Completely and Utterly Unusable: Texas Testing Rates

COVID-19 Texas Testing problems

What a week!  After our team broke this story regarding anomalies with testing numbers coming out of the Texas DSHS on Sunday 8/9, with additional tweet series every day through Wednesday, the story began to gain traction with major news outlets. A quick recap:

  • It started with a paragraph via Bloomberg on Wednesday, followed by a full blown story reporting: Texas officials have launched an investigation into why statewide Covid-19 data showed a record surge in positive-test rates even as hospital admissions and other metrics indicate the virus’s spread is slowing.”
  • On Friday, The COVID Tracking Project picked up our story and did very comprehensive analysis concluding: “Right now, the [testing and positivity rate] data reported by Texas DSHS—both statewide and at the county level—is unusable for [the purpose of assessing the trajectory of the virus].”
  • Also on Friday, The Texas Tribune interviewed officials at the DSHS and reported “The state’s maximum capacity for processing electronic lab reports was 48,000 a day, but at the height of the pandemic in July, it frequently received more than that, “So we started to have a backlog,” Anton said. When DSHS upgraded its lab reporting system on Aug. 1, giving it the capacity to process more than 100,000 electronically-submitted test results per day, some of those backlogged tests also were counted for the first time”

So what is the issue with testing and why is positivity rate currently not usable in the State of Texas?

To summarize the main issue – it’s a category of counted tests called “Pending Assignments”

The backstory:

  • DSHS, for months, was beyond their capabilities to process incoming tests.
  • So they created a category, basically a “county,” called Pending Assignments, into which they filed tests they were unable to process that day.   This category’s count can be seen daily in 2 places:
    • Visible on the DSHS Dashboard, the “COVID-19 Test and Hospital Data” Section, Under the “Total Tests by County” tab
  • This category had grown to 1.1 Million tests by August 1.
  • These tests had been “counted” into the States reported number, which can be seen on Row 259 of same spreadsheet cited above, and on the Dashboard, under both “Testing Data” and “Total Tests by County.”
  • These Tests were counted into the State Totals.   However, they were not evaluated for Positive or Negative, or whether they were antibody or Antigen tests, and obviously weren’t assigned to their County. 

Counting these “Pending Assignment” Tests (PA) into the State Total had 2 big effects:

  1. While the Pending Assignment category was increasing in number, not counting for Positive or Negative had the effect of lowering the reported positivity rate in June and July especially.
  2. And when the Pending Assignment category started decreasing in number, a percentage of these tests were Antibody or Antigen tests, and had to be deducted out of the “Testing Data” and “Total Tests by County” because they were not actually “Molecular (PCR) Tests.”

The work to reduce the PA Backlog, starting August 1, had 3 big effects:

  • Effect #1 – Not enough bandwidth to do BOTH PA Backlog reduction and August Test processing
    • It is fully apparent from the correlation between reduction in “Pending Assignment” (PA) and Daily Reported testing, that the DSHS lacked sufficient bandwidth to focus on both reducing the backlog AND reporting new August Tests.  They had to pick what to focus on.  From August 1 to August 13, the chose PA.    
    • Figure 1 below shows as the Backlog went down starting August 1, so did Reported Daily Testing. 
    • On August 14, the afternoon after the Governor’s Investigation was announced, we saw a record number of tests reported, and the Backlog go back up.  
    • 4 days later, the backlog has basically remained flat, while Reported Daily Testing is almost back up to its previous levels. 
    • Also, since the announcement of the Governor’s investigation, we have seen on the 2 days where the Reported testing was generally low, PA went down (that day’s work focused on PA).  On the 2 days where Reported Testing was through the roof, PA went up (that day’s work focused on new Tests).
    • The reduction in the orange line in Figure 1, impacted the bottom half of the positivity rate equation.

Figure 1

  • Effect #2 – Reductions to Total Texas Test Counts due to removing Antigen and Antibody tests
    • The other effect previously mentioned is that a percentage of these counted but unprocessed tests, likely in the mid to high single digit percentages, are actually Antigen and Antibody Tests. 
    • Therefore, because these aren’t molecular tests, they must be deducted from the Texas total Tests reported, even though they were added to/counted weeks and months ago. 
    • Those reductions also hurt the bottom half of the Positivity Rate equation

The net effect of all this is that Positivity Rate was probably higher than reported in June and July, while in August much lower than the unsupportable 24% that was being reported as recently as August 13.  As of August 16, that Positivity Rate has declined below 12%.  These unnatural fluctuations of both reported testing and positivity rate can be seen in Figures 2 & 3 below.

Figure 2

Figure 3

The bottom line is that Pending Assignments – currently 478,325 of them – and the work on this backlog, is impacting the reported Positivity Rate in Texas.  Both on the case side and the test side of the equation.  It is currently an unusable statistic.  Until the backlog goes to zero, and then at least 7 days after that point, will a 7-day Positivity Rate have any relevance to assessing the trajectory of the virus.  A more representative picture of the state’s trajectory is likely something like Figure 4 – an early July peak for cases, and mid to late July for hospitalizations. 

Figure 4

Many thanks to @JoeWo2020 for pointing me to the infamous line 258, and for all his hard work and analysis throughout!

Also want to thanks the those that provided additional assistance and reporting on this story @dpconnally @NahasNewman, @wyattsheepie @jcarrollhtx; @n_pat_kelly @PeterJ_Walker @CohoKelly