Statistical analysis
Sample size calculations were based on the assumed rate of patients in
the group of individuals unexposed to DMARDs (0.15). With 5%
significance level and 80% power, 1,701 subjects in the exposed group
and 850 in the unexposed group were needed to detect a minimum relative
risk (RR) of 1.3.
To evaluate the associations between different treatments and the
presence of hsCOVID-19 symptoms, Poisson regression models with robust
variance estimation were used to estimate RR and 95% confidence
intervals (95%CI) [4]. Models were adjusted by sex, age, diabetes,
pulmonary disease, cardiovascular disease, chronic kidney disease, and
active cancer or treatment. Model 1 aimed to estimate the association
between treatments grouped by drug type (1) bDMARDs; (2) sDMARDs, (3)
glucocorticoids, (4) chronic nonsteroidal anti-inflammatory drugs
(NSAIDs) and (5) Anti-hypertensive drugs. Then, associations between
hsCOVID-19 symptoms were estimated by each individual treatment (with
>100 exposed patients; reference category = “unexposed”;
Model 2). Finally, as anti-TNFα treatments were the major group of
bDMARDs, the effect of each anti-TNFα drug was estimated separately in
model 3. Model 3 also included the effect of anti-IL17 and
anti-IL23(-12), but anti-IL6 could not be analysed as a separate group
as there were not hsCOVID-19 symptoms reported among individuals exposed
to IL-6 antagonists. Interactions between different drug types were also
tested (model 4). Models 1 and 2 were furthermore replicated in the
subgroup of individuals aged ≤ 60 years (model 5 and 6) and
>60 years (model 7 and 8). All analyses were stratified by
sex. Finally, the main treatment indications, together with the studied
comorbidities (sex, age, cardiovascular disease, diabetes, pulmonary
disease, kidney disease and cancer) were used to create a matched
dataset for the exposure to each treatment Propensity score matching was
applied for the exposure to anti-TNFα, anti-proinflammatory ILs,
glucocorticoids (≤10 mg/day in women) and
chloroquine/hydroxychloroquine. Statistical analyses were performed
using R (R Foundation for Statistical Computing, Vienna, Austria)
version 3.5.2.