Re: Effect of hydroxychloroquine with or without azithromycin on the mortality of COVID-19 patients: a systematic review and meta-analysis' by Fiolet et al.

To the Editor,

We read with interest the paper of Fiolet et al. [1], which reviewed the effect of the hydroxychloroquine (HCQ) regimen alone or associated with azithromycin (HCQ+AZI) on mortality from coronavirus disease 2019 (COVID-19). For the effect of HCQ alone on mortality, we noted the following inaccuracies.

• 1.
  Fiolet et al. wrongly interpreted the relative risk (RR) in Geleris et al. [2] as that of mortality. The original work used the composite end point ‘intubation or death’.
• 2.
  They reported the RR associated with the whole cohort of patients in Magagnoli et al. [3]. This RR was affected by length-biased sampling, suffered from a large number of missing end points that was unbalanced across the groups (18% of patients on HCQ, 12% of patients on HCQ+AZI and 26% of controls were still hospitalized). Magagnoli et al. conducted a subanalysis to account for ‘the issue of length-biased sampling and differential rates of right-censored observations among the groups’ and provide a better estimate of the RR.
• 3.
  For the effect of HCQ+AZI on mortality, we noted that Fiolet et al. mistakenly interpreted the RR in Rivera et al. [4] to be associated with the HCQ+AZI regimen. The original work indicates this RR to be associated with ‘HCQ plus any other therapy’, which was not limited to azithromycin because about 22% of these patients were not treated with azithromycin.
• 4.
  For the reasons discussed in (2), the RR of the subanalysis from Magagnoli et al. should have been reported.

Furthermore, it should be noted that Fiolet et al. excluded the results reported by Arshad et al. [5], which provided further evidence in favour of the use of HCQ. Fiolet et al. argued that patients in the HCQ arm were twice as likely as controls to receive steroids. However, Arshad et al. accounted for this imbalance using propensity-score matching and found the mortality RR associated with corticoids to be 0.8, which is remarkably similar to the effect of dexamethasone on mortality evidenced by the Recovery trial (RR = 0.83) [6]. On the other hand, Rivera et al. was included in the meta-analysis even though the original work reported an odds ratio of 3 associated with the use of steroids, strongly suggesting that indication biases were insufficiently corrected. We acknowledge that the results in Arshad et al. do not come without biases and limitations, noticeable in their multivariate regression, but they provided detailed information regarding multivariate regression scores and matching tables, which, for example, were absent in Magagnoli et al. Such changes in the meta-analysis might reverse the conclusions of the original paper.

In observational studies, patient groups are heterogeneous and vary considerably from one study to another, i.e. therapeutic management (type of oxygen therapy, corticosteroid therapy, anticoagulation, time to implement these different treatments), different population (eating habits, smoking). Such an analysis does not describe the reality, which is often more complex, and this also constitutes a major bias.

Altogether, we raise serious concerns about the choices made for the study inclusion and selection. The main results reported by Fiolet et al. rely on four misrepresentations of original work.

In the light of these elements, we believe the authors should at minimum reconsider their conclusions.

Funding

No external funding was received.

Authors' contributions

NV contributed to conceptualization and wrote the original draft; AL, ATB and SG contributed to writing, review and editing; and NV acted as supervisor.

Transparency declaration

The authors have stated that there are no conflicts of interest.

Notes

Editor: L. Leibovici

References