For the past five weeks, new cases of COVID-19 have declined in the United States, according to official data reported by states and territories. This is also our fourth straight week of seeing fewer people hospitalized with COVID-19. Taken together—even assuming that many cases still go undetected—these two metrics suggest that far fewer people are being infected with COVID-19 now than at the second surge’s mid-July peak, when more than 450,000 Americans were diagnosed with the disease. And in unquestionably good news, reported COVID-19 deaths fell for the third week in a row.
This week, the number of people who received positive test results is below 300,000, the lowest weekly case number we’ve seen in the past nine weeks. Unfortunately, the number of total tests reported by states is also down again this week. This means that some of the drop in cases we’ve seen in August is likely the result of decreased testing: with fewer tests being completed, more infections will go undiagnosed. Without the hospitalization data—almost all of which is reported by hospitals and hospital associations, not by the public health authorities that report tests, cases, and deaths—we would have a much harder time understanding whether cases really were falling sharply. But can we trust the hospital numbers? We think so, with some caveats we’ll get into below.
Where are the tests going?
Testing has dropped most sharply in the South and West, the two regions hit hardest by the second surge of COVID-19 cases. In the Northeast, the epicenter of spring outbreaks, testing has leveled off, but at a higher per-capita average rate than the South or West ever reached. Testing in the Midwest has also plateaued.
The causes of our national testing drop remain unclear: Are fewer people seeking tests because fewer people feel sick? Have long test-turnaround times discouraged people from pursuing testing? Are restrictive testing criteria intended to conserve limited numbers of tests reducing overall test numbers? Is the rise of antigen testing playing a part in the decline of PCR tests? All of these factors likely play a role in testing levels, but it’s impossible to tell which ones are most influential in each region.
What we do know is that when the Northeast emerged from the devastating outbreaks of the early spring, regional testing continued to rise sharply. Months later, the Northeast region still has the highest per-capita testing—and the lowest new case numbers—in the United States. The region’s sustained and strong testing performance is no accident. Testing demand is in part driven by federal, state, and local public health communication, and by coordinated efforts to bring testing to the places where it’s needed most. In the South and West, however, the opposite pattern holds, with testing dropping in lockstep with new cases. We don’t yet know what this reduced visibility will mean for these regions as we move toward fall, but the reduction in testing does mean we need to rely heavily on hospitalization numbers as a semi-independent source of verification.
About that hospital data
Back in July, we wrote about the problems with state-reported hospital data that emerged when hospitals across the United States were given only days to adapt to a new reporting process. We also wrote about the subsequent gaps in the hospital data reported by the states and the Department of Health and Human Services. In the past ten days, we have seen hospital data reported by states appear to stabilize, with none of the major gaps in reporting that characterized the first month after the new reporting process took effect.
An internal analysis from The COVID Tracking Project comparing state-reported hospital data to the data published by HHS also suggests that although the federally published numbers are usually higher than state-reported numbers, the overall trends in hospitalizations are largely well matched between the two datasets. We will publish a more complete version of this analysis soon, but our initial analysis suggests that the trends in state-reported hospital data are once again a useful measure for checking the trends we see in state-reported cases—and for helping us understand the impact of COVID-19 on our healthcare systems.. Those trends do indicate that far fewer people in the South and West are hospitalized with COVID-19 now than a month ago.
After the whipsaw changes to hospital data last month, we’ve seen more and more readers express doubt in the validity of all state-reported data. That happens either because they’re concerned that federally imposed reporting changes have allowed agencies to manipulate the numbers, or because they believe that state governments over-report or under-report their data to suit their political aims. For these reasons, we thought it might be useful to explain a bit more about where the data we compile from states originates, and what processes it goes through before arriving in our dataset.
Where the data comes from
As we discuss in our data documentation, we compile data almost exclusively from official state and territorial sources. (The exceptions include examples like Hawaii’s ICU and ventilator data is routinely published on the Lieutenant Governor’s Instagram account. We corresponded with the Lieutenant Governor’s office to verify the account’s information before adopting this as our source.)
But before they get to the official state and territorial dashboards and databases, where do all these numbers come from? To dramatically simplify the complexity of the 56 datasets we compile from states and territories, here’s what we know:
Laboratories that process COVID-19 tests report their results data to, variously, local, county, and state/territorial public health authorities—and sometimes directly to the Department of Health and Human Services too. Some of the data they report includes detailed demographic information and some does not.
Hospitals that treat COVID-19 patients report data to, variously, their state/territory, their state hospital association, and the Department of Health and Human Services. Some of the data they report includes detailed demographic information and some does not.
The governments of US states and territories decide what information that they receive from hospitals and laboratories should be made public on their dashboards and in their public databases. Many of these public dashboards include the ability to view data by county or ZIP code.
Many US counties and cities also publish COVID-19 data they receive from local health departments. This information is collected by USA Facts, a non-partisan data organization, and this county-level database is used by the Centers for Disease Control and Prevention to provide county-level statistics for cases and deaths.
Because COVID-19 data in the United States is generated and reported in this highly distributed way, we believe that it would be quite difficult for states and territories to directly manipulate counts for tests, cases, and deaths without creating noticeable discrepancies with county and local datasets.
That said, we do see widely varying data definitions across the 56 jurisdictions we track, and some states have elected to publish much less data than others. Some states, for example, still do not publish probable cases of COVID-19, and probable COVID-19 deaths, despite CDC guidance to do so. Others only publish hospitalization numbers for lab-confirmed cases of COVID-19, rather than confirmed and suspected cases, as requested by the federal government. Decisions like these can reduce a state’s apparent number of cases, deaths, and hospitalized COVID-19 patients without actually changing the underlying (non-public) data.
On our newly redesigned website, we have reworked the way we present state and national data to try to reveal more about what data states and territories actually provide. We’ve begun publishing total tests in multiple units (specimens tested vs. people tested vs. testing encounters) where states provide them, and we will soon be adding counts for probable and confirmed cases and deaths where they are available. On the whole, however, we believe that throughout the country, local and state public health officials are working extremely hard to responsibly collect and publish useful data for their residents. Although the results of these efforts are not always precise or completely clear, we also believe that the trends in the data are largely trustworthy, taken in the appropriate context.
More “Weekly Update” posts
A long holiday weekend makes ambiguous testing data even harder to understand, but hospitalizations are dropping, which is good.
Though cases are rising in parts of the Midwest, hospitalizations in the West and South continued trending downward. The Labor Day holiday impacted data reporting lag times both this week and last, obscuring what had been positive trends in September.
After five straight weeks of sharp declines in new cases of COVID-19 in the United States, we’ve leveled off again. Reported tests are up about 5 percent, but are actually dropping in the Midwest, where three states—Iowa, North Dakota, and South Dakota—are driving an uptick in new cases. This week we also launched a new dataset that looks closely at the toll of COVID-19 in nursing homes and other long-term care facilities.