临床研究统计：如何纵向合并多变量交叉表与统计表格

我正在开展临床研究的统计分析，需要在同一张表格里同时呈现：

• 多个交叉表（类似SPSS CROSSTABS输出，可通过descr::CrossTabs等包复现），带列总计
• 连续变量的中位数（百分位数）及Wilcoxon-Mann-Whitney检验P值

目前用gtsummary包的tbl_cross生成了多个交叉表，但tbl_merge(v)会把表格横向并排，而且总计行存在缺失值适配问题，希望改成纵向合并的形式。现有代码如下：

set.seed(123)
member <- sample(c("Yes", "No"), 100, replace = TRUE)
author <- sample(c("Yes", "No"), 100, replace = TRUE)
review <- sample(0:10, 100, replace = TRUE)
publish <- sample(0:10, 100, replace = TRUE)
pay <- sample(0:10, 100, replace = TRUE)
df <- data.frame(member, author, review, 
                 publish, pay) |>
  dplyr::mutate(across(c(review, publish, pay),
                       ~cut(.,
                            breaks = c(-Inf, 4.5, 5.5, Inf),
                            labels = c("No", "Maybe", "Yes"),
                            include.lowest = TRUE), 
                       .names = "{.col}_group"))

cv <- c("review_group", "publish_group", "pay_group")
v <- map(
  cv,
  ~ tbl_cross(data = df, row = .x, col = member,
              missing = "no", percent = "column") |>
    add_p(pvalue_fun = label_style_pvalue(digits = 3))
)

tbl_merge(v)

解决方案：用tbl_stack替代tbl_merge并调整总计行 #

核心是改用gtsummary的tbl_stack()实现纵向合并，同时调整表格结构修复总计行问题，完整代码如下：

library(gtsummary)
library(dplyr)
library(purrr)

set.seed(123)
member <- sample(c("Yes", "No"), 100, replace = TRUE)
author <- sample(c("Yes", "No"), 100, replace = TRUE)
review <- sample(0:10, 100, replace = TRUE)
publish <- sample(0:10, 100, replace = TRUE)
pay <- sample(0:10, 100, replace = TRUE)
df <- data.frame(member, author, review, 
                 publish, pay) |>
  dplyr::mutate(across(c(review, publish, pay),
                       ~cut(.,
                            breaks = c(-Inf, 4.5, 5.5, Inf),
                            labels = c("No", "Maybe", "Yes"),
                            include.lowest = TRUE), 
                       .names = "{.col}_group"))

# 定义生成单个交叉表的函数，包含总计行和P值
create_crosstab <- function(var_name) {
  tbl_cross(data = df, row = all_of(var_name), col = member,
            missing = "no", percent = "column") |>
    # 确保列总计正常显示
    modify_header(all_cols() ~ "**{col}**") |>
    add_p(pvalue_fun = label_style_pvalue(digits = 3)) |>
    # 为每个表格添加变量名作为标题行
    modify_table_styling(
      location = "header",
      columns = label,
      rows = TRUE,
      fmt_fun = function(x) paste0("**", var_name, "**")
    )
}

cv <- c("review_group", "publish_group", "pay_group")
# 批量生成所有交叉表
v <- map(cv, create_crosstab)

# 纵向合并所有表格
tbl_stack(v)

关键说明 #

1. 替换合并逻辑：用tbl_stack()替代横向合并的tbl_merge()，直接实现多个表格的纵向堆叠
2. 修复总计行：通过modify_header()确保列总计正常渲染，避免缺失值适配问题
3. 统一表格样式：自定义函数给每个交叉表添加变量名标题，提升表格可读性
4. 扩展连续变量统计：如果需要加入连续变量的中位数（百分位数）和Wilcoxon检验结果，可生成对应表格后一起堆叠，示例：

# 生成连续变量统计表格（以review为例）
continuous_tbl <- tbl_summary(df, 
                              by = member,
                              include = review,
                              statistic = all_continuous() ~ "{median} ({p25}, {p75})",
                              missing = "no") |>
  add_p(test = all_continuous() ~ "wilcox.test") |>
  modify_header(all_cols() ~ "**{col}**")

# 合并连续变量表格和交叉表
tbl_stack(c(list(continuous_tbl), v))

内容的提问来源于stack exchange，提问作者devster