AIclinicalresearch/r-statistics-service/tools/t_test_paired.R

#' @tool_code ST_T_TEST_PAIRED
#' @name 配对 T 检验
#' @version 1.0.0
#' @description 配对设计的均值差异检验（前后对比）
#' @author SSA-Pro Team

library(glue)
library(ggplot2)
library(base64enc)

run_analysis <- function(input) {
  # ===== 初始化 =====
  logs <- c()
  log_add <- function(msg) { logs <<- c(logs, paste0("[", Sys.time(), "] ", msg)) }
  
  on.exit({}, add = TRUE)
  
  # ===== 数据加载 =====
  log_add("开始加载输入数据")
  df <- tryCatch(
    load_input_data(input),
    error = function(e) {
      log_add(paste("数据加载失败:", e$message))
      return(NULL)
    }
  )
  
  if (is.null(df)) {
    return(make_error(ERROR_CODES$E100_INTERNAL_ERROR, details = "数据加载失败"))
  }
  log_add(glue("数据加载成功: {nrow(df)} 行, {ncol(df)} 列"))
  
  p <- input$params
  guardrails_cfg <- input$guardrails
  
  before_var <- p$before_var
  after_var <- p$after_var
  
  # ===== 参数校验 =====
  if (!(before_var %in% names(df))) {
    return(make_error(ERROR_CODES$E001_COLUMN_NOT_FOUND, col = before_var))
  }
  if (!(after_var %in% names(df))) {
    return(make_error(ERROR_CODES$E001_COLUMN_NOT_FOUND, col = after_var))
  }
  
  # ===== 数据清洗 =====
  original_rows <- nrow(df)
  df <- df[!is.na(df[[before_var]]) & !is.na(df[[after_var]]), ]
  
  removed_rows <- original_rows - nrow(df)
  if (removed_rows > 0) {
    log_add(glue("数据清洗: 移除 {removed_rows} 行缺失值 (剩余 {nrow(df)} 行)"))
  }
  
  before_vals <- df[[before_var]]
  after_vals <- df[[after_var]]
  diff_vals <- after_vals - before_vals
  
  n <- length(diff_vals)
  
  # ===== 护栏检查 =====
  guardrail_results <- list()
  warnings_list <- c()
  method_used <- "t.test"
  use_wilcoxon <- FALSE
  
  # 样本量检查
  sample_check <- check_sample_size(n, min_required = 10, action = ACTION_WARN)
  guardrail_results <- c(guardrail_results, list(sample_check))
  log_add(glue("样本量检查: N = {n}, {sample_check$reason}"))
  
  # 差值正态性检验
  if (isTRUE(guardrails_cfg$check_normality) && n >= 3) {
    log_add("执行差值正态性检验")
    norm_check <- check_normality(diff_vals, alpha = 0.05, 
                                  action = ACTION_SWITCH, 
                                  action_target = "Wilcoxon signed-rank test")
    guardrail_results <- c(guardrail_results, list(norm_check))
    log_add(glue("差值正态性: p = {round(norm_check$p_value, 4)}, {norm_check$reason}"))
  }
  
  guardrail_status <- run_guardrail_chain(guardrail_results)
  
  if (guardrail_status$status == "blocked") {
    return(list(
      status = "blocked",
      message = guardrail_status$reason,
      trace_log = logs
    ))
  }
  
  if (guardrail_status$status == "switch") {
    use_wilcoxon <- TRUE
    log_add(glue("触发方法切换: {guardrail_status$reason}"))
    warnings_list <- c(warnings_list, "差值不满足正态性，自动切换为 Wilcoxon 符号秩检验")
  }
  
  if (length(guardrail_status$warnings) > 0) {
    warnings_list <- c(warnings_list, guardrail_status$warnings)
  }
  
  # ===== 核心计算 =====
  if (use_wilcoxon) {
    log_add("执行 Wilcoxon 符号秩检验")
    result <- wilcox.test(before_vals, after_vals, paired = TRUE, exact = FALSE)
    method_used <- "Wilcoxon signed rank test"
    
    # Wilcoxon 效应量 r
    z_value <- qnorm(result$p.value / 2) * sign(median(diff_vals))
    effect_r <- abs(z_value) / sqrt(n)
    effect_interpretation <- if (abs(effect_r) < 0.1) "微小" else if (abs(effect_r) < 0.3) "小" else if (abs(effect_r) < 0.5) "中等" else "大"
    
    output_results <- list(
      method = method_used,
      statistic = jsonlite::unbox(as.numeric(result$statistic)),
      p_value = jsonlite::unbox(as.numeric(result$p.value)),
      p_value_fmt = format_p_value(result$p.value),
      effect_size = list(
        r = jsonlite::unbox(round(effect_r, 4)),
        interpretation = effect_interpretation
      )
    )
  } else {
    log_add("执行配对 T 检验")
    result <- t.test(before_vals, after_vals, paired = TRUE)
    method_used <- "Paired t-test"
    
    # Cohen's d for paired samples
    mean_diff <- mean(diff_vals)
    sd_diff <- sd(diff_vals)
    cohens_d <- mean_diff / sd_diff
    effect_interpretation <- if (abs(cohens_d) < 0.2) "微小" else if (abs(cohens_d) < 0.5) "小" else if (abs(cohens_d) < 0.8) "中等" else "大"
    
    log_add(glue("t = {round(result$statistic, 3)}, df = {round(result$parameter, 1)}, p = {round(result$p.value, 4)}, Cohen's d = {round(cohens_d, 3)}"))
    
    output_results <- list(
      method = method_used,
      statistic = jsonlite::unbox(as.numeric(result$statistic)),
      df = jsonlite::unbox(as.numeric(result$parameter)),
      p_value = jsonlite::unbox(as.numeric(result$p.value)),
      p_value_fmt = format_p_value(result$p.value),
      conf_int = as.numeric(result$conf.int),
      effect_size = list(
        cohens_d = jsonlite::unbox(round(cohens_d, 4)),
        interpretation = effect_interpretation
      )
    )
  }
  
  # 添加描述性统计
  output_results$descriptive <- list(
    before = list(
      variable = before_var,
      n = n,
      mean = round(mean(before_vals), 3),
      sd = round(sd(before_vals), 3),
      median = round(median(before_vals), 3)
    ),
    after = list(
      variable = after_var,
      n = n,
      mean = round(mean(after_vals), 3),
      sd = round(sd(after_vals), 3),
      median = round(median(after_vals), 3)
    ),
    difference = list(
      mean = round(mean(diff_vals), 3),
      sd = round(sd(diff_vals), 3),
      median = round(median(diff_vals), 3)
    )
  )
  
  # ===== 生成图表 =====
  log_add("生成配对比较图")
  plot_base64 <- tryCatch({
    generate_paired_plot(df, before_var, after_var, diff_vals)
  }, error = function(e) {
    log_add(paste("图表生成失败:", e$message))
    NULL
  })
  
  # ===== 生成可复现代码 =====
  original_filename <- if (!is.null(input$original_filename) && nchar(input$original_filename) > 0) {
    input$original_filename
  } else {
    "data.csv"
  }
  
  reproducible_code <- glue('
# SSA-Pro 自动生成代码
# 工具: 配对 T 检验
# 时间: {Sys.time()}
# ================================

library(ggplot2)

# 数据准备
df <- read.csv("{original_filename}")
before_var <- "{before_var}"
after_var <- "{after_var}"

# 数据清洗
df <- df[!is.na(df[[before_var]]) & !is.na(df[[after_var]]), ]

# 配对 T 检验
before_vals <- df[[before_var]]
after_vals <- df[[after_var]]
result <- t.test(before_vals, after_vals, paired = TRUE)
print(result)

# Cohen d (效应量)
diff_vals <- after_vals - before_vals
cohens_d <- mean(diff_vals) / sd(diff_vals)
cat("Cohen d =", round(cohens_d, 3), "\\n")

# 可视化
df_long <- data.frame(
  id = rep(1:nrow(df), 2),
  time = rep(c("Before", "After"), each = nrow(df)),
  value = c(before_vals, after_vals)
)
ggplot(df_long, aes(x = time, y = value, group = id)) +
  geom_line(alpha = 0.3) +
  geom_point() +
  theme_minimal() +
  labs(title = "Paired Comparison")
')
  
  # ===== 返回结果 =====
  log_add("分析完成")
  
  return(list(
    status = "success",
    message = "分析完成",
    warnings = if (length(warnings_list) > 0) warnings_list else NULL,
    results = output_results,
    plots = if (!is.null(plot_base64)) list(plot_base64) else list(),
    trace_log = logs,
    reproducible_code = as.character(reproducible_code)
  ))
}

# 辅助函数：生成配对比较图
generate_paired_plot <- function(df, before_var, after_var, diff_vals) {
  n <- nrow(df)
  
  # 创建长格式数据
  df_long <- data.frame(
    id = rep(1:n, 2),
    time = factor(rep(c("Before", "After"), each = n), levels = c("Before", "After")),
    value = c(df[[before_var]], df[[after_var]])
  )
  
  p <- ggplot(df_long, aes(x = time, y = value)) +
    geom_line(aes(group = id), alpha = 0.3, color = "gray60") +
    geom_point(aes(group = id), alpha = 0.5, size = 2) +
    stat_summary(fun = mean, geom = "point", size = 4, color = "#ef4444", shape = 18) +
    stat_summary(fun = mean, geom = "line", aes(group = 1), color = "#ef4444", size = 1.2) +
    theme_minimal() +
    labs(
      title = paste("Paired Comparison:", before_var, "vs", after_var),
      subtitle = paste("n =", n, ", Mean change =", round(mean(diff_vals), 2)),
      x = "Time Point",
      y = "Value"
    )
  
  tmp_file <- tempfile(fileext = ".png")
  ggsave(tmp_file, p, width = 6, height = 5, dpi = 100)
  base64_str <- base64encode(tmp_file)
  unlink(tmp_file)
  
  return(paste0("data:image/png;base64,", base64_str))
}