Effect of anakinra in COVID-19 - Authors' reply.

We thank Emma Kooistra and colleagues for their interest in our study. They raise several interesting points. A certain degree of difference in baseline characteristics is an inherent limitation of cohort studies. Kooistra and colleagues point out a between-group difference in age, which we agree is a risk factor for poorer outcomes and might have biased our results in favour of the anakinra group. However, they neglect other variables that would be expected to have the opposite effect, including more severe impairment in respiratory function in patients receiving high-dose anakinra, a feature that is also indicative of clinical severity and is associated with poorer outcomes. Kooistra and colleagues also discuss serum ferritin, and correctly state that serum ferritin concentrations higher than 2000 ng/mL increase the likelihood of macrophage activation syndrome (MAS). However, serum ferritin is more drastically elevated in classic MAS compared with COVID-19. Other clinical features of MAS, including pancytopenia and haemophagocytosis, are also not typically observed in COVID-19-related hyperinflammation. Indeed, it has become increasingly clear that the hyperinflammatory reaction observed in some patients with COVID-19 resembles MAS, yet represents a separate, unique condition. Analogies between MAS and COVID-19 were useful in highlighting individual predispositions to the development of hyperinfammation, and in instructing use of anakinra, which has emerged as a promising therapy for COVID-19 in several publications following our first report.4, 5 However, one should be mindful that hyperinflammation in COVID-19 does not equate to MAS; the validity of the 2000 ng/mL ferritin threshold in COVID-19 is not clear-cut.

Concerning the statistical design of our study, we agree with Kooistra and colleagues that propensity score matching is a preferable option for correcting unavoidable differences in clinical features of patients in cohort studies, outside a controlled trial setting. However, the size of our cohorts at the time that we started our study precluded effective deployment of propensity score matching. We welcome evidence generated using this statistical approach in future studies, as well as data from randomised clinical trials. If data from randomised trials become available, these will supersede our findings. However, although superior to observational studies at generating evidence, randomised trials are not as pragmatic under truly emergency circumstances. Observational studies informed clinical decision making at the cusp of the COVID-19 pandemic. Several ongoing clinical trials will follow and generate controlled data, some of which might only become available after the end of the pandemic, with little or no impact on patient management.