Here are notes and links I personally find useful, all gathered in one place. Mostly, I kept looking at webcams to see whether people are practicing social distancing and decided to put them all in one page. I also gather data when nervous, so I also have plots of things like testing, activity, and cases over time so I can see the info directly myself (for example, Knox County Health Department plots cases per day with a trendline, but so far their trendline is only a straight line - I was curious if a more flexible fitting might be more informative). I am not an epidemiologist, so please do not draw any advice from this page (and I’ve been careful NOT to do projections or anything like that – there is a lot that goes into modeling, see this video by my colleague Nina Fefferman, who unlike me is an expert in this, for more on models). This is just me poking around with the data to make myself feel better – the raw code is here if you’d like to examine it yourself.
Other sources for data and other information:
And sections of my page:
The following plots are created in R using data from https://covid19datahub.io (Guidotti, E., Ardia, D., (2020), “COVID-19 Data Hub”, Working paper, doi: 10.13140/RG.2.2.11649.81763), as well as information from Google and the state of Tennessee dataset page and plotting and analysis using the
dplyr packages. Plots created on 2020-11-22 08:54:56. I use a seven day rolling average for most plots using the
geom_ma function of
The number of new should be not be going up if things are well-controlled, and ideally should be going down. I used to plot active cases, but how those are tracked has changed (it used to be based on recoveries, now it’s just a two weeek lag of active). This uses data from https://www.tn.gov/health/cedep/ncov/data/downloadable-datasets.html. The number of new cases per day seems to align fairly well, but not perfectly, between the state and county data.
I plot five regions (though for simplicity, for some of the plots I omit East TN, and regions with no data automatically do not get plotted for certain measures)
Information from the last two datasets comes from the hard work of Alex Zukowski, who requests the testing data and who aggregates the reported data. Zukowski, Taylor A. “University of Tennessee, Knoxville COVID-19 Dataset.” COVID-19 in The University of Tennessee, Knoxville, 2020, sites.google.com/view/utk-covid19/data. I recommend going to https://sites.google.com/view/utk-covid19 for more on this dataset and the work going into gathering it.
For many of these plots, I am now showing the seven day averages.
Here are the same plots, but normalized so it’s per 100,000 people in each area per day (making it easier to see patterns in less populous areas). I here use the Harvard Global Health Institute’s standards for guidance of control. Yellow means “rigorous test and trace programs advised”, orange means “stay-at-home orders and/or rigorous test and trace programs advised”, and red means “stay-at-home orders necessary”.
Testing is a key part of getting a handle on the pandemic. Here are Knox County, Oak Ridge’s counties (Anderson and Roane), and all the East TN counties combined (the same ones used for regional hospital reporting for bed availability). This plots shows the daily number of tests per 100,000 residents. It also includes testing done by UT student health, using data compiled by Alex Zukowski (Zukowski, Taylor A. “University of Tennessee, Knoxville COVID-19 Dataset.” COVID-19 in The University of Tennessee, Knoxville, 2020, sites.google.com/view/utk-covid19/data). For the UT data, I am plotting against number of students (30,000).
Another way to look at testing (or disease spread) is to look at the proportion of positive tests (“positivity rate”): they can go up if a greater proportion of symptomatic people are being tested and/or if the frequency of the disease is increasing. Updated WHO guidelines recommend testing should be 5% positive or lower (black line). The lines show the seven day averages. UT’s testing positivity rate (using student health data is so high it obscures the county data, so it is reported separately).
Many of the local schools are adopting a hybrid approach, where families can choose whether to send their kids to school or do online learning (sometimes with reduced options). The New York Times reports on July 14, 2020, that, “As education leaders decide whether to reopen classrooms in the fall amid a raging pandemic, many are looking to a standard generally agreed upon among epidemiologists: To control community spread of the coronavirus, the average daily infection rate among those who are tested should not exceed 5 percent.” Many districts nationally are not hitting that goal, but a question is whether local ones will. The first district to open in our area was Oak Ridge public schools, who had a start date for students in person on July 29, 2020. Following the procedure of the NYT‘s article, I am plotting the seven day average of positivity rate for tests for the counties including Oak Ridge’s counties (Roane and Anderson). The black line is the seven day rolling mean (total new confirmed / total new tests, not a mean of this ratio per day), the blue line is this smoothed, the dotted line is the 5% threshold, and the green rectangle is the goal and starts at the start of school – the positivity should be in that rectangle to meet epidemiologists’ goals. Oak Ridge is giving families options of whether to enroll their children in online or in person school: their plan, and the way to choose either mode, is here. They have updated their in person plan to require masks when possible but not social distancing; when individuals test positive, the possibly affected parts of the building will be closed for 2-5 days. We’ve opted for the online option for our family.
Harvard’s Edmond J. Safra Center for Ethics and the Harvard Global Health Institute have created (July 19, 2020) guidelines for schools reopening (see also their full report PDF). They distinguish students by age group: “With COVID-19, people 18 and younger have far lower risk of death, hospitalization, and severe outcomes and are also less likely to get infected. Within this group, students in the younger age band of 10 and under also transmit at lower rates. This last point about lower rates of transmission may also pertain to people 15 and younger, a point that research should clarify in coming weeks. Keeping levels of risk low for young children via pandemic resilient teaching and learning spaces is more readily achievable than doing so for high school age students and the adult educators and staff in the school building.” They provide suggestions on building ventilation, social distancing, and more, and mention the utility of looking at multiple metrics (such as positivity rate, hospitalization, and more), but focus on daily new cases per 100,000 people (with a caution about missing cases if testing is not adequate; for example, if positivity rate is above 10%). The plot below has summaries of their recommendations for priority for which grade levels can meet in person, if there are conditions for pandemic resistant teaching and learning. The black line is the rolling seven day average of new cases per day for the counties containing Oak Ridge (Roane and Anderson), the blue line is this smoothed.
Tennessee now breaks out covid test results for school age children (age 5-18). The number of tests in this age group and the underlying population size aren’t known to me, so things like positivity rate and proportion of students infected cannot be determined, but the raw number of students infected can be shown (doing a seven day average):
As of the last time the data were updated (likely 4 days ago), regional hospitals had 47 ICU beds available of 278 total (so was at 83 percent capacity), and 161 available ventilators out of 279 total, (so was at 42 percent capacity). The hospitals overall had 806 beds available of 3163 total (so was at 75 percent capacity). This is based on 19 acute care hospitals in the East TN region; based on the 14 counties these hospitals are in, these serve at least 1,235,720 people. When a line hits 100%, the local hospitals are theoretically full for that resource (for all patients, not just covid patients), though there is surge capacity on top of this. Note that these data are updated only weekly, so current conditions maybe be much better or worse than these plots show. Data on capacity from Knox County’s dashboard, data on hospitalizations and testing over time from the state data.
Trends in hospitalization of covid patients over time. The vertical dotted line shows the last day with updated capacity information, when there were 47 ICU beds available (ignoring surge capacity) for people in the East Tennessee region. The seven day average is shown.
Which age groups are being infected is an important question as local schools open up. These are data from Knox county alone. Note that one important bias here could be testing rate: if two year olds aren’t being tested as often as forty year olds, the age 0-10 results will be lower than the actual rate of infection, for example.
Another question is what is happening at UTK. There is official information available at https://www.utk.edu/coronavirus/guides/data-monitoring-and-contingency-options, of which I show one figure below. There are also reports in the news media, such as this article from Aug. 18 showing that 23 of 91 football players (over 25%) have had positive covid tests since returning but before the start of classes. New York has set a threshold of 100 cases or 5% positive tests for colleges to move online; UT has had over that many active cases; its positivity rate is not officially public (as of now), though this site provides such information. The numbers of cases and tests have dramatically dropped recently at UT; one troubling suggestion for why is reports of “pacts” by students not to get tested locally so as to avoid notifying the university. This violates the agreement students made before classes started and puts the local community, both on and off campus, at greater risk. UT is trying tests of sewage and saliva tests: upcoming saliva tests are scheduled at https://calendar.utk.edu/covid-19_testing.
For a detailed look at UT, check out https://sites.google.com/view/utk-covid19, for work done on an individual basis by a UTK student, Alex Zukowski. UT’s official dashboard is at https://www.utk.edu/coronavirus/guides/data-monitoring-and-contingency-options/ but has less detail. I use the daily testing counts UT used to provide to Alex Zukowski as well as UT’s weekly reports of the totals for that week. To convert to daily data, I divide the weekly reports by seven.
UT is now publishing results of saliva tests. ALL students residing in dorms, fraternities, or sororities have agreed to testing, but in some cases over half are missing mandatory testing. No sanctions have been announced yet. This table shows testing results of saliva tests:
Testing at UT is also a question. The plot below shows reported number of tests at the student health center, based on data aggregated by Alex Zukowski (see above). Note that these are the raw number of reported tests: a value of 20 means out of all 30,0000 UTK students, only 20 individuals were tested that day at the student health center. Not included here are athletes, who have much more frequent testing.
The dotted line on the plot shows the date of a news story about covid pacts among students to avoid getting tested locally so they could avoid quarantine or isolation. Spread by infected individuals matters, even if they personally are asymptomatic, so personal actions to avoid testing and quarantine / isolation can lead to disease spread to others who might not be so lucky, so it’s very disappointing to hear of this. The drop in testing after that line may be purely coincidental, but it is a remarkable change in trend. Other universities test more intensely; for example, U. of Illinois tests several thousand per day, not a few dozen. Not being an epidemiologist I do not want to weigh in on what strategy is optimal, but it is a remarkable contrast in approaches. UT is also starting to report on mandatory saliva-based testing, where they ask everyone residing in a dorm for a saliva sample, pool 3-5 individuals in a tube, test each tube, and then bring back individuals from pools that tested positive for individual testing. From 19.1 to 36.1 percent of the students listed as being in these dorms do not contribute samples.
As for the county data, the positivity rate can also be important: if all the tests show up positive, likey symptomatic, cases, then there are likely other people being missed. The New York Times reports on July 14, 2020, that, “As education leaders decide whether to reopen classrooms in the fall amid a raging pandemic, many are looking to a standard generally agreed upon among epidemiologists: To control community spread of the coronavirus, the average daily infection rate among those who are tested should not exceed 5 percent.” The horizontal black line below shows that guidance. The CDC provides thoughts about possible testing strategies at colleges and universities here but not a firm number for how much testing they believe is necessary.
We can also use the UTK saliva data to estimate the number of new cases: the proportion of students who ultimately test postive out of those tested by the saliva scans. Given poor compliance, the true proportions may be much higher than this (if students want to endanger others by not isolating, they may avoid testing if they think they have been exposed). Depending on the date of the test, between 1 of every 41 to 1 of every 531 of students who contributed samples had active covid infections.
At the date of the most recent set of samples, from 2020-11-09, UTK reported 82 active cases overall (this presumably includes the 14 positive diagnostic tests from the sampling). If the pool of students from this sample, which has a participation rate of 73.9 percent, is representative of the 30,000 person student body as a whole, that would mean there would be 732 actual active cases among the students on that date.
I am converting the saliva test info to new cases per 100K to correspond with the Harvard Global Health Institute’s standards for guidance of control (yellow is community spread; red is the highest risk possible in their guidance – for a county, that would mean “stay-at-home orders necessary” according to them, though how that translates into actions on a college campus is different). These guidelines are for daily new cases but the saliva data are active cases at a point in time. I am adopting the assumption that Knox County uses that new cases are only active for 14 days, so I’m computing the new case estimate as 1/14 the active case estimate. The green, yellow, and orange bands are the same as for Knox and Anderson+Roane counties above; the red band looks much thicker because it has the same lower bound as the others but no maximum, and some of the estimates for UTK are outside the range for the county-wide data. I include the 95% confidence interval for the estimate of new cases per 100,000 people, though this is an underestimate of the uncertainty (are dorms good random samples or are results clustered, are students who refuse testing similar to those who get tested, etc.).
A different way to consider this is what percent of the community has active covid infections at any one time. There are two ways to get this. One is assuming that every person who has active covid 1) knows this through testing and 2) reports this to UT. This uses the active case numbers reported by UT (gotten through the intermediary of Alex Zukowski). The other way is to assume that the random saliva testing at the dorms is a better way to get this estimate. This avoids issues of asymptomatic or presymptomatic individuals not getting tested, but it could be biased in other ways: maybe the ~25% of students who skip mandatory testing are more likely to have covid (or maybe they’re studying at home and not using the dorm at all), maybe the dorms selected are selected based on the results of sewage testing for covid, etc. My guess is that the saliva testing is the better estimate.
Given the estimated percentage of individuals with active cases, what is the chance that in a given group of people at UT, there is someone there with covid? This is hard to know, as people with active cases who know about them are hopefully isolating themselves, and others are isolating who have been exposed and who may have active cases but do not know that yet. If we assume that the saliva samples are the best estimates of proportion of active cases (since it does not rely on students deciding to go to student health due to symptoms), we can use that to estimate the true number of active cases. We can then compare that to the number of people isolating due to active cases to figure out how many active cases are not in the self-reported situation. They could still be isolating because they have close contacts who are isolating, or they might not, so we can use both those assumptions. For example, on the last date of saliva samples, 2020-11-09, there were 82 individuals at UT with active reported covid infections, 390 individuals isolating (including presumably all the reported active cases), but based on the saliva proportion of active cases there would be 732 active cases across the 30,000 students. The point estimate of the number of people who may be active and not in isolation may be between 342 and 650 individuals, depending on how many of the ones who are active but don’t know it are self isolating anyway. Remember that all these numbers are very uncertain, especially those based on small sample sizes, but we can use this to get a gut sense of the probability of encountering someone who actively has covid in different group sizes – perhaps important for showing the importance of avoiding large unmasked groups, especially before returning home to family over the winter break. For example, for a thirty person party some weekend, there is between a 44 to 67% chance that at least one person at that party actively has covid during the party (the lower estimate assumes that every person in isolation, whether they know it or not, is one of the people with active covid, but that still leaves many to continue spreading it (and people outside UT’s campus also have covid, too)). Even repeated small group interactions pose a risk. If someone does not have covid, and repeatedly meets with random groups of three other people who might have it, there is only a 3.4% chance of someone else in each group having covid, but by the time the person has met with five such groups there is a 16% chance that the person will have been exposed to someone with covid in at least one of these groups.
UT’s webcam of the Rock, a campus landmark, to see how UT’s community is choosing to protect others. You will have to click to start the stream. Some members of the community engage in hate speech on the Rock – it is often quickly replaced by more positive messages, but it may be present on the Rock temporarily.