使用gtsummary与survey包时遇问题：找不到Variable列

问题背景 #

使用survey::svyglm构建准二项逻辑回归模型，模型本身运行正常，但调用gtsummary::tbl_regression生成表格并添加显著性星号时，报错提示"找不到'Variable列'"，报错溯源到add_significance_stars函数。

模型构建代码 #

PSYDISTMODEL2015to2019 <- survey::svyglm(
  spdmon ~ EcstasyVariable + PSILCY2+ lsdflag + PEYOTE2 +MESC2 + MarriageVariable+  EducationVariable+ CATAG6+irsex+ income+ NEWRACE2+ sexident +KetamineVar +  DMTVariable+PoppersVariable+
    RiskyVariable+ cocflag+ herflag+ pcpflag+  InhalantVariable + PainRelieverVariable+ TranquilizerVariable +
    StimulantVariable+ SedativeVariable+ mrjflag+year, 
  data=NSDUH2015to2019Data,
  family=quasibinomial(), 
  design=totalDesign
)
summary(PSYDISTMODEL2015to2019)
DistressModelOdds2015to2019 <- exp(cbind(OR = coef(PSYDISTMODEL2015to2019), confint(PSYDISTMODEL2015to2019)))
DistressModelOdds2015to2019 <- round(DistressModelOdds2015to2019, digits =2)
DistressModelOdds2015to2019[c(2,3),]

# 模型相关输出均正常

原表格生成代码（报错版本） #

#Past Month Psy Distress
PSYDIST2015to2019OverallTable <- tbl_regression(PSYDISTMODEL2015to2019,exponentiate = TRUE, include = c(
  "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
  "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
  "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar")
    
  , show_single_row = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  ))  %>%  add_significance_stars(pattern = "{estimate}{stars} ({conf.low}, {conf.high})",
                                  hide_ci = TRUE, hide_se = TRUE) %>% 
  modify_table_styling(
    columns = estimate,
    rows = p.value < 0.05 & estimate < 1,
    text_format = "bold")  %>% modify_header(estimate ~ "**aOR (95% CI)**") %>%
  modify_header(label ~ "**Lifetime Use**") %>%
  modify_footnote(estimate ~ "aOR = adjusted odds ratio; CI = confidence interval", abbreviation = TRUE) 

报错信息 #

Error in `scope_tidy()`:
! The `.$x` data frame does not have the required "variable" column.
Backtrace:
  1. ... %>% ...
  6. gtsummary::add_significance_stars(...)
  7. gtsummary::modify_footnote(x, any_of(pattern_cols[1]) ~ p_footnote)
  8. gtsummary:::.combine_update_and_dots(...)
  9. broom.helpers::.formula_list_to_named_list(...)
 10. broom.helpers:::.single_formula_to_list(...)
 11. broom.helpers::.select_to_varnames(...)
 12. broom.helpers::scope_tidy(var_info, data)

解决方案 #

1. 升级依赖包 #

首先确保gtsummary和broom.helpers是最新版本，旧版本对survey模型的兼容性存在bug：

install.packages(c("gtsummary", "broom.helpers"))

2. 调整add_significance_stars参数 #

报错核心原因是设置hide_ci=TRUE后，conf.low和conf.high列被移除，但pattern里还在调用这些列，导致后续函数找不到对应字段。修改方式二选一：

方式一：保留CI列，不设置hide_ci=TRUE

PSYDIST2015to2019OverallTable <- tbl_regression(
  PSYDISTMODEL2015to2019,
  exponentiate = TRUE,
  include = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  ),
  show_single_row = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  )
) %>%
  add_significance_stars(
    pattern = "{estimate}{stars} ({conf.low}, {conf.high})",
    hide_se = TRUE
  ) %>%
  modify_table_styling(
    columns = estimate,
    rows = p.value < 0.05 & estimate < 1,
    text_format = "bold"
  ) %>%
  modify_header(
    estimate ~ "**aOR (95% CI)**",
    label ~ "**Lifetime Use**"
  ) %>%
  modify_footnote(
    estimate ~ "aOR = adjusted odds ratio; CI = confidence interval",
    abbreviation = TRUE
  )

方式二：不在pattern中包含CI，通过表头说明

PSYDIST2015to2019OverallTable <- tbl_regression(
  PSYDISTMODEL2015to2019,
  exponentiate = TRUE,
  include = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  ),
  show_single_row = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  )
) %>%
  add_significance_stars(
    pattern = "{estimate}{stars}",
    hide_se = TRUE,
    hide_ci = TRUE
  ) %>%
  modify_table_styling(
    columns = estimate,
    rows = p.value < 0.05 & estimate < 1,
    text_format = "bold"
  ) %>%
  modify_header(
    estimate ~ "**aOR (95% CI)**",
    label ~ "**Lifetime Use**"
  ) %>%
  modify_footnote(
    estimate ~ "aOR = adjusted odds ratio; CI = confidence interval",
    abbreviation = TRUE
  )

3. 分步调试定位问题 #

如果上述方法无效，分步执行代码，先生成基础表格确认正常，再逐步添加修饰：

# 生成基础表格
base_table <- tbl_regression(
  PSYDISTMODEL2015to2019,
  exponentiate = TRUE,
  include = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  ),
  show_single_row = c(
    "EcstasyVariable", "PSILCY2", "lsdflag", "PEYOTE2", "MESC2", "cocflag", "herflag",
    "pcpflag", "InhalantVariable", "PainRelieverVariable", "TranquilizerVariable", "StimulantVariable",
    "SedativeVariable", "mrjflag", "PoppersVariable", "DMTVariable", "KetamineVar"
  )
)

# 检查基础表格是否正常
print(base_table)

# 逐步添加修饰
final_table <- base_table %>%
  add_significance_stars(hide_se = TRUE) %>%
  modify_table_styling(columns = estimate, rows = p.value < 0.05 & estimate < 1, text_format = "bold") %>%
  modify_header(estimate ~ "**aOR (95% CI)**", label ~ "**Lifetime Use**") %>%
  modify_footnote(estimate ~ "aOR = adjusted odds ratio; CI = confidence interval", abbreviation = TRUE)

内容的提问来源于stack exchange，提问作者user29407182