To the Editor,
We read with interest the meta-analysis published in Clinical Microbiology and Infection by Fiolet et al.  entitled “Effect of hydroxychloroquine with or without azithromycin on the mortality of COVID-19 patients: a systematic review and meta-analysis”. This meta-analysis concluded that the combination of hydroxychloroquine (HCQ) and azithromycin (AZ) was associated with increased mortality and that HCQ alone had no effect on mortality. We believe that this study is fatally flawed. To start with, its conclusion is impossible in light of the treatment in our center of patients with HCQ-AZ, with reduction of mortality in the population at risk (>60 years) by a factor two .
In this context, we sought to understand how the authors reached their conclusions. First, none of the authors has extensive experience in the treatment of infectious diseases, and a generic systematic review of literature does not replace expert understandings of study methods and pitfalls, as we have described . The authors report several meta-analyses but omitted ours . Our analysis highlighted a major factor of heterogeneity between studies:
Rote “Big-data” studies, with complete disconnection between data analysis and clinician expertise, and clinical studies, in which the analysis is conducted by clinicians with a precise and detailed protocol. Our work notably contributed to highlighting the Lancetgate  and showed that Big-data studies are associated with conflicts of interest and absence of detailed 29 treatment protocols. The fatal flaw of the Fiolet et al. analysis is that it used subjective, capricious and specious criteria decisions about which studies to include.
Large valid observational studies reporting significant benefit and published during the inclusion period and that used standard accepted methods to control for confounding factors (propensity-score matching) were not included, notably Arshad et al. in the USA (n = 2,541), Bernaola et al. in Spain (n = 1,645), and our study of 199 patient pairs in France. One inclusion criterion mentioned by the authors  is “cases confirmed by RT-PCR”. This however is in contradiction to inclusion of Skipper et al., with “Only 58% of participants received SARS-CoV-2 testing” and the RECOVERY trial for which PCR confirmation was not mandatory, as well as that it used toxic doses (2400 mg HCQ within the first 24 hours).
Worse, Fiolet et al. included data from the study by Rivera et al. which itself is fatally flawed. That study, of cancer patients, included “Participation by anonymous individual health-care practitioners located in Argentina, Canada, the EU, the UK, and the USA is also allowed. The mechanism of data collection can be retrospective (after the course of COVID-19) or concurrent, at the discretion of the respondent.”
This is not a sampling frame for any type of epidemiologic study. There is no assurance that individual practitioners didn’t select patients on which to report because the patients didn’t do well on antiviral therapies. Haphazard subject collection is not a representative epidemiologic sampling method and has no justification for use. Second, the Rivera et al. data show dramatic differences in HCQ and AZ use for nontreated vs treated subjects by baseline disease severity, and the authors did not report results on HCQ+AZ use but on HCQ+other medication use, which is not adjusted adequately for severity. Simply put, patients with worse conditions were given more medications and were more likely to die of their cancers. Third, Fiolet et al. use results values in their forest plot that do not appear in the paper or supplement of Rivera et al.
Two new retrospective studies further contradict the authors’ conclusions. The study of 8,075 patients in Belgium (Catteau, 2020) and 3,451 patients in Italy (Di Castelnuovo, 2020) report clear benefits of chloroquine derivatives on mortality (Figure). In this context, we present (Figure) an update of our meta-analysis . The clinical studies include 4,121 patients from 7 countries, with summary odds ratio 0.60, 95% confidence interval 0.50-0.73, p=10-7).
Results were consistent among studies (I2 =20%, p=0.23). Among the 35,985 patients included in the Big data studies, a significant but smaller benefit was found (0.84, 0.75-0.94, p=.003) with appreciable heterogeneity (I2 =76%, p<.001). History and reasoning demonstrates that the authors’ conclusions are erroneous. We suggest that the authors leave their political motivations at the door before undertaking scientific work.
New Jersey Study of 1,274: Hydroxychloroquine in the treatment of outpatients with mildly symptomatic COVID-19; A multi-center observational study where hydroxychloroquine exposure was associated with a decreased rate of subsequent hospitalization