Abstract
Introduction Dexamethasone decreases mortality in coronavirus disease 2019 (COVID-19) patients on intensive respiratory support (IRS) but is of uncertain benefit if less severely ill. We determined whether early (within 48 h) dexamethasone was associated with mortality in patients hospitalised with COVID-19 not on IRS.
Methods We included patients admitted to Veterans Affairs hospitals between June 7, 2020-May 31, 2021 within 14-days after SARS-CoV-2 positive test. Exclusions included recent prior corticosteroids and IRS within 48 h. We used inverse probability of treatment weights (IPTW) to balance exposed and unexposed groups, and Cox proportional hazards models to determine 90-day all-cause mortality.
Results Of 19 973 total patients (95% men, median age 71, 27% black), 15 404 (77%) were without IRS within 48 h. Of these, 3514/9450 (34%) patients on no oxygen received dexamethasone and 1042 (11%) died; 4472/5954 (75%) patients on low-flow nasal cannula (NC) received dexamethasone and 857 (14%) died. In IPTW stratified models, patients on no oxygen who received dexamethasone experienced 76% increased risk for 90-day mortality (hazard ratio [HR] 1.76, 95% confidence interval [CI] 1.47 to 2.12); there was no association with mortality among patients on NC (HR 1.08, 95% CI 0.86 to 1.36).
Conclusion In patients hospitalised with COVID-19, early initiation of dexamethasone was common and was associated with no mortality benefit among those on no oxygen or NC in the first 48 h; instead, we found evidence of potential harm. These real-world findings do not support the use of early dexamethasone in hospitalised COVID-19 patients without IRS.
Introduction
Corticosteroids have emerged as an effective therapy for critically ill patients with COVID-19. The large United Kingdom RECOVERY trial RCT of corticosteroids in COVID-19 patients demonstrated an overall 2.8% absolute decrease in mortality for patients treated with dexamethasone compared to usual care [1]. When stratified by respiratory support at randomisation, dexamethasone was associated with greater benefit amongst those on invasive mechanical ventilation (IMV) versus supplemental oxygen (inclusive of non-invasive mechanical ventilation [NIV]); dexamethasone was not significantly associated with mortality in those not on oxygen. Dissemination of these and other results led to rapid uptake in use of corticosteroids for COVID-19 patients, particularly those receiving more intensive respiratory support (IRS) such as high-flow nasal cannula (HFNC), NIV and IMV [2–6].
However, whether corticosteroids are beneficial in all patients with COVID-19 remains uncertain. The association between corticosteroids and outcomes among a wider group of patients with COVID-19 – including a larger proportion without IRS than in the RECOVERY trial – has been mixed [7–11]. Variability in the effect of corticosteroids may be due to numerous factors. A recent Cochrane review concluded that systemic corticosteroids “probably reduce all-cause mortality slightly” but that there is an “urgent need for good-quality evidence for specific subgroups of disease severity, for which we propose level of respiratory support at randomisation.” [12]
We determined the association between corticosteroids and 90-day all-cause mortality using real-world clinical data from the Department of Veterans Affairs (VA), the largest integrated healthcare system in the United States. In a racially and geographically diverse, national cohort of hospitalised COVID-19 patients, we first assessed patterns of corticosteroid receipt. As nearly all patients on IRS received corticosteroids, mainly dexamethasone, we focused on those who were without IRS. We used propensity score weighting to account for confounding by indication. We hypothesised that dexamethasone would not be associated with mortality benefit in patients without IRS.
Methods
Study design and population
We conducted an observational cohort study of 27 168 patients admitted to a VA hospital within 14 days after a positive polymerase chain reaction (PCR) or antigen test for SARS-CoV-2 between June 7, 2020 and May 31, 2021 (to allow 90-day follow-up on all) [13, 14]. Before June 7 corticosteroids were mostly initiated after 48 h. Index date was defined as date of presentation, including emergency room and time under observation status if not admitted directly. We determined length of stay by concatenating episodes of care separated by <24 h, with first episode on the index date as day one. Due to changes in COVID-19 incidence and treatment protocols over time, we divided the observation period into seven time phases (table 1). Additional methodological details are in the online Supplement.
Characteristics of patients stratified by highest oxygen support during first 48 h of hospitalisation for COVID-19
Exclusions
Of 27 168 patients, we excluded 7195, yielding a cohort of 19 973 patients (fig. 1 and Supplement). The most common exclusion was length of stay less than 48-hours as these patients had insufficient time to receive dexamethasone, followed by any systemic corticosteroid exposure prior to index date. This was defined as any corticosteroids within 14 days, or receipt of corticosteroids for ≥14 days in the preceding 45 days. For mortality analyses, we further excluded 454 patients because they were at sites where no or all patients received dexamethasone (n=277), received hydroxychloroquine (n=89), or received vasopressors in the first 48 h (n=90), as these patients may have had an alternative indication for corticosteroids.
Derivation of Study Population. Time period of admission from June 7, 2020 through May 31, 2021.
Exposures, outcomes, and covariates
All data came from VA electronic health record (EHR) extracts, which provide directly analyzable demographics, comorbidities, medications, vital signs and laboratory results as well as notes that require text processing.
Dexamethasone exposure
Exposure was defined as at least one dose of oral or parenteral dexamethasone within 48 h after index date as determined from bar code medication administration (BCMA) data. We also determined administration of other systemic corticosteroids (prednisone, prednisolone, methylprednisolone and/or hydrocortisone).
Outcome
The primary outcome was 90-day all-cause mortality, ascertained using inpatient records and VA death registry data to capture deaths outside of hospitalisation.
Respiratory support
We stratified patients by highest level of respiratory support during the initial 48 h of hospitalisation into the following categories: 1) no oxygen support; 2) low-flow oxygen via nasal cannula (NC) that was not identified as a high-flow or other delivery device; 3) other supplemental oxygen/NIV, including face mask, non-rebreather mask, or other delivery not identifiable as low-flow NC or high-flow; 4) high-flow oxygen/HFNC; and 5) IMV. When no evidence of oxygen supplementation was found, patients were classified as without oxygen (category 1). IMV was identified by structured data sources (ICD-10 procedure and Current Procedural Terminology [CPT] codes). Categories 2–4 were assessed from unstructured text notes using natural language processing (NLP), validated with manual chart review to identify key terms indicative of respiratory support (Supplement).
Covariates
We obtained age, race, ethnicity, sex, comorbidities, additional medications, vital signs and laboratory results. We calculated the Charlson Comorbidity Index (CCI) [15] and the Veterans Health Administration COVID-19 (VACO) Index (table 2 and Supplement table 1) [16]. We focused on routinely collected laboratory tests that have been associated with increased mortality in COVID-19 [17] including albumin, liver function, lactate, white blood cell count, and creatinine (table 2 and Supplemental table 1). We selected the worst laboratory, temperature, blood pressure, and pulse oximetry within the initial 48 h. To account for potential effects of co-prescribed medications, we included use of remdesivir and prophylactic anticoagulants within the initial 48 h [14]. Intensive care unit (ICU) admission was determined using VA bedsection codes [14, 18]. As there was generally very little missing data (<5%), an explicit level for missingness was used for selected covariates.
Characteristics of patients without oxygen or on NC after inverse probability of treatment weighting (IPTW) for estimating the average treatment effect in the total population (ATE models)
Statistical analysis
We first compared COVID-19 patients by the five respiratory support categories using summary statistics (table 1). Because nearly all patients on IMV or HFNC received dexamethasone, there was insufficient variability to allow generation of propensity score weights. Category 3, Other/NIV, was heterogenous with respect to respiratory support used and illness severity. For these reasons, as well as the greater clinical equipoise, we restricted our analysis to patients without IRS (specifically, no oxygen or only low-flow NC support).
In those without IRS, we compared mortality by exposure to dexamethasone overall and stratified by NC. To account for confounding by indication, we generated propensity scores for the probability of receiving dexamethasone in the first 48 h using logistic regression. Models included covariates associated with dexamethasone exposure and mortality: comorbidities, laboratory results, vital signs, site utilisation patterns, co-medications and the time phases (table 2 and Supplement table 1). We constructed inverse probability of treatment weights (IPTW) from propensity scores for each patient to create pseudo-populations with balanced distributions of covariates [19]. In our primary analysis, we used average treatment effect (ATE) weights, reflecting the overall population from which the sample was taken. We used stabilised weights and trimmed from analysis the ten patients with the most extreme high and low weights [20]. We calculated standardised mean differences (SMD) between treatment groups and considered 0.2 or less as balanced. Using days since index date as the time scale, we compared differences in survival using weighted Kaplan-Meier (KM) plots [21] and estimated ATE using Cox proportional hazards models to generate hazard ratio (HR) and confidence limits using a robust variance estimator. We included the VACO Index in outcome models to further account for residual confounding [16].
Subgroup and sensitivity analyses
In subgroup analyses, we excluded patients admitted to the ICU within the first 48 h, and restricted to those age 70 and older. In sensitivity analyses, we limited the window between a positive SARS-CoV-2 test to within 24 or 48 h of index date; In addition, we considered exposure to any systemic corticosteroids within the first 48 h. We also evaluated the average treatment effect among the treated (ATT) population that received dexamethasone in weighted Cox proportional hazards models, and constructed unweighted, but multivariable adjusted models for all primary and subgroup analyses (Supplement tables 2 and 3).
IPTW Cox proportional hazards models for 90-day mortality associated with early dexamethasone exposure in patients hospitalised for COVID-19 without IRS
Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, North Carolina, USA) and R version 4.0.4. Statistical significance was defined as p<0.05. Our study was approved by the Institutional Review Boards of VA Puget Sound Health Care System, VA Connecticut Healthcare System and Yale University, all of whom granted waivers of consent. Study findings are reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplement table 4).
Results
Patient characteristics, dexamethasone exposure and respiratory support
Patients hospitalised during the seven phases (n=19 973) were mostly male (95%). Median age was 71 years (interquartile range [IQR] 62–77); 55% were non-Hispanic white, 27% non-Hispanic black, and 9% Hispanic (table 1). Most patients (83%) were admitted within one day after positive SARS-CoV-2 test. More than half overall (60%) received corticosteroids within 48 h, of whom 95% received dexamethasone. Concurrent remdesivir and prophylactic anticoagulants initiated within 48 h of admission were more common in those who received dexamethasone than in those who did not (remdesivir 43% versus 13%; anticoagulants 46% versus 10%, respectively).
When stratified by highest level of respiratory support in the first 48 h of admission, 77% of patients were on either no oxygen (47%) or NC only (30%) (table 1). Dexamethasone was administered to 34% without oxygen, 75% on NC, 69% on other supplemental oxygen/NIV, 91% on HFNC, and 90% on IMV. Use of dexamethasone generally increased over time (fig. 2). Overall, unadjusted 90-day mortality was 17% and varied by respiratory support (fig. 2).
Proportion of patients exposed to dexamethasone (a) and unadjusted Kaplan Meier survival curves for 90-day mortality (b) according to respiratory support level. Note that the RECOVERY trial was halted on June 8, 2020, with press release of results on June 16, 2020. a) No Oxygen. b) Oxygen by NC.
Dexamethasone and mortality in patients without IRS
Amongst patients without IRS, the median duration of inpatient dexamethasone administration was 5 days (IQR 3–8) in patients without oxygen, and 6 days (IQR 4–9) in patients on NC. These were similar to hospital length of stay (table 1). Only 341 (3.6%) and 115 (1.9%) patients, respectively, received only one day of inpatient dexamethasone.
After propensity score weighting, our samples (pseudopopulations) were well-balanced (table 2 and Supplement table 1). Among patients without oxygen, weighted KM curves (fig. 3) show that those who received dexamethasone had higher mortality over 90-days than those who did not, with differences beginning to appear 10-days after index date. In ATE estimates (table 3), patients without oxygen who received dexamethasone had an 76% increased hazard of 90-day mortality (HR 1.76, 95% CI 1.47 to 2.12).
IPTW Kaplan Meier survival curves for 90-day mortality by corticosteroid use among those on no oxygen or NC. ATE: Average Treatment Effect; IPTW: Inverse probability of treatment weighting; NC: nasal cannula.
In patients on NC, 90-day mortality was similar in those who did and did not receive dexamethasone, as shown in weighted KM curves (fig. 3). ATE estimates demonstrated a non-significant 8% increased mortality risk (HR 1.08, 95% CI 0.86 to 1.37) in patients on NC who received dexamethasone. When combining patients on no oxygen or NC, dexamethasone was associated with approximately 60% or more increased mortality risk.
Subgroup and sensitivity analyses
Results were similar in subgroup analyses excluding patients admitted to ICU within the first 48 h and limiting the sample to patients age 70 and older (table 3). Findings were also consistent considering SARS-CoV-2 testing within 24 or 48 h of index date, exposure to all corticosteroids, and when using ATT or multivariable Cox models (Supplement tables 2 and 3). Among patients on NC, HRs were similar using ATT estimates, but dexamethasone was associated with a statistically significant increase in mortality in multivariable Cox models (HR 1.31, 95% CI 1.08–1.60, Supplemental table 3).
Discussion
In this US national cohort of hospitalised patients with COVID-19, dexamethasone use was common and increased over time. Among patients on IRS, 90% received dexamethasone within 48 h of admission. Focusing on patients without IRS, where there is less evidence supporting corticosteroid use, we found that among patients without oxygen in the first 48 h, dexamethasone was administered in 34% and was associated with 76% increased 90-day mortality. Among those on NC in the first 48 h, dexamethasone was administered in 75% and was associated with no mortality benefit. This real-world evidence does not support the use of dexamethasone in hospitalised COVID-19 patients without IRS in the first 48 h.
While we cannot rule out residual confounding, our findings were robust employing several different approaches and in subgroup and sensitivity analyses, including limiting the time window for SARS-CoV-2 test result, exposure to any systemic corticosteroid, restricted to patients over age 70 and excluding those who were admitted to the ICU within 48 h. Results were consistent using ATE, reflecting the overall population from which the sample was taken, and using ATT, reflecting the population who received dexamethasone. They were also consistent controlling for potential confounders such as demographics, phase of the pandemic, site prescribing patterns, comorbidities, vital signs, laboratory values and co-administration of medications including remdesivir [16].
Importantly, patients without IRS in the initial 48 h represent the majority (77%) of COVID-19 admissions in the cohort; thus, our findings have important clinical implications on the potential unintended consequences of widespread dexamethasone adoption for COVID-19 amongst patients who are without IRS. We found that uptake of dexamethasone for COVID-19 patients hospitalised in the VA was rapid after release of the RECOVERY trial results in early June 2020. By mid-July 2020 most facilities had increased the proportion of patients administered dexamethasone to 90% of patients on HFNC or IMV within 48 h of admission, exceeding national estimates from other health systems [6]. However, sites also increased use of dexamethasone for patients with less severe COVID-19, including those not on oxygen or only on NC (fig. 2), suggesting indication creep.
Our results provide real world evidence of practice patterns and extend findings from RECOVERY [1]. We provide clinically actionable evidence demonstrating significantly and substantially increased mortality in hospitalised COVID-19 patients not on oxygen who received early dexamethasone. Moreover, our results inform an area of significant clinical uncertainty, namely the use of dexamethasone in COVID-19 patients with less severe respiratory failure. Clinical guidelines issued by the US National Institutes of Health provide a moderate recommendation for corticosteroids in hospitalised COVID-19 patients “on supplemental oxygen.” [22] While dexamethasone was associated with improved outcomes in patients on oxygen support in RECOVERY, this category included all forms of oxygen, except for IMV, but inclusive of NIV. We further stratified patients by level of oxygen support during the initial 48 h of hospitalisation, addressing a significant knowledge gap [12]. We found a lack of benefit associated with dexamethasone in patients on only low-flow NC within 48 h of admission, suggesting that use of corticosteroids in this population should be re-considered and requires further prospective study.
Even before COVID-19, the impact of corticosteroids has been inconsistent in other causes of pneumonia including influenza, community-acquired pneumonia (CAP), and the original severe acute respiratory syndrome (SARS) [8, 23–26]. The impact of corticosteroids likely depends on multiple factors, including patient age and other characteristics, heterogeneity in host response to infection, etiology of pneumonia, time since onset of infection and presence and severity of acute respiratory distress syndrome (ARDS) [8, 10, 27–31]. While corticosteroids may decrease host inflammatory response, potentially modulating lung injury, they may also have harmful side effects or unintended consequences on adaptive immune responses that may be important to resolution of infection and increase risk of secondary infection.
There remain unanswered questions regarding the use of dexamethasone for patients hospitalised with COVID-19, particularly those without IRS. For some patients, initiation within 48 h of hospitalisation may be too early and could impair viral clearance [32]. While most patients in our cohort had positive SARS-CoV-2 testing within one day of hospitalisation, we do not know how long symptoms preceded seeking medical attention and testing. Corticosteroids may also have a differential effect depending on degree of inflammation [33], but often extensive missing data and selection bias in obtaining tests such as C-reactive protein (CRP) and interleukin-6 (IL-6) make this difficult to explore in real-world data. Further, it is unclear whether the dexamethasone regimen used in RECOVERY is optimal or whether the formulation, dose and duration of corticosteroids should vary by factors such as patient age or severity of COVID-19 [2–4, 32]. While most corticosteroid use in our cohort was dexamethasone, we found consistent results when including all systemic corticosteroids, and also when restricting to individuals over age 70, although other reports have found a potential for increased harm in older persons [31]. It is also unknown whether corticosteroids have a differential effect in breakthrough COVID-19 after vaccination or different variants of SARS-CoV-2.
There are several limitations to our study. First, the study was observational. While we used detailed clinical data that included measures reflecting illness severity and administration of co-medications in a large population well balanced by propensity for treatment, residual confounding for severity of illness could have contributed to greater mortality in those exposed to dexamethasone. Some laboratory results could have occurred after dexamethasone exposure, as both were ascertained within 48 h. However, the impact of dexamethasone on acute laboratory results is likely limited and this approach allowed an equal time window to detect worst results in patients exposed and unexposed to dexamethasone. Although respiratory support algorithms were manually reviewed and validated, some misclassification may have occurred; but substantial separation in Kaplan-Meier curves showing increasing mortality with greater respiratory support, provides face validity. We also cannot rule out alternative indications for dexamethasone beyond COVID-19 in patients not on oxygen or on NC. However, we excluded those on corticosteroids prior to admission and patients on vasopressors within the initial 48 h. Further, we did not calculate dose and only assessed days of inpatient dexamethasone exposure. However, most patients had length of inpatient dexamethasone treatment equal to their hospital length of stay, and very few received only one dose of dexamethasone (<4%). Finally, our cohort consisted predominantly of male Veterans, but had excellent racial and geographic variability.
In summary, we found no evidence of mortality benefit at 90-days associated with early initiation of dexamethasone in patients hospitalised with COVID-19 among those on no oxygen or NC within the first 48 h of admission, and instead found evidence of potential harm. These findings come from a large population with detailed clinical data providing real-world evidence that was consistent using different analytic approaches to control for confounding and remained robust in a variety of subgroup and sensitivity analyses. Given the frequent and continued administration of dexamethasone to a substantial proportion of patients who are not on oxygen or are only on low-flow NC, the real-world evidence presented here highlights the non-beneficial and potentially harmful expansion in use of dexamethasone in hospitalised COVID-19 patients without IRS. Future work should also evaluate dexamethasone and associated outcomes among hospitalised patients with COVID-19 breakthrough infections and different SARS-CoV-2 variants.
Footnotes
Author contributions: KC, JT, MG, BJ, VM, MEO, CTR, MRB, ACJ and KMA conceived (formulated or helped in the evolution) of the study. JT, RD, PRA and CTR curated the data. RD and JT performed the formal analysis. KC, PRA, ACJ and KMA acquired funding. KC, RD, JT, PRA, MEO, CTR, ACJ, and KMA designed the methodology. RD, JT, and SS managed and coordinated the project. KC, JT, PRA, BJ, and KMA contributed to validation. KC, RD, JT, and KMA wrote the first draft of the manuscript. All authors fulfill ICJME criteria for authorship; all authors participated in interpretation of the data; in critically revising the manuscript for important intellectual content; and approved the final version to be published. All authors agree to be accountable for the work and ensure that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. KC and KMA are joint principal investigators. KC, RD, JT, and KMA are guarantors. The corresponding author attests that all listed authors meet authorship criteria. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Financial support: VA/HSR&D C19-20-406(KC/KMA), VA/RR&D 1I0IRX003666-01(KC), MVP000(ACJ), VA/HSR&D 13-457(PRA) Veterans Affairs Rehabilitation Research and Development; Grant: 1I0IRX003666-01; Health Services Research and Development; DOI: http://dx.doi.org/10.13039/100007217; Grant: 13-457, C19-20-406, MVP000.
This article has supplementary material available from erj.ersjournals.com
Data sharing: Owing to US Department of Veterans Affairs (VA) regulations and our ethics agreements, the analytic data sets used for this study are not permitted to leave the VA firewall without a data use agreement. This limitation is consistent with other studies based on VA data. However, VA data are made freely available to researchers with an approved VA study protocol. For more information, please visit https://www.virec.research.va.gov or contact the VA Information Resource Center at VIReC{at}va.gov.
- Received September 21, 2021.
- Accepted November 17, 2021.
- Copyright ©The authors 2021.
This version is distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0. For commercial reproduction rights and permissions contact permissions{at}ersnet.org
References
