UPDATE: This article was published on September 2. On September 3, Metro announced a change to their 7-day Positive Test Rate calculation which more closely aligns with the way the state is calculating it, as well as our understanding and calculations in the article. We are appreciative of this change and applaud all efforts to increase transparency in the calculations used in reopening guidance.

Recently, we wrote about some discrepancies between various sources for the number of available hospital beds in Davidson County. Metro Nashville appears to be using one number of available beds; the state appears to be using another (and this coincides with other sources, and statements made by Metro Coronavirus Task Force leader Alex Jahangir earlier in the year).

These sorts of problems, however, are not limited to the hospital bed capacity. Today, we tackle another example of a metric tracked by Metro Nashville which is at odds with other data sources: The 7-Day Positive Test Rate. As of this writing (September 2), the official figure reported by Metro is 9.0%, just below the stated goal of 10.0% or below.

As with hospital bed capacity, this calculation is pretty simple, and doesn't require a ton of math beyond a simple division: All we need to know is the number of positive test results in the last 7 days, for the numerator, and the total number of test results (positive OR negative) in the last 7 days, for the denominator.

If it were only that simple!

One may think that Metro's daily press updates might contain the data we need to recreate their figure. After all, these updates contain a table each day that looks like this:

Of course, these are cumulative figures - we care about the 7-day rate. Fortunately, Metro archives all of these updates, so we can go back 7 days to August 26 to see what the numbers looked like then. The difference between the two reports is the new activity over a 7-day period; seems intuitive that this would match the Metro-reported figure.

Taking the difference between the two reports, we can see that Metro has reported 710 new "positive/probable cases" in the last 7 days, and has reported 8,401 "negative results." If we divide the positives (710) by the total results (710 + 8,401), we get... 7.79%. That's quite a bit lower than the 9.0% on the homepage!

In fact, the official Metro number has been higher than our calculated number fairly consistently since the 7-Day Positive Test Rate was introduced as a reopening metric on August 6, and has almost always been higher since July 9. Given that this has been introduced as a metric that Metro is tracking to assess reopening, it seems important for metro to get this right, or at least make clear what they're doing.

So, whatever Metro is doing, it's clearly not the intuitive calculation we've just described. Fortunately, there are other data elements and sources that we can use to test alternative hypotheses. The most straightforward comes from the Tennessee Department of Health: Each day, they publish an update of data by county in a couple different downloadable files.

• The first, "Public Dataset Daily Data Snapshot", contains current data for each county, which includes the 7-day Positive Test Rate in a column labeled "PERC_POST_LAST7". It also contains the new total tests and new positive tests for the current day, and the sum of total tests and positive tests for the last 7 days. The last two figures can be used to directly calculate their 7-day Positive Test Rate.

We have a problem here, though: the state's number for Davidson County is just 4.68%. This doesn't tie to either of the numbers we presented before, and notably, is almost half of the rate reported by Metro to assess reopening guidelines.

• A second dataset from the state on the same web page, "County New", has much of this data historically day-by-day. From this, we can not only verify that our understanding of the state's 7-Day Positive Test Rate calculation is correct, we can also check their 7-day test totals.

The three rightmost columns (N, O, and P) have been added by us to demonstrate that this data ties to the first dataset that we examined for 9/1. Specifically, column N sums the positive tests in the last 7 days and gets 995; column O sums the total tests in the last 7 days and gets 21,242; column P divides these two figures. All of the data matches from the first file. This means we can look at how the state's 7-Day Positive Test Rate has evolved over time, and how it has differed from Metro's number.

Rather than getting us closer to an understanding of Metro's 9.0% rate, however, this exercise has taken us farther away. The state's figure of 4.68% is even further away than the 7.79% figure we calculated from Metro's daily updates. We track all three of these figures over time, sourcing them in the ways described above, here.

What else can we do to try to match Metro's figure? Well, there is one other option, which Metro only made available to the public beginning on August 21. The Davidson County COVID-19 Dashboard provides a tab called "COVID-19 Epidemiology and Surveillance Updates", which takes the form of a PDF report with additional data not available elsewhere. It also lags the current day's data by one day, so as of this writing on September 2, the report is dated September 1. Still, since we're tracking this data over time, it may be useful.

Page 7 of this report contains a section that looks like this:

The numbers on the left under "COVID Tests," you'll notice, match with what Metro provides on their daily press updates. However, despite referring to these as "tests," it appears they would more accurately be described as "de-duplicated test results," given the additional information on the right-hand side: "COVID-19 PCR Tests." In this section, there are higher figures for all 3 values ("Negative PCR", "Positive PCR", and "Total PCR Tests").

This data is not archived historically on Metro's website; our team has been saving this file each day to keep a historical account of how these figures change. Above, we show the "PCR Test" data on September 1. 7 days prior, on August 25, this data looked like this:

If we take the differences in the "PCR Tests" data, we get 978 positives out of X total tests, resulting in a 7-Day Positive Test Rate of 6.10% - lower than both Metro numbers we had before, and closer to (but still higher than) the state's number.

So, after all of this, we are still at a loss. Nothing we have done has resulted in a 9.0% positive test rate for Davidson County over the past 7 days. Our final check is to look at the daily test result counts themselves. While this is unlikely to get us to a satisfying conclusion, we may at least be able to discern patterns in the differences between test totals reported by the state and those reported by Metro.

As a precursor, though: We know that Metro receives data from the state, and not the other way around. We know this because on June 28, the state Department of Health released a statement that they would not be able to report due to technical difficulties. Metro, however, reported their daily update as usual on June 28; it was one day later, on June 29, that they were unable to report, stating: "The Metro Public Health Department receives COVID-19 test results from the Tennessee Department of Health for our Daily Update." So the appropriate comparison is between Metro's data today and the state's data yesterday.

Here are the new positive and negative figures derived from each of these sources for the 7 days preceding September 1 (which is the last day for which we have all of the aforementioned data sources available to review):

This table makes it more clear that what they call "Tests" (orange) in their daily press updates are actually de-duplicated cases. But the comparison between the state's data (red/left) and Metro's daily PCR tests a day later (yellow/right) only serves to add more confusion. When we started this journey comparing the daily press updates to the state data, we thought it was clear that there was a filtering process resulting in lower numbers. Now, however, we believe there must be both a filter on the state data and additions made to the data from Metro's own sources. In several recent days, new positive tests reported by Metro have exceeded those reported by the state for the same time period. On 8/31 and 9/1, new positives reported by Metro were higher than that reported by the state, but new negatives were lower. This suggests filtering and merging from a data source that is not publicly available.

Whatever that process, we're not able to reproduce Metro's reopening metric, neither through the data that they themselves provide, nor through the data that the state Department of Health provides. Moreover, those two data sources are increasingly at odds with one another throughout August and into September. A simple problem of division with at least three answers that we can come up with, none of which match how Metro is calculating the metric (which is higher than all of them).

We're not sure why.

COVID-19 has presented myriad challenges for state and local governments and public health departments. Many were never equipped to handle real-time reporting to the public prior to April; even those that were had to learn a new lexicon before they could responsibly report on COVID (cases, hospitalizations, ventilators, COVID-Like Illness...). Pair this with a decentralized system of healthcare and government, and it's no wonder the format and level of detail of the data varies from one location to the next.

Nashville's Metro Public Health Department, for its part, launched a "Roadmap for Reopening" in mid-April: A set of "data-driven, not date-driven" metrics which, when met, would assure residents and local leaders that it was safe to move to the next "Phase" of reopening. So far, so good.

But among these metrics are two which deal with hospital capacity: One which tracks the percentage of floor beds available in area hospitals, and another which tracks the percentage of ICU beds. This calculation seems pretty straighforward, right? You just take the number of patients currently in hospitals (the "hospital census"), and divide that by the number of beds, to get a percentage. Metro declared that 25% of beds should be available at any one time, before reducing the thresholds down to 20% sometime in May.

So, about that calculation: You just need a numerator and denominator. Simple, right? Well... no, because Metro doesn't publish those numbers, as other cities do. Memphis, for example, provides an easy-to-read table as part of Mayor Strickland's daily updates there:

From the above, we know that there are 2,286 total patients in Memphis-area hospitals, and 190 available beds. This means that there are 2,476 total beds, and that 92% are taken by patients of all types. We also know that 171 beds are taken by "COVID Positive" patients, or 7% of the total.

What are the relevant figures in Nashville? We don't know, of course, because the city doesn't publish them (we wish they did). But with some investigation, we can make reasonable estimates for what bed and patients counts are now, and have been in the past.

First, let's look at some publicly available data. The Tennessee Department of Health publishes a statewide listing of hospitals and bed counts in each, most recently updated in June 2020 (during the pandemic). Tally up the number of beds in Davidson County, and you get 4,264 beds. We also have this private database of hospital locations, bed counts, and average utilization; filtering down to the 13 hospitals in Nashville yields a similar total of 4,186 beds. Two sources (one public, one private), similar totals; we seem to be on the right track.

But we can further triangulate this capacity figure using another source: statements by local health officials to the press. Jump back to the beginning of the COVID-19 timeline, and you find this Nashville Post article from March 26, which quotes Metro Coronavirus Task Force leader Alex Jahangir as saying that Nashville "has about 3,000 available beds and 1,600 intensive care unit beds, and they are making plans to build emergency facilities if necessary." It's unclear, here, whether the ICU beds are considered in the total bed count; nevertheless, we either have 3,000 beds with room to expand, or 4,600 beds. Jahangir subsequently announced that the city was expanding hospital capacity to keep up with demand during late July, suggesting that this number is flexible and can change over time.

Finally, we DO know the number of COVID patients in hospital, as this information is published on the Davidson County COVID-19 Dashboard. As of August 20, there were 160 COVID-19 patients in hospital, a number that has ranged from about 130 in mid/late June, up to a peak of about 220 in late July. This hardly supports Jahangir's late July claim that "hospitals are getting full of COVID patients," based on what we know about the city's overall capacity; 220 is something between 4.8% and 7.3% of total hospital capacity based on Jahangir's own statements from March (depending on how you want to interpret them).

The above sources and figures were used to create this very website's Nashville Hospital Capacity Tracker - it's our best estimate of how many hospital beds are in Nashville, and of those, how many are available, how many are taken by COVID patients, and how many are taken by non-COVID patients.

Recently, though, statements by Dr. Jahangir have cast doubt over how the city is calculating the capacity metrics used on its website. Here are a couple notable examples:

• In an interview Tuesday evening with WSMV-TV, Jahangir stated that "three weeks ago, 13% of all people hospitalized in Nashville, 222 individuals, were hospitalized because of COVID." This implies a total patient count of roughly 1,708 on the date in question, which was July 22, according to the county's dashboard:

But on that same date, Metro Nashville reported that available bed capacity was just 17%. This means that the 1,708 total patients made up 83% of beds. Do the math, and the implication is that there were 2,057 beds available - far below all of the earlier estimates from data sources, as well as from Jahangir himself in March. Jahangir continues in his August 18 interview: "Today that number is 9% [of total patients are COVID patients], or I believe 150 people, 160 people." This works out, again, to something like 1,700 total hospital patients - a number, you'll note, that is virtually unchanged from the "peak" in late July! - on a date which Metro Nashville reported, again, that available bed capacity was just 17%. These two virtually identical numbers suggest that Metro Nashville has been using a total bed count of 2,000 - 2,100 in their calculation of hospital bed capacity - again, far below the numbers that we can identify from other sources.

• Jahangir repeats these numbers in the Mayor's daily press briefing on August 18 as well (skip to 6:00).

So, what is going on here? Did the number of beds decrease since late March (per Jahangir) and since June (per the TN Department of Health)? If so, does Nashville have the ability to increase its hospital capacity once again in the event of a surge?

Or, perhaps Nashville is counting only staffed beds (that is, beds which can be attended to by nurses and physicians), and not the total number of physical beds, in its calculation of capacity? Memphis' numbers refer to this in a footnote, so perhaps this explains the discrepancy. But once again: Does Nashville have the ability to staff these additional 1,000 - 2,500 beds in the event of a surge?

Or, in the most pessimistic view: Is Nashville manipulating the denominator in their calculation of hospital bed capacity to achieve a desired result? We won't, and don't, accuse them of that without evidence.

We don't know the answers to these questions. It doesn't have to be this way - Nashville draws their data from the same state database - and a calculation involving simple division could very easily be explained to the public and clarified. Memphis provides this context, which is critical to our understanding of whether their hospital system is equipped to handle additional case loads from COVID-19.

Until someone demands answers, though, we're left wondering.

We are Nashville for Rational COVID Policy - a non-partisan advocacy group of concerned Nashvillians. We are small business owners, medical professionals, concerned parents, and workers who want to ensure that Nashville's COVID policy is clear, transparent, grounded in data - and that it's working to minimize ALL societal costs (in terms of lives, as well as livelihoods) for everyone in Nashville.

Our goals and mission for the people of Nashville are:

1. To maintain transparency and accountability in public policy related to COVID mitigation efforts. Policy measures should be clearly defined, and changes should be communicated well in advance, giving the public sufficient time to respond. Further, measures should be grounded in data that is publicly available, and which most accurately captures the level of community spread and associated risk that exists in Nashville as of today.
2. To make known the substantial costs associated with sweeping mitigation efforts. Business closures, school closures, and restrictions on private establishments are detrimental to business owners, employees, parents and children alike - not only financially, but also in terms of mental health, education, and in terms of second-order public health effects. These measures should not be held in place a moment longer than necessary, and their benefit should be assessed against their cost as part of a public dialogue, not by decree.
3. We are non-partisan and are not affiliated with any other political causes, candidates, or parties. We strongly believe that COVID presents risks and challenges - but that prudent public policy involves balancing those risks against the costs of mitigation efforts, so as to minimize the total damage to our community in view of the current data and information.

We look forward to a productive relationship with city leaders and health officials to ensure that the health, safety, and livelihoods of all residents of Nashville are taken into account to arrive at rational, data-driven policy.