The Vaccine Won’t be The End…

An algorithm of hope.

This past week, The Knowledge Society hosted a global artificial Intelligence hackathon. We were given one week to create a solution to a problem using AI, and then pitch our idea to some super cool people.

The process:

Brainstorming

• What do people care about?
• What’s an issue that we can tackle using the resources we have available?
• Is AI the right technology to solve this issue, and is there a simpler solution?
• What do we need to do to ensure this is executed properly?
• Is there a solution that already exists, and if so, how effective is it?

Some issues we were passionate about:

• Police Brutality
• COVID 19
• The Pink Tax
• Climate change
• Mental health
• Cancer

Research

Our team then learned more about existing solutions to these issues, and how scalable our possible solution could be. After talking to many directors and joining several calls, we recognized the importance of having a niche. We found an area that had not yet had an accurate and reliable solution, which was regarding the distribution of the COVID 19 vaccine.

Presenting… Our Hackathon Project!

A lot of people see the vaccine as the end to COVID, but what people don’t realize is that if we don’t optimize the effects of the vaccine we won’t be that far ahead.

Last month, Canada announced they had ordered 72 million COVID vaccines.

The real issue is not the development of the vaccine but its distribution.

Who should get the vaccine first and why?

To optimize the prioritization, artificial intelligence will be key. To make sure the vaccine performs to its full potential, it’s essential that the people who need it the most, get it first. And we’re here to make sure that happens. The World Health Organization has already come up with a preliminary plan however, our solution is to create an AI vaccine distribution model to give the vaccine to the right people at the right time. We would have most accurate outcome that would ensure the lowest infection rates and mortality rates, over the vaccination period.

In order to do this, we created a year-long simulation for 1000 people living in Newmarket Ontario. Keep in mind this is just a proof of concept, which is why the simulation is pretty small.

As of now, there are a few possibilities of how vaccine distribution will work…

a) This graph demonstrates what the future will look like in our simulation without a vaccine:

b) This graph demonstrates what using a first come first serve method will look like, in our simulation once the vaccine is ready to be used:

c) Our model:

Compared with the previous outcomes, though our AI model may not be the obvious choice, its important to consider the contrast in mortality rate.

With the first come first serve method, the mortality rate is 4%…

With our AI solution, the mortality rate drops by 50%!!!

So, how did we do it?

Proof of Concept

We made a simulation to demonstrate our proof of concept on a smaller dataset. The World Health Organization has begun making a preliminary plan that includes 15 different categories. In our algorithm, we only used 6, as well as the different variables mentioned below. We made up some of the data for our simulation, however, it was influenced by data we found from demographic Canada.

Variables

Based on the World Health Organization’s criteria we took into account many variables for each person, which includes:

• Pre-existing medical conditions
• The number of people living in the same household
• Occupation (mainly to prioritize essential workers)
• Pregnancy
• Age
• Population density
• Their network, ie, where they work, go to school (if they do), where they shop, etc.

Network

We explored the demographics of where our trial would take place — Newmarket, Ontario, and here’s what we researched:

• amount of people at each workplace
• the number of people in a school (elementary, and secondary)
• the number of people coming in and out of different stores
• Amount of people in each household, and how many

From this, we were able to determine how much you were exposed to others. Your exposure is part of the input which gives us your “vaccine score”.

Vaccine Score

By using the variables and network, the algorithm decides on a vaccine score.

The vaccine score can be anywhere between 0 and 1. The closer to 0 the vaccine score is, the higher risk the person is based on the variables. Those who are high risk will have higher priority. This means they will be a part of the group that goes first when the first batch arrives. The vaccine score is our main output.

Data Analysis:

This chart shows the input for each person, giving them the output of their vaccine score dependent on the variables and their weights.

This chart shows:

• Their vaccine score (output)
• If they’re in healthcare
• If they’re a frontline worker
• If they’re a teacher
• Their house hold
• If they’re female
• They’re COVID start and end date if applicable
• Where they shop
• They’re workplace

This is the data our team put into the spreadsheet to sort lowest to highest vaccine score. Once the data from this was sorted, our team began to analyze it.

We worked on finding the correlations between their score and the World Health Organizations categories. The goal with this is to find patterns in the data which could be given to the W.H.O to update and further analyze their existing procedures.

The AI in our solution:

In our neural network, we have our input, which are the variables, and the data; produced, we have an output. The outputs are the vaccine scores, and graphs; which help contrast the differences between no vaccine, a first-come, first-serve method, and using our AI to produce the best outcome.

In between the input and output, we have the hidden layers that have different weights and biases. The program starts with input and assigns random weights to get a random outcome, which is multiplied by either 0, 1, 2, or 4.

Lots of “random”…

Because we didn’t have an output, we had to create a special “loss function”

(loss function — how far off is the neural network form the goal)

In our algorithm, there are 5 layers. Each layer follows the steps to get a loss. The loss is then put back as input, to continue the cycle, to get the smallest loss function. The graphs show the loss functions of each possibility, and the AI trained itself 150 times to find the lowest loss function (which would be the lowest infection rates and deaths) to prioritize and group people.

For example: If a healthcare worker also had asthma, and on the first run, the algorithm gave them a high vaccine score (meaning they are at low risk), a graph is produced, demonstrating what would happen if this was the case: the amount of deaths, and infections, if they did not prioritize this group of people/ person.

Every day, there are 5 main steps taking place in the algorithm:

1. Checking your likelihood of contracting the virus at at work, school, and when shopping (grocery, and essential shopping)
2. Interactions at home
3. Vaccine score (most important part the of simulation)
4. If they are eligible for a vaccine, and if it’s available (depending on vaccine score)
5. Your stage of health, and your likelihood to recover

Reflection:

After completing my first hackathon I learned so much about how to execute an idea. Our team began with a problem we cared about and we brought that idea to life. It definitely came with some challenges though and technical difficulties, however despite the outcome, I found the learning from this hackathon very valuable.

Things to work on:

• Last minute our graphs comparing first come first serve method, and our AI model, did not come out as planned. There was no contrast between the graphs, and we could have helped solve this issue by changing some of the variables, or testing our algorithm more times
• After hearing lots of feedback, when we explained our idea, we weren’t clear, therefore we had a lot of misunderstanding. A simpler problem and solution statement would have been helpful.

Conclusion

The COVID vaccine is on it’s way however the pandemic is far from over. The issue is the deployment of the vaccine which our AI solution helps with. To recap:

Input:

• age, work, interactions, health conditions etc.

Output:

• A vaccine score rating the patients priority
• Graphs to compare the mortality, and infections with each possible outcome.

Though our model is only being applied to COVID, our team has realized that that AI is the future of healthcare. The algorithm we created for this challenge can also be applied for people waiting in line for organ transplants, or for any prioritization of a limited resource.

How do you think AI with revolutionize healthcare?

Our team:

Kate Denney, Sophia Laird, Spring Fu, Christina Vadivelu, Hana Ahmad

We would love some feedback on this article, and the concept! Thank you :)