Early in the Pandemic, KPRC's Joel Eisenbaum interviewed me as an example of "Citizen Statisticians" who were following Texas DSHS Covid-19 Data that Texas DSHS. Back then, dashboards were not as thorough, user-friendly, or versatile as they are now. He asked me about the lockdown and what one should look for in the data to decide whether new (more lax or more strict) restrictions are warranted. I know he was looking for an answer in layman's terms to explain the data to his followers. I gave him an answer that showed more mathematical respect for his audience.
Compare baseline data to a new condition (Before vs. After). In other words: Are the trends the same or different? How close are they? This is a t-test; with spreadsheets and accessible data, the calculation is at our fingertips.
This is a common question in any field where statistical evaluations are made: product quality, performance of one treatment vs. another, comparison of one policy against another. And, with all the effort going into the collection of Covid-19 data, why not use it here?
This will also give me a chance to showcase recommended sources of data. Below, I will be performing calculations for four different Trauma Service Areas (TSAs): Q (Houston), R(Galveston), O (Austin), and I(El Paso). The plots come from https://covid-texas.csullender.com . This site aggregates data from the Texas Department of Health Services. Plots for each TSA contain a blue curve for Covid-19 inpatients, an orange curve for Covid-19 ICU patients, and a side-bar that provides the actual value of both metrics at for the date corresponding to the slider position. The plots for a sub-group of TSA's are shown below in Figure 1.
We will use Covid-19 Hospital Patients as a "conservative" alternative measure of cases. It is conservative for two reasons: first, in contrast to cases, it measures the number of people who have the most severe symptoms; and second, it is a trailing indicator. When a person gets a test, their symptoms have recently arisen or are in response to being notified from another "contact" who has tested positive. Being admitted to a hospital and occupying a bed on the Covid-19 floor (or ICU) can be one to two weeks later.
KPRC broke the news on May 21, 2021 that two (asymptotic) Delta Covid-19 cases had been identified in Galveston County. At this time, the CDC did not recognize Delta as on a Variant of Concern. This did not happen until June 20, 2021 when it was found to contribute to an increasing fraction of U.S. cases. By Independence Day, the Medical Establishment, the press, and Government Officials began to ring the Delta alarm. This along with another report based on a Galveston County Church Group Weekend Field Trip with many testing positive for the Delta Variant upon return. On the order of 500 individuals participated in the activity with many unvaccinated.
Where in the state did cases take off first? Could a statistical test have been used to give officials a "numerical heads-up" that changes were just around the corner. To answer these questions, I built a Table with entries that represent the two periods of time for during the month of June for a number of Trauma Service Areas (TSAs). Our "statistical" experiment will look at the first two weeks of June and compare them with the last two weeks.
Figure 2 shows the two sets of highlighted Covid-19 Hospitalized Patients data for TSA-C (Witchita Falls). These cells were inputs for the Spreadsheet t-test Function executed in the
first column. The same operation was set up for the subsequent rows. The t-test uses the data to calculate the probability that the two sets are not the same (i.e. the null hypothesis fails). Typically, a threshold of 5% is used to distinguish a pass (data is the same) from a failure (data is different). As shown, the cells with yellow highlights, F, G, and R show different trends before and after 6/15. The data is plotted below: Figure 3 with Before not = After [F, G, R]
Figure 4 are examples where Before = After [E, Q, and O].
Quite honestly, Figure 4 is a lot easier to accept than Figure 3. Given the noise of the unsmoothed curves, the test results are interesting but would not convince Lina Hidalgo or anyone else that Delta was lurking in the population waiting to spread over the Holiday festivities. Now, based on the situation depicted by Figure 1, the Dawn of the Delta Surge is upon us; hopefully, we can build on our experience of the last twelve months to minimize the number and severity of new infections.
Comments