In epikit: Miscellaneous Helper Tools for Epidemiologists

The goal of {epikit} is to provide miscellaneous functions for applied epidemiologists. This is a product of the R4EPIs project; learn more at https://r4epis.netlify.app/.

Installation

You can install {epikit} from CRAN (see details for the latest version):

install.packages("epikit")

Click here for alternative installation options

If there is a bugfix or feature that is not yet on CRAN, you can install it via the {drat} package:

You can also install the in-development version from GitHub using the {remotes} package (but there's no guarantee that it will be stable):

# install.packages("remotes")
remotes::install_github("R4EPI/epikit")

library("epikit")

The {epikit} was primarily designed to house convenience functions for applied epidemiologists to use in tidying their reports. The functions in {epikit} come in a few categories:

Age categories

A couple of functions are dedicated to constructing age categories and partitioning them into separate chunks.

age_categories() takes in a vector of numbers and returns formatted age categories.
group_age_categories() will take a data frame with different age categories in columns (e.g. years, months, weeks) and combine them into a single column, selecting the column with the lowest priority.

library("knitr")
library("magrittr")

set.seed(1)
x <- sample(0:100, 20, replace = TRUE)
y <- ifelse(x < 2, sample(48, 20, replace = TRUE), NA)
df <- data.frame(
  age_years = age_categories(x, upper = 80), 
  age_months = age_categories(y, upper = 16, by = 6)
)
df %>% 
  group_age_categories(years = age_years, months = age_months)

Quick proportions with conficence intervals

There are three functions that will provide quick statistics for different rates based on binomial estimates of proportions from binom::binom.wilson()

attack_rate()
case_fatality_rate()
mortality_rate()

attack_rate(10, 50)
case_fatality_rate(2, 50)
mortality_rate(40, 50000)

In addition, it's possible to rapidly calculate Case fatality rate from a linelist, stratified by different groups (e.g. gender):

library("outbreaks")
case_fatality_rate_df(ebola_sim_clean$linelist, 
  outcome == "Death", 
  group = gender,
  add_total = TRUE,
  mergeCI = TRUE
)

Inline functions

The inline functions make it easier to print estimates with confidence intervals in reports with the correct number of digits.

fmt_ci() formats confidence intervals from three numbers. (e.g. fmt_ci(50, 10, 80) produces r fmt_ci(50, 10, 80)
fmt_pci() formats confidence intervals from three fractions, multiplying by 100 beforehand.

The _df suffixes (fmt_ci_df(), fmt_pci_df()) will print the confidence intervals for data stored in data frames. These are designed to work with the outputs of the rates functions. For example, fmt_ci_df(attack_rate(10, 50)) will produce r fmt_ci_df(attack_rate(10, 50)). All of these suffixes will have three options e, l, and u. These refer to estimate, lower, and upper column positions or names.

fmt_count() will count a condition in a data frame and present the number and percent of TRUE values. For example, if you wanted to count the number of women patients from Rokupa hospital, you would write: fmt_count(ebola_sim_clean$linelist, gender == "f", hospital == "Rokupa Hospital") and it would produce: r fmt_count(ebola_sim_clean$linelist, gender == "f", hospital == "Rokupa Hospital")

Confidence interval manipulation

The confidence interval manipulation functions take in a data frame and combine their confidence intervals into a single character string much like the inline functions do. There are two flavors:

merge_ci_df() and merge_pci_df() will merge just the values of the confidence interval and leave the estimate alone. Note: this WILL remove the lower and upper columns.
unite_ci() merges both the confidence interval and the estimate into a single character column. This generally has more options than merge_ci()

This is useful for reporting models:

fit <- lm(100/mpg ~ disp + hp + wt + am, data = mtcars)
df  <- data.frame(v = names(coef(fit)), e = coef(fit), confint(fit), row.names = NULL)
names(df) <- c("variable", "estimate", "lower", "upper")
print(df)

# unite CI has more options
unite_ci(df, "slope (CI)", estimate, lower, upper, m100 = FALSE, percent = FALSE)

# merge_ci just needs to know where the estimate is
merge_ci_df(df, e = 2)

Give me a break

If you need a quick function to determine the number of breaks you need for a grouping or color scale, you can use find_breaks(). This will always start from 1, so that you can include zero in your scale when you need to.

find_breaks(100) # four breaks from 1 to 100
find_breaks(100, snap = 20) # four breaks, snap to the nearest 20
find_breaks(100, snap = 20, ceiling = TRUE) # include the highest number

Pull together population counts

To quickly pull together population counts for use in surveys or demographic pyramids the gen_population() function can help. If you only know the proportions in each group the function will convert this to counts for you - whereas if you have counts, you can type those in directly. The default proportions are based on Doctors Without Borders general emergency intervention standard values.

# get population counts based on proportion, stratified
gen_population(groups = c("0-4","5-14","15-29","30-44","45+"), 
               strata = c("Male", "Female"), 
               proportions = c(0.079, 0.134, 0.139, 0.082, 0.067))

Type in counts directly to get the groups in a data frame.

# get population counts based on counts, stratified - type out counts
# for each group and strata
gen_population(groups = c("0-4","5-14","15-29","30-44","45+"), 
               strata = c("Male", "Female"), 
               counts = c(20, 10, 30, 40, 0, 0, 40, 30, 20, 20))

Table modification

These functions all modify the appearance of a table displayed in a report and work best with the knitr::kable() function.

rename_redundant() renames redundant columns with a single name. (e.g. hopitalized_percent and confirmed_percent can both be renamed to %)
augment_redundant() is similar to rename_redundant(), but it modifies the redundant column names (e.g. hospitalized_n and confirmed_n can become hospitalized (n) and confirmed (n))
merge_ci() combines estimate, lower bound, and upper bound columns into a single column.

df <- data.frame(
  `a n` = 1:6,
  `a prop` = round((1:6) / 6, 2),
  `a deff` = round(pi, 2),
  `b n` = 6:1,
  `b prop` = round((6:1) / 6, 2),
  `b deff` = round(pi * 2, 2),
  check.names = FALSE
)
knitr::kable(df)
df %>%
  rename_redundant("%" = "prop", "Design Effect" = "deff") %>%
  augment_redundant(" (n)" = " n$") %>%
  knitr::kable()