Study population and setting
This retrospective cohort study evaluated post-acute sequelae of laboratory-confirmed COVID-19 (PASC, also termed Long-COVID) among non-hospitalized and hospitalized individuals in the Veteran Health Administration (VHA), a large national healthcare system serving military veterans with 1,255 health facilities across the United States. Individuals were included if they had a health system encounter in 2019, were alive on March 1, 2020, had a positive COVID-19 test between March 1 and November 30, 2020, and survived at least 30 days following their positive test. Follow-up for post-acute sequelae of COVID-19 occurred through January 31, 2021. Individuals with COVID-19 who survived without hospitalization for 30 days following their first positive test (non-hospitalized COVID-19 cases; N=73,435) were selected and individually matched to 70 VHA users without a positive COVID-19 test (N=4,990,835). A second set of analyses compared individuals who were hospitalized from 5 days before to 30 days after their first positive COVID-19 test (hospitalized COVID-19 cases; N=13, 654) to individuals who were hospitalized within the 5 days before and 30 days after a positive test for seasonal influenza between October 1, 2016 and February 29, 2020 (hospitalized influenza cases; N=13,997). The outcomes for both sets of comparisons were the incident risk of death, outpatient admission, 379 diagnoses from ICD-10 codes, 380 medication classes, and 62 laboratory tests (all from beyond 30 days from positive test). A third analysis compared the risk of pre-specified acute COVID-19 outcomes (based on the joint CDC/NIH workshop on PASC) for three mutually exclusive groups (all compared to the same VHA reference group without COVID-19): non-hospitalized COVID-19 patients, hospitalized COVID-19 patients not requiring ICU admission, and COVID-19 patients requiring ICU admission. All analyses were weighted by propensity scores constructed from pre-defined covariates, including age, race, sex, receipt of long-term care, proxies of healthcare utilization, and deprivation index at patients’ residency address, and algorithmically identified covariates from the diagnoses, medications, and laboratory testing values that showed evidence of difference between the comparison groups.
Summary of Main Findings
In the non-hospitalized group, individuals with COVID-19 (median follow-up: 126 days, IQR: 81-203; 88% male; median age 60.7 years) compared to VHA users without COVID-19 (median follow-up 130 days, IQR: 82-205; 90% male; median age 66.7 years) had increased risk of death (hazard ratio (HR): 1.59, 95% CI: 1.46-1.73) and outpatient encounters (HR: 1.20, 95% CI: 1.19-1.21), corresponding to 8.39 (95% CI: 7.09-9.58) excess deaths and 33.22 (95% CI: 30.89-35.58) excess outpatient encounters, each per 1000 patients at 6 months. The most pronounced absolute differences in incident diagnoses were for respiratory signs and symptoms (excess burden per 1000 people at 6 months (EB): 28.5 episodes), hypertension (EB: 15.2), sleep-wake disorders (EB: 14.5), nervous system signs and symptoms (EB: 14.3), musculoskeletal pain (EB: 13.9), malaise and fatigue (EB: 12.6), and disorders of lipid metabolism (EB: 12.3). The COVID-19 group was also more likely to have been prescribed many medications, including bronchodilators, opioid analgesics, and anticoagulants, and had pronounced differences in laboratory values, such as lower hemoglobin, lower hematocrit, and higher hemoglobin A1c.
In the hospitalized group, individuals hospitalized with COVID-19 (median follow-up: 150, IQR: 84-217; 94% male; median age 70.3 years) compared to individuals hospitalized for season influenza (median follow-up: 157, IQR: 87-220; 94% male; median age 70.1 years) also had increased risk of death (HR: 1.51; 95% CI: 1.30-1.76) and outpatient encounters (HR: 1.12; 95% CI: 1.08-1.17), corresponding to 28.79 (95% CI: 19.52-36.85) excess deaths and 6.37 (95% CI: 4.01-9.03) excess outpatient encounters, each per 1000 patients at 6 months. Additionally, compared to hospitalized influenza cases, hospitalized COVID-19 cases had an EB (per 1000 people at 6 months) for several diagnoses, including respiratory failure (EB: 70.4), disorders of lipid metabolism (EB: 43.5), fluid and electrolyte disorders (EB: 40.2), malaise and fatigue (EB: 36.5), as well as increased use of anticoagulants, laxatives, vitamin C, vitamin D, gastric medications, and lower levels follow-up levels of serum albumin, serum total protein, higher platelet counts, and higher serum calcium, among others.
When comparing the outcomes for the three mutually exclusive groups (non-hospitalized COVID-19, hospitalized COVID-19 without ICU admission, COVID-19 with ICU admission) compared to VHA users, many incident diagnoses were at heightened risk in all groups, including acute coronary disease, arrhythmias, acute kidney injury, chronic kidney disease, memory problems, and thromboembolic events. For each of these, the magnitude of risk increased with the severity of the initial clinical course of COVID-19.
This was a very large and comprehensive study, evaluating data on diagnoses, medication use, and laboratory values following laboratory-confirmed COVID-19, ascertained from integrated electronic health records. The large VHA source population provided adequate power to assess the mid-term outcomes of COVID-19 by sub-categories of disease course severity, and also provided a large pool of individuals without COVID-19 diagnoses available for selection as appropriate controls. The authors used robust methods to control for potential confounding, including propensity score weighting of comparison groups and multivariable adjustment. The analytic plan was rigorous, e.g. accounting for competing risks, varying model assumptions in sensitivity analyses, testing negative outcome and exposure controls, and using Bonferroni-adjusted p-value to address spurious results arising from multiple comparisons. COVID-19 was not associated with any of the negative outcome controls–outcomes that were not expected to have a causal association with COVID-19; e.g. neoplasms or accidental injuries, strengthening confidence in the validity of the analyses. Likewise, there were no associations between the negative exposure controls (comparing the influenza cohorts tested in odd and even months), further reducing concern about unmeasured confounding and model misspecification. The study reported the incidence rate of each outcome among those with and without COVID-19, and the difference in the incidence rates between groups in addition to the hazard ratio, providing information not just on the magnitude of the relative hazard but also the extent of the excess burden of each outcome in the study population.
The study population was primarily male (~90% across all groups), and may not reflect long-term outcomes or differences that are more pronounced in women. Further, these results are specific to a population with health insurance and access to care, and it is possible the post-acute outcomes might be worse among other populations impacted by COVID-19. The study evaluated all outcomes separately; therefore, it does not provide information on an overall incidence of PASC or of grouped organ-specific conditions among this population. The study was limited to patients diagnosed with COVID-19 before November 2020, prior to the widespread transmission of COVID-19 variants of concern in the US; it is possible that infection with these variants may lead to different long term sequelae. The authors do not mention missing data, so it is assumed that analyses are restricted to those with complete data and may be biased as a result. Although several methods were used to control for and evaluate confounding, residual confounding due to unmeasured factors is possible. While it is possible that some of the control participants may have had COVID-19 that was undiagnosed or tested outside the VHA system, this misclassification would minimize a true difference between populations, resulting in smaller estimates of the hazard ratio and excess burden.
This is one of the largest studies of post-acute COVID-19 to date, providing data on a range of mid-term outcomes following COVID-19, including among non-hospitalized patients, hospitalized patients, and those requiring care in the ICU, relative to a comparison group of similar veterans without COVID-19 diagnoses during the same time period. The study is also among the first to address questions about how the post-acute outcomes of COVID-19 compare to the syndromes following other viral infections, by directly comparing post-30 day outcomes of those hospitalized with COVID-19 to those hospitalized with influenza.