conmat: Programmatic generation of synthetic contact matrices

What are contact matrices

• Estimates of the numbers of contact between individuals in various categories
• Used in models to investigate:
  • Effect of NPIs
  • Understand effects of demographic changes
  • “More accurately” estimate transmission

Contact diary studies

Contact diary studies follow individuals and ask them:

• how many contacts of a certain type,
• the duration of contact,
• the location of the contact,
• the frequency of the contact

POLYMOD

Published in 2008, Mossong et. al undertook a contact diary study of 7290 participants across Europe.

It remains the most widely cited contact diary study.

Contact matrix projection

Prem et. al formed a model that predicts the number of contacts based on the age structure of the population.

This model then allows for prediction to settings that are outside the original POLYMOD countries.

What is provided:

Our model of contact

Not much information to train model. We model the number of contacts an individual in bin i has with bin j:

\begin{aligned} c_{ij} = \beta_{0,i} + \beta_1(|i-j|) + &\beta_2(|i-j|^2) + \beta_3(i\times j) + \beta_4(i+j) +\\ &\beta_5\max(i, j) + \beta_6\min(i, j) \end{aligned}

Model terms

Projection onto target demography

Importance of contact diary study

We’ve used POLYMOD-trained models for over a decade.

There are 44 other surveys available in one collection on Zenodo, and probably more out there in the wild.

Example of using a different study

china_survey <- socialmixr::get_survey("https://doi.org/10.5281/zenodo.3878754") |>
    summarise(contacts = sum(cnt_home))

china_pop <- read_csv("./data/china_pop_age_dist.csv")
china_pop_cm <- conmat::as_conmat_population(
  data = china_pop,
  age = lower.age.limit,
  population = population
)

Example of using a different study

china_survey <- socialmixr::get_survey("https://doi.org/10.5281/zenodo.3878754") |>
    summarise(contacts = sum(cnt_home))

china_pop <- read_csv("./data/china_pop_age_dist.csv")
china_pop_cm <- conmat::as_conmat_population(
  data = china_pop,
  age = lower.age.limit,
  population = population
)

model <- conmat::fit_single_contact_model(
    contact_data = china_survey,
    population = china_pop_cm
)

predicted_contacts <- conmat::predict_contacts(
    model = model,
    population = china_pop_cm,
    age_breaks = c(seq(0, 80, by = 5), Inf)
)

Why input survey matters

These matrices are for a Chinese population, using a Chinese survey (left) and POLYMOD (right), for the home setting.

Why input survey matters

These matrices are for a Chinese population, using a Chinese survey (left) and POLYMOD (right), for the home setting.

Summary

conmat is a new, open-source, programmable system to generate synthetic contact matrices

Summary

conmat is a new, open-source, programmable system to generate synthetic contact matrices

Available right now:

remotes::install_github("idem-lab/conmat")

or pre-computed matrices on Zenodo: https://zenodo.org/records/12776714

Acknowledgements

SPECTRUM-SPARK Seed Funding