Setting:
The study utilized the VA electronic health care databases. The VA provides health care to US Veterans and operates the largest national integrated healthcare system in the United States with 1,255 health care facilities, including 170 VA medical centers and 1,074 outpatient sites located across the United States. Veterans enrolled have access to the Department of Veterans Affairs comprehensive medical benefits package including inpatient hospital care; outpatient services; preventive, primary, and specialty care; prescriptions; mental healthcare; home healthcare; geriatric and extended care; medical equipment; and prosthetics. VA electronic health care databases are update daily.
Cohort:
US Veterans who encountered the VHA between January 01, 2019 and December 31, 2019 were selected (N=5,808,018). Among those alive on March 01, 2020 (N=5,606,309), the COVID-19 group consisted of those with a COVID-19 positive test between March 01, 2020 and March 15, 2021 (n=98,661). We then selected those who were alive on the 30th day after their first positive test (N=181,384). COVID-19 positive patients were separated into 3 groups based on the care received during the 30 days after the positive test: 1) non-hospitalized (N=155,987); 2) hospitalized (N=19,359); 3) received intensive care (N=6038). To generate a comparison group without COVID-19, we selected 4,534,600 participants who did not have a COVID-19 positive test, and randomly assigned every 25 of them the same cohort enrollment date as one participant in the COVID-19 group. We then further selected from the control group who were alive during the first 30 days after the date of enrollment (VHA group n=4,397,509). Participants were followed until May 01, 2021 (Supplementary Figure 2).
Data sources:
Data used in this study was collected from the VA Corporate Data Warehouse (CDW)23-28. The CDW Outpatient Encounters and Inpatient Encounters domains provided information related to diagnoses, procedures and hospitalization records29. The CDW Outpatient Pharmacy domain and CDW Bar Code Medication Administration domain were used to collect prescription information. CDW Patient domain was used to collect demographic information. The CDW Laboratory Results domain was used to collect laboratory test information, and the COVID-19 Shared Data Resource was used to collect COVID-19 tests. The Area Deprivation Index (ADI), a composite measure of income, education, employment, and housing was obtained from the University of Wisconsin30.
Negative outcome controls:
The use of negative controls in observational studies may help detect the presence of both suspected and unsuspected spurious biases; the application of negative controls will test if shared biases in outcome ascertainment, residual confounding, analytic approach, or other latent biases might have influenced the results31,32. Here we followed the approach outlined by Lipsitch and collaborators to test accidental injuries and neoplasms as negative outcome controls31, where based on current knowledge, we would expect no association between COVID-19 infection and these 2 negative outcome controls.
Post-acute sequelae of COVID-19
We examined a set of 33 post-acute COVID-19 outcomes; these outcomes were selected based on prior studies1,33, review of the literature2,3, and the most recent US National Institute of Health workshop on PASC. Outcomes were defined based on ICD10 codes recorded from inpatient or outpatient encounters, medication records, or laboratory tests when appropriate using definitions validated for use with electronic health records33-44. Detailed definitions of the outcomes are presented in Supplementary table 11. Cardiovascular outcomes included acute coronary disease, arrythmias, bradycardia, chest pain, heart failure, myocarditis and tachycardia; coagulation outcomes included thromboembolism; dermatologic outcome included hair loss and skin rash; endocrine outcome included diabetes mellitus, hyperlipidemia and obesity; gastrointestinal outcome included constipation, diarrhea and GERD, general outcome include fatigue; kidney outcome include acute kidney injury and chronic kidney disease; mental health outcome included anxiety, depression, mood disorder, sleep disorder and substance abuse; musculoskeletal outcome included joint pain and muscle weakness; neurologic outcome included headache, memory problems, smell problems and stroke; pulmonary outcome included cough, hypoxemia and shortness of breath. Occurrence of incident clinical manifestation was defined as the occurrence of a manifestation that did not occur within past one year before cohort enrollment. PASC was defined as the presence of at least one incident clinical manifestation in excess of the non-infected controls.
Covariates:
Covariates for analyses included age, race (White, Black, and Other), sex, receipt of long-term care, Area Deprivation Index based on patient addresses and proxies of healthcare utilization such as number of outpatient encounters, number of hospital admissions, number of outpatient prescriptions and number of outpatient serum creatinine measurements in the year before enrollment. We also included comorbidities such as chronic lung disease, cancer, cardiovascular disease, cerebrovascular disease, dementia, diabetes mellitus, hypertension, hyperlipidemia, depression, anxiety, chronic kidney disease, hepatitis C and peripheral artery disease. In addition, covariates included overweight, obesity, smoking status (never, former, and current) and the Charlson comorbidity index were also adjusted for. We also adjusted for US geographic region (West, Mid-west, South and Northeast) where the care was received, and additional health system characteristics including total number of beds, number of COVID-19 tests administered, COVID-19 positivity rate, and average hospital bed occupancy during the week of participant enrollment.
Statistical analyses:
Characteristics of the VHA users without COVID-19, and those with COVID-19 according to care setting of the acute infection (non-hospitalized, hospitalized, and admitted to intensive care) were described.
Excess burden of PASC, defined as having at least one sequela in excess of VHA users without COVID-19, was estimated using Poisson regressions, where burden was defined as the number of incident sequelae occurring during follow-up. The excess burdens of having 1, 2, to 33 PASC at 6 months, as well as the total excess number of PASC, were estimated in the overall cohort and by care setting of the acute infection.
We then estimated the excess burden of incident individual sequela. For each outcome examined, we built a cohort of participants without a history of the outcome. Cox models adjusting for covariates were used to estimate the hazard ratio of each COVID-19 care setting compared to VHA users, and the survival probability for the 4 groups at 6 months. Cause specific hazard models were used where occurrence of death was considered as competing risk during the analyses. Excess burden per 1000 patients at 6 months were computed as the difference in survival probability between each COVID-19 care setting and the VHA users. Burden of outcomes in the overall COVID-19 population was computed as the weighted sum of the burden of the three care settings based on the proportion of COVID-19 patients in each care setting. Analyses were also conducted to estimate the excess burden within subgroups by age, race, sex, and baseline health status. Burden differences between subgroups of age≤60 and >70, Black and White, female and male, and 0 and >3 comorbidity score were then estimated.
All analyses were done using SAS Enterprise Guide version 7.1 (SAS Institute, Cary, NC). Data visualizations were performed in R 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria). The study was approved by the Institutional Review Board of the Department of Veterans Affairs St. Louis Health Care System, St. Louis, MO.