epiwave

This package is the first implementation of a novel algorithm to estimate infection incidence timeseries. It relies underneath on existing sampling algorithms from the greta software. we get infection incidence, no other package does this. estimating r effective. we could benchmark but pred accuracy not super important for users.

Package to estimate epidemiological parameters using a Gaussian Process for infection timeseries

epiwave estimates a time-series of the daily number of new infections, using a time-series of daily reported case numbers and an estimated distribution of the time delay between the onset of infection and case notification. This model is unique because it integrates multiple data sources to estimate the underlying infection dynamics. Besides case, hospitalisation, and death notification data, these data sources could include wastewater data and serological data. From the estimated time-series of daily new infections we can then estimate useful epidemic parameters such as the effective reproduction number over time. This package will be especially useful when case incidence data, the most commonly used source of information for estimating effective reproduction number, is unreliable or challenging to use, but other data sources are available.

Installation

epiwave can be installed from GitHub with:

remotes::install_github('idem-lab/epiwave')

Usage

Each jurisdiction’s data is prepared the same way: define_observation_data() per stream (e.g. cases, hospitalisations), bundled into one define_observation_model() call per jurisdiction. Jurisdictions are then combined explicitly, named by jurisdiction, via stack_jurisdictions(). Jurisdictions sharing one fit are partially pooled (shared GP kernel hyperparameters, and a shared hierarchical day-of-week prior where requested).

library(epiwave)
library(epiwave.params)
library(distributional)

dates <- seq(as.Date("2024-01-01"), by = "day", length.out = 150)

# delay distributions, shared across jurisdictions
cases_delay <- as_discrete_pmf(distributional::dist_gamma(shape = 3, rate = 0.5))
hosp_delay <- as_discrete_pmf(distributional::dist_weibull(shape = 2.51, scale = 10.17))

# jurisdiction A: cases and hospitalisations, observed days 10-90
observation_model_a <- define_observation_model(
  target_infection_dates = dates,
  cases = define_observation_data(
    timeseries_data = data.frame(date = dates[10:90], value = rpois(81, lambda = 50)),
    delay_from_infection = cases_delay,
    proportion_infections = 0.5,
    dow_model = TRUE),
  hospitalisations = define_observation_data(
    timeseries_data = data.frame(date = dates[10:90], value = rpois(81, lambda = 5)),
    delay_from_infection = hosp_delay,
    proportion_infections = 0.05)
)

# jurisdiction B: same two streams, observed days 60-140 -- coverage doesn't
# need to match jurisdiction A
observation_model_b <- define_observation_model(
  target_infection_dates = dates,
  cases = define_observation_data(
    timeseries_data = data.frame(date = dates[60:140], value = rpois(81, lambda = 30)),
    delay_from_infection = cases_delay,
    proportion_infections = 0.5,
    dow_model = TRUE),
  hospitalisations = define_observation_data(
    timeseries_data = data.frame(date = dates[60:140], value = rpois(81, lambda = 3)),
    delay_from_infection = hosp_delay,
    proportion_infections = 0.05)
)

# jurisdictions are combined explicitly, named by jurisdiction
stacked <- stack_jurisdictions(
  jurisdiction_a = observation_model_a,
  jurisdiction_b = observation_model_b
)

fit_object <- fit_waves(
  observations = stacked,
  infection_model_type = "gp_growth_rate"
)

For a single jurisdiction, skip stack_jurisdictions() entirely and pass define_observation_model()’s output straight to fit_waves().

Citation

When using this package, please cite the underlying statistical software, greta as well as the package itself:

Contribution

This is a work in progress. If you see any mistakes in the package (branch main), let us know by logging a bug on the issues page.

Code of Conduct

Please note that the epiwave project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

Support

This project was supported by the Australia-Aotearoa Consortium for Epidemic Forecasting and Analytics.