How do COVID-19 admissions evolve over time?

Anthony B. Masters

Feb 9·3 min read

‘Protect the NHS’ is the appeal from the UK government, ordering people to stay at home. Pressure on healthcare services is important to understand during a pandemic.

How do COVID-19 admissions and patient levels change over time?

Evolving over time

We can plot COVID-19 admissions and COVID-19 patients in hospitals in England. NHS England use the following definitions:

• COVID-19 admission: admissions with positive tests plus in-patient diagnoses. For hospital admissions, the person must have a positive SARS-CoV-2 test in the prior 14 days. NHS England counts in-patient diagnoses against the day before their diagnosis.
• COVID-19 patients (or hospital cases): the number of confirmed COVID-19 patients in hospital at 8am. This measure includes all hospitals.

Prof Johnson (Bristol) produces graphs showing admissions and patients over time.

Using R, we are going to produce a similar graph using ggplot2.

Straight from the API

Public Health England provide an interface for their data-set. That means we can download an R package, rather than upload data via spreadsheets.

library(tidyverse)
library(lubridate)
library(scales)
library(ukcovid19)

Depending on your version, you may need to use remotes to install the PHE package:

library(remotes)
remotes::install_github("publichealthengland/coronavirus-dashboard-api-R-sdk")

That allows to draw numbers straight from the interface.

First, we have to set the ‘structure’ of our data request:

covid19_structure <- list(
  date = "date",
  areaType = "areaType",
  areaName = "areaName",
  areaCode = "areaCode",
  newAdmissions = "newAdmissions",
  hospitalCases = "hospitalCases")

The two measures of interest are newAdmissions and hospitalCases .

Next, we draw data for the UK total, four nations, and English regions. For brevity, this is what the draw for the four nations looks like:

covid19_hospital_nations_df <- ukcovid19::get_data(
  filters = c("areaType=nation"),
  structure = covid19_structure)

We then bind those three draws together, transforming the date column and dropping missing rows.

covid19_hospital_df <- bind_rows(
  covid19_hospital_uk_df,
  covid19_hospital_nations_df,
  covid19_hospital_regions_df) %>%
  mutate(date = as_date(date)) %>%
  drop_na()

Making the graphs

We can create a static graph for England.

We set the caption, and create the breaks for the colourful scale:

covid19_hospital_caption <- "Source: Public Health England COVID-19 Dashboard API."
covid19_scale_breaks <- c("2020-04-01", "2020-07-01", "2020-10-01", "2021-01-01")
covid19_scale_breaks <- covid19_scale_breaks %>% as_date()

This ‘phase portrait’ graph shows COVID-19 admissions and patients. The graph shows the third dimension — time — through colour.

This is the core of the graph:

covid19_hospital_eng_gg <- covid19_hospital_df %>%
  filter(areaName == "England") %>%
  ggplot(aes(x = newAdmissions,
             y = hospitalCases)) +
  geom_point(aes(color = date),
             size = 3,
             alpha = 0.2) +

Graphs need labels, which what the labs function is for:

labs(title = "There were over 4,000 COVID-19 admissions in England on 12 January 2021.",
      subtitle = "Plot of COVID-19 admissions versus COVID-19 patients in English hospitals over time.",
      x = "COVID-19 admissions (new admissions plus in-patient diagnoses)",
      y = "COVID-19 patients in hospital",
      caption = covid19_hospital_caption) +

Scales then allow us to set limits. Since we are using a colour scale, we want the legend to be long. We can amend that through the guides function:

scale_x_continuous(expand = c(0,0),
                     limits = c(0, 4500)) +
  scale_y_continuous(expand = c(0,0),
                     limits = c(0, 35000)) +
  scale_color_gradientn(name = "Date",
                        colors = rainbow(6),
                        breaks = covid19_scale_breaks,
                        labels = date_format("%b-%y")) +
  guides(color = guide_colorbar(barwidth = unit(10, "cm")))

That code produces this graph:

We could also create an animated graph:

Since we are drawing data straight from the PHE interface, these graphs are easy to replicate.

The full R code for the graphs is available on R Pubs and GitHub.