Last updated on 2025-05-06 | Edit this page

Week 2: Access delays to estimate transmission and severity

This practical is based in the following tutorial episodes:

Welcome!

  • A reminder of our Code of Conduct. If you experience or witness unacceptable behaviour, or have any other concerns, please notify the course organisers or host of the event. To report an issue involving one of the organisers, please use the LSHTM’s Report and Support tool.

Read This First

Instructions:

  • Each Activity has five sections: the Goal, Questions, Inputs, Your Code, and Your Answers.
  • Solve each Activity in the corresponding .R file mentioned in the Your Code section.
  • Paste your figure and table outputs and write your answer to the questions in the section Your Answers.
  • Choose one group member to share your group’s results with the rest of the participants.

During the practical, instead of simply copying and pasting, we encourage learners to increase their fluency writing R by using:

  • The double-colon notation, e.g. package::function() to specify which package a function comes from, avoid namespace conflicts, and find functions using keywords.
  • Tab key ↹ to autocomplete package or function names and display possible arguments.
  • Execute one line of code or multiple lines connected by the pipe operator (%>%) by placing the cursor in the code of interest and pressing the Ctrl+Enter.
  • R shortcuts to insert the pipe operator (%>%) using Ctrl/Cmd+Shift+M, or insert the assignment operator (<-) using Alt/Option+-.

If your local configuration was not possible to setup:

Paste your !Error messages here

Practical

This practical has two activities.

Activity 1: Transmission

Estimate \(R_{t}\), new infections, new reports, growth rate, and doubling/halving time using the following available inputs:

  • Incidence of reported cases per day
  • Reporting delay

As a group, Write your answers to these questions:

  • What phase of the epidemic are you observing? (Exponential growth phase, near peak, or decay end phase)
  • Is the expected change in daily reports consistent with the estimated effective reproductive number, growth rate, and doubling time?
  • Interpret: How would you communicate these results to a decision-maker?
  • Compare: What differences do you identify from other group outputs? (if available)

Inputs

Group Incidence Link
1 COVID 30 days https://epiverse-trace.github.io/tutorials-middle/data/covid_30days.rds
2 Ebola 35 days https://epiverse-trace.github.io/tutorials-middle/data/ebola_35days.rds
3 Ebola 60 days https://epiverse-trace.github.io/tutorials-middle/data/ebola_60days.rds
4 COVID 60 days https://epiverse-trace.github.io/tutorials-middle/data/covid_60days.rds
Disease Reporting delays
Ebola The time difference between symptom onset and case report follows a Lognormal distribution with uncertainty. The meanlog follows a Normal distribution with mean = 1.4 days and sd = 0.5 days. The sdlog follows a Normal distribution with mean = 0.25 days and sd = 0.2 days. Bound the distribution with a maximum = 5 days.
COVID The time difference between symptom onset and case report follows a Gamma distribution with uncertainty. The mean follows a Normal distribution with mean = 2 days and sd = 0.5 days. The standard deviation follows a Normal distribution with mean = 1 day and sd = 0.5 days. Bound the distribution with a maximum = 5 days.

Solution

Code

Ebola (sample)

R 

# Load packages -----------------------------------------------------------
library(epiparameter)
library(EpiNow2)
library(tidyverse)

# Read reported cases -----------------------------------------------------
dat_ebola <- readr::read_rds(
  "https://epiverse-trace.github.io/tutorials-middle/data/ebola_35days.rds"  #<DIFFERENT PER GROUP>
) %>%
  dplyr::select(date, confirm = cases)

# Define a generation time from {epiparameter} to {EpiNow2} ---------------

# access a serial interval
ebola_serialint <- epiparameter::epiparameter_db(
  disease = "ebola",
  epi_name = "serial",
  single_epiparameter = TRUE
)

# extract parameters from {epiparameter} object
ebola_serialint_params <- epiparameter::get_parameters(ebola_serialint)

# adapt {epiparameter} to {EpiNow2} distribution inferfase
# preferred
ebola_generationtime <- EpiNow2::Gamma(
  shape = ebola_serialint_params["shape"],
  scale = ebola_serialint_params["scale"]
)
# or
ebola_generationtime <- EpiNow2::Gamma(
  mean = ebola_serialint$summary_stats$mean,
  sd = ebola_serialint$summary_stats$sd
)


# Define the delays from infection to case report for {EpiNow2} -----------

# define delay from symptom onset to case report
# or reporting delay
ebola_reportdelay <- EpiNow2::LogNormal(
  meanlog = EpiNow2::Normal(mean = 1.4, sd = 0.5),
  sdlog = EpiNow2::Normal(mean = 0.25, sd = 0.2),
  max = 5
)

# define a delay from infection to symptom onset
# or incubation period
ebola_incubationtime <- epiparameter::epiparameter_db(
  disease = "ebola",
  epi_name = "incubation",
  single_epiparameter = TRUE
)

# incubation period: extract distribution parameters
ebola_incubationtime_params <- epiparameter::get_parameters(
  ebola_incubationtime
)

# incubation period: discretize and extract maximum value (p = 99%)
# preferred
ebola_incubationtime_max <- ebola_incubationtime %>%
  epiparameter::discretise() %>%
  quantile(p = 0.99)
# or
ebola_incubationtime_max <- ebola_incubationtime %>%
  quantile(p = 0.99) %>%
  base::round()

# incubation period: adapt to {EpiNow2} distribution interface
ebola_incubationtime_epinow <- EpiNow2::Gamma(
  shape = ebola_incubationtime_params["shape"],
  scale = ebola_incubationtime_params["scale"],
  max = ebola_incubationtime_max
)

# collect required input
ebola_generationtime
ebola_reportdelay
ebola_incubationtime_epinow


# Set the number of parallel cores for {EpiNow2} --------------------------
withr::local_options(list(mc.cores = parallel::detectCores() - 1))


# Estimate transmission using EpiNow2::epinow() ---------------------------
# with EpiNow2::*_opts() functions for generation time, delays, and stan.
ebola_estimates <- EpiNow2::epinow(
  data = dat_ebola,
  generation_time = EpiNow2::generation_time_opts(ebola_generationtime),
  delays = EpiNow2::delay_opts(ebola_incubationtime_epinow + ebola_reportdelay),
  stan = EpiNow2::stan_opts(samples = 1000, chains = 3)
)


# Print plot and summary table outputs ------------------------------------
summary(ebola_estimates)
plot(ebola_estimates)
COVID (sample)

R 

# Load packages -----------------------------------------------------------
library(epiparameter)
library(EpiNow2)
library(tidyverse)

# Read reported cases -----------------------------------------------------
dat_covid <- read_rds(
  "https://epiverse-trace.github.io/tutorials-middle/data/covid_30days.rds"  #<DIFFERENT PER GROUP>
) %>%
  dplyr::select(date, confirm)

# Define a generation time from {epiparameter} to {EpiNow2} ---------------

# access a serial interval
covid_serialint <- epiparameter::epiparameter_db(
  disease = "covid",
  epi_name = "serial",
  single_epiparameter = TRUE
)

# extract parameters from {epiparameter} object
covid_serialint_params <- epiparameter::get_parameters(covid_serialint)

# adapt {epiparameter} to {EpiNow2} distribution inferfase
# preferred
covid_generationtime <- EpiNow2::LogNormal(
  meanlog = covid_serialint_params["meanlog"],
  sdlog = covid_serialint_params["sdlog"]
)
# or
covid_generationtime <- EpiNow2::LogNormal(
  mean = covid_serialint$summary_stats$mean,
  sd = covid_serialint$summary_stats$sd
)


# Define the delays from infection to case report for {EpiNow2} -----------

# define delay from symptom onset to case report
# or reporting delay
covid_reportdelay <- EpiNow2::Gamma(
  mean = EpiNow2::Normal(mean = 2, sd = 0.5),
  sd = EpiNow2::Normal(mean = 1, sd = 0.5),
  max = 5
)

# define a delay from infection to symptom onset
# or incubation period
covid_incubationtime <- epiparameter::epiparameter_db(
  disease = "covid",
  epi_name = "incubation",
  single_epiparameter = TRUE
)

# incubation period: extract distribution parameters
covid_incubationtime_params <- epiparameter::get_parameters(
  covid_incubationtime
)

# incubation period: discretize and extract maximum value (p = 99%)
# preferred
covid_incubationtime_max <- covid_incubationtime %>%
  epiparameter::discretise() %>%
  quantile(p = 0.99)
# or
ebola_incubationtime_max <- covid_incubationtime %>%
  quantile(p = 0.99) %>%
  base::round()

# incubation period: adapt to {EpiNow2} distribution interface
covid_incubationtime_epinow <- EpiNow2::LogNormal(
  meanlog = covid_incubationtime_params["meanlog"],
  sdlog = covid_incubationtime_params["sdlog"],
  max = covid_incubationtime_max
)

# collect required input
covid_generationtime
covid_reportdelay
covid_incubationtime_epinow


# Set the number of parallel cores for {EpiNow2} --------------------------
withr::local_options(list(mc.cores = parallel::detectCores() - 1))


# Estimate transmission using EpiNow2::epinow() ---------------------------
# with EpiNow2::*_opts() functions for generation time, delays, and stan.
covid_estimates <- EpiNow2::epinow(
  data = dat_covid,
  generation_time = EpiNow2::generation_time_opts(covid_generationtime),
  delays = EpiNow2::delay_opts(covid_reportdelay + covid_incubationtime_epinow),
  stan = EpiNow2::stan_opts(samples = 1000, chains = 3)
)


# Print plot and summary table outputs ------------------------------------
summary(covid_estimates)
plot(covid_estimates)

Outputs

Group 1: COVID 30 days

With reporting delay plus Incubation time: image

With reporting delay plus Incubation time:

SH 

> summary(covid30_epinow_delay)
                            measure               estimate
                             <char>                 <char>
1:           New infections per day  13193 (5129 -- 33668)
2: Expected change in daily reports      Likely increasing
3:       Effective reproduction no.      1.5 (0.92 -- 2.5)
4:                   Rate of growth 0.099 (-0.049 -- 0.26)
5:     Doubling/halving time (days)         7 (2.7 -- -14)
Group 2: Ebola 35 days

With reporting delay plus Incubation time: image

With reporting delay plus Incubation time:

SH 

> summary(ebola35_epinow_delays)
                            measure               estimate
                             <char>                 <char>
1:           New infections per day            5 (0 -- 26)
2: Expected change in daily reports      Likely decreasing
3:       Effective reproduction no.     0.66 (0.13 -- 2.2)
4:                   Rate of growth -0.039 (-0.18 -- 0.12)
5:     Doubling/halving time (days)      -18 (5.5 -- -3.9)
Group 3: Ebola 60 days

With reporting delay plus Incubation time: image

With reporting delay plus Incubation time:

SH 

> summary(ebola60_epinow_delays)
                            measure                  estimate
                             <char>                    <char>
1:           New infections per day                0 (0 -- 0)
2: Expected change in daily reports                Decreasing
3:       Effective reproduction no.    0.038 (0.0013 -- 0.39)
4:                   Rate of growth -0.16 (-0.32 -- -0.00055)
5:     Doubling/halving time (days)      -4.4 (-1300 -- -2.2)
Group 4: COVID 60 days

With reporting delay plus Incubation time: image

With reporting delay plus Incubation time:

SH 

> summary(covid60_epinow_delays)
                            measure               estimate
                             <char>                 <char>
1:           New infections per day     1987 (760 -- 4566)
2: Expected change in daily reports      Likely decreasing
3:       Effective reproduction no.     0.81 (0.43 -- 1.3)
4:                   Rate of growth -0.047 (-0.2 -- 0.092)
5:     Doubling/halving time (days)      -15 (7.5 -- -3.5)

Interpretation

Interpretation template:

  • From the summary of our analysis we see that the expected change in reports is Likely decreasing with the estimated new infections, on average, of 1987 with 90% credible interval of 760 to 4566.

  • The effective reproduction number \(R_t\) estimate (on the last date of the data), or the number of new infections caused by one infectious individual, on average, is 0.81, with a 90% credible interval of 0.43 to 1.30.

  • The exponential growth rate of case reports is, on average -0.047, with a 90% credible interval of -0.2 to 0.01.

  • The doubling time (the time taken for case reports to double) is, on average, -15.0, with a 90% credible interval of 7.5 to -3.5.

Interpretation Helpers:

  • About the effective reproduction number:
    • An Rt greater than 1 implies an increase in cases or an epidemic.
    • An Rt less than 1 implies a decrease in transmission, which could lead to extinction if sustained.
  • An analysis closest to extinction has a central estimate of:
    • Rt less than 1
    • growth rate is negative
    • doubling or halving time negative, which indicate a decline in cases
  • However, given the uncertainty in all of these estimates, there is no statistical evidence of extinction if the 90% credible intervals of:
    • Rt include the value 1,
    • growth rate include the value 0,
    • doubling or halving time include the value 0.
  • From table:
  • From figure:
    • The estimate of Reports fits the input incidence curve.
    • The forecast of New infections and Reports per day assumes no change in the reproduction number. For that reason, the forecast section of “Effective reproduction no.” is constant.
    • When we include at delays both the incubation and reporting delay,
      • In Reports, the forecast credible intervals increases.
      • New infections per day, uncertainty increases in an equivalent size to the delays
  • From comparing COVID and Ebola outputs:
    • The finite maximum value of the generation time distribution define the range of the “estimate based on parial data”.

Activity 2: Severity

Estimate the naive CFR (nCFR) and delay-adjusted CFR (aCFR) using the following inputs:

  • Reported cases (aggregate incidence by date of onset)
  • Onset to death delay

As a group, Write your answers to these questions:

  • What phase of the epidemic are you observing? (Exponential growth phase, near peak, or decay end phase)
  • Does the time series include all the possible deaths to observe from known cases?
  • How much difference is there between the nCFR and aCFR estimates?
  • Interpret: How would you communicate these results to a decision-maker?
  • Compare: What differences do you identify from other group outputs? (if available)

Inputs

Group Data Action to data input Link
1 COVID-19 Diamond Princess Keep dates before March 1st https://epiverse-trace.github.io/tutorials-middle/data/diamond_70days.rds
2 COVID-19 Diamond Princess Estimate from complete time series https://epiverse-trace.github.io/tutorials-middle/data/diamond_70days.rds
3 MERS Korea 2015 Adapt from incidence to {cfr} https://epiverse-trace.github.io/tutorials-middle/data/mers_linelist.rds

Solution

Code

COVID (sample)

R 

# Load packages -----------------------------------------------------------
library(cfr)
library(epiparameter)
library(tidyverse)


# Read reported cases -----------------------------------------------------
covid_dat <- read_rds(
  "https://epiverse-trace.github.io/tutorials-middle/data/diamond_70days.rds" 
)

covid_dat


# Create incidence object ------------------------------------------------
covid_incidence <- covid_dat %>%
  incidence2::incidence(
    date_index = "date",
    counts = c("cases", "deaths"),
    complete_dates = TRUE
  )

plot(covid_incidence)


# Confirm {cfr} data input format ----------------------------------------

# does input data already adapted to {cfr} input? Yes.
covid_adapted <- covid_dat
# does input data require to be adapted to {cfr}? No.
# covid_adapted <- covid_incidence %>%
#   cfr::prepare_data(
#     cases = "cases",
#     deaths = "deaths"
#   )

covid_adapted

# Access delay distribution -----------------------------------------------
covid_delay <- epiparameter::epiparameter_db(
  disease = "covid",
  epi_name = "onset-to-death",
  single_epiparameter = TRUE
)


# Estimate naive and adjusted CFR ----------------------------------------

# Estimate Static Delay-Adjusted CFR
covid_dat %>%
  filter(
    date < ymd(20200301)
  ) %>% 
  cfr::cfr_static()

# Estimate Static Delay-Adjusted CFR
covid_dat %>%
  filter(
    date < ymd(20200301)
  ) %>% 
  cfr::cfr_static(
    delay_density = function(x) density(covid_delay, x)
  )


# CFR estimates in whole data --------------------------------------------

# Estimate Static Delay-Adjusted CFR
covid_dat %>% 
  cfr::cfr_static()

# Estimate Static Delay-Adjusted CFR
covid_dat %>%
  cfr::cfr_static(
    delay_density = function(x) density(covid_delay, x)
  )
MERS (sample)

R 

# Load packages -----------------------------------------------------------
library(cfr)
library(epiparameter)
library(tidyverse)


# Read reported cases -----------------------------------------------------
mers_dat <- readr::read_rds(
  "https://epiverse-trace.github.io/tutorials-middle/data/mers_linelist.rds"
)

mers_dat


# Create incidence object ------------------------------------------------
mers_incidence <- mers_dat %>%
  incidence2::incidence(
    date_index = c("dt_onset", "dt_death"),
    complete_dates = TRUE
  )

plot(mers_incidence)


# Confirm {cfr} data input format ----------------------------------------

# does input data already adapted to {cfr} input? No.
# mers_adapted <- mers_dat
# does input data require to be adapted to {cfr}? Yes.
mers_adapted <- mers_incidence %>%
  cfr::prepare_data(
    cases = "dt_onset",
    deaths = "dt_death"
  )

mers_adapted

# Access delay distribution -----------------------------------------------
mers_delay <- epiparameter::epiparameter_db(
  disease = "mers",
  epi_name = "onset-to-death",
  single_epiparameter = TRUE
)


# Estimate naive and adjusted CFR ----------------------------------------

# Estimate Static CFR
mers_adapted %>%
  # filter(
  #   #<COMPLETE>
  # ) %>%
  cfr::cfr_static()

# Estimate Static Delay-Adjusted CFR
mers_adapted %>%
  # filter(
  #   #<COMPLETE>
  # ) %>%
  cfr::cfr_static(
    delay_density = function(x) density(mers_delay, x)
  )

Outputs

Covid Diamond Princess 2020 Mers Korea 2015
image image
Data Input Filter Category estimate low high
covid date < 2020-03-01 0.009 0.003 0.018
covid date < 2020-03-01 + delay density 0.026 0.010 0.053
covid no filter 0.020 0.011 0.033
covid delay density 0.020 0.011 0.033
mers no filter 0.074 0.036 0.132
mers delay density 0.138 0.072 0.229

Interpretation

Interpretation template:

  • As of 15th June 2015, the MERS cases in the population have a delay-adjusted case fatality risk of 13.8% with a 95% confidence interval between 7.2% and 22.9%.

Intepretation helpers:

  • The MERS incidence curve seems to be in a decay phase. However, it is expected to have death reports in upcoming dates, as observed in the COVID Diamond Princess data.
  • For COVID, until the end of February (on March 1st), the aCFR central estimate is closer to the CFR estimates on April 15th.
  • For MERS, the aCFR estimate is almost the double of the nCFR estimate.

Complementary notes:

  • cfr::static() assumption and limitations
    • One key assumption of cfr::static() is that reporting rate and fatality risk is consistent over the time window considered.
    • Early data from national surveillance systems had limitations (e.g., limited testing, changing case definitions, often only most severe being tested -a.k.a., preferential assessertainment-). So neither method (aCFR nor nCFR) gives the ‘true’ % of fatal symptomatic cases.
    • Compared to the Diamond Princess data, for national surveillance systems cfr::static() is therefore most useful over longer timeseries.
    • Alternativelly, cfr can also estimate the proportion of cases that are ascertained during an outbreak using cfr::estimate_ascertainment().

Continue your learning path

{EpiNow2} Case studies and use in the literature

{cfr} Estimating the proportion of cases that are ascertained during an outbreak

end