This page last updated on 2021-09-16 11:00:06.
COVID is real and serious. The best path for stopping COVID-19 is:
- only consider evidence-justified measures
- prioritize measures that have the most impact
- encourage voluntary steps at every turn
- use the lightest regulatory touch possible
- focus mainly on protecting us from each other
There are two bad paths:
One is mainly denialism, which trivializes COVID-19. Denialists generally oppose any COVID-prevention measure.
The other is creating panic to justify arbitrary authority. This makes things worse: We shift scarce resources and finite human attention away from things that work, and we implement more costly measures than needed to assure reasonable safety. Also, people stop believing in COVID-prevention when they see how much the panciked cried wolf.
What does “week” mean?
Some plots use a week number. This is the epidemiological week, as defined by the CDC. This page was generated on 2021-09-16, which is day 5 of epidemiological week 37.
Hospital utilization for various areas
Hospitalizations are generally the most accurate way of characterizing COVID spread. Presumably, each patient admitted for COVID is due to actual symptoms, as assessed by a doctor. While case counts (testing) and fatalities can provide supplemental guidance, confounding factors blunt their usefulness and accuracy.
Before I go into more detail, here's a plot for the whole state:
Texas as a whole has never come close to exceeding its total hospital bed count. That said, there are two important things to consider when digesting that statement:
Some areas have had medical systems overwhelmed. Notable examples are South Texas (summer 2020) and El Paso (fall 2020). Also, some hospitals in otherwise sound areas can get overwhelmed, such as smaller hospitals in outlying urban areas. Outside of such temporal situations, COVID hospitalizations occupy only a modest fraction of beds.
Not all beds are appropriate for treating COVID patients. That said, hospitals have much more flexibility in adapting spaces than what is commonly understood.
Impacts of major shifts in COVID-19-related rules or of mass events should be clearly apparent by 12 days after the event. Many events were widely feared to cause COVID surges, but the data does not substantiate that. These include the BLM demonstrations and certain national holidays. Thanksgiving might be different, with its much higher emphasis on indoor activities.
Texas's COVID-19 hospitalization counts are simply any hospitalization where the patient tests positive for COVID-19. Texas does not differentiate if the hospitalization is due to COVID or incidental, such as an asymptomatic COVID patent who is hospitalized for a knee operation.
Some states do differentiate this. North Dakota's stats suggest the vast majority of COVID-19 hospitalizations are due to COVID-19.
Even for the COVID-incidental hospitalizations, the fact that the patent is COVID-positive still means a host of protective measures must be taken, which competes with “due to COVID” patients for resources, like enhanced PPE, negative-pressure ventilation, etc.
My selection of areas is arbitrary, based on my interest. This section is currently a mishmash.
“TSA” below refers to Texas's Trauma Service Areas (more info). TSAs are the finest granularity that DSHS publishes on hospitalization data. Below are separate plots specifically about Dallas. Those plots use data collected directly by Dallas, and they paint a different story for ICU beds than the state data.
The next plots just show COVID-related hospitalizations. Lacking important context, they can be used to cause panic.
This is the same data as prior, but bucketed into epidemiological weeks.
This shows day-to-day percent changes in hospitalizations. When the red line is below 0, COVID-related hospitalizations are declining. That is where we want to be.
Here's the percent of all available hospital beds that have COVID-positive patients. Per Governor Abbott's executive order EA-32, if this exceeds 15% for seven days in a row, that tightens up COVID-prevention measures.
Important: the denominator is the number of beds available. Even if cases are rising, the percent can be driven down simply by adding to total bed capacity.
Here's the most recent 20 days of the data, conveyed a bit differently.
This plot shows the percent of each TSA region's hospital beds that are used by COVID-positive patients. The bars are red if they exceed the governor's 15% limit.
Fatalities by week for Dallas County
Fatalities are an appealing stat because it's hard to fake a death. The accuracy is illusory, however. Deaths are declining even though hospitalizations and cases are rising. That is probably because treatments are improving, hospitals are (were?) less taxed, and other reasons.
That is to be celebrated. However, it does not justify dropping spread-prevention measures. That would overwhelm the medical system, which in turn would cause fatality surges. In other words, a lowered fatality rate depends on a not-overwhelemd medical system.
A note: The most recent fatality counts will always be too low. It takes time for cases to be added to this data. Creating a death certificate and getting that report to the DSHS takes a variable amount of time.
Positive COVID-19 tests by week for various jurisdictions
While testing is cited a lot, it has limitations in its usefulness. Because test stats roughly correspond with hospitalization trends, they might be accurate on a gross level, but that's it. Things you should be aware of:
Testing rates are affected by factors poorly related to spread. They may include test availability, test locations, total test count, evolving perceptions of when to get tested, test-eligibility changes (this mainly happened early on), testing surges before nationwide social events (e.g., Thanksgiving), community outreach, and more.
Some factors might encourage communities with substantially different rates of infection to get tested. For example, early on, uncertainty with the cost of testing may bias testing to communities with more wealth, which could include a higher percent of people who travel, which could include people who came to or from notorious viral hot spots, like NYC early on. Also, in Dallas County, there have been testing-promotion outreaches to specific communities.
Also, all this data is of PCR tests (the slow tests). It does not show antigen tests (the fast tests).
Sometimes you'll read about much higher test counts than that is indicated below. That happens when PCR and antigen tests are combined into one number. This inflates numbers with double-counting. Why? The antigen test is fast but less accurate. The CDC and others still recommend re-confirming positive antigen tests with slower but accurate PCR tests.
Let's suppose someone gets both an antigen and PCR test today, and gets a positive on both. The antigen test's results will likely show in today's or tomorrow's case numbers. Then in a few days, that person's PCR test result will come back and again show in the numbers. That is, one person's case will show in two days' counts. That is double-counting that can be avoided by only using PCR test results.
Dallas County Judge Clay Jenkins uses the inflated, double-counted test numbers. Here's an example tweet from November 25, 2020, where he inflates the case count by 13%:
NEW: Dallas County Reports 1,368 New Positive 2019 Novel Coronavirus (COVID-19) Cases and 6 Deaths
Including 183 Probable Cases pic.twitter.com/yjOGxLJjqc
— Clay Jenkins (@JudgeClayJ) November 25, 2020
The next plot has to use poor data to calculate percent positivity.
Percent positivity is simply a day's positive results (numerator) divided by all tests administered (denominator). The problem is DSHS gives bad data: the number it reports for a given day's case (positive) count is all positive tests reported that day, not the positive results of tests administered on that day.
Example: You took a COVID test on July 5, and your positive result came back on July 10. DSHS should add that positive to the July 5 case counts. Instead, it adds your case to the July 10 counts.
A rolling average help some. The bars in this plot represent the imperfect data, and the red line is the rolling average.