Table 1.
Key studies examining COVID-19 outcomes in rheumatic disease patients.
| Reference | Location | Study Population | COVID-19 Ascertainment | Primary Finding | Limitations |
|---|---|---|---|---|---|
| Williamson et al. (OpenSAFELY) (5) | United Kingdom | General population (>17 million adults) | Positive molecular testing | RA/SLE/Psoriasis associated with higher risk of COVID-19 related death (HR 1.19, 95% CI: 1.11 to 1.27). | Multiple testing; adjusting for causal intermediates*; unmeasured confounding†; inclusion of non-laboratory confirmed COVID-19 cases; missing data; lumping of RA, SLE, and psoriasis; lack of information about DMARDs and disease activity. |
| Ye et al. (3) | Wuhan, China | COVID-19 patients with rheumatic disease versus without rheumatic disease | Positive molecular testing (n=20) or positive IgM and IgG (n=1) | Respiratory failure more common in rheumatic disease patients than comparators (38% vs. 10%, p<0.01). | Small sample size (n=21 rheumatic disease patients); unmeasured confounding; collider bias.‡ |
| D’Silva & Serling-Boyd et al. (4); Serling-Boyd & D’Silva et al. (6) | Boston, MA | COVID-19 patients with rheumatic disease versus without rheumatic disease | Positive molecular testing | Higher odds of mechanical ventilation in rheumatic disease patients versus comparators in first two months of pandemic (OR 3.11, 95% CI: 1.07 to 9.05). Improved risk of mechanical ventilation six months into the pandemic. | Collider bias may bias results towards null; adjusting for causal intermediates. |
| D’Silva & Jorge et al. (9) | United States | COVID-19 patients with rheumatic disease versus without rheumatic disease | Positive molecular testing or diagnostic code | Higher risks of hospitalization and acute renal failure mediated by comorbidities. Higher risk of venous thromboembolism (RR 1.60, 95% CI: 1.14 to 2.25), regardless of comorbidities. | Unmeasured confounding; inaccuracies in ICD-10 coding; lack of geographic information; collider bias. |
| Pablos et al. (30) | Spain | COVID-19 patients with rheumatic disease versus without rheumatic disease | Positive molecular testing | Higher odds of severe COVID-19 with glucocorticoids (OR 2.20, 95% CI: 1.36 to 3.54). | Limited to hospitalized patients; small sample sizes in subgroups; collider bias. |
| Gianfrancesco et al. (31) | International Physician-Reported Registry | Hospitalized versus non-hospitalized rheumatic disease patients with COVID-19 | Physician-reported diagnosis | Prednisone doses ≥10 mg daily associated with higher odds of hospitalization (OR 2.05, 95% CI: 1.06 to 3.96). TNF inhibitors associated with lower odds of hospitalization (OR 0.40, 95% CI: 0.19 to 0.81). | Selection bias (more severe cases more likely to be captured); unmeasured confounding; large number of unresolved cases (35%) at time of publication. |
COVID-19, Coronavirus Disease 2019; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; HR, hazard ratio; DMARD, disease-modifying antirheumatic drug; OR, odds ratio; CI, confidence interval; RR, relative risk; TNF, tumor necrosis factor; ICD-10, International Classification of Diseases, 10th Revision.
Causal intermediates are covariates that are causally influenced by the exposure and also causally affect the outcome of interest. For example, if the exposure is rheumatic disease and the outcome is COVID-19 related death, a comorbidity such as diabetes may be a causal intermediate as rheumatic disease treatments such as glucocorticoids can cause diabetes, which in turn can cause more severe COVID-19.
Unmeasured confounding refers to covariates that may not be measured and/or adjusted for in analyses. For example, glucocorticoid use may be an unmeasured confounder in the OpenSAFELY study.
Collider bias occurs when analyses are restricted by a collider variable, which is a variable that is a common effect of the exposure and outcome. In studies where the exposure is COVID-19 infection and the outcomes are COVID-19-related outcomes such as hospitalization, mechanical ventilation, or death, conditioning on the common effect of COVID-19 may bias results towards the null.