使用tbl_strata+tbl_summary分层表格异常：统计方式不符预期

问题描述 #

需要生成按group_category和Sex双重分层的表格，要求所有变量以mean(se)统计。但使用tbl_strata结合tbl_summary的分层代码时，delta_IL1B变量却被以n(%)统计，而仅按group_category单分组的代码中该变量统计正常。

单分组正常代码 #

cyto_table <- ccf.data.merge %>% #data frame
  select(group_category, delta_TNFa, delta_IFNy, delta_IL1B, delta_IL2, delta_IL6, delta_IL8) %>% #variables I want for my table
      tbl_summary(
        by = group_category, #group_category = treatment A and B
        statistic = list(all_continuous() ~ "{mean} ({se})"), missing = "no") %>%
      modify_header(label ~ "**Variable**") %>%
      bold_labels()
cyto_table

分层异常代码 #

cyto_table <- ccf.data.merge %>%
  select(group_category,Sex,delta_TNFa, delta_IFNy, delta_IL1B, delta_IL2, delta_IL6, delta_IL8) %>% #variables I want for my table
  dplyr::mutate(
    Sex = factor(Sex, labels = c("Female", "Male"))) %>% #I am not sure the reasoning behind this line. the original code had this, I'm not gonna argue.
  tbl_strata(
    strata = group_category, #here the code is layering the able to be stratified by the treatments (A and B).
    ~.x %>% #I believe this line is making a function for tbl_summary to pass through.
      tbl_summary(
        by = Sex, #finally Sex is added here so that the data is then split by Sex within each treatment.
        statistic = list(all_continuous() ~ "{mean} ({se})"), missing = "no") %>%
      modify_header(label ~ "**Variable**") %>%
      bold_labels())
cyto_table

异常原因 #

核心原因是分层后的子数据集里，delta_IL1B被tbl_summary自动识别为分类变量：

• 当按group_category分层后，某个子组内的delta_IL1B可能只有唯一值（比如全部为0），或者数值数量过少，触发了tbl_summary的自动类型判断逻辑，将其归类为分类变量。
• 单分组时，整个数据集的delta_IL1B是连续分布的，所以被正确识别为连续变量，使用mean(se)统计。

解决方法 #

1. 强制指定变量类型 #

在tbl_summary中通过type参数强制将delta_IL1B标记为连续变量，忽略自动识别结果：

cyto_table <- ccf.data.merge %>%
  select(group_category,Sex,delta_TNFa, delta_IFNy, delta_IL1B, delta_IL2, delta_IL6, delta_IL8) %>%
  dplyr::mutate(Sex = factor(Sex, labels = c("Female", "Male"))) %>%
  tbl_strata(
    strata = group_category,
    ~.x %>%
      tbl_summary(
        by = Sex,
        # 强制指定delta_IL1B为连续变量
        type = list(delta_IL1B ~ "continuous"),
        statistic = list(all_continuous() ~ "{mean} ({se})"), 
        missing = "no") %>%
      modify_header(label ~ "**Variable**") %>%
      bold_labels())
cyto_table

2. 检查分层子数据集的变量分布 #

可以临时在tbl_strata中添加打印代码，查看每个分层组内delta_IL1B的数值情况，确认是否存在唯一值或类型异常：

cyto_table <- ccf.data.merge %>%
  select(group_category,Sex,delta_TNFa, delta_IFNy, delta_IL1B, delta_IL2, delta_IL6, delta_IL8) %>%
  dplyr::mutate(Sex = factor(Sex, labels = c("Female", "Male"))) %>%
  tbl_strata(
    strata = group_category,
    ~{
      # 打印当前分层组的delta_IL1B统计信息
      cat("当前分层组：", unique(.x$group_category), "\n")
      print(summary(.x$delta_IL1B))
      .x %>%
        tbl_summary(
          by = Sex,
          statistic = list(all_continuous() ~ "{mean} ({se})"), 
          missing = "no") %>%
        modify_header(label ~ "**Variable**") %>%
        bold_labels()
    })

根据打印结果，如果发现子组内delta_IL1B确实无变异，可以考虑合并组或者在统计时注明情况。

3. 提前统一变量类型 #

在数据预处理阶段，明确将delta_IL1B转换为数值型，避免分层时出现类型转换异常：

cyto_table <- ccf.data.merge %>%
  select(group_category,Sex,delta_TNFa, delta_IFNy, delta_IL1B, delta_IL2, delta_IL6, delta_IL8) %>%
  dplyr::mutate(
    Sex = factor(Sex, labels = c("Female", "Male")),
    # 强制转换为数值型
    delta_IL1B = as.numeric(delta_IL1B)
  ) %>%
  tbl_strata(
    strata = group_category,
    ~.x %>%
      tbl_summary(
        by = Sex,
        statistic = list(all_continuous() ~ "{mean} ({se})"), 
        missing = "no") %>%
      modify_header(label ~ "**Variable**") %>%
      bold_labels())
cyto_table

内容的提问来源于stack exchange，提问作者Tara Mahmood