In njtierney/conmat: Builds Contact Matrices using GAMs and Population Data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

conmat

The goal of conmat is to make it easy to generate synthetic contact matrices for a given age population.

What is a contact matrix?

Contact matrices describe the degree of contact between individuals of given age groups.

For example, this matrix describes the number of contacts between individuals

#| label: show-contact
#| echo: FALSE

name_vec <- c("0-4", "5-9", "10-14")
cmat <- matrix(
  data = rep(NA, 9),
  nrow = 3,
  ncol = 3,
  byrow = TRUE,
  dimnames = list(
    name_vec,
    name_vec
  )
)

diag(cmat) <- c(10, 11, 13)
cmat[upper.tri(cmat)] <- 3:5
cmat[lower.tri(cmat)] <- 3:5

cmat

The rows and columns represent the age groups of the people. On the main diagonal we see that we have a higher number of contacts - showing that people of similar ages tend to interact more with one another.

We can use the information in these matrices to model how diseases such as COVID-19 spread in a population through social contact.

Why do we need synthetic contact matrices?

Contact matrices are produced from empirical data resulting from a contact survey, which requires individuals to diary the amount and manner of contact a person has in a day.

However, these surveys are highly time-consuming and expensive to run, meaning that only a handful of these empirical datasets exist globally.

We can use statistical methods to create synthetic contact matrices, which are new contact matrices that have been generalised to new countries based on existing surveys.

Why do we need conmat?

Existing methods only provide outputs of the contact matrices for each country, or at best, for urban and rural areas for a given country.

We need methods that allow for flexibly creating synthetic contact matrices for a specified age population, as the age population distribution of many countries (e.g., Australia), are quite heterogeneous, and assuming it is homogeneous would result in inaccurate representation of community infection in many regions.

Installation

You can install the stable version from CRAN with:

install.packages("conmat")

You can install the development version with:

install.packages("conmat", repos = "https://idem-lab.r-universe.dev")

Or alternatively you can use remotes

# install.packages("remotes")
remotes::install_github("idem-lab/conmat")

Example

First we want to fit the model to the POLYMOD data, which contains various survey and population data.

library(conmat)
polymod_contact_data <- get_polymod_contact_data(setting = "work")
polymod_survey_data <- get_polymod_population()

The contact data is a data frame containing the age from and to, and the number of contacts for each of the specified settings, "home", "work", "school", "other", or "all" as well as the number of participants. By default, polymod_contact_data contains data from "all", but we're going to use the "work" set of data, as it produces an interesting looking dataset. Each row contains survey information of the number of contacts. Specifically, the number of contacts from one age group to another age group, and then the number of participants in that age group.

The survey data, polymod_survey_data contains the lower age limit and the population in that age group.

polymod_survey_data

Predicting the contact rate

We can create a model of the contact rate with the function fit_single_contact_model

set.seed(2022 - 09 - 06)
contact_model <- fit_single_contact_model(
  contact_data = polymod_contact_data,
  population = polymod_survey_data
)

This fits a generalised additive model (GAM), predicting the contact rate, based on a series of prediction terms that describe various features of the contact rates.

contact_model

We can use this contact model to then predict the contact rate in a new population.

As a demonstration, let's take an age population from a given LGA in Australia (this was the initial motivation for the package, so there are some helper functions for Australian specific data).

fairfield <- abs_age_lga("Fairfield (C)")
fairfield

We can then pass the contact model through to predict_contacts, along with the fairfield age population data, and some age breaks that we want to predict to.

set.seed(2022 - 09 - 06)
synthetic_contact_fairfield <- predict_contacts(
  model = contact_model,
  population = fairfield,
  age_breaks = c(seq(0, 85, by = 5), Inf)
)

synthetic_contact_fairfield

Plotting

Let's visualise the matrix to get a sense of the predictions with autoplot. First we need to transform the predictions to a matrix:

synthetic_contact_fairfield %>%
  predictions_to_matrix() %>%
  autoplot()

Applying the model across all settings.

You can also fit a model for all of the settings all at once with a series of functions, fit_setting_contacts, and predict_setting_contacts. This means we can do the above, but for each setting, "home", "work", "school", "other", and "all". We would recommend this when using conmat, as it is a pretty common use case. However for demonstration purposes we wanted to show how it works for a single matrix here first. We also provide details on how to fit the model to each of these settings in parallel. For more details on that workflow, see the "getting started" vignette.

Data sources

This package provides data and helper functions for the data, for use in calculating contact matrices. The data sources are from the Australian Bureau of Statistics (ABS), as we were using these a lot when we created the package. In the future we might wrap these data sources and helpers into another package, but for the time being they are here. Below are a couple of examples of data provided, see the "data sources" vignette and helpful at the website for full details.

You can extract the age population structure for the LGA, Brisbane, like so:

abs_age_lga("Brisbane (C)")

Note that you need to use the exact LGA name - you can look up LGA names in the data set abs_lga_lookup:

abs_lga_lookup

Or get the information for states like so:

abs_age_state(state_name = "QLD")

Note

The contact matrices created using this package are transposed when compared to the contact matrices discussed by Prem and Mossong. That is, the rows are "age group to", and the columns are "age group from".