LOCF插补VISUAL52变量缺失值报错排查及描述统计方法验证咨询

Hey there, let's work through your questions one by one to get things sorted out:

The complete() function from the mice package is exclusively designed for mids objects—the specialized output you get when running multiple imputation with mice(). But your imp_locf is just a plain numeric vector (created by data.table::nafill()), not a mids object. R can't find a matching method to run complete_ on a basic numeric vector, hence the error.

If you just want to add the imputed values back to your original dataset, you don't need complete() at all—just assign the vector directly:

hw3$VISUAL52_imputed <- imp_locf

First, a quick critical note: Your dataset has repeated measures for each individual (VISUAL0, VISUAL4, etc. are time-points for the same subject). If you're using LOCF as it's intended for longitudinal data—filling missing VISUAL52 values with the last non-missing VISUAL measurement from that same individual—your original approach only operating on the VISUAL52 column is incorrect (it doesn't use the subject's prior time-point data). Here are valid implementations:

Option 1: Use zoo::na.locf() (row-wise for individual time-series) #

Convert each subject's VISUAL measurements into a row-wise time-series, apply LOCF, then extract the filled VISUAL52 value:

library(zoo)
# Create a matrix of all VISUAL time-points ordered by time
visual_timepoints <- as.matrix(hw3[, paste0("VISUAL", c(0, 4, 12, 24, 52))])
# Apply LOCF row-by-row, take the last value (which corresponds to VISUAL52)
hw3$VISUAL52_locf <- apply(visual_timepoints, 1, function(x) na.locf(x, na.rm = FALSE)[length(x)])

Option 2: Tidyverse approach (dplyr + tidyr) #

Reshape your data to long format, group by individual, apply LOCF, then reshape back to wide format:

library(dplyr)
library(tidyr)

hw3_with_locf <- hw3 %>%
  # Add a unique ID for each subject (since your dataset doesn't include one)
  mutate(subject_id = row_number()) %>%
  # Reshape to long format: one row per subject-timepoint
  pivot_longer(cols = starts_with("VISUAL"),
               names_to = "timepoint",
               values_to = "visual_score") %>%
  # Ensure timepoints are ordered correctly to avoid LOCF misalignment
  mutate(timepoint = factor(timepoint, levels = paste0("VISUAL", c(0, 4, 12, 24, 52)))) %>%
  arrange(subject_id, timepoint) %>%
  # Apply LOCF within each subject's time-series
  group_by(subject_id) %>%
  mutate(visual_locf = na.locf(visual_score, na.rm = FALSE)) %>%
  ungroup() %>%
  # Reshape back to wide format and extract the filled VISUAL52 value
  pivot_wider(names_from = timepoint,
              values_from = c(visual_score, visual_locf)) %>%
  select(subject_id, VISUAL52_locf = visual_locf_VISUAL52) %>%
  # Merge the imputed column back to the original dataset
  right_join(hw3, by = "subject_id")

Option 3: If you meant column-wise LOCF (fill down within VISUAL52 only) #

If your goal was to fill missing VISUAL52 values with the last non-missing value from the row above (not subject-specific prior data), you can also use zoo::na.locf() directly on the vector:

hw3$VISUAL52_locf <- na.locf(hw3$VISUAL52, na.rm = FALSE)

Your current Hmisc::numSummary() works, but there are more flexible or report-friendly options tailored to grouped continuous data:

Option 1: dplyr for fully custom grouped stats #

This lets you define exactly which statistics you want, with clean, readable output:

library(dplyr)
hw3 %>%
  group_by(treat) %>%
  summarise(
    sample_size = sum(!is.na(VISUAL52)),
    mean_score = mean(VISUAL52, na.rm = TRUE),
    se_mean = sd(VISUAL52, na.rm = TRUE)/sqrt(sample_size),
    q0 = quantile(VISUAL52, 0, na.rm = TRUE),
    q25 = quantile(VISUAL52, 0.25, na.rm = TRUE),
    q50 = quantile(VISUAL52, 0.5, na.rm = TRUE),
    q75 = quantile(VISUAL52, 0.75, na.rm = TRUE),
    q100 = quantile(VISUAL52, 1, na.rm = TRUE)
  )

Option 2: finalfit (you already loaded this!) #

Perfect for generating stats ready for clinical/academic reports, with minimal code and publication-friendly formatting:

library(finalfit)
hw3 %>%
  summary_factorlist(dependent = "VISUAL52",
                     explanatory = "treat",
                     cont = c("mean", "sd", "quantiles"),
                     quantiles = c(0, 0.25, 0.5, 0.75, 1))

Option 3: psych::describeBy() for comprehensive stats #

If you want extra metrics like skewness, kurtosis, and range alongside the basics:

library(psych)
describeBy(hw3$VISUAL52, group = hw3$treat, na.rm = TRUE)

内容的提问来源于stack exchange，提问作者12666727b9