Today, the Bell Ringer will start a public affairs column called TheCOVID Report with David Blake, a professor in the Department of Neuroscience and Regenerative Medicine at Augusta University. Blake was educated in biomedical engineering at Johns Hopkins Medical School for his Ph.D., and his studies were quantitative neurophysiology. His academic work was primarily electrical engineering coursework in signal processing and graduate-level statistics. He has spent most of the last 25 years using signal processing and statistics to study how signals in the brain become perceptions and cognition. This work has often involved statistical analysis of noisy, large datasets to find consistent relationships between measures of cognition and measures from neurons. His current lab work primarily studies how deep brain stimulation can improve cognition. In 2020, he was sent home for six weeks during a pandemic with access to a large open dataset that related COVID-19 cases to hospital data and deaths, and he set about to use this data to inform people on social media about COVID-19. His column in the Bell Ringer will provide the public with valuable statistical analysis during the current pandemic.* * *Monday, July 27Hello, everyone. I intend to make this blog a daily event on weekdays. I have been using data analytics to study the progression of COVID-19 in the nation and in the community. I will be making daily reports on how things are going, but today I wanted to introduce the methods I use. My purpose here is to aid public understanding of this outbreak and of data analytics.COVID-19 is an unusual virus in many ways, and among them is its tendency to produce asymptomatic infections. Of course, it is also a deadly virus, and I don’t want to downplay that at all, but today, let’s focus on it producing asymptomatic cases. In Wuhan, China, long after the outbreak had been contained, they found a half dozen cases in one day. That alarmed the Chinese, so to gain the public’s trust they tested 11 million people over two weeks and found 206 cases. These were cases that found no reason to go get tested. So, if we want to know the number of new infections per day that those 206 represent, we also need to know that if you are infected, you will test positive for roughly 12 days. So, those 206 infections can also be represented as 17 (206 divided by 12) new infections per day. In very low level outbreaks, we can therefore estimate there are three unknown cases for each case that produces fever and coughing. This aspect of COVID-19, the large fraction of asymptomatic cases, makes it tough to know how many people are infected.Many data scientists have looked at this problem, and a subset of us have decide it is fair to estimate the number of infections from 1) the number of confirmed cases and 2) the percent of those tests that are returned positive, or the test positivity. If test positivity rises, the argument goes, then the ratio between actual infections and confirmed cases should also rise. Like the case in Wuhan, there are now a number of studies we can use to normalize such a model. These studies generate surveillance data that has some way, other than confirmed cases, of estimating the total number of people infected. In Wuhan they simply tested everyone. In many places, they performed antibody surveillance work to estimate the cumulative number of people infected in that area. In New York City, in two hospitals, they tested each pregnant woman for 10 days.If you collect these data points, and for each, calculate the ratio between confirmed cases and actual infections, and plot this as a function of test positivity, you get the following graph. A function was worked out to relate these two that can be used in practice, and that is the line. It is not perfect. The function I use includes the point (0,0), while it is clear that for a near zero test positivity, like in Wuhan, there are still three infections for each case.You may fairly wonder why this would be necessary. Why can’t we just use confirmed cases? Testing is a limited resource, especially in outbreaks, and testing facilities need to ration tests to the most important people, who are vulnerables and healthcare workers. When tests are limited, test positivity tends to rise, and the relation between confirmed cases and the magnitude of our outbreak is destroyed. Using the sampling function I illustrate above, we can still have an estimate of the magnitude of our outbreak even if our testing cannot keep up. Currently, in Georgia, our testing cannot keep up. But using this function, we can make a plot of the outbreak in the state of Georgia. In the top panel (below), I show confirmed cases in Georgia, and in the bottom plot, the percent of the state infected. I overlay the percent infected with a time shifted plot of the average death rate to see if we are getting better at treating COVID-19 and saving more lives. You can see this way of looking at the data suggests the death rate in Georgia will be rising quite a bit in the next three weeks to match the increase in our outbreak.I want to end this first blog post with a few references I used to collect that surveillance data I illustrate above, and to web sites I use to collect testing and death data on COVID-19. You may also be interested to know that I freely provide source code to anyone who asks, for any reason, so if you want to see the code (it is in Python), just ask. My email address is below.Surveillance datahttps://www.caixinglobal.com/2020-05-26/mass-testing-finds-more-than-200-asymptomatic-covid-19-cases-in-wuhan-101559009.htmlhttps://www.nejm.org/doi/full/10.1056/NEJMc2009316https://www.medrxiv.org/content/10.1101/2020.05.04.20090076v2https://news.iu.edu/stories/2020/06/iupui/releases/17-fairbanks-isdh-second-phase-covid-19-testing-indiana-research.htmlhttps://www.axios.com/coronavirus-new-york-antibody-test-f4fbed78-646f-4b46-90b8-5e8ca75380e4.htmlhttps://health.oregonstate.edu/covid-19/traceData on cases and testinghttps://covidtracking.com/datahttps://github.com/CSSEGISandData/COVID-19/tree/master/csse_covid_19_data Contact David Blake at Send Email to Dave Blake.