In agroimpacts/covid19interventions: Package for a COVID-19 related spatial interventions database for the U.S.

knitr::opts_chunk$set(echo = FALSE)
library(covid19interventions)
library(shiny)
library(DT)
library(ggplot2)
library(tidyverse)
library(changepoint)
library(shinythemes)
library(DT)
library(knitr)

data(mobility)
data(county_intervention_cases)

data(county_interventions)
data(df_joined)

Overview

• COVID-19 is an infectious and easily communicable virus
• The U.S. confirmed its first case of COVID-19 on Jan. 21, 2020
• On Feb. 26, 2020, the first suspected case of a community spread in the U.S. reported
• Local government bodies were first to enforce stay-at-home orders to prevent the spread of COVID-19 before statewide measures implemented

Goal

• Identify counties that enforced stay-at-home measures before their state to address the COVID-19 pandemic
• Understand if the county and state level intervention to minimize direct human contact affects the curve

Methodology

• Initially collected data for all types of county-level interventions
• Focused on the two least and most populated counties in every state
• The process was slow- most counties did not have any interventions in place
• In the interest of time, limited interventions to stay-at-home orders only
• Google's Mobility dataset utilized to expedite the collection process by observing the change in movement at the county level

Change in Mobility

• Movement information released by Google to support data-driven decision making in managing COVID-19
• Collected through Google products such as Google Maps
• Data reports percent change in movement across places in comparison to the median trend seen between Jan. 3, 2020 and Feb. 5, 2020
• Used to identify counties with the highest drop in workplace visits, potentially as a result of a county-level stay-at-home order

Changepoint Analysis

• Changepoint detection technique used to determine the change in workplace mobility curve before a statewide stay-at-home was order implemented
• Specifically, At-Most-One-Change (AMOC) method used to identify a single point of significant change
• Mean of workplace mobility calculated pre and post changepoint for every county
• Counties with the lowest mean for workplace visits post changepoint prioritized in the data collection process

Change in Mobility by County

fluidPage(
    sidebarLayout(
          sidebarPanel(
            htmlOutput("state_selector"),
            htmlOutput("county_selector"), width = "5"),
           mainPanel(plotOutput("cpplot"),
                     textOutput("pre_cp"),
                     textOutput("post_cp"),
                     width = 7)))


output$state_selector <- renderUI({
selectInput(
  inputId = "state", 
  label = "State:",
  choices = as.character(unique(df_joined$sub_region_1)),
  selected = "California")
  })

output$county_selector <- renderUI({
  available <- df_joined[df_joined$sub_region_1 == input$state, "sub_region_2"]
  selectInput(
    inputId = "county", 
    label = "County:",
    choices = unique(available),
    selected = "Los Angeles County")
})

  output$cpplot <- renderPlot({
    cp_calc <- df_joined %>% filter(sub_region_1 == input$state & 
                               sub_region_2 == input$county &
                               date < order_date &
                               !is.na(workplace_int)) %>%
    select(workplace_int) %>% 
    as.matrix() %>% 
    as.numeric() %>% 
    cpt.mean(method='AMOC')
    plot(cp_calc, cpt.col='blue', 
       ylab = "Workplace Mobility Change from Baseline (%)",
       xlab = "Days since February 15")
    output$pre_cp <- renderText({paste("Pre Changepoint Mean: ",
                                       round(param.est(cp_calc)$mean[1], 2))})
    output$post_cp <- renderText({paste("Post Changepoint Mean:",
                                        round(param.est(cp_calc)$mean[2]))})

})

Final Interventions Database

knitr::include_graphics("output/database.PNG")

Interventions

county_interventions_cases_sl <- county_intervention_cases %>% 
  select(admin1, SAH_State_Date,cases, cases_2day_before, cases_week_after, cases_2week_after, cases_4week_after) %>% 
  distinct(admin1, SAH_State_Date,cases, cases_2day_before, cases_week_after, cases_2week_after, cases_4week_after)

#as.data.frame(county_interventions_cases_sl)                      
datatable(county_interventions_cases_sl)

Before Interventions

knitr::include_graphics("output/state_before_plot.PNG")

During Interventions

knitr::include_graphics("output/state_during_plot.PNG")

1 Week After Interventions

knitr::include_graphics("output/state_1w_after.PNG")

2 Weeks After Interventions

knitr::include_graphics("output/state_2w_after.PNG")

4 Weeks After Interventions

knitr::include_graphics("output/state_4w_after.PNG")

Rationale

• Population density is one of the biggest factors in a pandemic
• Policy for COVID-19 has been decided on by governors in particular
• Expressing a Governor's political party will build overall trends in decision making for the US as a whole
• Intervention types across the US were very different from Stay at home minors, to shelter in place
• Interventions also occurred across both March of 2020 and April 2020

Population vs Intervention Date

• This chart shows the intervention date with the total population of the state. N/A for the party is for territory without a governor
• There is a clear difference in the political party's by the state for the state's intervention date
• Democrats tend to be earlier by the republicans were later in March and early April.

knitr::include_graphics("output/000036.png")

Political Party vs State Intervention Date

• Overall it shows the bipartisan policy between states for a stay at home order
• The chart shows the clear dominance geographically of the stay at home order
• Note this table leaves off 9 states that have no SAH order (of any type)

knitr::include_graphics("output/0000b4.png")

Flattening the Curve

• Did SAH interventions impact the curve?

• Methodology

• "Linearize" exponential rise of cases using natural log\ \ y = A$e^{kt}$ → ln(y) = kt + ln(A) \
• Smooth natural log scale graph
• Calculate slope k between all the datepoints
• Locate breakpoints in k curve to find dates of greates slope change
• Calculate the "lag" between the SAH order date and the last breakpoint

Alaksa State

knitr::include_graphics("output/example3.png")

San Diego County, California

knitr::include_graphics("output/example4.png")

Future Work

• Investigate impact of different intervention measures
• Relate political party to date of SAH intervention
• Consider additional variables, such as rurality
• Quantitative analysis on lag dates
• Pair lag data with mobility data (Google + Unacast Scoreboard)
• Interesting opportunities for analysis as intervention measures are uplifted

Thank you

• University of Arizona
• Lyndon
• Lei

agroimpacts/covid19interventions documentation built on May 6, 2020, 12:35 a.m.