Correlation Between COVID-19 Cases and Deaths in Four Texas Counties

Abstract

Rationale: Predicting deaths from COVID-19 in the near-term has important public health implications. National models may not be applicable at the county level, where limited test availability and/or delays in test results may alter the relationship between COVID-19 diagnoses and deaths. Methods: Publicly available data for daily new COVID-19 cases and deaths from March 4th, 2020 to December 1st, 2020 in Dallas County was obtained from the Texas Department of State Health Services website on December 17th, 2020. COVID-19 cases were reported by local health departments based on the date of test results, while deaths were reported based on death certificates. Due to the lag in case and death reporting, the last two weeks prior to the date of download were excluded. A linear regression was performed using the 7-day rolling average of newly reported cases vs the 7-day rolling average of new deaths utilizing different lag periods. The lag period resulting in the highest R2 value was identified. A similar analysis was subsequently performed in three other Texas counties. Results: Dallas County, which has a population of 2.636 million, had 114,981 confirmed COVID-19 cases and 1708 COVID-19 related deaths over the study period. As shown in Figure 1A, The maximum R2 value was observed at a lag period of 10 days (R2 = 0.8158, p < 0.001). Spikes in cases were seen in July and late November, with deaths following shortly after (Figure 1B). Similar results were seen in Tarrant and Bexar counties, with a maximum R2 value occurring at a lag period of 12 and 7 days (R2 = 0.7323, R2 = 0.7800), respectively. However, Harris County had a maximum R2 value at a lag of only 2 days (R2 = 0.7324). Discussion: Potential contributors to the lag between diagnosis and death include the disease process itself as well as county specific delays in testing and/or testing reporting. In particular, in locations with large surges, cases may overwhelm testing capabilities such that mean case count is under reported, and more cases are identified late in the disease process. Conclusions: In all four counties, peaks in deaths from COVID-19 closely followed peaks in reported cases. In three of four counties, the lag was 7-12 days, consistent with the expected lag between diagnosis and death. In Harris county however, the lag was only 2 days, supporting the idea that national models may not be applicable at a county level.