COVID-19 Symptoms and Deaths Among Healthcare Workers, United States

Shao Lin; Xinlei Deng; Ian Ryan; Kai Zhang; Wangjian Zhang; Ese Oghaghare; DeeDee Bennett Gayle; Benjamin Shaw


Emerging Infectious Diseases. 2022;28(8):1624-1641. 

In This Article


Study Design and Controls

Our study population included all COVID-19 infection cases reported by the CDC. We used a case–control design to compare demographics and symptoms between HCW deaths and HCW nondeaths (control 1). HCW nondeaths were the primary control group, representing the source population and controlling for important confounders, including occupation, education, medical knowledge, and access to medical care. To compare our findings with previous research, we added 2 other reference groups from the US general population: non-HCW deaths (control 2), which is commonly used by other studies, and non-HCW nondeaths (control 3).

Data Acquisition

We obtained data on laboratory-confirmed COVID-19 cases, probable cases, and deaths across the United States from the Restricted Access Dataset operated by the CDC. In January 2020, COVID-19 data collection commenced, and COVID-19 was added to the nationally notifiable condition list; on April 5, 2020, COVID-19 was classified as immediately notifiable, urgent (within 24 hours) (interim-20-ID-01). All states and territories were encouraged to enact laws in their jurisdictions to submit case notifications to CDC. CDC also requested that public health departments report all COVID-19 cases using standardized case report forms and case definitions for laboratory-confirmed or probable cases. This surveillance system includes patient-level data reported by all US territories and states. This study covers the timeframe January 1, 2020–October 12, 2021.

We obtained demographic and medical information for each record in this dataset, including COVID-19 case status (confirmed or probable case), date of first positive specimen collection, and demographics (sex, age group, race, ethnicity, and county and state of residence) (Table 1). We also obtained information on presence of severe COVID-19 clinical indicators and of less severe symptoms (Table 2). CDC suppressed data cells reporting <5 records and uncommon combinations of demographic characteristics (recoded to NA) to prevent releasing personally identifiable data.

Outcomes and Predictors

The health outcomes in this study were COVID-19–related deaths. Among HCW deaths and control groups 1, 2, and 3, a total of 97.8% were confirmed COVID-19 cases, and 2.2% were probable cases. We calculated fatality as the number of COVID-19 deaths divided by all COVID-19 cases in the United States. We used 20 predictors in the analysis, including demographic variables, severe COVID-19 clinical indicators, and less severe reported symptoms.

Statistical Analysis and Confounders

We first compared all 20 predictor variables between HCW deaths and the 3 control groups using χ2 tests. We then developed logistic regression models by regressing fatality against each symptom predictor while controlling for potential confounders, including sex, age group, race, ethnicity, and periods of different SARS-CoV-2 variants and COVID-19 vaccines. We selected these confounders because they were associated with SARS-CoV-2 infection and various symptoms based on the literature and our data. We defined viral variant periods when specific SARS-CoV-2 variants were dominant in the United States:[18] the original variants were dominant until March 20, 2021; the Alpha variant during March 21–May 30, 2021; the Delta variant during May 31–December 10, 2021; and Omicron since December 11, 2021. However, Omicron was not included because its dominance fell outside our study period (January 1, 2020–October 12, 2021). In addition, the first vaccine was given in America on December 14, 2020.[19] To account for these confounders, we included 3 dummy variables representing the periods of different SARS-CoV-2 variants and when vaccinations started in the United States and controlled these variables in each symptom model. Finally, we examined and compared the temporal trends of confirmed cases and deaths among HCWs and the general population. To reduce the instance of false-positive findings due to multiple testing, we conducted sensitivity analyses using the Bonferroni test method (Table 1, Table 2 and Table 3). We accomplished all data cleaning, analysis, and results using R version 3.6.1 (