Risk Factors for Hospitalization, Mechanical Ventilation, or Death Among 10 131 US Veterans With SARS-CoV-2 Infection

Data Source and Study Population

The VA supports the largest integrated national health care system in the United States and provides care for more than 6 million veterans annually. The VA uses a single, national comprehensive electronic health care information network. We derived data from the VA’s Corporate Data Warehouse, a data repository of electronic medical records, developed by the VA Informatics and Computing Infrastructure (VINCI) to facilitate research. To support COVID-19 research, VINCI analysts created and are continually updating the COVID-19 Shared Data Resource,²⁵ which includes analytic variables extracted from the Corporate Data Warehouse for all VA enrollees tested for SARS-CoV-2. Using this resource, we identified all VA enrollees (N = 88 747) who had nasopharyngeal swabs tested for SARS-CoV-2 nucleic acid by polymerase chain reaction in inpatient or outpatient VA facilities (including VA nursing homes) between February 28 and May 14, 2020, for any indication, excluding those who were VA employees. Most tests were performed in VA laboratories using the US Food and Drug Administration–approved RealTime (Abbott Laboratories) or Xpert-Xpress (Cepheid) SARS-CoV-2 assays; some were sent to commercial or state public health laboratories, especially during the early days of the pandemic. Cohort members were followed up through June 22, 2020, for study outcomes allowing for a minimum follow-up of 39 days. This study was approved by the institutional review board of the Veterans Affairs Puget Sound Healthcare System, which granted a waiver of informed consent because this was a retrospective cohort study based on an existing database. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Definition of Positive or Negative SARS-CoV-2 Status and Index Date

Patients were defined as positive for SARS-CoV-2 if they had at least 1 positive polymerase chain reaction test (n = 10 131 [11.4%]) during the ascertainment period. Patients were defined as negative for SARS-CoV-2 if all polymerase chain reaction tests were negative (n = 78 616 [88.6%]). Final adjudication of SARS-CoV-2 status was performed by the VA National Surveillance Tool, which is intended to be the single, authoritative data source for the determination of positive and negative cases within the Veterans Health Administration. The index date for all analyses was defined as the date of the earliest positive test (for patients with SARS-CoV-2) or the date of the earliest negative test (for patients with no SARS-CoV-2), unless the patient had been admitted to a VA hospital during the preceding 15 days, in which case the date of admission served as the index date.

Adverse Outcomes

We determined the following 3 outcomes: (1) hospitalization at the index date or within 15 days of the index date, (2) mechanical ventilation at the index date or within 60 days, and (3) all-cause mortality at any time after the index date. Deaths that occurred both in and out of the VA are comprehensively captured in the Corporate Data Warehouse through a variety of sources, including VA inpatient files, VA Beneficiary Identification and Records Locator System, Social Security Administration death files, and the Department of Defense.²⁶ Episodes of mechanical ventilation and hospitalization that occurred outside the VA were captured only if the VA paid for these hospitalizations at non-VA facilities under the VA Community Care program.

Baseline Characteristics Evaluated for Associations With Adverse Outcomes

We included characteristics that were evaluated in prior studies or that we hypothesized might be associated with adverse outcomes. Baseline sociodemographic characteristics included age, sex, race, ethnicity, body mass index (calculated as weight in kilograms divided by height in meters squared), urban vs rural residence (based on zip codes), and regional per capita COVID-19–related mortality, operationalized as the number of COVID-19–related deaths per million residents in each participant’s state of residence as of June 11, 2020 (Table 1).²⁷

Comorbid conditions were extracted by VINCI analysts based on International Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) recorded in VA electronic health records during the 2-year period on or before the index date.²⁵ We used the Charlson Comorbidity Index (CCI) to estimate the overall burden of comorbidity (Table 2).

We also included documented symptoms thought to be related to SARS-CoV-2, identified by VINCI analysts based on a combination of natural language processing of text notes in patients’ electronic medical records and searching for relevant ICD-10 codes,²⁵ occurring on or within 30 days prior to the index date (Table 3). We do not report associations with loss of smell or taste, given that these symptoms were not widely recognized during the ascertainment period and thus rarely reported.

We analyzed 13 routinely available laboratory blood tests (Table 4). For each test, we extracted the value closest to the index date, on or within 10 days before the index date, or, if absent, within 5 days after the index date (2539 of 2905 [87.4%] were performed within 2 days of the index date).

Statistical Analysis

Using the Kaplan-Meier method, we calculated 30-day hospitalization, mechanical ventilation, and mortality rates from the index date through June 22, 2020. Participants who did not experience the outcome of interest were censored at the end of follow-up.

All analyses were stratified by the VA medical center where patients were tested for SARS-CoV-2. We used Cox proportional hazards models to compare patients with and without SARS-CoV-2 with respect to risk of adverse outcomes. We also used Cox proportional hazards models to identify independent risk factors for each outcome among patients with SARS-CoV-2, adjusting for sociodemographic characteristics, comorbid conditions, and presenting symptoms, as listed in Table 1, Table 2, and Table 3. Laboratory tests were not included in multivariable adjustment due to concerns about overadjustment, given that these were felt to be part of the causal pathway of the disease rather than predisposing risk factors. Laboratory tests were categorized based on quintiles (ie, ≤25th, >25th to 50th, >50th to 75th, >75th to 90th, and >90th percentiles), with an additional category for missing tests, which were relatively rare. In secondary analyses, we used competing risks analysis for the outcomes of hospitalization or ventilation to account for the competing risk of death.

Multivariable population attributable fractions (PAFs) for each major risk factor were estimated by finding the mean over randomly selected permutations of the PAF when other risk factors were sequentially removed from the model. The number of permutations was sufficient to approximate the true mean to within 0.1%. Confidence intervals were calculated using Monte Carlo simulation (500 iterations over 5000 samples).

Analysis was conducted in Stata MP version 15 (StataCorp), R 64-bit version 3.6.1 (R Project for Statistical Computing), with the averisk package version 1.0.3. Statistical significance was set at P < .05, and all tests were 2-tailed.

Sociodemographic Characteristics and Adverse Outcomes in Patients Who Tested Positive for SARS-CoV-2

Veterans who tested positive for SARS-CoV-2 had a mean (SD) age of 63.6 (16.2) years; 9221 (91.0%) were men, 944 (9.3%) were Hispanic individuals, 5022 (49.6%) were White individuals, and 4215 (41.6%) were Black individuals (Table 1). They were more commonly from urban rather than rural areas (7714 [76.1%] vs 2412 [23.8%]) and had a high prevalence of obesity (4542 [44.8%]). They originated from all 50 US states and Puerto Rico, with the greatest number from New York (1555 [15.3%]), New Jersey (757 [7.5%]), Louisiana (598 [5.9%]), and Pennsylvania (563 [5.6%]) (Figure, A; eTable 3 in the Supplement).

Increasing age was the characteristic most strongly associated with risk of hospitalization, mechanical ventilation, and death. Compared with patients younger than 50 years of age (30-day mortality, 0.4%), those aged 50 to 64 years (30-day mortality, 4.1%; aHR, 9.27; 95% CI, 4.51-19.08), 65 to 79 years (30-day mortality, 13.8%; aHR, 27.47; 95% CI, 13.48-55.99), and 80 years and older (30-day mortality, 29.7%; aHR, 60.80; 95% CI, 29.67-124.61) had progressively higher mortality (Figure, C). Compared with White patients, Black patients were more likely to be hospitalized (aHR, 1.13; 95% CI, 1.04-1.23) and to receive mechanical ventilation (aHR, 1.52; 95% CI, 1.25-1.85) but no more likely to die (aHR, 1.04; 95% CI, 0.88-1.21). Compared with women, men were likely to be hospitalized (aHR, 1.22; 95% CI, 1.04-1.42) or to receive mechanical ventilation (aHR, 2.07; 95% CI, 1.30-3.32), but the association of male sex with mortality did not reach statistical significance (aHR, 1.38; 95% CI, 0.93-2.06), likely reflecting the small number of women in the sample. Areas with high regional COVID-19 disease burden were associated with increased risk of death (eg, ≥700 vs <130 deaths per 1 million residents: aHR, 1.21; 95% CI, 1.02-1.45). Hispanic ethnicity (mortality: aHR, 1.03; 95% CI, 0.79-1.35), having overweight (mortality, body mass index 30.0-34.9 vs 18.5-24.9: aHR, 0.90; 95% CI, 0.77-1.06) or obesity (mortality, body mass index ≥35 vs 18.5-24.9: aHR, 0.97; 95% CI, 0.77-1.21), and urban residence (mortality: aHR, 0.92; 95% CI, 0.80-1.07) were also not associated with increased risk of adverse outcomes.

Comorbid Conditions and Adverse Outcomes in Patients Who Tested Positive for SARS-CoV-2

Veterans who tested positive for SARS-CoV-2 had a high overall burden of comorbidity (Table 2), with less than one-third having no coexisting comorbid conditions (3139 [31.0%]). A higher CCI score was strongly associated with increasing risk of hospitalization (eg, ≥5 vs 0: aHR, 1.82; 95% CI, 1.61-2.05), mechanical ventilation (eg, ≥5 vs 0: aHR, 2.15; 95% CI, 1.61-2.87), and death (eg, ≥5 vs 0: aHR, 1.93; 95% CI, 1.54-2.42). Comorbid conditions that were significantly associated with hospitalization included diabetes (aHR, 1.17; 95% CI, 1.08-1.26), hypertension (aHR, 1.15; 95% CI, 1.05-1.26), chronic kidney disease (aHR, 1.21; 95% CI, 1.11-1.32), cirrhosis (aHR, 1.27; 95% CI, 1.08-1.49), and alcohol dependence (aHR, 1.24; 95% CI, 1.11-1.39). Comorbid conditions that were significantly associated with mechanical ventilation included diabetes (aHR, 1.40; 95% CI, 1.18-1.67), hypertension (aHR, 1.30; 95% CI, 1.03-1.64), obstructive sleep apnea (aHR, 1.22; 95% CI, 1.01-1.46), and obesity hypoventilation (aHR, 1.99; 95% CI, 1.19-3.31). Congestive heart failure (aHR, 1.30; 95% CI, 1.10-1.54), chronic kidney disease (aHR, 1.25; 95% CI, 1.08-1.45), and cirrhosis (aHR, 1.55; 95% CI, 1.16-2.07) were the only comorbid conditions significantly associated with mortality. Chronic obstructive pulmonary disease (aHR, 1.02; 95% CI, 0.88-1.19), hypertension (aHR, 0.95; 95% CI, 0.81-1.12), and smoking (eg, current vs never: aHR, 0.87; 95% CI, 0.67-1.13) were not associated with mortality.

Documented Symptoms and Adverse Outcomes in Patients Who Tested Positive for SARS-CoV-2

The most common documented symptoms included fever (4187 [41.3%]), cough (2620 [25.9%]), and dyspnea (1907 [18.8%]) (Table 3). Symptoms that were significantly associated with hospitalization included fever (aHR, 1.91; 95% CI, 1.78-2.06), dyspnea (aHR, 2.18; 95% CI, 2.02-2.36), nausea (aHR, 1.43; 95% CI, 1.23-1.67), diarrhea (aHR, 1.29; 95% CI, 1.14-1.46), abdominal pain (aHR, 1.39; 95% CI, 1.19-1.63), and fatigue (aHR, 1.32; 95% CI, 1.20-1.46). Symptoms that were significantly associated with mechanical ventilation included fever (aHR, 2.31; 95% CI, 1.95-2.75), dyspnea (aHR, 2.95; 95% CI, 2.49-3.49), nausea (aHR 1.56; 95% CI, 1.11-2.19), and diarrhea (aHR, 1.57; 95% CI, 1.21-2.02). Only fever (aHR, 1.51; 95% CI, 1.32-1.72) and dyspnea (aHR, 1.78; 95% CI, 1.53-2.07) were significantly associated with mortality.

Laboratory Test Results and Adverse Outcomes Among Patients Who Tested Positive for SARS-CoV-2

Associations of laboratory tests with outcomes were only determined among 2905 hospitalized patients because they are not routinely ascertained in nonhospitalized patients. Many laboratory test results were associated with mechanical ventilation and mortality in a dose-response manner, including elevated creatinine (>3.80 mg/dL vs ≤0.98 mg/dL [to convert to millimoles per liter, multiply by 88.4], mechanical ventilation: aHR, 3.30; 95% CI, 2.25-4.84; mortality: aHR, 3.79; 95% CI, 2.62-5.48), elevated serum aspartate aminotransferase (>89 U/L vs ≤25 U/L [to convert to microkatals per liter, multiply by 0.0167], mechanical ventilation: aHR, 2.92; 95% CI, 2.13-4.02; mortality: aHR, 3.00; 95% CI, 2.21-4.07), elevated neutrophil to lymphocyte ratio (>12.70 vs ≤2.71, mechanical ventilation: aHR, 2.84; 95% CI, 2.06-3.92; mortality: aHR, 2.88; 95% CI, 2.12-3.91), elevated total white blood cell count (>11 200/μL vs ≤4770/μL [to convert to ×10⁹, multiply by 0.001], mechanical ventilation: aHR, 2.34; 95% CI, 1.74-3.14; mortality: aHR, 2.16; 95% CI, 1.62-2.87), elevated neutrophil count (>10 140/μL vs ≤3180/μL [to convert to ×10⁹, multiply by 0.001], mechanical ventilation: aHR, 2.65; 95% CI, 1.90-3.69; mortality, aHR, 2.03; 95% CI, 1.47-2.80), reduced lymphocyte count (≤500/μL vs >1400/μL [to convert to ×10⁹, multiply by 0.001], mechanical ventilation: aHR, 1.98; 95% CI, 1.44-2.73; mortality: aHR, 2.00; 95% CI, 1.46-2.74), reduced albumin (>3.9 g/dL vs ≤2.7 g/dL [to convert to grams per liter, multiply by 10.0], mechanical ventilation: aHR, 1.90; 95% CI, 1.36-2.67; mortality: aHR, 2.05; 95% CI, 1.46-2.88), and elevated alanine aminotransferase (≤18 U/L vs >68 U/L [to convert to microkatals per liter, multiply by 0.0167], mechanical ventilation: aHR, 1.74; 95% CI, 1.26-2.41; mortality, aHR, 1.86; 95% CI, 1.35-2.57) (Table 4), but not serum bilirubin, platelet count, hemoglobin, and international normalized ratio (eTable 4 in the Supplement).

PAFs of Major Risk Factors for 30-Day Mortality

Most deaths (63.4%) were associated with older age groups relative to the reference group (ie, aged 18-49 years): 6.2% (95% CI, 6.1%-6.3%) were associated with age 50 to 64 years, 28.9% (95% CI, 20.9%-36.6%) with age 65 to 79 years, and 28.3% (95% CI, 20.1%-30.8%) with age of 80 years or older (Figure, D). Male sex (relative to female sex) contributed 12.3% (95% CI, 5.8%-19.1%). Comorbidity burden contributed 2.0% (95% CI, 0.1%-4.0%) for CCI score of 1 or 2, 2.6% (95% CI, 1.9%-5.8%) for CCI score of 3 or 4, and 6.5% (95% CI, 6.3%-6.6%) for CCI score of 5 or greater. Finally, fever contributed 5.0% (95% CI, 3.5%-6.8%) and dyspnea, 4.0% (95% CI, 2.6%-5.2%), with negligible contributions from other risk factors.

Limitations

This study has limitations. Our results in the predominantly male veteran population may not be generalizable to other populations and groups, especially women. We used ICD-10 codes for the determination of comorbid conditions. However, most ICD-10–based definitions have been widely used and validated in VA studies. Novel natural language processing plus ICD-10 codes were used for the definition of SARS-CoV-2 symptoms, although the performance characteristics of these definitions are not yet known. We captured deaths that occurred both within and outside the VA; however, hospitalizations or mechanical ventilations that occurred outside the VA and were not paid for by the VA were not captured. Our results are limited to those patients who were tested within the VA system. Therefore, our results likely reflect institutional policies and practices related to testing. Strengths of our study include its national scope, large number of patients, relatively long follow-up for a range of outcomes, and analysis of many potential risk factors.