An overview
Data cleaning is a critical step of data analysis, especially considering the messy nature of real-world data, which often includes duplicates, errors, incomplete entries, irrelevant formats, etc. Addressing these issues is essential for producing accurate, reliable, and reproducible results. However, data cleaning can pose a substantial barrier in data analysis due to the time-consuming nature of the process.
{cleanepi} is an R package designed specifically to address this challenge by offering tools to clean, curate, and standardize datasets. Tailored specifically for epidemiological data and compatible with data frame-like structures, {cleanepi} offers a suite of functions designed to streamline common data cleaning tasks.
This vignette provides a comprehensive guide to the functionalities encapsulated within {cleanepi}. It provides users with detailed insights into each function’s purpose and practical usage, equipping them with the tools necessary to navigate and manipulate cluttered datasets effectively.
General data cleaning tasks
The main function in {cleanepi} is
clean_data()
that can perform the following tasks:
- Clean up column names and convert them to more readable formats.
This includes many sub-tasks such as replacing a space, dot, or hyphen
between two words with underscore; converting camel-cases to
snake-cases; substituting foreign characters with their corresponding
English characters; and splitting a long word into multiple short words
by capital characters within, if any, and connecting them with
underscores.
- Find and remove duplicated rows across all or some specific columns.
- Remove empty rows and columns as well as constant columns,
i.e. columns with the same value across all rows.
- Replace missing entries with
NA
. - Check whether the sequence of date events are correct in all rows of the input data.
- Convert
character
columns intoDate
if the column actually contains values of typeDate
to some extent (default is at least 60% of the values areDate
). - Detect and remove rows with subject ids that do not comply with the expected format.
- Perform dictionary-based cleaning: replace keys in specific columns with their corresponding values stored in a data dictionary provided as a data frame, detect and replace misspelled values with their correct ones.
- Convert numbers written in characters into numeric.
- Calculate the time span between columns or values of type Date.
- Convert numeric values into Date.
In addition, the package also has two surrogate functions:
1. scan_data()
: scan the input data to determine the
percent of missing, numeric, character, logical and date values in every
character column of the input data frame.
2. print_report()
: print the data cleaning report.
# IMPORTING THE TEST DATASET
test_data <- readRDS(
system.file("extdata", "test_df.RDS", package = "cleanepi")
)
study_id | event_name | country_code | country_name | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex |
---|---|---|---|---|---|---|---|
PS001P2 | day 0 | 2 | Gambia | 01/12/2020 | 06/01/1972 | Dec 01, 2020 | 1 |
PS002P2 | day 0 | 2 | Gambia | 28/01/2021 | 02/20/1952 | Jan 01, 2021 | 1 |
PS004P2-1 | day 0 | 2 | Gambia | 15/02/2021 | 06/15/1961 | Feb 11, 2021 | -99 |
PS003P2 | day 0 | 2 | Gambia | 11/02/2021 | 11/11/1947 | Feb 01, 2021 | 1 |
P0005P2 | day 0 | 2 | Gambia | 17/02/2021 | 09/26/2000 | Feb 16, 2021 | 2 |
PS006P2 | day 0 | 2 | Gambia | 17/02/2021 | -99 | May 02, 2021 | 2 |
PB500P2 | day 0 | 2 | Gambia | 28/02/2021 | 11/03/1989 | Feb 19, 2021 | 1 |
PS008P2 | day 0 | 2 | Gambia | 22/02/2021 | 10/05/1976 | Sep 20, 2021 | 2 |
PS010P2 | day 0 | 2 | Gambia | 02/03/2021 | 09/23/1991 | Feb 26, 2021 | 1 |
PS011P2 | day 0 | 2 | Gambia | 05/03/2021 | 02/08/1991 | Mar 03, 2021 | 2 |
# SCAN THE DATA
scan_result <- scan_data(test_data)
Field_names | missing | numeric | date | character | logical |
---|---|---|---|---|---|
study_id | 0 | 0.0 | 0.0 | 1 | 0 |
event_name | 0 | 0.0 | 0.0 | 1 | 0 |
country_name | 0 | 0.0 | 0.0 | 1 | 0 |
date.of.admission | 0 | 0.0 | 1.0 | 0 | 0 |
dateOfBirth | 0 | 0.1 | 0.9 | 0 | 0 |
date_first_pcr_positive_test | 0 | 0.0 | 1.0 | 0 | 0 |
In {cleanepi}, every cleaning operation is encapsulated within a module, with detailed descriptions provided in the package design vignette. The parameters required for the main functions of each module are outlined in the sections below.
The clean_data()
function, described above, can take the
following arguments:
-
data: a
data.frame
orlinelist
. -
standardize_column_names: a list of parameters to
be used for standardizing the column names. These arguments are the
inputs for the
standardize_column_names()
function. -
replace_missing_values: a list of parameters to be
used when replacing the missing values by
NA
. Elements of this list are the inputs for thereplace_missing_values()
function. -
remove_duplicates: a list with the arguments that
define the columns and other parameters to be considered when looking
for duplicates. They are the input values for the
remove_duplicates()
function. -
remove_constants: a list with the parameters that
define whether to remove constant data or not. The values are the input
for the
remove_constants()
function. -
standardize_dates: a list of parameters that will
be used to standardize the date values from the input data. They
represent the input values for the
standardize_dates()
function. -
standardize_subject_ids: a list of parameters that
are needed to check whether the subject ids comply with the expected
format. These arguments are the input values of the
check_subject_ids()
. -
to_numeric: a list with the parameters needed to
convert the specified columns into numeric. When provided, the
parameters will be the input values for the
convert_to_numeric()
function. -
dictionary: a data frame that will be used to
substitute the current values in the specified columns with those in the
dictionary. It is the main argument for the
clean_using_dictionary()
function. -
check_date_sequence: a list of arguments to be used
when determining whether the sequence of the target date events is
respected across all rows of the input data. The elements of this list
are the input for the
check_date_sequence()
function.
It is important to note that only cleaning operations that are explicitly defined will be performed.
The below code chunk shows how to define a set of cleaning operations that we want to perform on the input data.
# PARAMETERS FOR REPLACING MISSING VALUES WITH NA
replace_missing_values <- list(target_columns = NULL, na_strings = "-99")
# PARAMETERS FOR COLUMN NAMES STANDARDIZATION
standardize_column_names <- list(keep = NULL, rename = NULL)
# PARAMETERS FOR DUBLICATES DETECTION AND REMOVAL
remove_duplicates <- list(target_columns = NULL)
# PARAMETERS FOR STANDARDING DATES
standardize_dates <- list(
target_columns = NULL,
error_tolerance = 0.4,
format = NULL,
timeframe = as.Date(c("1973-05-29", "2023-05-29")),
orders = list(
world_named_months = c("Ybd", "dby"),
world_digit_months = c("dmy", "Ymd"),
US_formats = c("Omdy", "YOmd")
)
)
# PARAMETERS FOR STANDARDING SUBJECT ids
standardize_subject_ids <- list(
target_columns = "study_id",
prefix = "PS",
suffix = "P2",
range = c(1, 100),
nchar = 7
)
# CONVERT THE 'sex' COLUMN INTO NUMERIC
to_numeric <- list(target_columns = "sex", lang = "en")
# PARAMETERS FOR CONSTANT COLUMNS, EMPTY ROWS AND COLUMNS REMOVAL
remove_constants <- list(cutoff = 1)
# LAOD THE DATA DICTIONARY FOR DICTIONARY-BASED CLEANING
dictionary <- readRDS(
system.file("extdata", "test_dictionary.RDS", package = "cleanepi")
)
# CLEAN THE INPUT DATA FRAME
cleaned_data <- clean_data(
data = test_data,
remove_constants = remove_constants,
replace_missing_values = replace_missing_values,
remove_duplicates = remove_duplicates,
standardize_dates = standardize_dates,
standardize_subject_ids = standardize_subject_ids,
to_numeric = to_numeric,
dictionary = dictionary
)
#> ℹ Cleaning column names
#> ℹ Replacing missing values with NA
#> ℹ Removing constant columns and empty rows
#> ℹ Removing duplicated rows
#> ℹ No duplicates were found.
#> ℹ Standardizing Date columns
#> ℹ Checking subject IDs format
#> ! Detected invalid subject ids at lines: "3, 5, 7".
#> ℹ You can use the `correct_subject_ids()` function to correct them.
#> ℹ Converting the following column into numeric: sex
#>
#> ℹ Performing dictionary-based cleaning
It returns the cleaned dataset. The report generated from the data
cleaning operations is a list object
that is attached to
the cleaned data and can be accessed using the attr()
or
attributes()
function. The report show details of outcome
of each cleaning operation that was performed during the process. Users
can access the report using the code below:
# ACCESS THE DATA CLEANING REPORT
report <- attr(cleaned_data, "report")
# SUMMARIZE THE REPORT OBJECT
summary(report)
#> Length Class Mode
#> colnames 2 data.frame list
#> missing_values_replaced_at 1 -none- character
#> constant_data 4 data.frame list
#> out_of_range_dates 3 data.frame list
#> multi_format_dates 5 data.frame list
#> incorrect_subject_id 2 data.frame list
#> converted_into_numeric 1 -none- character
They can also display the report in a web browser using the
print_report()
function as shown below:
print_report(cleaned_data)
Specific data cleaning tasks
{cleanepi} luckily provides users with the flexibility to call a specific function if they wish to perform that particular task individually. This approach allows users to have more control over the data cleaning process and to apply additional data cleaning functions as needed.
For example, some data cleaning operations, such as renaming columns,
removing empty rows and columns, removing columns with the same values
across all rows, and standardizing date columns, play a central role in
standardizing the epidemiological data. These operations can be
performed within the framework of the clean_data()
function
or alone by calling the relevant function.
This setup offers users both convenience and flexibility, as they can perform all the cleaning operations at once, or execute them individually according to their specific needs.
Remove constant data
Some datasets contain constant columns (columns with the same values
across all rows), and/or empty rows and columns (rows or columns where
all values are missing i.e NA
). The
remove_constants()
function can be used to remove such
“noise”. The function takes the following argument:
- data: the input data frame or linelist,
-
cutoff: a numeric, between
0
and1
, to be used when removing empty rows and columns. When provided, only rows and columns where the percent of missing data is greater than this cut-off will removed. Rows and columns with 100% missing values will be remove by default.
# IMPORT THE INPUT DATA
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
# INTRODUCE AN EMPTY COLUMN
data$empty_column <- NA
# remove the constant columns, empty rows and columns
dat <- remove_constants(
data = data,
cutoff = 1
)
study_id | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex |
---|---|---|---|---|
PS001P2 | 01/12/2020 | 06/01/1972 | Dec 01, 2020 | 1 |
PS002P2 | 28/01/2021 | 02/20/1952 | Jan 01, 2021 | 1 |
PS004P2-1 | 15/02/2021 | 06/15/1961 | Feb 11, 2021 | -99 |
PS003P2 | 11/02/2021 | 11/11/1947 | Feb 01, 2021 | 1 |
P0005P2 | 17/02/2021 | 09/26/2000 | Feb 16, 2021 | 2 |
PS006P2 | 17/02/2021 | -99 | May 02, 2021 | 2 |
PB500P2 | 28/02/2021 | 11/03/1989 | Feb 19, 2021 | 1 |
PS008P2 | 22/02/2021 | 10/05/1976 | Sep 20, 2021 | 2 |
PS010P2 | 02/03/2021 | 09/23/1991 | Feb 26, 2021 | 1 |
PS011P2 | 05/03/2021 | 02/08/1991 | Mar 03, 2021 | 2 |
The remove_constants()
function returns a dataset where
all constant columns, empty rows and columns are iteratively removed.
Note that when the first iteration of constant data removal results in a
dataset with new empty rows and/or columns and constant columns, this
process will be repeated several times until there is no more constant
data. Rows and columns that were deleted at any iterations will be
mentioned in the report object.
Cleaning column names
The syntax used to name the columns of a dataset during its creation
depends on many factors such as the language, the naming convention,
etc. We provide, in {cleanepi}, the
standardize_column_names()
function to clean up column
names and convert them to more intuitive formats. It performs many
sub-tasks including: replacing a space, dot, or hyphen between two words
with underscore; converting camel-cases to snake-cases; substituting
foreign characters with their corresponding English characters; and
splitting along word into multiple short words by capital characters
within, if any, and connecting them with underscores. The function can
take the following arguments:
- data: the input data frame or linelist
-
keep: a vector of column names to maintain as they
are. When dealing with a linelist, this can be set to
linelist_tags
, to maintain the tagged column names. The Default isNULL
. -
rename: a vector of column names to be renamed in
the form of
new_name = "old_name"
. If not provided, all columns will undergo standardization.
# IMPORT AND PRINT THE INITAL COLUMN NAMES
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
print(colnames(data))
#> [1] "study_id" "event_name"
#> [3] "country_code" "country_name"
#> [5] "date.of.admission" "dateOfBirth"
#> [7] "date_first_pcr_positive_test" "sex"
# KEEP 'date.of.admission' AS IS
cleaned_data <- standardize_column_names(
data = data,
keep = "date.of.admission"
)
print(colnames(cleaned_data))
#> [1] "study_id" "event_name"
#> [3] "country_code" "country_name"
#> [5] "date.of.admission" "date_of_birth"
#> [7] "date_first_pcr_positive_test" "sex"
# KEEP 'date.of.admission' AS IS, BUT RENAME 'dateOfBirth' AND 'sex' TO
# 'DOB' AND 'gender' RESPECTIVELY
cleaned_data <- standardize_column_names(
data = data,
keep = "date.of.admission",
rename = c(DOB = "dateOfBirth", gender = "sex")
)
print(colnames(cleaned_data))
#> [1] "study_id" "event_name"
#> [3] "country_code" "country_name"
#> [5] "date.of.admission" "DOB"
#> [7] "date_first_pcr_positive_test" "gender"
By providing the function with these parameters, the users can redefine the name of the columns and get easy to work with column names. This enables a more comprehensive naming system that is tailored to the needs of the user.
Replacing missing entries with NA
It is common to have missing values in an input dataset. By default,
R expects missing values to be represented by NA
, However,
this is not always the case as some dataset can contain a specific
character string that denotes the missing value. In the presence of such
a scenario, users can call the replace_missing_values()
function to substitute these missing values with NA
. This
will make the data suitable for any data science operations. The
function takes the following arguments:
- data: the input data frame or linelist.
-
target_columns: a vector of column names. If
provided, the substitution of missing values will only be executed in
those specified columns. When the input data is a
linelist
object, this parameter can be set tolinelist_tags
if you wish to replace missing values withNA
on tagged columns only. The default value isNULL
i.e. replace missing values across all columns. -
na_strings: a vector of character strings that
represents the missing values in the columns of interest. By default, it
utilizes
cleanepi::common_na_strings
. However, if the missing values string in the columns of interest is not included in this predefined vector, it can be used as the value for this argument.
# VISUALIZE THE PREDEFINED VECTOR OF MISSING CHARACTERS
print(cleanepi::common_na_strings)
#> [1] "missing" "NA" "N A" "N/A"
#> [5] "#N/A" "NA " " NA" "N /A"
#> [9] "N / A" " N / A" "N / A " "na"
#> [13] "n a" "n/a" "na " " na"
#> [17] "n /a" "n / a" " a / a" "n / a "
#> [21] "NULL" "null" "" "\\?"
#> [25] "\\*" "\\." "not available" "Not Available"
#> [29] "NOt available" "not avail" "Not Avail" "nan"
#> [33] "NAN" "not a number" "Not A Number"
# REPLACE ALL OCCURENCES OF "-99" WITH NA IN THE "sex" COLUMN
cleaned_data <- replace_missing_values(
data = readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi")),
target_columns = "sex",
na_strings = "-99"
)
# REPLACE ALL OCCURENCES OF "-99" WITH NA FROM ALL COLUMNS
cleaned_data <- replace_missing_values(
data = readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi")),
target_columns = NULL,
na_strings = "-99"
)
study_id | event_name | country_code | country_name | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex |
---|---|---|---|---|---|---|---|
PS001P2 | day 0 | 2 | Gambia | 01/12/2020 | 06/01/1972 | Dec 01, 2020 | 1 |
PS002P2 | day 0 | 2 | Gambia | 28/01/2021 | 02/20/1952 | Jan 01, 2021 | 1 |
PS004P2-1 | day 0 | 2 | Gambia | 15/02/2021 | 06/15/1961 | Feb 11, 2021 | NA |
PS003P2 | day 0 | 2 | Gambia | 11/02/2021 | 11/11/1947 | Feb 01, 2021 | 1 |
P0005P2 | day 0 | 2 | Gambia | 17/02/2021 | 09/26/2000 | Feb 16, 2021 | 2 |
PS006P2 | day 0 | 2 | Gambia | 17/02/2021 | NA | May 02, 2021 | 2 |
PB500P2 | day 0 | 2 | Gambia | 28/02/2021 | 11/03/1989 | Feb 19, 2021 | 1 |
PS008P2 | day 0 | 2 | Gambia | 22/02/2021 | 10/05/1976 | Sep 20, 2021 | 2 |
PS010P2 | day 0 | 2 | Gambia | 02/03/2021 | 09/23/1991 | Feb 26, 2021 | 1 |
PS011P2 | day 0 | 2 | Gambia | 05/03/2021 | 02/08/1991 | Mar 03, 2021 | 2 |
Standardizing Dates
The default date format in R is Ymd
(the ISO8601
format). However, it is very common to encounter date values that are
written differently from this. Also, there are cases where a column in a
data frame contains both values of type Date
,
character
or others.
The standardize_dates()
function provides a
comprehensive set of options for converting date columns into a
specified format and handling various scenarios, such as different date
formats and mixed data types in a column.
Entries which cannot be processed result in NA
. An error
threshold can be used to define the maximum number of resulting
NA
(i.e. entries without an identified date) that can be
tolerated. If this threshold is exceeded, the initial values will be
returned.
The function expects the following arguments:
-
data: A data frame or linelist (required).
- target_columns: A vector of the names of the columns to be converted (optional). When not provided, the function will attempt to detect date columns and perform the conversion if the conditions above are met.
- format: A format of the values in the specified columns (optional). If not provided, the function will attempt to infer the format.
-
timeframe: The expected time frame within which the
date values should fall. Values outside of this range will be set to
NA
(optional). -
error_tolerance: The maximum percentage of
NA
values (non date values) that can be allowed in a converted column. Default is 40% i.e.0.4
. - orders: A character vector or list of codes for fine-grained parsing of dates. It is used for parsing of mixed dates. If a list is supplied, that list will be used for successive tries in parsing. Default is:
orders <- list(
quarter_partial_dates = c("Y", "Ym", "Yq"),
world_digit_months = c("ymd", "ydm", "dmy", "mdy", "myd", "dym", "Ymd", "Ydm",
"dmY", "mdY", "mYd", "dYm"),
world_named_months = c("dby", "dyb", "bdy", "byd", "ybd", "ydb", "dbY", "dYb",
"bdY", "bYd", "Ybd", "Ydb"),
us_format = c("Omdy", "YOmd")
)
⚠️ The
error_tolerance
must be used with caution. When it is set, and the percentage of non-date values (NA i.e. values that were not converted into date) in a character column is greater than this threshold, the column will be returned as it is. The values outside of this timeframe can be accessed from the report object via itsout_of_range_dates
element.
The value for the
orders
argument can be modified to suit the user’s needs. Other date formats can be specified too. For instance, if you want to prioritize American-style dates with numeric months, you can switch the second and third elements of the default orders as shown below:
# GIVE PRIORITY TO AMERICAN-STYLE DATES
us_ord <- orders[c(1L, 3L, 2L)]
# ADD A FORMAT WITH HOURS TO THE EXISTING orders
# THIS WILL ALLOW FOR THE CONVERSION OF VALUES SUCH AS "2014_04_05_23:15:43"
# WHEN THEY APPEAR IN THE TARGET COLUMNS.
orders$ymdhms <- c("Ymdhms", "Ymdhm")
This function provides users with the flexibility to standardize date
columns in their dataset according to specified requirements, including
format
, timeframe
, and
error tolerance
for conversion from character to date
columns.
# STANDARDIZE VALUES IN THE 'date_first_pcr_positive_test' COLUMN
test_data <- readRDS(
system.file("extdata", "test_df.RDS", package = "cleanepi")
)
head(test_data$date_first_pcr_positive_test)
#> [1] "Dec 01, 2020" "Jan 01, 2021" "Feb 11, 2021" "Feb 01, 2021" "Feb 16, 2021"
#> [6] "May 02, 2021"
res <- standardize_dates(
data = test_data,
target_columns = "date_first_pcr_positive_test",
format = NULL,
timeframe = NULL,
error_tolerance = 0.4,
orders = list(
world_named_months = c("Ybd", "dby"),
world_digit_months = c("dmy", "Ymd"),
US_formats = c("Omdy", "YOmd")
)
)
This function returns the input dataset where the (specified) columns are converted into Date if the conditions are met.
study_id | event_name | country_code | country_name | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex |
---|---|---|---|---|---|---|---|
PS001P2 | day 0 | 2 | Gambia | 01/12/2020 | 06/01/1972 | 2020-12-01 | 1 |
PS002P2 | day 0 | 2 | Gambia | 28/01/2021 | 02/20/1952 | 2021-01-01 | 1 |
PS004P2-1 | day 0 | 2 | Gambia | 15/02/2021 | 06/15/1961 | 2021-02-11 | -99 |
PS003P2 | day 0 | 2 | Gambia | 11/02/2021 | 11/11/1947 | 2021-02-01 | 1 |
P0005P2 | day 0 | 2 | Gambia | 17/02/2021 | 09/26/2000 | 2021-02-16 | 2 |
PS006P2 | day 0 | 2 | Gambia | 17/02/2021 | -99 | 2021-05-02 | 2 |
PB500P2 | day 0 | 2 | Gambia | 28/02/2021 | 11/03/1989 | 2021-02-19 | 1 |
PS008P2 | day 0 | 2 | Gambia | 22/02/2021 | 10/05/1976 | 2021-09-20 | 2 |
PS010P2 | day 0 | 2 | Gambia | 02/03/2021 | 09/23/1991 | 2021-02-26 | 1 |
PS011P2 | day 0 | 2 | Gambia | 05/03/2021 | 02/08/1991 | 2021-03-03 | 2 |
It also adds two or three elements to the report object:
- A data frame with the columns where date values were standardized,
- A data frame with the values that fall outside of the specified timeframe,
- A data frame featuring date values that can comply with more than one specified format.
# STANDARDIZE VALUES IN ALL COLUMNS
res <- standardize_dates(
data = test_data,
target_columns = NULL,
format = NULL,
timeframe = NULL,
error_tolerance = 0.4,
orders = list(
world_named_months = c("Ybd", "dby"),
world_digit_months = c("dmy", "Ymd"),
US_formats = c("Omdy", "YOmd")
)
)
#> ! Found <numeric> values that could also be of type <Date> in column:
#> dateOfBirth.
#> ℹ It is possible to convert them into <Date> using: `lubridate::as_date(x,
#> origin = as.Date("1900-01-01"))`
#> • where "x" represents here the vector of values from these columns
#> (`data$target_column`).
# GET THE REPORT
report <- attr(res, "report")
Below are the possible formats for the date columns based on the
specified potential formats from the orders
argument.
field | idx | world_named_months | world_digit_months | US_formats |
---|---|---|---|---|
date.of.admission | 1 | NA | 2020-12-01 | 2020-01-12 |
date.of.admission | 4 | NA | 2021-02-11 | 2021-11-02 |
date.of.admission | 9 | NA | 2021-03-02 | 2021-02-03 |
date.of.admission | 10 | NA | 2021-03-05 | 2021-05-03 |
dateOfBirth | 1 | NA | 1972-01-06 | 1972-06-01 |
dateOfBirth | 7 | NA | 1989-03-11 | 1989-11-03 |
dateOfBirth | 8 | NA | 1976-05-10 | 1976-10-05 |
dateOfBirth | 10 | NA | 1991-08-02 | 1991-02-08 |
Standardizing subject ids
Detecting incorrect, duplicated, and missing subject ids
The check_subject_ids()
function is designed to identify
rows from the input dataset where the ids don’t comply with the expected
subject ids format. It expects the following parameters:
- data: A data frame or linelist (required).
- target_columns: The name of the column containing the subject ids in the dataset (required).
- nchar: The expected number of character in the subject ids (optional).
- prefix: A string. If subject ids have a specific prefix, it is used as a value for this argument. This is optional and can be omitted if there is no prefix.
- suffix: A string. If subject ids have a specific suffix, it is used as a value for this argument. It can be ignored otherwise.
- range: A vector of two elements. If there is an expected range of numbers within the subject ids, define it using this parameter. It is optional and can be omitted if there is no specific range.
By providing these parameters, the function becomes a versatile tool for data cleaning, ensuring that the user is alerted on the presence of unexpected, missing and duplicated subject ids. When using the function, make sure to tailor the parameters according to the specific requirements of your dataset and the expected characteristics of the subject ids.
# DETECT AND REMOVE INCORRECT SUBJECT IDs
res <- check_subject_ids(
data = readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi")),
target_columns = "study_id",
prefix = "PS",
suffix = "P2",
range = c(1L, 100L),
nchar = 7L
)
#> ! Detected invalid subject ids at lines: "3, 5, 7".
#> ℹ You can use the `correct_subject_ids()` function to correct them.
# EXTRACT REPORT
report <- attr(res, "report")
# DISPLAY THE INCORRECT SUBJECT IDs
print(report$incorrect_subject_id)
#> idx ids
#> 1 3 PS004P2-1
#> 2 5 P0005P2
#> 3 7 PB500P2
The check_subject_ids()
function returns the input
dataset and send a warming when there are some incorrect ids.
In addition to detecting undesirable subject ids, the function will
also look for missing and duplicated ids. As the result of this, the
report made from this operation might contain two extra elements:
missing_ids (a vector of row indexes where there is a
missing ids) and duplicated_ids (a data frame of rows
with the duplicated ids). Use the print_report()
function
to display the report made from this operation.
Correct wrong subject ids
After the detection of the incorrect subject ids using the
check_subject_ids()
, use the
correct_subject_ids()
to replace non complying ids with the
correct ones. The function requires a data frame with the following two
columns:
- from: a column with the incorrect subject ids,
- to: a column with the values to be used to substitute the incorrect ids.
# IMPORT THE INPUT DATA
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
# GENERATE THE CORRECTION TABLE
correction_table <- data.frame(
from = c("P0005P2", "PB500P2", "PS004P2-1"),
to = c("PB005P2", "PB050P2", "PS004P2"),
stringsAsFactors = FALSE
)
# PERFORM THE CORRECTION
dat <- correct_subject_ids(
data = data,
target_columns = "study_id",
correction_table = correction_table
)
Checking date sequence
The check_date_sequence()
function verifies the order of
sequences in date event columns within a dataset. It ensures that the
values in the specified date columns follow the desired chronological
order. Here are the arguments accepted by the function:
- data: A data frame or linelist (required).
-
target_columns: A vector containing the names of
the date columns of interest. These columns should be listed in the
expected order of occurrence that reflects the chronological sequence of
events. For example,
target_columns = c("date_of_infection", "date_of_admission", "date_of_death")
.
By utilizing these arguments, the check_date_sequence()
function facilitates the validation of date sequences within a dataset,
ensuring data integrity and accuracy for further analysis. Additionally,
it offers flexibility by allowing users to choose whether to remove rows
with incorrect sequences or store them for further examination in the
report object.
# IMPORT THE DATA AND STANDARDIZE THE TARGET DATE COLUMNS
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
data <- standardize_dates(
data,
target_columns = c("date_first_pcr_positive_test", "date.of.admission")
)
# DETECT ROWS WITH INCORRECT DATE SEQUENCE
res <- check_date_sequence(
data = data,
target_columns = c("date_first_pcr_positive_test", "date.of.admission")
)
#> ! Detected 2 incorrect date sequences at lines: "6, 8".
#> ℹ Enter `attr(dat, "report")[["incorrect_date_sequence"]]` to access them,
#> where "dat" is the object used to store the output from this operation.
# DISPLAY THE INCORRECT SEQUENCES OF DATE
incorrect_sequence <- attr(res, "report")[["incorrect_date_sequence"]]
print(incorrect_sequence)
#> row_id date_first_pcr_positive_test date.of.admission
#> 6 6 2021-05-02 2021-02-17
#> 8 8 2021-09-20 2021-02-22
The check_date_sequence()
function returns the input
dataset, augmented with an attributes named as
incorrect_date_sequence
if there are rows with incorrect
date sequences. This attribute highlights any discrepancies found in the
date sequences, enabling users to take appropriate actions. Use the
print_report()
function to display the report made from
this operation.
Converting character columns into numeric
In certain scenarios, the input data contains columns where the
numbers are written in letters. For instance, the ages of a study
participants can be written in letters. Similarly, a column can contain
values written in both numbers and letters (an age column where some
values are written in numbers and others in letters). The
convert_to_numeric()
function offers the framework to
convert all numbers written in letters from a given column into numeric.
It makes call of the main function from the {numberize}
package and takes the following arguments:
- data: A data frame or linelist (required).
-
target_columns: A vector containing the names of
the columns of interest. When dealing with a linelist, this can be set
to
linelist_tags
if the tagged columns are the one to be converted into numeric. Whentarget_columns = NULL
, the function uses the output form thescan_data()
function to identify the columns where the proportion of numeric values is at least twice as the percentage of character values. Those columns will be the columns of interest and the character values in them will be converted into numeric.
Note that any string in such column that can not be converted into numeric will be set toNA
in the resulting data. -
lang: A character string with the language in which
the letters are written. Currently one of
"en", "fr", or "es"
for English, French or Spanish respectively.
# CONVERT THE 'age' COLUMN IN THE TEST LINELIST DATA
dat <- readRDS(system.file("extdata", "messy_data.RDS", package = "cleanepi"))
head(dat$age, 10L)
#> [1] "37" "seventy-four" "17" "3" "37"
#> [6] NA "6" "26" "eleven" "44"
dat <- convert_to_numeric(
data = dat,
target_columns = "age",
lang = "en"
)
head(dat$age, 10L)
#> [1] 37 74 17 3 37 NA 6 26 11 44
Converting numeric values into date
Some columns in a data frame might contain numeric values that
represents the number of days elapsed between two events. For instance,
the recruitment day of individuals in a study can be stored as a numeric
column where the numeric values are the count of days between when they
are recruited and when there were admitted in the hospital. The actual
dates when the individuals were recruited can be retrieved using the
convert_numeric_to_date()
function. This function can take
the following parameters:
- data: the input data frame or linelist.
-
target_columns: a vector or a comma-separated list
of columns names to be converted from numeric to date. When the input
data is a
linelist
object, this parameter can be set tolinelist_tags
if tagged variables are the target columns. - ref_date: a reference date
-
forward: a Boolean that indicates whether the
counts started after the reference date (
TRUE
) or not (FALSE
). The default isTRUE
.
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi")) %>%
standardize_dates(target_columns = "date.of.admission")
# CREATE THE RECRUITMENT DATE COLUMNS
data$recruitment_date <- sample(20:50, nrow(data), replace = FALSE)
# RETRIVE THE DATE INDIVIDUALS WERE RECRUITED FROM THEIR DATE OF ADMISSION
dat <- convert_numeric_to_date(
data = data,
target_columns = "recruitment_date",
ref_date = "date.of.admission",
forward = TRUE
)
# RETRIVE THE DATE INDIVIDUALS WERE RECRUITED FROM 2019-10-13
dat <- convert_numeric_to_date(
data = data,
target_columns = "recruitment_date",
ref_date = as.Date("2019-10-13"),
forward = FALSE
)
The function returns the input data where the values in the target columns(s) are converted into Date. This enables the usage of {cleanepi}’s functions that operates on Date columns as well as the powerful functions that can be used to manipulate Dates from the {base} R or {lubridate} packages.
Finding duplicated rows
The find_duplicates()
function serves the purpose of
identifying duplicated rows within a given dataset. It accepts the
following parameters:
- data: The input data frame or linelist.
-
target_columns: A vector containing either column
names or column indices from which duplicated rows will be identified.
If
NULL
is passed, duplicates will be detected across all columns of the dataset. Notably, if the input dataset is alinelist
object,target_columns
can be set tolinelist_tags
specifically to identify duplicates across the tagged variables only.
By leveraging the find_duplicates()
function with
appropriate parameters, users can efficiently pinpoint duplicated rows
within their datasets, either across all columns or selectively across
tagged variables in a linelist
object.
# IMPORT A `linelist` DATA
data <- readRDS(
system.file("extdata", "test_linelist.RDS", package = "cleanepi")
)
# SHOW THE TAGGED VARIABLES
linelist::tags(data)
#> $date_onset
#> [1] "dt_onset"
#>
#> $date_reporting
#> [1] "dt_report"
#>
#> $gender
#> [1] "sex"
#>
#> $outcome
#> [1] "outcome"
# FIND DUPLICATES ACROSS ALL COLUMNS EXCEPT THE SUBJECT ids COLUMN
all_columns <- names(data)
target_columns <- all_columns[all_columns != "id"]
dups <- find_duplicates(
data = data,
target_columns = target_columns
)
#> ℹ No duplicates were found.
# FIND DUPLICATES ACROSS TAGGED VARIABLES
dups <- find_duplicates(
data = data,
target_columns = "linelist_tags"
)
#> ! Found 57 duplicated rows in the dataset.
#> ℹ Use `attr(dat, "report")[["duplicated_rows"]]` to access them, where "dat" is
#> the object used to store the output from this operation.
Upon execution, the find_duplicates()
function
identifies all duplicated rows either based on all columns or those
specified, and stores them in the report, where in addition to the
existing columns, it appends two extra columns to the dataset:
-
row_id
: Contains indices of the duplicated rows from the original input dataset. -
group_id
: Contains unique identifiers assigned to each duplicated group, which is defined as a set of rows sharing identical values in the designated columns of interest.
By including these extra columns, users gain insights into the specific rows identified as duplicates and their corresponding group identifiers, enabling efficient analysis and management of duplicated data within the dataset.
# DISPLAY THE DUPLICATES
report <- attr(dups, "report")
duplicates <- report$duplicated_rows
Removing duplicates
To eliminate duplicated rows from a dataset, the
remove_duplicates()
function can be employed. This function
internally utilizes the find_duplicates()
function and
expects the following parameters:
- data: A data frame or linelist from which duplicated rows will be removed.
-
target_columns: A vector containing either column
names or indices specifying the columns from which duplicated rows will
be identified. If set to
NULL
, the function will remove duplicates across all columns. If the input dataset is alinelist
object, setting this parameter tolinelist_tags
will identify and remove duplicates across the tagged variables only.
# REMOVE DUPLICATE ACROSS TAGGED COLUMNS ONLY.
res <- remove_duplicates(
data = readRDS(
system.file("extdata", "test_linelist.RDS", package = "cleanepi")
),
target_columns = "linelist_tags"
)
#> ! Found 57 duplicated rows in the dataset.
#> ℹ Use `attr(dat, "report")[["duplicated_rows"]]` to access them, where "dat" is
#> the object used to store the output from this operation.
Upon execution, the remove_duplicates()
function returns
the input dataset without duplicated rows removed (if found). The
details about the duplicates removal operation are stored in the report
object that is attached to the output object. When duplicates are found,
this report will contain the following elements:
- duplicated_rows: A data frame with the detected duplicates.
- removed_duplicates: A data frame with the duplicated rows that have been removed.
- duplicates_checked_from: A vector of column names from which duplicates were identified.
By examining these elements within the report, users gain insights into the specific duplicated rows, those that were removed, and the columns used to identify the duplicates, thus facilitating transparency and documentation of the duplicates removal process.
# ACCESS THE REPORT
report <- attr(res, "report")
# SUMMARIZE THE REPORT OBJECT
summary(report)
#> Length Class Mode
#> duplicated_rows 6 grouped_df list
#> duplicates_checked_from 1 -none- character
#> removed_duplicates 5 grouped_df list
Use the print_report()
function to display the report
made from this operation.
Dictionary based data substituting
The clean_using_dictionary()
function offers a
convenient way to replace the options in a data frame or linelist with
their corresponding values stored in a data dictionary. The function
expects the following arguments:
- data: The input data frame or linelist that contains the options to be replaced.
-
dictionary: The data dictionary in a form of a data
frame that contains the complete labels for these options. The structure
of this data dictionary file should adhere to the standards expected by
the {matchmaker}
package, as the
clean_using_dictionary()
function relies on functions from this package.
options | values | grp | orders |
---|---|---|---|
1 | male | sex | 1 |
2 | female | sex | 2 |
The add_to_dictionary()
function is a useful tool for
expanding the coverage of a data dictionary by defining options that are
present in the input data but not originally included in the dictionary.
This function enables users to dynamically update the dictionary to
accommodate new values encountered in the dataset. In addition to the
current data dictionary the function takes the arguments defined
below:
option, value, grp, order: the values for the options to be added in the data dictionary. The example below shows how this function is used.
By employing the add_to_dictionary()
function, users can
ensure that the data dictionary remains comprehensive and aligned with
the evolving nature of the input dataset, thereby enhancing the accuracy
and completeness of data interpretation and analysis. In the example
below, we add -99
to our test data dictionary,
test_dictionary
.
# READING IN THE DATA
data <- readRDS(
system.file("extdata", "test_df.RDS", package = "cleanepi")
)
# ADD THE EXTRA OPTION TO THE DICTIONARY
dictionary <- add_to_dictionary(
dictionary = dictionary,
option = "-99",
value = "unknow",
grp = "sex",
order = NULL
)
options | values | grp | orders |
---|---|---|---|
1 | male | sex | 1 |
2 | female | sex | 2 |
-99 | unknow | sex | 3 |
# PERFORM THE DICTIONARY-BASED SUBSTITUTION
cleaned_df <- clean_using_dictionary(
data = data,
dictionary = dictionary
)
study_id | event_name | country_code | country_name | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex |
---|---|---|---|---|---|---|---|
PS001P2 | day 0 | 2 | Gambia | 01/12/2020 | 06/01/1972 | Dec 01, 2020 | male |
PS002P2 | day 0 | 2 | Gambia | 28/01/2021 | 02/20/1952 | Jan 01, 2021 | male |
PS004P2-1 | day 0 | 2 | Gambia | 15/02/2021 | 06/15/1961 | Feb 11, 2021 | unknow |
PS003P2 | day 0 | 2 | Gambia | 11/02/2021 | 11/11/1947 | Feb 01, 2021 | male |
P0005P2 | day 0 | 2 | Gambia | 17/02/2021 | 09/26/2000 | Feb 16, 2021 | female |
PS006P2 | day 0 | 2 | Gambia | 17/02/2021 | -99 | May 02, 2021 | female |
PB500P2 | day 0 | 2 | Gambia | 28/02/2021 | 11/03/1989 | Feb 19, 2021 | male |
PS008P2 | day 0 | 2 | Gambia | 22/02/2021 | 10/05/1976 | Sep 20, 2021 | female |
PS010P2 | day 0 | 2 | Gambia | 02/03/2021 | 09/23/1991 | Feb 26, 2021 | male |
PS011P2 | day 0 | 2 | Gambia | 05/03/2021 | 02/08/1991 | Mar 03, 2021 | female |
Calculating time span in different time scales (“years”, “months”, “weeks”, or “days”)
The timespan()
function computes the time span between
two elements of type Date. The resulting time span can be expressed in
“years”, “months”, “weeks”, or “days”, depending on the user-specified
unit. The functions can take the following arguments:
- data: The input dataset (required).
-
target_column: A string with the name of the target
column (require). The values in this column are expected to be in the
form of
Ymd
i.e.2024-01-31
. The time span will be calculated between these values and theend_date
defined below. -
end_date: It can be either a character that is the
name of another column of type Date from the input data or a vector of
Date values or a single Date value (required). This should also be in
the ISO8601 format (“2024-01-31”) and its default value is today’s date
Sys.Date()
. - span_unit: This parameter determines the unit in which the time span is expressed (required). It can be calculated in “years”, “months”, “weeks”, or “days”. By default, the time span is calculated in “years” if this parameter is not provided.
-
span_column_name: A string for the name of the
column added to the input data. The default is
span
. -
span_remainder_unit: A parameter used to determine
the unit in which the remainder of the time span calculation will be
returned. The possible units are: “days” or “weeks” or “months”. By
default, the function returns decimal values
i.e.
span_remainder_unit = NULL
.
With these arguments, the function offers flexibility in determining the time span in different units. It facilitates various analytics tasks where the time span computation is a necessary component, providing users with the ability to customize the output according to their specific requirements.
# IMPORT DATA, REPLACE MISSING VALUES WITH 'NA' & STANDARDIZE DATES
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi")) %>%
replace_missing_values(
target_columns = "dateOfBirth",
na_strings = "-99"
) %>%
standardize_dates(
target_columns = "dateOfBirth",
error_tolerance = 0.0,
format = "%m/%d/%Y" # nolint: nonportable_path_linter.
)
#> ℹ The target column will be standardized using the format: "%m/%d/%Y".
# CALCULATE INDIVIDUAL AGE IN YEARS FROM THE 'dateOfBirth' COLUMN AND SEND THE
# REMAINDER IN MONTHS
age <- timespan(
data = data,
target_column = "dateOfBirth",
end_date = Sys.Date(),
span_unit = "years",
span_column_name = "age_in_years",
span_remainder_unit = "months"
)
# CALCULATE THE TIME SPAN IN DAYS BETWEEN INDIVIDUALS DATE OF ADMISSION AND THE
# DAY THEY TESTED POSITIVE
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
dat <- data %>%
replace_missing_values(
target_columns = "dateOfBirth",
na_strings = "-99"
) %>%
standardize_dates(
target_columns = c("date_first_pcr_positive_test", "date.of.admission"),
error_tolerance = 0.0,
format = NULL
) %>%
timespan(
target_column = "date_first_pcr_positive_test",
end_date = "date.of.admission",
span_unit = "days",
span_column_name = "elapsed_days",
span_remainder_unit = NULL
)
The timespan()
function augments the input dataset by
adding one or two extra columns containing age-related information.
These additional columns are as follows:
Calculated time span in the specified scale: Contains the calculated time span in the specified unit (“years”, “months”, “weeks”, or “days”).
Remainder from the time span calculation: Indicates the remaining number of “days” or “weeks” or “months” after calculating the time span, representing the fractional part of the time span calculation. This column is included if needed, and provides additional granularity in the time span representation.
study_id | event_name | country_code | country_name | date.of.admission | dateOfBirth | date_first_pcr_positive_test | sex | elapsed_days |
---|---|---|---|---|---|---|---|---|
PS001P2 | day 0 | 2 | Gambia | 2020-12-01 | 06/01/1972 | 2020-12-01 | 1 | 0 |
PS002P2 | day 0 | 2 | Gambia | 2021-01-28 | 02/20/1952 | 2021-01-01 | 1 | 27 |
PS004P2-1 | day 0 | 2 | Gambia | 2021-02-15 | 06/15/1961 | 2021-02-11 | -99 | 4 |
PS003P2 | day 0 | 2 | Gambia | 2021-02-11 | 11/11/1947 | 2021-02-01 | 1 | 10 |
P0005P2 | day 0 | 2 | Gambia | 2021-02-17 | 09/26/2000 | 2021-02-16 | 2 | 1 |
PS006P2 | day 0 | 2 | Gambia | 2021-02-17 | NA | 2021-05-02 | 2 | -74 |
PB500P2 | day 0 | 2 | Gambia | 2021-02-28 | 11/03/1989 | 2021-02-19 | 1 | 9 |
PS008P2 | day 0 | 2 | Gambia | 2021-02-22 | 10/05/1976 | 2021-09-20 | 2 | -210 |
PS010P2 | day 0 | 2 | Gambia | 2021-03-02 | 09/23/1991 | 2021-02-26 | 1 | 4 |
PS011P2 | day 0 | 2 | Gambia | 2021-03-05 | 02/08/1991 | 2021-03-03 | 2 | 2 |
Using {cleanepi} functionalities with pipe operators
# IMPORT THE INPUT DATASET
data <- readRDS(system.file("extdata", "test_df.RDS", package = "cleanepi"))
# IMPORT THE DATA DICTIONARY
test_dictionary <- readRDS(
system.file("extdata", "test_dictionary.RDS", package = "cleanepi")
)
# SCAN THROUGH THE DATA
scan_res <- scan_data(data)
# PERFORM DATA CLEANING
cleaned_data <- data %>%
standardize_column_names(keep = NULL, rename = c(DOB = "dateOfBirth")) %>%
replace_missing_values(target_columns = NULL, na_strings = "-99") %>%
remove_constants(cutoff = 1.0) %>%
remove_duplicates(target_columns = NULL) %>%
standardize_dates(
target_columns = NULL,
error_tolerance = 0.4,
format = NULL,
timeframe = as.Date(c("1973-05-29", "2023-05-29"))
) %>%
check_subject_ids(
target_columns = "study_id",
prefix = "PS",
suffix = "P2",
range = c(1L, 100L),
nchar = 7L
) %>%
convert_to_numeric(target_columns = "sex", lang = "en") %>%
clean_using_dictionary(dictionary = test_dictionary)
#> ℹ No duplicates were found.
#> ! Detected invalid subject ids at lines: "3, 5, 7".
#> ℹ You can use the `correct_subject_ids()` function to correct them.
# ADD THE DATA SCANNING RESULT TO THE REPORT
cleaned_data <- add_to_report(
x = cleaned_data,
key = "scanning_result",
value = scan_res
)
Printing the report
print_report(
data = cleaned_data,
report_title = "{cleanepi} data cleaning report",
output_directory = ".",
output_file_name = NULL,
format = "html",
print = TRUE
)