In THF-evaluative-analytics/THFstyle: This package aims to make plots look like Health Foundation branded

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This documentation will guide you on how to use the THFstyle package to create charts with The Health Foundation (THF) branding. THFstyle package standardises the colours, shapes, text and sizes of charts by just adding it together with simple functions from the ggplot2 package. We provide below two examples of types of charts produced by the Improvement Analytics Unit (IAU) as part of their evaluations using THF_theme.

Loading packages required

library(THFstyle)
library(broom)
library(dplyr)
library(ggplot2)
library(tidyr)
library(boot)

Example 1: IAU standard coefplot

The below piece of code creates a forest plot for the coeffiecients of a regression model. Here we assume that we have two types of outcomes (counts and proportions) and we show the confidence intervals for the relative log-scaled risk.

Create a dataset with coefficient estimates

We will start by creating a dataframe with coefficient estimates from regressions. We do this by running regressions for binary and count data and combining the estimates that we are interested in.

model_carb <- glm(carb ~ mpg  + cyl  + hp + vs , data=mtcars, family = poisson(link = "log"))

model_carb_tidy <- broom::tidy(model_carb, conf.int = TRUE) %>%   
  filter(term=="vs") %>%
  mutate(outcome="carb") %>%
  mutate(measure="count") %>%
  as.data.frame()

model_am <- glm(am ~ vs , data=mtcars, family=binomial(link = "logit")) 

model_am_tidy <- broom::tidy(model_am, conf.int = TRUE) %>%
  filter(term=="vs") %>%
  mutate(outcome="am") %>%
  mutate(measure="proportion") %>%
  as.data.frame()

coef_data <- rbind(model_carb_tidy, model_am_tidy) %>%
  rename(ci_low=conf.low, ci_high=conf.high)

Create a standard IAU forest plot using THFtheme

We can now create a forest plot using the data we just created. We use theme_THF to make the background and axes look right and scale_colour_THF to get the right colours.

ggplot2::ggplot(data = coef_data, aes(y = outcome,
                                  x = estimate, xmin = ci_low, xmax = ci_high, color = measure)) +
     ggplot2::geom_vline(xintercept = 1, linetype = "dashed", size = 0.6, colour = "grey75") + 
     ggplot2::geom_errorbarh(height = 0.2) + 
     ggplot2::geom_point(shape=21, fill="white", size = 3) + 
     ylab('') +
     xlab("Relative risk, log-scaled") +
     theme_THF() +
  scale_colour_THF()

Example 2: IAU standard barchart

The below piece of code creates a barchart for counts of some binary variables. Here we can assume that binary variables represent comorbity flags and we show the proportions for two groups of populations (intervention and control).

Create a dataset with binary variables

nuclear_dat <- boot::nuclear %>% 
  select(intervention=pt, pr, ct, bw)

 # create proportion of subjects with comorbs in each intervention group ----
 prop_dat <- nuclear_dat %>%
    pivot_longer(-intervention, names_to = 'variable') %>% 
    group_by(intervention, variable, value) %>%
    summarise(count = n()) %>%
    mutate(prop = count/sum(count)) %>% 
    filter(value == 1) %>%
    ungroup()

  # take intervention proportions only to create an ordering variable ----
  order_dat <- prop_dat %>%
    filter(intervention == 1) %>%
    mutate(order = prop) %>%
    dplyr::select(variable, order)

  # merge ordering variable onto proportion data ----
  bar_data <- left_join(prop_dat, order_dat, by = "variable")

Create a standard IAU coefficient plot using THFtheme

ggplot2::ggplot(data = bar_data , aes(x = reorder(variable, -order), y = prop, width = 0.5, fill = factor(intervention))) + 
    scale_fill_THF(name = "", breaks=c(1,0), labels = c("intervention", "control")) +
    geom_bar(position = position_dodge(width=0.6), stat = "identity") +
    scale_y_continuous(expand = c(0,0),"", breaks = seq(0, 1 , by = 0.1), limits = c(0, 1)) +
    scale_x_discrete("") +
    theme_THF()  

THF-evaluative-analytics/THFstyle documentation built on June 24, 2020, 12:26 a.m.