Covid-19 analysis

Github repo: https://github.com/roboton/corona

library(tidyverse)
library(rvest)
library(magrittr)
library(lubridate)

Scrape data from worldometers

cor_html <- read_html("https://www.worldometers.info/coronavirus/#countries")
tests_html <- read_html("https://www.worldometers.info/coronavirus/covid-19-testing/")

Clean and join country-level case detail and test data

cor_data <- cor_html %>% html_node("#main_table_countries") %>%
  html_table() %>%
  # removes commas in numbers
  mutate_if(is.character, function(x) { gsub("[,:]", "", x) }) %>%
  # turn everything but country column into a numeric
  mutate_at(vars(-`Country,Other`), as.numeric) %>%
  # compute a mortality rate
  replace_na(list("TotalDeaths" = 0)) %>%
  mutate(`Case Fatality rate`= TotalDeaths * 100 / TotalCases)

tests_data <- tests_html %>%
  html_nodes("table") %>%
  # Extract the first table
  extract2(1) %>%
  html_table(header = T) %>%
  # Missing variable names so fill in
  as_tibble(.name_repair = "unique") %>%
  # Get of rid of commas
  mutate_if(is.character, function(x) { gsub("[,:]", "", x) }) %>%
  # turn columns except country into numeric
  mutate_at(vars(-Country), as.numeric) %>%
  # rename South Korea to S. Korea
  mutate(Country=if_else(Country == "South Korea", "S. Korea", Country))

(join_data <- cor_data %>%
  left_join(tests_data, by=c("Country,Other" = "Country")))

Analysis

Crude mortality rate

join_data %>%
  filter(!is.na(Population)) %>%
  mutate(`Crude mortality rate`= 100 * TotalDeaths / Population) %>%
  ggplot(aes(`Crude mortality rate`,
             reorder(`Country,Other`, `Crude mortality rate`),)) +
  geom_point() + ylab("Country") +
  scale_x_continuous(labels = function(x) format(x, scientific = FALSE))

Positive test rate

join_data %>%
  filter(!is.na(Population)) %>%
  mutate(`Positive test rate`= 100 * TotalCases / `Tests Performed`) %>%
  ggplot(aes(`Positive test rate`,
             reorder(`Country,Other`, `Positive test rate`),
             color=`Tests per Million People`)) +
  geom_point() +
  scale_x_continuous(labels = function(x) format(x, scientific = FALSE)) +
  ylab("Country")

Case fatality rate by test saturation

One misleading stat we often see are mortality rates based on non-uniform testing. In some countries, testing capabilities are limited so only the very sick are being tested. If you’re only testing the very sick it will give you a mortality rate that is biased upward. Below we take a look at the mortailty rate (as defined by number of deaths / number of positive cases) by the amount of tests per million people within a country.

join_data %>%
  select(
    `Tests per Million People`, `Case Fatality rate`, `Country,Other`, Population,
    TotalCases) %>%
  filter(complete.cases(.)) %>% 
  filter(`Case Fatality rate` > 0) %>%
  mutate(low_high = if_else(`Tests per Million People` < 1000, "low", "high")) %>%
  ggplot(aes(`Tests per Million People`, `Case Fatality rate`, color=TotalCases,
             group=low_high)) +
  geom_point() +
  geom_line(stat="smooth", method = "lm", alpha=0.3) +
  #geom_hline(yintercept=1, alpha=0.5, color="green") +
  geom_text(aes(label=`Country,Other`),hjust=0, vjust=2, size=3) +
  #geom_line(stat="smooth", method="lm", alpha=0.2, color="blue") +
  ylim(0, 7)