clinpubr: Clinical Publication

Overview

clinpubr is an R package designed to streamline the workflow from clinical data processing to publication-ready outputs. It provides tools for clinical data cleaning, significant result screening, and generating tables/figures suitable for medical journals.

Key Features

• Clinical Data Cleaning: Functions to handle missing values, standardize units, convert dates, and clean numerical/categorical variables.
• Result Screening: Screening results of regression and interaction analysis with common variable transformations to identify key findings.
• Publication-Ready Outputs: Generate baseline characteristic tables, forest plots, RCS curves, and other visualizations formatted for medical publications.

Installation

You can install clinpubr from CRAN with:

install.packages("clinpubr")

Optional Dependencies

Some functions require additional packages for full functionality. The package will automatically prompt you to install missing packages when needed. If you want to install the package with all dependencies, you can use:

install.packages("clinpubr", dependencies = TRUE)

Basic Usage

Cleaning Tools

Example 1.1: Standardize Values in Medical Records

library(clinpubr)

# Sample messy data
messy_data <- data.frame(values = c("１２．３", "0..45", "  67 ", "", "ａｂａｎｄｏｎ"))
clean_data <- value_initial_cleaning(messy_data$values)
print(clean_data)
#> [1] "12.3"    "0.45"    "67"      NA        "abandon"

Example 1.2: Check Non-numerical Values

# Sample messy data
x <- c("1.2(XXX)", "1.5", "0.82", "5-8POS", "NS", "FULL")
print(check_nonnum(x))
#> [1] "1.2(XXX)" "5-8POS"   "NS"       "FULL"

This function filters out non-numerical values, which helps you choose the appropriate method to handle them.

Example 1.3: Extracting Numerical Values from Text

# Sample messy data
x <- c("1.2(XXX)", "1.5", "0.82", "5-8POS", "NS", "FULL")
print(extract_num(x))
#> [1] 1.20 1.50 0.82 5.00   NA   NA

print(extract_num(x,
  res_type = "first", # Extract the first number
  multimatch2na = TRUE, # Convert illegal multiple matches to NA
  zero_regexp = "NEG|NS", # Convert "NEG" and "NS" (matched using regex) to 0
  max_regexp = "FULL", # Convert "FULL" (matched using regex) to some specified quantile
  max_quantile = 0.95
))
#> [1] 1.20 1.50 0.82   NA 0.00 1.47

Other Cleaning Functions

• to_date(): Convert text to date, can handle mixed format.
• unit_view() and unit_standardize(): Provide a pipeline to standardize conflicting units.
• cut_by(): Split numerics into factors, offers a variety of splitting options and auto labeling.
• And more…

Screening Results to Identify Potential Findings

data(cancer, package = "survival")

# Screening for potential findings with regression models in the cancer dataset
scan_result <- regression_scan(cancer, y = "status", time = "time", save_table = FALSE)
#> Taking all variables as predictors
knitr::kable(scan_result)

Generating Publication-Ready Tables and Figures

Example 3.1: Automatic Type Infer and Baseline Table Generation

cohort <- data.frame(
  age = c(17, 25, 30, NA, 50, 60),
  sex = c("M", "F", "F", "M", "F", "M"),
  value = c(1, NA, 3, 4, 5, NA),
  dementia = c(TRUE, FALSE, FALSE, FALSE, TRUE, FALSE)
)
res <- exclusion_count(
  cohort,
  age < 18,
  is.na(value),
  dementia == TRUE,
  .criteria_names = c(
    "Age < 18 years",
    "Missing value",
    "History of dementia"
  )
)
#> Warning in exclusion_count(cohort, age < 18, is.na(value), dementia == TRUE, :
#> Criterion 'Age < 18 years' resulted in NA values. These rows have been excluded
#> by default. Consider adding an explicit check for missing values (e.g.,
#> is.na(variable)) as a preceding criterion.
knitr::kable(res) # Display the table

Example 3.2: Automatic Type Infer and Baseline Table Generation

var_types <- get_var_types(mtcars, strata = "vs") # Automatically infer variable types
print(var_types)
#> $factor_vars
#> [1] "cyl"  "vs"   "am"   "gear"
#> 
#> $exact_vars
#> [1] "cyl"  "gear"
#> 
#> $nonnormal_vars
#> [1] "drat" "carb"
#> 
#> $omit_vars
#> NULL
#> 
#> $strata
#> [1] "vs"
#> 
#> attr(,"class")
#> [1] "var_types"

tables <- baseline_table(mtcars,
  var_types = var_types, contDigits = 1, save_table = FALSE,
  filename = "baseline.csv", seed = 1 # set seed for simulated fisher exact test
)
knitr::kable(tables$baseline) # Display the table

Example 3.3: RCS Plot

data(cancer, package = "survival")

# Performing cox regression, which is inferred by `y` and `time`
p <- rcs_plot(cancer, x = "age", y = "status", time = "time", covars = c("sex", "ph.karno"), save_plot = FALSE)
#> Warning in predictor_effect_plot(data = data, x = x, y = y, time = time, : 1
#> incomplete cases excluded.
plot(p)

Example 3.4: Interaction Plot

data(cancer, package = "survival")

# Generating interaction plot of both linear and RCS models
p <- interaction_plot(cancer,
  y = "status", time = "time", predictor = "age",
  group_var = "sex", save_plot = FALSE
)
plot(p$lin)

plot(p$rcs)

Example 3.5: Regression Forest Plot

data(cancer, package = "survival")
cancer$dead <- cancer$status == 2 # Preparing a binary variable for logistic regression
cancer$`age per 1 sd` <- c(scale(cancer$age)) # Standardizing age

# Performing multivairate logistic regression
p1 <- regression_forest(cancer,
  model_vars = c("age per 1 sd", "sex", "wt.loss"), y = "dead",
  as_univariate = FALSE, save_plot = FALSE
)
plot(p1)

p2 <- regression_forest(
  cancer,
  model_vars = list(
    Crude = c("age per 1 sd"),
    Model1 = c("age per 1 sd", "sex"),
    Model2 = c("age per 1 sd", "sex", "wt.loss")
  ),
  y = "dead",
  save_plot = FALSE
)
plot(p2)

Example 3.6: Subgroup Forest Plot

data(cancer, package = "survival")
# coxph model with time assigned
p <- subgroup_forest(cancer,
  subgroup_vars = c("age", "sex", "wt.loss"), x = "ph.ecog", y = "status",
  time = "time", covars = "ph.karno", ticks_at = c(1, 2), save_plot = FALSE
)
plot(p)

Example 3.7: Classification Model Performance

# Building models with example data
data(cancer, package = "survival")
df <- kidney
df$dead <- ifelse(df$time <= 100 & df$status == 0, NA, df$time <= 100)
df <- na.omit(df[, -c(1:3)])

model0 <- glm(dead ~ age + frail, family = binomial(), data = df)
model1 <- glm(dead ~ ., family = binomial(), data = df)
df$base_pred <- predict(model0, type = "response")
df$full_pred <- predict(model1, type = "response")

# Generating most of the useful plots and metrics for model comparison
results <- classif_model_compare(df, "dead", c("base_pred", "full_pred"), save_output = FALSE)
#> Assuming 'TRUE' is [Event] and 'FALSE' is [non-Event]

knitr::kable(results$metric_table)

plot(results$roc_plot)

plot(results$pr_plot)

plot(results$calibration_plot)

plot(results$dca_plot)

Example 3.8: Importance Plot

# Generating a dummy importance vector
set.seed(5)
dummy_importance <- runif(20, 0.2, 0.6)^5
names(dummy_importance) <- paste0("var", 1:20)

# Plotting variable importance, keeping only top 15 and splitting at 10
p <- importance_plot(dummy_importance, top_n = 15, split_at = 10, save_plot = FALSE)
plot(p)
#> Warning: Removed 1 row containing missing values or values outside the scale range
#> (`geom_bar()`).

Documentation

For detailed usage, refer to the package vignettes (coming soon) or the GitHub repository.

Contributing

Bug reports and feature requests are welcome via the issue tracker.

License

clinpubr is licensed under GPL (>= 3).