key: cord-0757047-qf0a9xj1
authors: Ward, Daniel; Gørtz, Sanne; Ernst, Martin Thomsen; Andersen, Nynne Nyboe; Kjær, Susanne K.; Hallas, Jesper; Christensen, Steffen; Christiansen, Christian Fynbo; Israelsen, Simone Bastrup; Benfield, Thomas; Pottegård, Anton; Jess, Tine
title: The effect of immunosuppressants on the prognosis of SARS-CoV-2 infection
date: 2021-09-02
journal: Eur Respir J
DOI: 10.1183/13993003.00769-2021
sha: 3f518221a6bbfd7e55d0577fcefee6467999e2cf
doc_id: 757047
cord_uid: qf0a9xj1

BACKGROUND: Immunosuppression may worsen SARS-CoV-2 infection. We conducted a nationwide cohort study of the effect of exposure to immunosuppressants on the prognosis of SARS-CoV-2 infection in Denmark. METHODS: We identified all SARS-CoV-2 test-positive patients from February to October 2020 and linked health care data from nationwide registers, including prescriptions for the exposure, immunosuppressant drugs. We estimated relative risks of hospital admission, intensive care unit (ICU) admission, and death (each studied independently up to 30 days from testing) with a log linear binomial regression adjusted for confounders using a propensity score-based matching weights model. RESULTS: A composite immunosuppressant exposure was associated with a significantly increased risk of death (adjusted relative risk 1·56 [95% confidence interval 1.10–2.22]). The increased risk of death was mainly driven by exposure to systemic glucocorticoids (aRR 2.38 [95% CI 1.72–3.30]), which were also associated with an increased risk of hospital admission (aRR 1.34 [95% CI 1.10–1.62]), but not ICU admission (aRR 1.76 [95% CI [0.93–3.35]); these risks were greater for high cumulative doses of glucocorticoids than for moderate doses. Exposure to selective immunosuppressants, tumour necrosis factor inhibitors, or interleukin inhibitors, was not associated with an increased risk of hospitalisation, ICU admission, or death, nor was exposure to calcineurin inhibitors, other immunosuppressants, hydroxychloroquine, or chloroquine. CONCLUSIONS: Exposure to glucocorticoids was associated with increased risks of hospital admission and death. Further investigation is needed to determine the optimal management of COVID-19 in patients with pre-morbid glucocorticoid usage, specifically whether these patients require altered doses of glucocorticoids.

Coronavirus disease , caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), manifests with varying clinical severity. 1, 2 An inflammatory response with virus-specific T cells clears the virus and leads to recovery in most patients, however an aberrant inflammatory response can lead to severe disease. 3 Severe cases are predominantly characterised by viral pneumonia and may feature multiorgan inflammatory involvement, including elevated pro-inflammatory cytokines such as interleukins IL-6 and IL-8, and tumour necrosis factor (TNF). [3] [4] [5] Patients receiving immunosuppressant therapies for conditions including inflammatory diseases and solid organ transplantation are susceptible to intercurrent viral and bacterial infections, 6, 7 and although evidence is lacking regarding their effect on COVID-19, expert groups concerned that immunosuppression may worsen the prognosis have advised withholding or reducing immunosuppressants during intercurrent COVID-19. [8] [9] [10] Immunosuppressants differ in their mechanisms of action and may therefore have differing effects on the disease course of COVID-19, and effects may vary with the severity of disease, and timing in the disease course. Certain immunosuppressants may have beneficial effects in COVID-19 by regulating the elevated inflammatory response associated with severe disease. Randomised controlled trials (RCTs) have demonstrated improved survival of COVID-19 patients treated with corticosteroids. 11, 12 A number of clinical trials of biological immunosuppressants including anti-IL-6 agents have been performed without conclusive evidence of improved outcomes, 13 but preliminary reports of a large RCT indicate improved survival in patients treated with tocilizumab. 14 The majority of RCTs and a meta-analysis of chloroquine or hydroxychloroquine to treat COVID-19 did not support efficacy. 15, In addition to efficacy studies of immunosuppressants as treatment for COVID-19, their safety also requires investigation to guide optimal management of comorbid diseases during the pandemic, as the presence of pre-existing immunosuppression may influence the prognosis of intercurrent COVID-19. Currently published studies of patients with COVID-19 receiving immunosuppressants for underlying conditions have been limited by small sample sizes or surveillance bias. [16] [17] [18] [19] We therefore aimed to conduct a nationwide cohort study of the effect of exposure to immunosuppressants on the risk of hospital admission, intensive care unit (ICU) admission, and death among all SARS-CoV-2 test-positive patients in Denmark from February to October 2020.

We conducted a nationwide cohort study using the Danish COVID-19 cohort, 20 based on data from the Danish Microbiology Register, a national register of all test results from all clinical microbiology departments in Denmark. 21 We defined the cohort as all individuals with a positive result for SARS-CoV-2 polymerase chain reaction (PCR) on an oro-or nasopharyngeal swab or lower respiratory tract specimen, from the first detected case on 26 February until 18 October 2020 (30 days before data extraction on 18 November 2020). We used individuals' first positive test date in the Danish Microbiology Register (the index date) and a pseudonymised unique identifier to link individual-level health care data from other Danish national registers. We obtained information on prescription drugs dispensed at retail pharmacies from the Danish National Prescription Register, 22 and information on diagnoses, and medical procedures (including the administration of intravenous drugs) from the Danish National Patient Register, a register of hospital activities. 23 We obtained the date of death from the Danish Register of Causes of Death, if present. 24 

The exposure was immunosuppressants drugs including hydroxychloroquine and chloroquine (immunomodulators which are suspected to alter the immune response in COVID- 19) , and systemic glucocorticoids, which in moderate to high doses can cause immunosuppression (see Appendix Table 1 for drug level ATC codes and procedure codes). The validity of the registration of immunosuppressants in our data sources has not been analysed, but studies have demonstrated a high validity of other procedures codes, such as antineoplastic procedures. 25 The exposure assessment window was 120 days preceding the index date, as packs contained up to 120 tablets, and treatments given more than 120 days before infection are unlikely to cause ongoing immunosuppression. We used a minimum daily dose of systemic glucocorticoids equivalent to 7•5 mg prednisone per day, to exclude doses unlikely to cause significant immunosuppression (Appendix Table 2 ). 26 As the prescribed daily dose is not available in the Danish National Prescription Register, 22 we estimated the daily dose as the sum of the amount of glucocorticoids dispensed to an individual during the exposure assessment window divided by the number of days from the first prescription to the index date. Unexposed patients did not receive any immunosuppressant during the exposure assessment window.

We studied immunosuppressants as a composite exposure in our main analysis. In secondary analyses, seeking to investigate the effect of classes of immunosuppressants, while maintaining sufficient sample size to detect an effect, we broke down immunosuppressants into smaller categories. Biological and targeted immunosuppressants indicated in severe immune-mediated inflammatory diseases (IMID) or to prevent transplant rejection (TNF inhibitors, interleukin inhibitors, selective immunosuppressants, and rituximab) comprised one group. Conventional disease modifying anti-rheumatic drugs, as well as other immunosuppressants (calcineurin inhibitors, other immunosuppressants, hydroxychloroquine, and chloroquine) formed a second group. Systemic glucocorticoids formed a third group.

The study outcomes were hospital admission, ICU admission, and death, each event studied separately and independently. We included events occurring up to 30 days from patients' first positive test date, as well as hospital and ICU admission up to 7 days before that date to include relevant events occurring before testing, while reducing unrelated events occurring after recovery. Previous studies have indicated that a small percentage of patients were hospitalised before testing, 27 and approximately 80% of deaths occur within 14 days of hospital admission. 28 

We controlled for confounding by including covariates for the exposure and outcomes in a propensity score (PS) model. These covariates were selected based on background knowledge, despite incomplete knowledge of the relation between all covariates, while excluding instrumental variables or mediators. We included demographic variables (age and sex), number of past hospital contacts, diagnoses, and co-medications (including medications as proxies for disease, such as for diabetes) as covariates of immunosuppressive treatment and prognosis of SARS-CoV-2 infection (ATC and ICD-10 codes listed in Appendix Table 3 ). To control for confounding by indication, we included diagnoses such as inflammatory diseases (included with in skin diseases, and gastrointestinal diseases categories), organ transplantation, and certain malignancies that indicate treatment with immunosuppressants. Procedures and non-immunosuppressant medications used to treat IMID were included as proxies of underlying disease severity.

Clinical characteristics of the cohort were assessed, with standardised mean differences (SMD) less than 0.1 considered well balanced. We estimated the PS as the probability of treatment conditional on observed covariates. 29 We used a PS-weighting model where exposed subjects' weights were calculated as (minimum(PS,1-PS))/PS, and unexposed subjects' weights were calculated as (minimum(PS, 1-PS))/(1-PS), known as 'matching weights'. This gave a better covariate balance than inverse probability of treatment weighting (IPTW) as initially planned (Appendix Figure 1 and Appendix . 30 Weights were truncated at the 1 st and 99 th centile. We removed antianaemic drugs from the final PS model due to imbalance; adjusting for it in the log binomial regression model gave similar results (Appendix Tables 4-12) .

We estimated crude and adjusted (weighted) relative risks (and 95% confidence intervals with robust variance estimates) of the outcomes for exposed patients compared to unexposed patients using a log linear binomial regression model. We preferred this model to a survival analysis with competing risks model because a high number of events such as death often occurred very close to the date of testing, and hospital and ICU admission could occur before testing, resulting in negative time-to-event. For the analyses of subgroups of immunosuppressants, we fitted separate PS models for each of the subgroups, selecting variables from the list of covariates (Appendix Table 21 ). Exposure to combinations of the described groups of immunosuppressants was relatively rare and unlikely to alter results, so we did not study their effect. We performed a post-hoc analysis of the dose-effect of systemic glucocorticoids. To create two exposure groups of approximately equal size to maintain statistical power, we categorised the prednisolone-equivalent cumulative dose within 120 days preceding the index date as moderate dose (<2000 mg) or high dose (≥2000 mg), which were each compared to unexposed patients.

To control for residual confounding, we performed an analysis comparing current users exposed 120 days preceding the index date to former users exposed to immunosuppressants 121-365 days preceding the index date. To study the effect of immunosuppressants in patients with more severe COVID-19, we restricted the cohort to hospital admissions coded with COVID-19 as the primary diagnosis, studying the outcomes ICU admission or death.

To reduce selection bias due to patients immunosuppressants, amongst other clinically vulnerable people, being prioritised for testing (mainly before a policy change in Denmark 21 April 2020), we made separate analyses of the cohort tested before or after 21 April 2020, and made calculations to estimate the effect of selection bias (see Appendix Methods). We used the statistical software Stata 16.1 (StataCorp LLC, College Station, TX).

From 26 February-18 October 2020, there were 36,727 individuals with positive SARS-CoV-2 PCR tests in Denmark, of which 527 were exposed to immunosuppressants and 36,200 were unexposed. There were 66 exposed to selective immunosuppressants, 105 to TNF inhibitors, 25 to interleukin inhibitors, 29 to calcineurin inhibitors, 218 to other immunosuppressants, 31 to hydroxychloroquine or chloroquine, 136 to systemic glucocorticoids, and zero to rituximab. The median age of exposed patients was 57 years (IQR 42 to 73), and the median age of unexposed patients was 39 years (IQR 23-55), with a greater prevalence of comorbid diagnoses in the exposed population (Table 1 ). In total, there were 715 deaths, and 492 (69%) of those were during hospital stays. There were 425 ICU admissions, and 105 (25%) of those patients died, all occurring within 28 days of ICU admission. Few patients were exposed to both glucocorticoids and selective immunosuppressants (<5), TNF inhibitors (<5), interleukin inhibitors (<5), calcineurin inhibitors (<5), or other immunosuppressants (14) .

Among patients exposed to the composite measure of immunosuppressants there were 165 hospital admissions, 25 ICU admissions, and 57 deaths, and among the unexposed there were 3373 hospital admissions, 400 ICU admissions and 658 deaths ( Table 2 and Appendix Table 4 ). After weighting in our PSbased model, there were 346 exposed to immunosuppressants, and 339 unexposed, with a well-balanced distribution of covariates (Table 1 and Appendix Figure 1 ). The distribution of antianaemic drug usage was not balanced in the weighting model, but including it as a variable in our regression model had little effect (Appendix Table 4 ), so we removed it from the final model. The crude relative risk of hospital admission was 3·36 (95% CI 2·95 to 3·83), of ICU admission was 4·29 (95% CI 2·89 to 6·37), and of death was 5·95 (95% CI 4·60 to 7·69) ( Table 2 ). The after weighting in our PS-based model, the adjusted relative risk (aRR) of hospital admission was 1•13 (95% CI 0•95 to 1•33), the aRR of ICU admission 1•16 (95% CI 0•66 to 2•03), and the aRR of death 1•56 (95% CI 1•10 to 2•22) ( Table 2) .

For patients exposed to selective immunosuppressants, TNF inhibitors or interleukin inhibitors, compared to unexposed patients, the aRR of hospital admission was 0•83 (95% CI 0•51 to 1•34), the aRR of ICU admission was 0•92 (95% CI 0•23 to 3•71), and the aRR of death was 1•17 (95% CI 0•38 to 3•62) ( Table 3 and Appendix Tables 5, 6, and 7). For patients exposed to calcineurin inhibitors, other immunosuppressants, hydroxychloroquine or chloroquine, compared to unexposed patients, the aRR of hospital admission was 0•82 (95% CI 0•60 to 1•12), the aRR for ICU admission was 1•03 (0•43 to 2•49), and the aRR for death was 0•93 (0•47 to 1•85). For patients exposed to systemic glucocorticoids, compared to unexposed patients, the aRR for hospital admission was 1•34 (95% CI 1•10 to 1•62), the aRR for ICU admission was 1 Comparing current users of immunosuppressants to former users, the aRR of hospital admission was 1•13 (95% CI 0•83 to 1•52), the aRR of ICU admission was 1•21 (95% CI 0•68 to 2•15), and the aRR of death was 1•21 (95% CI 0•68 to 2•15) ( Table 5 and Appendix Table 9 ). When restricting to admitted patients with COVID-19 as their primary diagnosis, the risk of death was not significantly increased in patients exposed to immunosuppressants (aRR 1•30, 95% CI 0•94 to 1•82) nor was the risk of ICU admission (aRR 0•89, 95% CI 0•50 to 1•56) (Appendix Table 10 ).

Prior to the change in testing strategy on 21 April 2020, there were 199 exposed to immunosuppressants and 7794 unexposed; from 21 April-18 October 2020, there were 328 exposed, and 28,406 unexposed (Appendix Table 11 and 12). For hospital admission the aRR was 0•99 (95% CI 0•82 to 1•20) in the first period, and 1•34 (95% CI 1•00 to 1•80) in the second period, the aRR of ICU admission was 0•65 (95% CI 0•29 to 1•46) and 3•23 (95% CI 1•50-6•98), and the aRR of death was 1•06 (95% CI 0•70 to 1•63) and 2•60 (95% CI 1•52 to 4•46) respectively.

Using a nationwide cohort of 36,727 individuals tested positive for SARS-CoV-2, of whom 527 were exposed to immunosuppressants, we assessed the effect of immunosuppressants on the prognosis of intercurrent SARS-CoV-2 infection. A composite immunosuppressant exposure was associated with a significantly increased risk of death, which was mainly driven by a doubling of risk associated with systemic glucocorticoids. Glucocorticoids were also associated with a 34% increased risk of hospital admission, while the risk of ICU admission was not significantly increased ( Table 3 ). The risks of hospitalisation, ICU admission, or death associated with selective immunosuppressants, TNF inhibitors, or interleukin inhibitors were not significantly increased or decreased, nor were they in patients exposed to calcineurin inhibitors, other immunosuppressants, hydroxychloroquine, or chloroquine (Table 3 ). These findings are in agreement with two multinational studies of COVID-19 patients: glucocorticoids were associated increased risk of ICU admission or death in patients with comorbid inflammatory bowel diseases; glucocorticoids were associated with greater risk of hospital admission in patients with comorbid rheumatic diseases. 18, 20 The finding of an increased risk of death associated with glucocorticoids early in the course of COVID-19 contrasts with studies finding that high dose glucocorticoids reduces mortality in patients with severe disease 10, 11 . Nonetheless, patients not requiring supplemental oxygen in the RECOVERY trial did not benefit from dexamethasone and the effect could be compatible with harm (RR 1•19, 95% CI 0•91 to 1•55). 10 This deleterious effect of glucocorticoids early in the disease course could be due to a suppressed adrenal stressresponse, as well as their suppressive effect on interferon production, resulting in impaired innate responses to viral infection. Chronic glucocorticoid exposure also has pleiotropic metabolic effects including impaired glucose handling and skeletal muscle catabolism among other effects that may contribute to adverse outcomes. By contrast, the initiation of glucocorticoids in severe disease appears to suppress the dysregulated inflammatory response which otherwise leads to multi-organ involvement and coagulation. The effect seen in our study appears to be dose related, but these subgroups were small, so interpretation of dose effects must be tentative. By contrast, treatment with high dose glucocorticoids reduces mortality in patients with severe COVID-19 disease. The majority of patients in our study were not admitted to hospital, and would have had milder COVID-19 not requiring oxygen therapy, similar to that subset of the RECOVERY trial. These findings prompt the important question of how to improve outcomes of COVID-19 in patients taking glucocorticoids. Whether patients on glucocorticoids require increased doses during COVID-19, as in other intercurrent illnesses, or reduced doses, requires further investigation. Important strengths of this nationwide cohort study include the use of prescription and hospital activity data from national registers. Our study reduced surveillance bias, which is the limitation of studies based on spontaneously reported cases, by including all of SARS-CoV-2 test positive person in Denmark. We maximised power by using the full cohort, without restricting to specific patient populations. This facilitated extensive control of confounders, including the diverse diseases that indicate the use of immunosuppressants and glucocorticoids, further improving the reliability of our results. Controlling for covariates using a propensity score weighting model optimised the covariate distribution in a subset of the population with clinical equipoise for immunosuppressant exposure. Our analysis of bias suggested that the risk associated with immunosuppressants may be greater than estimated, as selection bias that attenuated the relative risk estimates (see Appendix Methods). Selection bias had a greater effect in the period before 21 April, when patients on immunosuppressants were prioritised for testing, which may have contributed to the lower relative risks estimated compared to after that date (Appendix Table 9 and 10).

We also recognise limitations to our study. Our conclusions on the effects of classes of immunosuppressants are cautious, as the selected groups (other than glucocorticoids) included a number of drug classes, which may have divergent effects, impairing our ability to detect associations with individual drugs. As the number of exposed subjects was small, the matching weights model targeted the population average treatment effect in the treated, and this hinders the generalisability of the risk estimates to people without an underlying condition that could require immunosuppressant therapy. The number of covariates in our model was statistically limited by the number of outcome events, so there may be residual confounding caused by unmeasured disease severity. Residual confounding is suggested by the attenuation in the risk estimate for death associated with immunosuppressants when current users were compared to former users, which remained numerically increased but no longer statistically significant. Diagnostic coding is affected by differences in practices among clinicians, an inherent limitation when nationwide register data is used. Further studies may benefit from more detailed measures of severity. However, this is unlikely to completely account for the association of glucocorticoid exposure and severe outcomes, as other immunosuppressants such as TNF inhibitors are also treatments for severe IMIDs, but by contrast, those exposures were not significantly associated with severe outcomes.

In our cohort the majority of people with COVID-19 who died were never admitted to ICU, and a substantial number were not receiving hospital-based care when they died. Frailty may account for the greater number of deaths, and greater relative risks associated with immunosuppressants, compared to ICU admissions. Admission to ICU depends not only on clinical assessment of the admitted patient, but also on factors such as frailty, short life expectancy, as well as patient and family preferences; for example, a care home resident with such conditions might not be moved to hospital, thus would not be assessed for ICU admission. Further health system factors may also be important in the context of the pandemic 31 .

In conclusion, this nationwide cohort study found that pre-morbid exposure to glucocorticoids was associated with a worsened prognosis of SARS-CoV-2 infection. 18, 19 Studies are warranted to determine whether altered doses are beneficial, with attention to the severity of COVID-19 at treatment initiation. While other pharmacological interventions remain relevant research candidates, evidence from multiple sources indicate the importance of glucocorticoids on prognosis, the effect of which may depend on timing in the disease course. Our findings that other immunosuppressants were not significantly associated with severe outcomes are tentative, but in context, they support the continued use of steroid-sparing immunosuppressants for a broad patient population with ongoing health care needs during the pandemic. 

Neoplasms, blood and blood-forming organs Due to data protection laws exact counts of individuals between 1-4 are reported only as <5. Abbreviations: SMD: standardized mean difference; IMID: immune mediated inflammatory diseases; IQR: interquartile range. Due to imabalance of antianaemic drugs, this variable was removed from the final model. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects. Abbreviations: ATC; anatomical therapeutic chemicals, ICD: international classifications of diseases. Each cell with a heading in bold represents one variable, comprising the diseases or medications listed below, with corresponding ICD/ATC codes.

Appendix Table 4 . Relative risk of severe outcomes of SARS-CoV-2 infection for patients exposed to immunosuppressants compared to unexposed patients, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs. 1·25 (1·01-1·55) Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects. Table 7 . Relative risk of severe outcomes of SARS-CoV-2 infection for patients exposed to systemic glucocorticoids compared to unexposed patients, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects.

Appendix Table 8 . Relative risk of severe outcomes of SARS-CoV-2 infection in patients exposed to cumulative dose <2000mg (moderate dose) or ≥2000mg (high dose) compared to unexposed patients, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs. 1·50 (1·16 to 1·94) Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects.

Appendix Table 10 . Relative risk of severe outcomes of SARS-CoV-2 infection in patients admitted to hospital with COVID-19 exposed to immunosuppressants, compared to unexposed patients, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects.

Appendix Table 11 . Relative risk of severe outcomes of SARS-CoV-2 infection in patients exposed to immunosuppressants, 26 February-20 April, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs.

Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects.

Appendix Table 12 . Relative risk of severe outcomes of SARS-CoV-2 infection in patients exposed to immunosuppressants, 21 April-18 October, with crude, inverse probability of treatment weighting (IPTW) and matching weights models, with further adjustment for antianaemic drugs. Abbreviations: ICU; intensive care unit. Numbers of exposed and unexposed in the propensity score weighted dataset are weighted subjects and do not refer to individual subjects.

Appendix Figure 1 . Standardised mean differences of covariates with unadjusted (crude), inverse probability of treatment weighting and matching weights models, for patients exposed to immunosuppressants compared to unexposed.

Abbreviations: IPTW: inverse probability of treatment weighting.

October 2020, by exposure to immunosuppressants, and with inverse probability of treatment weighting (IPTW) and matching weights propensity score models. Abbreviations: IMID; immune mediated inflammatory diseases.

During the first months of the pandemic people in defined risk groups, such as patients receiving immunosuppressants, as well as elderly people and people with chronic diseases, who were expected to be more susceptible to a severe course of COVID-19 disease than the general population, were prioritised for testing. From 21 April 2020, testing was open to a wider population, however, it cannot be ruled out that persons in risk groups could be more inclined towards being tested. Therefore, we expected that patients exposed to immunosuppressants were overrepresented in our cohort. However, as virtually all individuals infected with SARS-CoV-2 who required hospital admission and a large part of those who died were tested, this overrepresentation only occurs to a small degree among individuals with the outcomes. This oversampling decreases the proportion of the exposed who have the outcome, compared to a random sample of

Clinical Characteristics of Coronavirus Disease 2019 in China

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Virology, transmission, and pathogenesis of SARS-CoV-2

An inflammatory cytokine signature predicts COVID-19 severity and survival

Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China

Diagnostic and therapeutic approach to infectious diseases in solid organ transplant recipients

Increased incidence of systemic serious viral infections in patients with inflammatory bowel disease associates with active disease and use of thiopurines

Pharmacotherapeutic considerations in solid organ transplant patients with COVID-19

Management of Patients With Crohn's Disease and Ulcerative Colitis During the Coronavirus Disease-2019 Pandemic: Results of an International Meeting

American College of Rheumatology Guidance for the Management of Rheumatic Disease in Adult Patients During the COVID-19 Pandemic: Version 3. Arthritis Rheumatol

Dexamethasone in Hospitalized Patients with Covid-19 -Preliminary Report

Association Between Administration of Systemic Corticosteroids and Mortality Among Critically Ill Patients With COVID-19: A Meta-analysis

Systematic review and meta-analysis of anakinra, sarilumab, siltuximab and tocilizumab for COVID-19

Tocilizumab in patients admitted to hospital with COVID-19 (RECOVERY): preliminary results of a randomised, controlled, open-label, platform trial. medRxiv

Effect of hydroxychloroquine with or without azithromycin on the mortality of COVID-19 patients: a systematic review and meta-analysis

Covid-19 in Immune-Mediated Inflammatory Diseases -Case Series from New York

Clinical course of COVID-19 in a series of patients with chronic arthritis treated with immunosuppressive targeted therapies

But Not TNF Antagonists, Are Associated With Adverse COVID-19 Outcomes in Patients With Inflammatory Bowel Diseases: Results From an International Registry

Characteristics associated with hospitalisation for COVID-19 in people with rheumatic disease: Data from the COVID-19 Global Rheumatology Alliance physician-reported registry

Existing data sources in clinical epidemiology: The Danish COVID-19 cohort

The Danish Microbiology Database (MiBa) 2010 to 2013

Data Resource Profile: The Danish National Prescription Registry

The Danish National Patient Registry: a review of content, data quality, and research potential

The Danish Register of Causes of Death

Validity of antineoplastic procedure codes in the Danish National Patient Registry

Standardised nomenclature for glucocorticoid dosages and glucocorticoid treatment regimens: Current questions and tentative answers in rheumatology

Characteristics and predictors of hospitalization and death in the first 11 122 cases with a positive RT-PCR test for SARS-CoV-2 in Denmark: a nationwide cohort

Features of 20 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: prospective observational cohort study

The central role of the propensity score in observational studies for causal effects

Alternative approaches for confounding adjustment in observational studies using weighting based on the propensity score: A primer for practitioners

Considerations for pharmacoepidemiological analyses in the SARS-CoV-2 pandemic

October 2020, by exposure to systemic glucocorticoids with inverse probability of treatment weighting (IPTW) and matching weights propensity score models. Abbreviations: ATC; anatomical therapeutic chemicals, GI: gastrointestinal, IMID: immune-mediated inflammatory diseases.

Appendix Table S1 . Notation for the numbers of individuals with or without the outcome hospital admission, by risk group and exposure status, for individuals tested and individuals infected with SARS-CoV-2.

All

As hospital admission is relatively rare for the total group of SARS-CoV-2 infected, we can approximate the risk ratio of Y (hospital admission) given I (exposure to immunosuppressants) with the odds ratio: which is the risk ratio of being tested for risk groups and health care workers compared to others, restricted to those without hospital admission. The risk of testing given no hospital admission possibly later in time is hard to interpret, so we will make an approximation. As around 80% of the hospital admissions happen on the same day as testing, we will make an approximation by changing the restriction from those without hospital admission to those without hospital admission on the day of the test (S):Thus, we have the following equation for the risk ratio (RR) of hospital admission (Y) given exposure (I) among those who have been tested (T=1):Where A is unexposed people, not in risk groups, without the outcome; B is unexposed people, in risk groups, without the outcome; C is unexposed people with the outcome; D exposed people without the outcome; and E is exposed people with the outcome.To apply the equation to our data we split the matching weights dataset for hospital admission, with 89 events among 208 exposed, and 77 events among 201 exposed (as per Appendix Table 15 ) into risk and non-risk groups based on rough definitions and to produce the values in Table S2 (corresponding to Appendix Table S1 ).Appendix Table S2 . The numbers of individuals with or without the outcome hospital admission, by risk group and exposure status, for individuals tested and individuals infected with SARS-CoV-2. For intensive care unit (ICU) admission and death, we can again use the premise that hospital admission is relatively rare so + will not change much by adding any number of unexposed to the numerator, or exposed with admission to hospital to the denominator. Thus, as + + + only depends on the total number of exposed and unexposed, the bias for ICU admission and death can be approximated by that for hospital admission. N.B, this bias analysis applies the main relative risk estimate with the composite exposure.

Appendix Figure 2 . Relative risk of severe outcomes of SARS-CoV-2 in patients exposed to immunosuppressants compared to unexposed patients, corrected for selection bias (the risk ratio of being tested, for risk groups and health care workers compared to the general population, ranging from 1 to 3).Abbreviations: ICU; Intensive care unit.