The Third Surge Worsens as Hospitalizations and Deaths Rise Nationwide. Plus a New Tool for Understanding Racial Data: This Week in COVID-19 Data, Oct 22

This week’s update is brought to you by Artis Curiskis, Alice Goldfarb, Erin Kissane, Kara Oehler, Jessica Malaty Rivera, Joanna Pearlstein, Sara Simon, Peter Walker, and Nadia Zonis.

The upswing in COVID-19 cases, hospitalizations, and deaths continued this week, as the United States recorded the highest number of hospitalizations in almost two months. Cases rose 13.5 percent, and though the nation as a whole exceeded 1 million tests per day on average this week, we saw a slight decline in tests from a week prior. Nearly 1,000 deaths from COVID-19 were reported on Wednesday alone, and that didn’t include Alabama, Florida, and Georgia, which were experiencing reporting issues. States reported 5,300 COVID-19 deaths since last Thursday, an increase of more than 10 percent week over week.

The number of hospitalized COVID-19 patients rose nearly eight percent from a week ago. Hospitals across the country saw patients increase, mirroring the widespread case growth. While the Midwest remains the region of greatest concern, Texas led the nation in increased hospitalizations over the past seven days.

Trouble continues in the Midwest 

North Dakota, Wisconsin, and South Dakota again claimed the highest numbers of cases per capita, with Montana close behind. In both Dakotas, the governors’ resistance to statewide lockdowns or mask mandates have left local officials pleading with citizens to take measures to slow the virus’s explosive spread. Last week the mayors of 16 South Dakota cities, including the capital, Pierre, published an open letter urging South Dakotans to wear masks, limit social interactions, and wash hands frequently, among other disease-prevention measures. 

On Tuesday, North Dakota’s health department said the state could not keep up with its contact tracing workload, tweeting, “Please be advised: NDDoH and local public health units are currently experiencing a backlog of COVID-19 case investigations, causing a delay in calls. If you have tested positive for COVID-19, please isolate and inform any close contacts you've had that they should get tested!”

To get a sense of the extent of the spread of COVID-19 in North Dakota, a comparison with San Francisco is instructive. There are more people in the city of San Francisco (883,000 people in 47 sq. miles) than in the state of North Dakota (762,000 people in 70,762 sq. miles). This week, more than 5,200 people in North Dakota tested positive for the coronavirus; San Francisco counted 213. Since the pandemic began, San Francisco has recorded just over 12,000 cases and 138 deaths; North Dakota has seen 34,165 cases and 323 deaths since March. This week San Francisco became the first urban county in California to move to the state’s yellow, or “minimal,” level for virus transmission. The state government has approved reopening or increasing capacity at non-essential offices, movie theaters, pools, and houses of worship, among other types of establishments in the city. 

Although the reports from Dakotas and Wisconsin are particularly bad, cases have surged across the Midwest, where the seven-day average now stands at 308 new cases per million people per day. If trends continue, there will soon be more cases per capita in the Midwest than have been recorded in any region since the outbreak began. (In the outbreaks concentrated in the Northeast this spring, infections far outpaced testing, so the Sunbelt surge over the summer may be a better point of comparison for the Midwest’s current troubles.)

Unsurprisingly, the uncontrolled virus transmission in the Midwest has begun to impact mortality levels across the region. Deaths have steadily risen since mid-summer, but have turned sharply upward this month; the seven-day average for deaths is up 64 percent since October 1.

Hospital data wobbles again

Over the summer, we reported extensively on irregularities in hospitalization data related to changes in reporting requirements issued by the US Department of Health and Human Services (HHS). More recently, hospitalization data reported by states and by HHS appeared to have stabilized, but as the fall surge worsens, we are once again seeing gaps in this crucial set of COVID-19 metrics.

In Missouri, for example, the currently hospitalized figure has been underreported since October 17; the state links the problem to hospitals’ scrambling to keep pace with HHS requirements. The state’s public health department posted the following in a public notice on its COVID-19 dashboard:

The total numbers of COVID-19 hospitalized patients were underreported on October 20 because of data changes from the US Department of Health and Human Services, and challenges with entering data to the TeleTracking portal used by hospitals to report data daily. Totals from October 17 onward are affected until further notice.

On October 21, researchers at The COVID Tracking Project identified five other states with  anomalies in their hospitalization figures. In Kansas, the number of ICU patients reported decreased from 80 to one without explanation, Wisconsin’s hospitalization figures stayed unexpectedly flat while the state’s other indicators worsened, and, as mentioned above, Georgia, Alabama, and Florida reported only partial updates to their hospitalization counts. Some or all of these reporting deficits may be linked to HHS reporting changes, but we haven’t yet determined a definitive reason—or set of reasons—for the apparent instabilities in hospital reporting.

Hair-raising moments in test positivity 

Another data-related issue that emerged this week concerned test positivity. The COVID Tracking Project has chosen not to calculate test positivity for several reasons, but many other organizations, including Johns Hopkins University, do use the data we compile to offer this calculation on their COVID-19 dashboards. Several news outlets noted that early this week Johns Hopkins was reporting a startling 48 percent test positivity rate for Iowa and an implausible 100 percent test positivity rate for Nevada. These apparent spikes illustrate the importance of careful interpretation and contextualization in reporting test positivity. It appears that on October 16, Iowa reclassified more than 12,000 tests previously reported in their PCR testing metric as antigen tests, removing them from the state’s PCR test figures. This change reduced the denominator in the state’s test positivity calculation, which made the state’s test positivity rate appear artificially high.

Meanwhile, Nevada launched a redesigned COVID-19 dashboard that no longer included the number of people who tested negative for COVID-19. This seemingly small change had a cascading effect on our total tests data, as we wrote in a brief public note on the state’s page:

On October 15, 2020, Nevada stopped reporting the number of unique people who tested negative on their dashboard without explanation. We carried forward the last reported number of people tested negative between October 15 and October 21 (our policy when a metric goes missing without notice is to “freeze” the number for up to seven days to allow states to recover from temporary reporting problems). Since our API uses the sum of positive+negative tests to calculate the total number of tests in Nevada, this caused the number of total tests on those dates to increase only by the number of positives. On October 21, 2020, we updated the number of Nevada’s negative tests for October 15 through October 20 based on a calculation of (Total people tested – Cases)

Further complicating the picture, our testing data reflects the number of tests reported by a state on a given day, rather than the number of tests actually conducted in that state on a given day. Only a few states make the actual date of testing figures available, usually well after the reported counts. This can lead to dramatic fluctuations in calculating test positivity using our data. As always, we recommend using seven-day averages to account for reporting problems, and looking into a state’s reporting history to identify any issues that can skew test positivity calculations.

This week in long-term care facility data 

Our team continues to compile weekly data on COVID-19’s spread through the country’s long-term care facilities. We’re alarmed that Florida went nearly three weeks without releasing long-term care death data, despite saying they publish numbers weekly. The state finally released data on October 22, though it’s dated October 16, reporting 599 new resident and staff deaths.

Prior to Florida’s update, we had found that Texas, Illinois, and Oregon accounted for 23 percent of long-term care COVID-19 deaths from October 8 to October 15, 2020.

We’re closely watching Louisiana because of a particular data quirk: Louisiana reports their long-term care data cumulatively, meaning we’re working with total numbers. It seems easy enough to assume we’d be able to track rises in this cumulative count, but unfortunately, it’s more complicated than that. Occasionally, we’ve found that a facility in the state will report cumulative COVID-19 case and death numbers one week, but the next week, those numbers might be gone, or the facility might be marked as “closed,” “not reported,” or “data pending.”

These disappearances from the dataset have made it difficult for us to get a full count for Louisiana. Week to week, with some facilities suddenly no longer reporting total counts, it can appear as if Louisiana’s cumulative numbers have dropped. In this week’s report, for example, data is missing for 13 facilities and is replaced by words like “closed.” Because of this, at least 733 resident cases, 109 resident deaths, and 402 staff cases are not included in Louisiana’s cumulative nursing home report this week.

In some of these situations, we’ve decided to carry over previous numbers to more accurately reflect our known counts. We’ve decided to do this especially because a handful of these “closed,” “not reported,” and “data pending” facilities have later reappeared in the count.

New sharable charts from the COVID Racial Data Tracker 

This week, the COVID Racial Data Tracker, a collaboration between The COVID Tracking Project and Boston University’s Center for Antiracist Research, launched shareable charts of per-capita cases and deaths, broken out by the race and ethnicity categories reported. You can now view infection and mortality data (cases and deaths per 100,000) for each state, as well as for the United States overall. This analysis is significant because it allows us to clearly see the comparisons between different racial and ethnic groups. We already knew that Black, Indigenous, Latinx, and other communities of color are being disproportionately affected by COVID-19. This simple snapshot makes evaluating the known disparities more accessible than ever before.

As always, you can download our entire race and ethnicity dataset and view our dashboard at the COVID Racial Data Tracker.

To keep up to date on our work, follow us on Twitter and join our low-frequency email list.