A total of 539 COVID-19 hospitalized patients were included in our cohort in Milan, from February 24 to May 17, 2020, of whom 174 died in hospital (day 14 probability of death: 29.5% – 95%CI: 25.5–34.0). We divided a subset of our cohort into three groups who started treatment a median of 1 day after admission: those receiving hydroxychloroquine alone (
N?=?197), those receiving hydroxycholoroquine?+?azithromycin (
N?=?94), and those receiving neither (controls) (
N?=?92). Of the latter group, ten started HIV antivirals (boosted-lopinavir or –darunavir), one teicoplanin, twelve immunomodulatory drugs, or corticosteroids, 23 heparin and 46 remained untreated. The percent of death in the three groups was 27%, 23%, and 51%. Mechanical ventilation was used in 4.3% of hydroxychloroquine, 14.2% of hydroxychloroquine?+?azithromycin, and 26.1% of controls. Unweighted and weighted relative hazards of mortality are shown in
Table 1. After adjusting for several key confounders (see table), the use of hydroxycholoroquine?+?azithromycin was associated with a 66% reduction in risk of death as compared to controls; the analysis also suggested more substantial effectiveness of hydroxychloroquine in patients with less severe COVID-19 disease (PO2/FiO2?>?300, interaction
p-value <0.0001). Our results are remarkably similar to those shown by Arshad et al.
Table 1Unadjusted and adjusted marginal relative hazards of in-hospital mortality
Some important weaknesses in Arshad et al.’s analysis have been pointed out (
), but not all of these apply to our study. Our propensity scores include some of the potential confounders that were missing in the analysis by Arshad (e.g., calendar day of admission, disease severity, cardiovascular disease (CVD), baseline plasma CRP); second, we have excluded people receiving other drugs which could have biased the effect of hydroxychloroquine when used in combination. Third, although residual confounding is a possibility (e.g., people with CVD were more frequent in control), people in the control group were more likely to undergo mechanical ventilation, which is a conservative bias. These results from two different real-life settings (Italy and USA), conflict with those of two large randomized trials (
,
). Although unmeasured confounding remains the most likely explanation for the discrepancies, a robust meta-analysis is still lacking, and we believe that hydroxychloroquine should be further tested in randomised trials. When best to start treatment is also a question that needs to be addressed in ad-hoc randomized studies.
Article info
Publication history
Published online: July 29, 2020
Received: July 10, 2020
Copyright
© 2020 The Author(s). Published by Elsevier Ltd on behalf of International Society for Infectious Diseases.