Elsa Health Monolith

This repository is a Monolithic that contains Providers-related Elsa Health projects.

Project Structure:

• apps/ - Main Applications
  • mobile-providers Lab-focused providers application
  • mobile-ctc CTC-focused providers application
  • mobile-addo ADDO-focused providers application.
  • edge Server for client analytics and supporting distributed architeture

Get Started

This project was created using Turborepo, so we get all the fancy stuff that it offers, like mananging workspaces. Since there are javascript related projects, this project uses yarn as a package manager.

Requirements:

• node >=14.*
• yarn >=1.22.*

Discussions & Feedbacks

Check out the Discussions page, look around at any discussions you might be interested in (or start one yourself). For new comers, check out this discussion

FAQ

Who are the intended users of this application?

While the tool can be used by all healthcare providers across all cadres, it is primiarly built for those at lower level facilities like dispensaries, laboratories, drug shops, and community healthcare providers.

What are the minimum requirements for the end user of the tool?

We expect that the user has had some training in healthcare services, even a minimal amount of training will be very helpful. The application also assumes the user knows the names of diseases or conditions, so that they can better interpret the results of the tool.

How do the algorithms work?

The tool uses probabilities to quantify the likelihood of a patient having a certain disease by running simulations of patinents with each of the diseases covered, and then comparing the results of the simulations to the data being observerd.

Does the application need internet access?

No. You do not need to have internet to use the built application. This does come with some down sides in the production setting (not being able to push updates), however, in our case, we have found that the benefits far outweigh the costs.

There is a plan for hybrid in the future.

How well does the disease identification work?

The disease identification models are built by health specialists like pediatricians, oncologists, dermatologits, etc and are focused on cause and effect mechanisms.

This means, the models will have varying accuracies depending on the deployment context and target group.

For our current deployment environemnt (Tanzania), we are seeing an average F1 score of 0.82.

I found a major issue! How do I let you know?

You should file an issue as soon as you can, and we will look into it. This is a great way to support our overall vision.

License

Apache License 2.0