Overview

This talk presented an overview of the Epiverse-TRACE ecosystem of R packages, and outlined the need for reliable, easy-to-use tooling for infectious disease outbreak analytics, as well as Epiverse-TRACE training material, community and the role Epiverse-TRACE can play as a co-creator in public health and open-source software development and maintenance.

Slides

A recording of the talk may become available in the future and if so a link will be provided to watch the presentation and questions in full.

Slides

Transcript

In this presentation, I’ll demonstrate what Epiverse tools can do and explain why they need to be accessible to everyone.

Let’s focus on a fictional situation. Imagine an outbreak is ongoing in Senegal. The good news in this scenario is that we have a vaccine with 80% efficacy, but the problem is that we only have 1 million doses, which isn’t enough to vaccinate the entire population. A key question for public health officials or decision-makers is how to distribute these limited doses.

To explain why different distribution methods can lead to different outcomes, I need to discuss the concept of herd immunity. In a partially immunized population, even individuals who are not vaccinated can receive indirect protection from those who are vaccinated. For example, in a fictional population of individuals A, B, C, and D, if A is infected, they could infect B, who might then infect C. However, if B is vaccinated, the transmission chain is broken, and C, though unvaccinated, remains protected. This indirect protection is a key element of herd immunity.

With this understanding, we can ask how to distribute the 1 million doses to maximize the overall impact. Let’s consider three scenarios:

1. Scenario A: Give all the doses to the youngest age group
2. Scenario B: Give all the doses to the oldest age group
3. Scenario C: Distribute the doses across the entire population without prioritizing any specific age group

To analyze these scenarios, we use the finalsize tool in R. This tool requires five inputs:

• The reproduction number, which indicates how contagious the pathogen is. For example, for COVID-19, it was around 2-3.
• The contact matrix, which shows how frequently different age groups interact. For instance, teenagers tend to interact more with their peers.
• The population pyramid, indicating the size of each age group. For example, in Senegal, the pyramid might look like a triangle, with many young individuals and fewer older individuals.
• Vaccine efficacy, describing how well the vaccine protects vaccinated individuals compared to unvaccinated ones.
• The vaccine distribution strategy, representing scenarios A, B, or C.

Running this tool produces output in the form of tables and visualizations. This table shows:

• Age groups in one column
• Susceptibility groups (vaccinated or unvaccinated) in another column
• The proportion of infected individuals in each group

While tables are informative, visualizations often make it easier to understand the results. For instance, a plot might show:

• The number of infections on the y-axis
• Age groups on the x-axis, with the youngest on the left and the oldest on the right
• Different colors representing the three scenarios: red for distributing doses evenly without prioritization, green for prioritizing older individuals, and blue for prioritizing younger individuals

From this visualization, two insights emerge:

1. Summing the total number of infections across all age groups reveals that distributing doses evenly (Scenario C) results in fewer infections overall than prioritizing either the youngest or oldest groups. This means that in this fictional scenario, it’s better to vaccinate as many people as possible without prioritizing a specific age group.

2. Interestingly, vaccinating everyone provides greater indirect protection to the youngest age group than prioritizing them directly. This result may seem counterintuitive, but it highlights the significant indirect benefits of herd immunity.

The model presented here is simple, assuming uniform vaccine efficacy and single-dose administration. However, the tool can handle much more complex scenarios. For example, we could account for age-specific fatality rates, regional differences in contact patterns, or vaccines requiring multiple doses. While I used a straightforward example, these tools allow for tailored, detailed analyses for any context.

It’s important to emphasize that these insights are specific to this fictional scenario in Senegal. If we applied the same analysis in another country, like France, the results could be entirely different due to differences in population structure and contact patterns. This variability underscores the importance of tailoring analyses to local contexts.

This brings me to the idea of localism, which is central to our work at data.org. Insights derived from one region may not be applicable elsewhere due to differences in population pyramids and interaction patterns. This is why it’s critical to make these tools open-source and freely accessible, enabling every country to conduct its own analyses.

Overview

The epidemiology R package developer community, including Epiverse-TRACE have been developing a set of analytical tools that could help support decision-makers during outbreaks.

This presentation provided an overview of the tasks that such tools can be applied to in the context of the ongoing mpox outbreaks. This presentation is based on its companion blog post, “2024 mpox outbreak: common analytics tasks and available R tools”.

Slides

Questions

• Are there any packages for incorporating phylogenetic data into reconstructing transmission chains?
  • Answer: outbreaker2, TransPhylo, phybreak

Abstract

Global health projects have long emphasized the digital transformation of Global South countries to align their contexts with global infrastructures. Much of this emphasis is done by pointing at these contexts as flawed: interoperability, standards and classifications, and human capacity as well are seen as insufficient, too local, and disorganized. Consequently, international funding is directed towards enhancing capacity and infrastructure in these regions to facilitate smoother data exchange, with digital transformation promising seamless transitions across different scales. What happens if we conduct an inversion of this narrative? Could we follow global infrastructures in their flaws when meeting “the local”?

Based on a robust tradition of reflections in Science and Technology Studies, Critical Global Health, and Critical Data Studies about context (De Laet M., 2000; Tsing, A., 2005; Adams, 2016; Biehl 2016; Montgomery, C. M., Kingori, P., Sariola, S., & Engel, N, 2017; Lee, F 2021; 2023; Hepp, A; Jarke J; Kramp L 2022), we propose a symmetrical understanding of local-global relations in Global Health projects and enterprises. For this, we offer some examples in which we can see the operation of this kind of understanding, drawing from our involvement with two global projects on emergency preparedness, targeting the production of technologies for the analysis and control of infectious diseases.

Slides

Slides

Abstract

It is common for R packages answering the same need to have different input and output formats. This may result in a large amount of spent time to reformat the inputs and outputs whenever a specific part of the data pipeline is swapped out to use a different R package. This time can come at a huge cost whenever results are needed quickly, such as in pandemic response. Using S3 classes providing standard formats that all downstream packages use may be a good solution to this issue, thus improving the interoperability within the global R package ecosystem. However, this approach comes with technical and social challenges. Here, I present the work we are doing to implement and encourage the adoption of standard S3 classes in epidemiology. I highlight key findings and challenges such as how to preserve backward compatibility in existing packages and give recommendation for future similar endeavors.

Slides

Questions

• How much of the tips presented here are applicable to S4?

• In which situations would you recommend S3 over R6?

• What do you think of the upcoming S7 framework?

Transcript

Hi, my name is Hugo. I’m the lead software architect at data.org. Today, I’d like to discuss the work we’re doing to build a harmonious ecosystem for epidemiology as part of the Epiverse-TRACE Project.

Scientific research increasingly relies on data science and computational tools across fields like epidemiology, climate science, and econometrics. However, maintaining and updating data pipelines has become complex. Changing a single step or using different software can require hours of data wrangling just to format inputs and outputs correctly. This maintenance is especially challenging during crises when rapid results are essential. When the next pandemic hits, we need to focus on actual science, not basic data wrangling.

Currently, we have some good isolated free software tools, but a robust ecosystem as a whole is what we truly need. This is precisely the goal of the Epiverse-TRACE project: to harmonize the epidemiology tooling ecosystem in R. We do this by making existing tools interoperable, supporting tools to adopt global standards (like those from the Digital Public Goods Alliance or organizations like rOpenSci), and fostering a sustainable community around these ideals.

Our goal isn’t to erase established communities. We recognize the value of diverse solutions within a rich ecosystem. However, interoperability within this ecosystem is essential. To achieve this, we involve the community, working with both user groups such as public health institutes and NGOs and developer communities.

Our approach has already shown success. We’ve packaged and released many unmaintained, non-portable codebases and more tools than I can cover here. Notably, two of these tools are already registered Digital Public Goods, and one is in the process of submission.

Building a sustainable network of collaborators is exciting and ambitious, but it also comes with challenges. Research and academia are competitive spaces, which can make collaboration difficult. Additionally, communication in a multi-stakeholder network with numerous collaborators can lead to delays and miscommunication. Sustaining this effort is challenging, especially with uncertain funding in this space.

To conclude, responding effectively to the next crisis, whether it’s the climate crisis or the next pandemic, will require interoperable tools built through collaborative, multi-stakeholder projects. While necessary, these complex communities bring challenges in communication, collaboration, and sustainability. Ultimately, what may appear to be a technical challenge often becomes a communication and social challenge as well.

With that, I’ll end with a picture of our core project team. Please feel free to talk with me if you’re interested in learning more. Thank you.

readepi slides

cleanepi slides

pipeline demo

• Coming soon

Questions & answers

Will {cleanepi} package have an option to create new columns and to perform other actions such as filtering and sorting data?

Some cleaning processes in the {cleanepi} package actually create new columns. {cleanepi} does data filtering but not data sorting, which can be done with other existing packages.

When will the {readepi} and {cleanepi} packages be available on CRAN?

These packages are still in development stage, and we intend to submit them to CRAN soon.

Does {cleanepi} has an option to show which IDs are been cleaned? This will help to inform the curators about which ID as wrong?

Yes, {cleanepi} reports any change to the end-user.

Will there be a package in epiverse for spatial analysis to visualize distribution of cases during epidemics?

Yes. We intend to create a package for visualization.

Have you thought about combining the epiverse packages into a single package (like tidyverse package for example) so the user won’t have to load the packages one by one?

Yes. In fact such package is already under development, see epiverse.

Can the epiverse be used to model severity or case fatality of outbreaks against counter factual to proof effectiveness of disease control interventions?

For this, we would say that Epiverse-TRACE packages such as {cfr} provide robust methods for computing unbiased and timely case fatality ratios. With such knowledge, it is up to the user to decide whether or not control measures are effective.

Is there any plan to account for genomic data?

Yes. We intend to create a package for reading and integrating genomic data with case data.

Recording

Go to the “Watch the recording” button on the event website to watch back this showcase.

Epi-training kit

serofoi

Diapositivas (Español)

Slides (English)

Talk slides

Abstract

Occasionally, R package developers might decide to supersede their packages with a new implementation to reflect their evolved thinking and experiences in package design. Examples include {ggplot2} and {reshape2}. In multiple contributor projects, this often comes with non-trivial decisions, including name changes, deprecations, and the preservation of git histories. These decisions are non-trivial, especially in epidemiology and public health where reproducibility is a key consideration. {bpmodels}, an R package for analyzing infectious disease data, is currently in the same situation. {bpmodels} was independently developed in 2019, but the Epiverse-TRACE Initiative, which aims to develop an interoperable ecosystem for outbreak analytics, now provides the capacity to maintain it. This involves a re-imagination of the package, including a name change to {epichains}, planned integration of existing and new functionalities, and object-oriented programming. This talk will highlight some conundrums faced during this process, which is ongoing, and provide the opportunity for mutual learning.

Talk slides

Talk slides

sivirep slides

epiCo slides

serofoi slides

Recording

Go to the “Watch the recording” button on the event website to watch back this showcase.

Abstract

There is an increasing trend of packaging and sharing tools in the epidemiology research community. But these tools remain difficult to use and to integrate in a data analysis pipeline. There is a need for a more integrated approach, ensuring that the various tools work well with one another. For example, minimal data wrangling should be needed to transform the output of one tool before passing it to the next tool down the data analysis pipeline. Similarly, various alternatives for a single step of the pipeline should as much as possible use the same inputs and return the same outputs. In this talk, I will present the Epiverse-TRACE project, which collaborates with many R package developers in epidemiology to integrate their tools in a unified universe. Indeed, the unique and challenging feature of Epiverse is that it doesn’t intend to create a unified universe from scratch, but instead aims at updating existing external pieces of software to better work together. This talk will explain how we identify the key parts that should be updated, and how we make these updates with minimal disruption to the individual package developers and established community of users.

Slides

Questions from the audience

How do you incentivize developers to collaborate with you?

As mentioned in the presentation, we argue that collaboration results in a lower maintenance load for each developer by externalising and sharing the maintenance load of common elements.

We also think this is the logical continuation of releasing an open-source piece of software. If the goal is to provide a service to the community, then this service is surely greater if we collaborate to make the various pieces interoperable.

As an anecdotal piece of evidence, Nick Tierney, the maintainer of the conmat R package, has been very keen to participate in this project and was immediately convinced of the relevance and benefit for users and the ecosystem as a whole.

How do you ensure the long-term sustainability of this project?

This is an important question in the domain of open-source and research software but this is not the specific focus of this project and we don’t provide specific solutions, besides conforming to best practices.

However, if anything, our projects should be more sustainable by construction since they result from a collaborative work and include multiple maintainers. This is also encoded in our projects by creating dedicated GitHub organisations, where all developers can participate on an equal footing, thereby also increasing the lottery factor.

See also

• The companion post on our blog

finalsize slides

epiparameter slides

episoap slides

Recording

Go to the “Watch the recording” button on the event website to watch back this showcase.