Package Design vignette for {cleanepi}

Concept and motivation

In this document, we will outline the design decisions that have steered the development strategies of the {cleanepi} R package, along with the rationale behind each decision and the potential advantages and disadvantages associated with them.

Data cleaning is an important phase for ensuring the efficacy of downstream analysis. The procedures entailed in the cleaning process may differ based on the data type and research objectives. Nonetheless, certain steps can be applied universally across diverse data types, irrespective of their origin.

Design decisions

The {cleanepi} R package is designed to offer functional programming-style data cleansing tasks. To streamline the organization of data cleaning operations, we have categorized them into distinct groups referred to as modules. These modules are based on overarching goals derived from commonly anticipated data cleaning procedures. Each module features a primary function along with additional helper functions tailored to accomplish specific tasks. It’s important to note that, except for few cases where the outcome from a helper function can impact on the cleaning task, only the main function of each module will be exported. This deliberate choice empowers users to execute individual cleaning tasks as needed, enhancing flexibility and usability.

cleanepi design diagram

At the core of {cleanepi}, the pivotal function clean_data() serves as a wrapper encapsulating all the modules, as illustrated in the figure above. This function is intended to be the primary entry point for users seeking to cleanse their data. It performs the cleaning operations as requested by the user through the set of parameters that need to be explicitly defined. Furthermore, multiple cleaning operations can be performed sequentially using the “pipe” operator (%>%). In addition, this package also has two surrogate functions:

1. scan_data(): Columns of type character might contain values of other types such as numeric, date, logical. This function enables users to assess the data types present in each character column of their dataset. The composition in data types of character columns will inform the user about what actions need to be performed on the data. Most frequent scenarios involve the presence of:
   • date values in either Date or numeric format (when date column is imported from MS Excel),
   • character values in a logical column (a not available within a column of TRUE or FALSE),
   • numbers written in letters.

When the input data contains character columns, the function returns a data frame with the same number of row as the character columns and six columns representing their column names, proportion of missing, numeric, date, character, and logical values respectively. We transpose the result relative to the input dataset (columns in the input are returned as row) to avoid horizontal scrolling in the case of datasets with a large number of character columns. The sum of the proportion across all columns is not always equal to 1 for the reason below: * When numeric values are found in a character column, they will be subjected to the followings: 1. conversion into Date using lubridate::as_date() with origin = as.Date("1900-01-01"). 2. conversion into Date using date_guess() function used by the standardize_dates() function. Numeric values that are successfully converted into date from either of the methods above are considered as potential dates. They will be added to the date count if they fall within the interval [50 years back from today’s date, today’s date].

There is no ambiguity for the columns of type Date, logical, and numeric. Values in such columns are expected to be of the same type. Hence, the function will not be applied on columns other than character columns. Consequently, it invisibly returns NA when applied on a dataset with no character columns, after printing out a message about the absence of character columns from the input dataset.

2. print_report(): By utilizing this function, users can visualize the report generated from each applied cleaning task, facilitating transparency and understanding of the data cleaning process.

Scope

{cleanepi} is an R package crafted to clean, curate, and standardize tabular datasets, with a particular focus on epidemiological data. In the architecture of {cleanepi}, the data cleaning operations are categorized into modules, each provides a specific data cleaning task. The modules in the current version of {cleanepi} encompass the:

• standardization of column names,
• detection and removal of duplicates,
• replacement of missing values with NA,
• standardization of subject IDs,
• standardization of date values,
• replacement of existing values with predefined ones (dictionary based substitutions),
• conversion of values when necessary,
• verification of the sequence order of date-events, and
• transformation of selected columns.

By compartmentalizing these operations into modules, {cleanepi} offers users a systematic and adaptable framework to address diverse data cleaning needs, especially within the realm of epidemiological datasets.

Input

The primary functions of the modules, as well as the core function clean_data(), accept input in the form of a data.frame or linelist. This offers flexibility for users regarding where they want to position {cleanepi} within the R package ecosystem for epidemic analysis pipelines, either to clean data before or after converting it to a linelist.

In addition to the target dataset, the core function clean_data() accepts other parameters which are specific to the cleaning module. Most of these parameters are provided in a form of a list. It subsequently invokes the primary functions specified for each module.

Output

Both the primary functions of the modules and the core function clean_data() return an object of the same type as the input dataset. Every cleaning operations applied to the input dataset can add an element to the report. The report generated from all cleaning tasks is attached to the output object as an attribute. It can be accessed using the attr() function in base R.

Modules in {cleanepi}

In this section, we provide a detailed description of the way that every module is built.

1. Standardization of column names

This module is designed to standardize the style and format of column names within the target dataset, offering users the flexibility to specify a subset of:

• focal columns to preserve in their original format, and

• columns to be renamed i.e. given a new name chosen by the user.

• Main function: standardize_column_names()

• Input:

  • A data.frame or linelist object.
  • Optionally, a vector of focal column names and a vector of column names to be renamed in the form of new_name = "old_name". If not provided, all columns will undergo standardization.

• Output:

  • The input object with standardized column names.

• Report:

  • A two-column table displaying the initial and current column names for each updated column in the original dataset.

• Mode:

  • Explicit

By incorporating the standardize_column_names() function, {cleanepi} streamlines the process of ensuring consistency and clarity in column naming conventions, thereby enhancing the overall organization and readability of the dataset.

2. Removal of empty rows and columns and constant columns

This module aims at eliminating irrelevant and redundant rows and columns, including empty rows and columns as well as constant columns. The main function was initially (i.e. up to version 1.0.2) built based on the {janitor} R package. We used janitor::remove_empty() and janitor::remove_conatant() to remove empty rows and columns and constant columns respectively. In janitor::remove_empty(), the empty rows are removed first, then the empty columns. This maximizes the chance of keeping more columns after this operation. As we noticed that the removal of the constant data might still result in a dataset with some empty row and/or columns and constant columns, we introduced the concept of iterative constant data removal in more recent versions of the package (> v.1.0.2). This means that the process of removing constant data is performed iteratively until there is no constant data. The report made from this operation informs about what rows and columns were removed at every iteration.

• Main function: remove_constants()
• Input: Accepts a data.frame or linelist object, along with:
  • A cut-off that determines the percent of missing values beyond which a row or column should be deleted.
• Output: Returns the input object after applying the specified operations.
• Report:
  • A data frame with four columns showcasing the removed empty and columns and constant columns at every iteration.
• Mode:
  • explicit

3. Detection and removal of duplicates

This module is designed to identify and eliminate duplicated rows.

• Main functions: find_duplicates(), remove_duplicates()
• Input: Accepts a data.frame or linelist object, along with optional parameters:
  • Vector of target columns (default is to consider all columns; possible to use linelist_tags to consider tagged variables only when the input is a linelist object).
• Output: Returns the input object after applying the specified operations.
• Report:
  • Two table(s) showcasing the found and removed duplicates.
  • A character vector with the columns used during duplicates removal.
• Mode:
  • explicit

Through the remove_duplicates() function, users can streamline their dataset by eliminating redundant rows, thus enhancing data integrity and analysis efficiency.

4. Replacement of missing values with NA

This module aims to standardize and unify the representation of missing values within the dataset.

• Main function: replace_missing_values()
• Input: Accepts a data.frame or linelist object, along with:
  • A vector of column names (if not provided, the operation is performed across all columns)
  • A string or a vector of strings representing the current missing values (default value is cleanepi::common_na_strings)
• Output: Returns the input object with all missing values represented as NA.
• Report: Generates a three-column table featuring index, column, and value for each missing item in the dataset.
• Mode:
  • explicit

By utilizing the replace_missing_char() function, users can ensure consistency in handling missing values across their dataset, facilitating accurate analysis and interpretation of the data.

5. Standardization of date values

This module is dedicated to convert date values in character columns into ISO8601 Date format, and ensuring that all dates fall within the expected user-provided timeframe.

• Main function: standardize_dates()
• Input: Accepts a data.frame or linelist object, along with:
  • A vector of targeted date columns (automatically determined if not provided)
  • Tolerance threshold that defines the % of missing (out of range values converted to NA) values to be allowed in a converted column. When % missing values exceeds or is equal to it, the original values are returned (default value is 40%)
  • format (default value is NULL)
  • timeframe (default value is null)
• Output: Returns the input object with standardized date values formatted as %Y-%m-%d.
• Report:
  • A two-column table listing the columns where date values were standardized.
  • A three-column table featuring index, column name, and values that fall outside the specified timeframe.
  • A data frame featuring date values that can comply with more than one specified format. Users can update the standardized date values with the correct ones when it’s appropriated.
• Mode:
  • explicit

By employing the standardize_dates() function, users can ensure uniformity and coherence in date formats across their dataset, while also validating the temporal integrity of the data within the defined timeframe.

6. Standardization of subject IDs

This module is tailored to verify whether the values in the column uniquely identifying subjects adhere to a consistent format. It also offers a functionality that allows users to correct the inconsistent subject ids.

• Main function: check_subject_ids()
• Input: Accepts a data.frame or linelist object, along with:
  • The name of the ID column
  • Strings for prefix, suffix,
  • A numerical range within the IDs
  • An integer representing the expected number of characters in a subject id.
• Output: Returns the input object with standardized subject IDs.
• Report: Generates a two-column table featuring index and value of each subject ID that deviates from the expected format.
• Mode:
  • explicit

The correct_subject_ids() function can be used to correct the identified incorrect subject ids. In addition to the input data, it expects a data frame with two columns from and to containing the wrong and the correct ids respectively.

By utilizing the functions in this module, users can ensure uniformity in the format of subject ids, facilitating accurate tracking and analysis of individual subjects within the dataset.

7. Dictionary based substitution

This module facilitates dictionary-based substitution, which involves replacing existing values with predefined ones. It replaces entries in a specific columns to certain values, such as substituting 1 with “male” and 2 with “female” in a gender column. It also interoperates seamlessly with the get_meta_data() function from {readepi} R package.

Note that the clean_using_dictionary() function will return a warning when it detects unexpected values in the target columns from the data dictionary. These unexpected values can be added to the data dictionary using the add_to_dictionary() function.

• Main function: clean_using_dictionary()
• Input: Accepts a data.frame or linelist object, along with a data dictionary featuring the following column names: options, values, and order.
• Output: Returns the input object where the specified options are replaced by their corresponding values.
• Report: Generates a three-column table with index, column, and value for each unexpected value encountered in a targeted column.
• Mode:
  • explicit

By leveraging the clean_using_dictionary() function, users can streamline and standardize the values within specific columns based on predefined mappings, enhancing consistency and accuracy in the dataset.

8. Conversion of values when necessary

This module is designed to convert numbers written in letters to numerical values, ensuring interoperability with the {numberize} package.

• Main function: convert_to_numeric()
• Input: Accepts a data.frame or linelist object, along with:
  • A vector of column names to be converted into numeric. When not provided, the target columns are those returned by the scan_data() function
  • A string that denotes the language used in the text. The current version supports English, French and Spanish.
• Output: Returns the input object with values in the target columns converted into numeric format.
• Report: Generates a three-column table with index, column, and value for each unrecognized value in the dataset (strings that could not be converted into numeric).
• Mode:
  • explicit

By employing the convert_to_numeric() function, users can seamlessly transform numeric representations written in letters into numerical values, ensuring compatibility with the {numberize} package and promoting accuracy in numerical analysis.

9. Verification of the sequence of date-events

This module provides functions to verify whether the sequence of date events aligns with expectations. For instance, it can flag rows where the date of admission to the hospital precedes the individual’s date of birth.

• Main function: check_date_sequence()
• Input: Accepts a data.frame or linelist object, along with:
  • A vector of date column names to be considered
• Output: Returns the input object and sends a message to inform about the presence of rows where the expected sequence of dates is not respected.
• Report: Generates a table listing the incorrect rows from the specified columns.
• Mode:
  • explicit

By using the check_date_sequence() function, users can systematically validate and ensure the coherence of date sequences within their dataset, promoting accuracy and reliability in subsequent analyses.

10. Transformation of selected columns

This module is dedicated to performing various specialized operations related to epidemiological data analytics. In the current version of the package, this module includes the following functions:

• Main function: timespan()
• Input: Accepts a data.frame or linelist object, along with:
  • The name of the column of interest
  • The reference date value or column
  • The time and remainder units (possible values are days, weeks, months, or years)
  • The names of the newly generated columns
• Output: Returns the input object with clean data and one or two additional columns with values in the specified time unit.
• Report: None.
• Mode:
  • explicit

By leveraging the timespan() function, users can efficiently compute and integrate time span information into their epidemiological dataset based on user-defined parameters, enhancing the analytics capabilities of the dataset.

Surrogate functions

1. scan_data(): This function is designed to generate a quick summary of the dataset, offering insights into the composition of each character column. It calculates the percentage of values belonging to different data types such as character, numeric, missing, logical, and date. This summary can help analysts and data scientists understand the structure and content of the dataset at a glance.

2. print_report(): This function is used for displaying the report detailing the result of the cleaning operations executed on the dataset. It likely presents information about the data cleaning processes performed, such as handling missing values, correcting data types, removing duplicates, and any other transformations applied to ensure data quality and integrity.

These surrogate functions play crucial roles in the data analysis and cleaning workflow, providing valuable information and documentation about the dataset characteristics and the steps taken to prepare it for analysis or modelling.

Related packages

{janitor}, {matchmaker}, {naniar}, {daiquiri} {epiCleanr}

Dependencies

The modules and surrogate functions will depend mainly on the following packages:

{numberize} used for the conversion of number from character to numeric, {dplyr} used in many way including filtering, column creation, data summary, etc, {magrittr} used here for its %>% operator, {linelist} used to perform some operations on linelist-type input objects, {matchmaker} utilized to perform the dictionary-based cleaning, {lubridate} used to create, handle, and manipulate objects of type Date, {reactable} mainly used here to customize the data cleaning report, {withr} utilized to handle the creation of temporary files and directory relevant for print_report() and, {readr} used to import data, {janitor} used to standardize column names in standardize_column_names().

The functions will require all other packages that needed in the package development process including:

{checkmate}, {kableExtra}, {bookdown}, {rmarkdown}, {testthat} (>= 3.0.0), {knitr}, {lintr}

Contribute

There are no special requirements to contributing to {cleanepi}, please follow the package contributing guide.