如何在R中批量执行受试者与基线的Mann-Whitney检验

1. 模拟示例数据 #

先构造一个与你描述结构一致的示例数据框，方便测试代码：

set.seed(123)
DF <- data.frame(
  Subject = rep(c("Patient1", "Patient1(B)", "Patient2", "Patient2(B)", "Patient3", "Patient3(B)"), each = 5),
  Avg_Score = rnorm(30, mean = 70, sd = 10),
  EOTM = rnorm(30, mean = 50, sd = 8),
  FSO_1 = rnorm(30, mean = 60, sd = 12)
)

2. 基础循环实现（逻辑直观） #

这种方法适合快速理解核心逻辑，在中等规模数据集上表现稳定：

# 提取所有非基线受试者ID（去除(B)标记）
subject_ids <- unique(gsub("\\(B\\)", "", DF$Subject))
subject_ids <- subject_ids[subject_ids != ""]

# 初始化存储结果的列表
p_results <- list()

# 循环处理每个受试者
for (id in subject_ids) {
  # 提取当前受试者的非基线数据
  non_baseline <- DF[DF$Subject == id, c("Avg_Score", "EOTM", "FSO_1")]
  # 提取对应基线组的数据
  baseline <- DF[DF$Subject == paste0(id, "(B)"), c("Avg_Score", "EOTM", "FSO_1")]
  
  # 对三个指标分别执行Mann-Whitney检验，提取p值
  p_avg <- wilcox.test(non_baseline$Avg_Score, baseline$Avg_Score)$p.value
  p_eotm <- wilcox.test(non_baseline$EOTM, baseline$EOTM)$p.value
  p_fso1 <- wilcox.test(non_baseline$FSO_1, baseline$FSO_1)$p.value
  
  # 将结果存入列表
  p_results[[id]] <- data.frame(
    Subject = id,
    Avg_Score_p = p_avg,
    EOTM_p = p_eotm,
    FSO_1_p = p_fso1
  )
}

# 合并列表为最终结果数据框DF2
DF2 <- do.call(rbind, p_results)

3. 大数据集优化实现（dplyr+purrr） #

如果数据集规模较大（如上万个受试者），推荐使用向量化分组处理，效率比显式循环更高：

library(dplyr)
library(purrr)

# 预处理数据：添加分组ID和基线标记
DF_processed <- DF %>%
  mutate(
    Group_ID = gsub("\\(B\\)", "", Subject),
    Is_Baseline = grepl("\\(B\\)", Subject)
  )

# 分组执行检验并生成结果
DF2 <- DF_processed %>%
  group_by(Group_ID) %>%
  do({
    # 分离当前组的基线和非基线数据
    baseline_data <- .[.$Is_Baseline == TRUE, c("Avg_Score", "EOTM", "FSO_1")]
    non_baseline_data <- .[.$Is_Baseline == FALSE, c("Avg_Score", "EOTM", "FSO_1")]
    
    # 批量计算三个指标的p值
    p_vals <- map_dfr(c("Avg_Score", "EOTM", "FSO_1"), function(col) {
      test_result <- wilcox.test(non_baseline_data[[col]], baseline_data[[col]])
      tibble(!!paste0(col, "_p") := test_result$p.value)
    })
    
    # 组合受试者ID和p值结果
    tibble(Subject = first(.$Group_ID)) %>%
      bind_cols(p_vals)
  }) %>%
  ungroup() %>%
  select(-Group_ID)

注意事项 #

• 若数据存在缺失值，可在wilcox.test中添加na.rm = TRUE参数忽略缺失值
• 默认是双侧检验，如需单侧检验可添加alternative = "less"或alternative = "greater"参数

内容的提问来源于stack exchange，提问作者dkcongo