is.na(estimate)
) %>%
  modify_header(estimate = "OR (95% CI)") %>%
  as_forest_plot( 
    col_names = c("estimate", "p.value"),
    col = forestplot::fpColors(box = "darkred"))

输出结果：

1. 核心问题分析 #

• modify_column_indent的行选择条件错误：原代码中rows = (header_row == TRUE)仅针对表头行，而非分类变量的子行
• 默认未显示参考水平：tbl_regression默认隐藏分类变量的参考水平，需手动开启
• as_forest_plot作为实验性函数，需要确保缩进设置能正确传递到forestplot的标签中

2. 修改后的代码 #

library(titanic)
library(tidyverse, warn.conflicts = FALSE)
library(gtsummary, warn.conflicts = FALSE)
library(bstfun, warn.conflicts = FALSE)
library(magrittr, warn.conflicts = FALSE)
library(janitor, warn.conflicts = FALSE)

titanic::titanic_train %>% 
  clean_names %>% 
  select(-c(name, parch, ticket, cabin, embarked)) %>% 
  mutate(pclass = factor(pclass)) %>% 
  glm(survived ~ . - passenger_id, data = ., family = "binomial") %>%
  # 开启参考水平显示
  tbl_regression(exponentiate = TRUE, show_reference_rows = TRUE) %>% 
  # 为分类变量的子行添加缩进（label与variable不同的行即为子行）
  modify_column_indent(columns = label, rows = variable != label) %>%
  modify_cols_merge(
    pattern = "{estimate} ({ci})",
    rows = !is.na(estimate)
  ) %>%
  # 补全参考水平的统计值，避免森林图空值
  modify_table_body(
    mutate,
    estimate = ifelse(is.na(estimate), 1, estimate),
    ci = ifelse(is.na(ci), "1.00 (1.00, 1.00)", ci)
  ) %>%
  modify_header(estimate = "OR (95% CI)") %>%
  as_forest_plot( 
    col_names = c("estimate", "p.value"),
    col = forestplot::fpColors(box = "darkred")
  )

3. 关键修改说明 #

• 显示参考水平：给tbl_regression添加show_reference_rows = TRUE参数，让分类变量的参考水平行显示出来
• 正确设置缩进：将modify_column_indent的行条件改为rows = variable != label，针对分类变量的子行（非主变量行）添加缩进
• 补全参考水平统计值：通过modify_table_body将参考水平的estimate设为1，CI设为(1.00,1.00)，保证森林图显示完整

内容的提问来源于stack exchange，提问作者Ryan