[ad_1]
A new data-driven model shows that wearing masks saves lives and the sooner you start, the better
Dr. Biplav Srivastava, a professor of computer science at the University of South Carolina, and his team have developed a data-driven tool that helps demonstrate the effect of wearing masks on COVID-19 cases and deaths. His model uses a variety of data sources to create alternative scenarios that can tell us “What could have happened?” whether a county in the United States had a higher or lower mask adherence rate. In this interview he explains how the model works, its limitations and what conclusions we can draw from it.
Computer scientist Biplav Srivastava provides a demo of the simulation to demonstrate that previous policies to recommend the use of masks make a bigger difference in the spread of the coronavirus.
What does this computer model do?
This is a nationwide tool that can show the effect masks can have. If it’s a county where people regularly wear masks, it’ll show you how many COVID-19 cases and deaths they’ve avoided. If you choose a county where people don’t wear masks, it will show you how many cases and how many deaths could have been avoided there.
How does it do it?
We need a lot of data to do this. The New York Times surveyed nearly every county in the United States over the summer and assigned a mask-wearing rating of 0-5 to each, so that’s the heart of the model. We also use data from the New York Times and Johns Hopkins for real-time case numbers; census data for demographics such as population size, average age, and more; and geographic data to measure the distance between counties.
It is based on a mathematical technique called robust synthetic control, which is often used in drug research, where there is a control group and a treatment group.
For example, let’s take a look at Wyandotte County, Kansas. It has a relatively high rating of wearing the mask of around 3.4. Why is the model designed to tell us “what if?” scenario, will examine what would have happened if the wearing mask score had been reduced to 3.0, which is our limit for “wearing the mask low”, but the user can experiment with other values as well just to see what happens. We arrived at 3.0 based on the analysis of habits of wearing masks nationwide. Actual values ranged from 1.4 to 3.85, with a national average of 2.98.
We can set a date when the wearing mask score changes to 3.0. If we set it to go from June 1 to October 1, it tells us Wyandotte County would have had 101.5% more cases and 150 more deaths during that time. It tells the user how many deaths have occurred or were prevented based on a death rate parameter that the user can set. In this example, it was set at 2%.
How does the model create the “what if?” scenario if it didn’t really happen? It does this by looking at other counties that are nearby and have similar demographics and case counts but a lower threshold for wearing the mask. Try to come up with a weighted average to form a synthetic control group similar to our county of interest (treatment group). The model then examines how divergent the two groups are in terms of case counts. The difference in case counts between the two groups is converted into a difference in deaths using the mortality rate parameter.
What does this tell us about the impact of policies on the use of masks?
It may be helpful to continue wearing the mask or to implement a mask policy at any time. But its impact is greatest when you do it early. When you run this pattern multiple times using different dates, you notice that the impact is reduced as you delay the implementation of a mask-wearing policy. So if a county had implemented a mask policy on June 1st, it would have prevented many cases. If it were to act on July 1st, it would have less impact. Had he acted in August, he would still have prevented cases, but a very small number.
What are the limits of this model?
This tool works better for some counties than others. In general, it works best with counties that are closer to the average, because they will have closer matches to compare. There is also a limitation in the sense that the New York Times mask adherence survey was conducted in the summer and things keep changing. So, if other researchers use this tool, they will have to take the changes into account.
But what you see is that when you implement a mask policy or the population regularly wear masks, it has a positive impact. And the sooner you do it, the more effective it is.
I would like to thank the work of my team, Sparsh Johri, Kartikaya Srivastava, Chinmayi Appajigowda and Lokesh Johri, in developing this program.
This article was first published in The Conversation.
About the author
Biplav Srivastava is a professor of computer science at the University of South Carolina.
.
[ad_2]
Source link
