Association Between Tumor Necrosis Factor Inhibitors and the Risk of Hospitalization or Death Among Patients With Immune-Mediated Inflammatory Disease and COVID-19

Key Points Question Is receipt of tumor necrosis factor (TNF) inhibitor monotherapy at the time of COVID-19 diagnosis associated with adverse COVID-19 outcomes compared with other treatment regimens among patients with immune-mediated inflammatory diseases (IMIDs)? Findings In this cohort study of 6077 patients with IMIDs and COVID-19, TNF inhibitors in combination with azathioprine/6-mercaptopurine therapy, methotrexate monotherapy, azathioprine/6-mercaptopurine monotherapy, or Janus kinase inhibitor monotherapy were each associated with significantly higher odds of hospitalization or death compared with TNF inhibitor monotherapy. Meaning This study’s findings support the continued use of TNF inhibitor monotherapy among individuals with IMIDs during the pandemic.


Introduction
Patients with COVID-19, caused by SARS-CoV-2, can have mild symptoms or experience a severe and/or life-threatening infection. 1 Comorbidities, such as lung disease, diabetes, and obesity, increase the risk of adverse COVID-19 outcomes. 2 Any benefits of treatments for immune-mediated inflammatory diseases (IMIDs) for COVID-19 outcomes remain a topic of interest. These treatments impact the immune system and are associated with a higher risk of infections overall. 3 This association raises concern about impaired immune response to SARS-CoV-2 among patients currently receiving treatment for IMIDs. However, many damaging consequences of SARS-CoV-2 infection are produced by a hyperinflammatory response. 4 Therefore, treatments that target an overactive immune response may have protective benefits against adverse COVID-19 outcomes. 1,4 Tumor necrosis factor (TNF) inhibitors, a class of biologic therapies that target the proinflammatory cytokine TNF, are first-or second-line treatments for many IMIDs. International registries of patients with IMIDs have provided initial information regarding COVID-19 outcomes among individuals who received TNF inhibitor therapies during the pandemic. An analysis of data from the Secure Epidemiology of Coronavirus Under Research Exclusion for Inflammatory Bowel Disease (SECURE-IBD) registry, which includes patients with inflammatory bowel disease (IBD) who were diagnosed with COVID-19, found that prevalent use compared with no use of TNF inhibitors at COVID-19 diagnosis was not associated with severe COVID-19 (odds ratio [OR], 0.9; 95% CI, 0.4-2.2). 5 A study of data from the Global Rheumatology Alliance (GRA) physician-reported registry of COVID-19 outcomes among people with rheumatic diseases found that prevalent use compared with no use of TNF inhibitors at COVID-19 diagnosis was associated with lower odds of COVID-19associated hospitalization (OR, 0.40; 95% CI, 0.19-0.81). 6 An analysis of data from the Psoriasis Patient Registry for Outcomes, Therapy and Epidemiology of COVID-19 Infection (PsoProtect) also found higher odds of hospitalization among patients treated with nonbiologic systemic therapies compared with biologic therapies, including TNF inhibitors (OR, 2.84; 95% CI, 1.31-6.18). 7 Although studies of individual registries have provided initial information, they were often underpowered to perform more granular analyses of commonly used medications, such as monotherapy vs combination therapy with immunomodulatory drugs, or analyses of medications that are used less frequently.
Pooling data across registries offers an opportunity to rapidly assess any association between TNF inhibitor therapies and COVID-19 outcomes among individuals with IMIDs and to evaluate the consistency of findings across studies and diseases. We pooled data from these 3 international registries of patients with IBD, psoriasis, and rheumatic diseases to evaluate the association between TNF inhibitor monotherapy and COVID-19-associated hospitalization or death compared with other commonly prescribed immunomodulatory regimens among individuals with IMIDs.

Registry Designs and Approvals
Details of the design of the GRA, SECURE-IBD, and PsoProtect registries have been described previously. [7][8][9][10] In brief, clinicians and trained staff directly report COVID- 19  reporting guideline for cohort studies.

COVID-19 Diagnosis
Among patients with rheumatic disease, a COVID-19 diagnosis was based on polymerase chain reaction (PCR), antibody serology testing, or metagenomic testing; computed tomographic scans; laboratory assays; or symptoms alone. Among patients with IBD, a COVID-19 diagnosis was based on PCR testing, symptoms with confirmatory antibody serology testing, or rapid antigen testing. Among patients with psoriasis, both confirmed and suspected COVID-19 diagnoses were reported; however, information regarding the type of diagnostic testing used was not collected.

Exposures and Outcome
To obtain sufficient statistical power, each exposure category was required to have more than 250 patients in the pooled analysis. Exposure was defined as a categorical variable that comprised the following categories: TNF inhibitor (including adalimumab, certolizumab pegol, etanercept, golimumab, and infliximab) monotherapy (reference category), TNF inhibitors in combination with methotrexate therapy, TNF inhibitors in combination with azathioprine/6-mercaptopurine therapy, methotrexate monotherapy, azathioprine/6-mercaptopurine monotherapy, and Janus kinase (Jak) inhibitor (including tofacitinib, baricitinib, and upadacitinib) monotherapy. The outcome of interest was hospitalization or death associated with COVID-19.

Inclusion and Exclusion Criteria
We included adults (age Ն18 years) with a diagnosis of inflammatory arthritis, IBD, or psoriasis who were reported to the GRA, SECURE-IBD, or PsoProtect registries, respectively, from March 12, 2020, to February 1, 2021. Our analysis included reconciled patients only. For the GRA registry, reconciled patients were defined as those with at least 1 of the following recorded outcomes: death, symptoms resolved at the time of data entry, not hospitalized more than 30 days after the initial diagnosis date, hospitalized and discharged, or not at risk of further interventions or death. For the SECURE-IBD and PsoProtect registries, patients were defined as reconciled after a minimum of 7 days or 14 days, respectively, or if sufficient time had passed to observe the disease course through the resolution of acute illness or death.
To limit confounding from other immunomodulatory medications, we excluded patients who received an exposure treatment regimen as well as concomitant immunomodulatory drugs, except when sulfasalazine, mesalamine, hydroxychloroquine or chloroquine, leflunomide, oral budesonide, or glucocorticoids were used as concomitant medications.

Statistical Analysis
We used descriptive statistics to summarize the demographic and clinical characteristics of the study population. Continuous variables were reported as means with SDs or medians with 25th and 75th percentiles, as appropriate. Categorical variables were reported as numbers with percentages. We performed registry-level analyses and a pooled analysis of data across the 3 registries to estimate independent associations between exposure categories and COVID-19 outcomes. Registry-level effect estimates were reported for exposure categories that included 10 or more patients.
Associations were estimated using multilevel multivariable mixed-effects logistic regression analysis and reported as odds ratios (ORs) with 95% CIs. We chose mixed-effects regression analysis for its ability to handle missing data using maximum-likelihood estimation and to fit nested random effects to account for multilevel clustering. 11 Covariates included in all models were age, sex, current smoking, IMID activity ( analyses, we included the following concomitant medications: sulfasalazine, hydroxychloroquine or chloroquine, and leflunomide for the GRA analysis and mesalamine, sulfasalazine, and oral budesonide for the SECURE-IBD analysis. If any of these concomitant medications were significant confounders (P < .05) in the registry-level analyses, they were also included as covariates in the pooled analysis; patients from registries that did not include the respective concomitant medications were assigned to the category of individuals with nonuse of these medications. Registry-level analyses also controlled for disease diagnosis; diagnoses in the GRA registry included rheumatoid arthritis (reference category), psoriatic arthritis, spondyloarthritis, and other inflammatory arthritis or more than 1 type of inflammatory arthritis, and diagnoses in the SECURE-IBD registry included Crohn disease (reference category), ulcerative colitis, and unspecified IBD.
We fitted country-level random effects to account for within-country correlations. To account for changes in COVID-19 treatment and health service use over time, we also fitted random effects for the calendar month of symptom onset (PsoProtect registry) or the calendar month during which the patient was diagnosed (GRA registry) or reported (SECURE-IBD registry). The pooled model also included registry-level random effects accounting for within-registry correlations. The hierarchical order of nested random effects in the pooled model was country followed by time and registry. To improve model fit, we removed influential statistical outliers identified in continuous variables (ie, age and glucocorticoid dose) from the analyses. As a result, 2 patients were removed who received a daily prednisone dose greater than 70 mg.
All analyses were conducted using Stata software, version 16.0 (StataCorp). The threshold for statistical significance was 2-sided P < .05.
Rheumatology clinics from 2 large health care systems (Mass General Brigham in Massachusetts and Mayo Clinic in Minnesota and Florida) had processes in place to systematically report all patients with COVID-19 to the GRA registry. To assess the extent of potential reporting bias arising from convenience sampling, ORs were derived after reweighting the covariate distribution of patients in the GRA registry to those of the 2 health care systems using the inverse odds of sampling weights technique, 12 and those values were compared with the original ORs using standardized difference. 13 In addition, to assess the robustness of results, a pooled sensitivity analysis was performed after excluding patients with a COVID-19 diagnosis that was based on symptoms alone.

Results
As of February 1, 2021, 8268 patients were reported to have received an exposure treatment regimen at COVID-19 diagnosis. Of those, 5220 patients were from the GRA registry, 2720 patients were from the SECURE-IBD registry, and 328 patients were from the PsoProtect registry. A total of 6077 patients from 74 countries met study eligibility criteria and were included in the analyses; of those, 3441 patients (56.6%) were from the GRA registry, 2336 patients (38.4%) were from the SECURE-IBD registry, and 300 patients (4.9%) were from the PsoProtect registry. Of the 2191 patients excluded from the analyses, most were excluded because they had a rheumatic disease diagnosis other than inflammatory arthritis (827 patients), were patients who were nonreconciled (581 patients), or received concomitant medications that were listed in the exclusion criteria (551 patients) ( Table 1).
The demographic and clinical characteristics of the 6077 patients included in the analysis are shown in Table 2 female, and 2468 patients (40.6%) were male. Race and ethnicity were not addressed in these analyses because information on race was not recorded in the PsoProtect registry, and information on race and ethnicity was not available for all countries in the GRA registry. The most common     diagnosis based on symptoms alone. Our pooled results remained consistent in a sensitivity analysis that excluded these patients (Figure).

Discussion
This cohort study found that TNF inhibitor monotherapy was associated with a lower risk of COVID-19-associated hospitalization or death among patients with IMIDs compared with other commonly used treatment regimens, including methotrexate, azathioprine/6-mercaptopurine, and Jak inhibitors. After controlling for active disease and common comorbidities, the odds of hospitalization or death among those who received TNF inhibitor combination therapies vs TNF inhibitor monotherapy depended on the type of additional medication used in the combination regimen.
Patients receiving TNF inhibitors in combination with azathioprine/6-mercaptopurine therapy had higher odds of hospitalization or death compared with those receiving TNF inhibitor monotherapy, Figure Registry-specific and pooled analyses, with TNFi monotherapy used as the reference category. Pooled estimates were obtained using hierarchical multivariable mixed-effects logistic regression analysis with registry and calendar month random effects nested within country. Pooled sensitivity analysis (n = 5213) excludes patients with a presumptive COVID-19 diagnosis (defined as a diagnosis based on symptoms alone). All ORs were adjusted for age, sex, current smoking, immune-mediated disease activity (remission vs active), important comorbidities (cardiovascular disease, diabetes, hypertension, obstructive lung disease, interstitial or other chronic lung disease, kidney disease, obesity [body mass index Ն30; calculated as weight in kilograms divided by height in meters squared], and cancer), and prednisone-equivalent glucocorticoid dose.  whereas individuals receiving TNF inhibitors in combination with methotrexate therapy had similar odds of hospitalization or death compared with those receiving TNF inhibitors alone.

JAMA Network Open | Infectious Diseases
The lower odds of unfavorable COVID-19 outcomes among patients receiving TNF inhibitors before SARS-CoV-2 infection has several possible explanations. Although the exact mechanism of SARS-CoV- 2-associated hyperinflammation remains uncertain, high serum TNF concentrations at   the time of COVID-19 admission have been associated with organ damage and worse COVID-19 outcomes. 15 Therefore, blocking TNF could inhibit this detrimental immune response. Multiple case series reporting favorable outcomes among patients receiving TNF inhibitor therapy support this assertion. 1,16,17 Upcoming results from clinical trials investigating the use of TNF inhibitors will enable further evaluation of the association between TNF inhibitor therapy and COVID-19 outcomes. 18,19 Other possible explanations for our findings include the consequences of non-TNF inhibitor immunosuppressive medications for COVID-19 outcomes. Thiopurine medications are associated with a higher risk of opportunistic viral infections. [20][21][22] A study examining data from a large registry of patients with IBD found that the receipt of thiopurines, including azathioprine and 6-mercaptopurine, was associated with a higher risk of serious viral infection, specifically infection from species of the Herpesviridae. 23 Although data regarding other viruses cannot be directly extrapolated to COVID-19, this higher risk highlights the potential for an association between thiopurine use and an increased risk of unfavorable outcomes after SARS-CoV-2 infection. Moreover, a recent study of data from the SECURE-IBD registry reported that thiopurine monotherapy and thiopurines in combination with TNF inhibitor therapy were associated with worse COVID-19 outcomes compared with TNF inhibitor monotherapy. 24 In contrast, researchers have postulated that methotrexate therapy may decrease the cytokine storm associated with COVID-19. 25,26   However, our results suggest worse outcomes associated with methotrexate monotherapy compared with TNF inhibitor monotherapy. This association could mean that TNF inhibitor therapy is exerting a protective benefit or that methotrexate therapy is exerting a harmful consequence.

JAMA Network Open | Infectious Diseases
Notably, the direction of association was the same for methotrexate used in combination with TNF inhibitors, although the effect estimate crossed the line of no effect, which is possibly associated with the use of lower methotrexate doses in combination therapy compared with monotherapy. 27,28 The timing of treatment initiation with Jak inhibitors may be an important factor associated with COVID-19 outcomes. The second iteration of the Adaptive COVID-19 Treatment Trial (ACTT-2) suggested a protective effect of treatment with baricitinib in combination with remdesivir therapy against unfavorable COVID-19 outcomes among some subgroups of patients with confirmed severe COVID-19. 29 However, population-based data from patients receiving Jak inhibitors before COVID-19 diagnosis suggest worse outcomes, which is consistent with the known association between this class of medications and reductions in the innate immune response, producing impaired viral clearance. 30 In our comparative analyses, we found that Jak inhibitor monotherapy was associated with higher odds of hospitalization or death than TNF inhibitor monotherapy.

Strengths and Limitations
This study has strengths. These strengths include the robust worldwide collaboration between 3 international registries, which enabled evaluation of a large geographically diverse sample of adults with IMIDs. To our knowledge, this study is the first to pool data across registries to evaluate COVID-19 outcomes among patients with IMIDs. Pooling data increased the power of the study, allowed for more granular analyses of medications, and improved generalizability across IMIDs.
Notably, our analyses controlled for active disease, which is only possible through the use of registry data because this variable is not typically available in administrative databases or electronic health records. Furthermore, clinicians or trained staff reported directly to each registry, which likely increased the accuracy of the information.
This study also has limitations. These limitations include the risk of reporting bias because the registries used convenience sampling. However, the results of our sensitivity analysis suggest that reporting bias was not a substantial concern in the GRA registry. The threshold for hospitalization and the ways in which patients are treated for COVID-19 differs over time and across regions. Such  differences have the potential to introduce bias if insufficiently accounted for in the analyses.

JAMA Network Open | Infectious Diseases
Although we attempted to account for associations in hospitalization or death owing to unmeasured temporal and geographical factors, residual confounding may remain. Additional factors that we were unable to account for included duration and previous lines of IMID therapy. Furthermore, the lack of a global COVID-19 registration system limited the feasibility of including a control group.
Although the case report forms were similar, the data domains across registries were not entirely uniform. For example, time and type of COVID-19 diagnosis, rheumatic disease activity, and certain comorbidities were recorded slightly differently across registries. These inconsistencies were addressed, to the extent possible, through the incorporation of registry-level random effects in multilevel modeling and through sensitivity analyses.

Conclusions
The results of this cohort study suggest that, among patients with IMIDs, receipt of TNF inhibitor monotherapy may be associated with a lower risk of COVID-19-associated hospitalization or death compared with other immunomodulatory treatment regimens. These findings support the continued use of TNF inhibitor monotherapy during the pandemic and warrant further research investigating the association of other biologic therapies with COVID-19 outcomes. Treatment with TNF inhibitor combination therapy was associated with a more favorable safety profile when methotrexate rather than azathioprine/6-mercaptopurine was used, suggesting that clinicians would benefit from weighing the risks vs benefits of deescalating treatment or changing medications when a patient is receiving concomitant TNF inhibitors and azathioprine/6-mercaptopurine.