Key-points Low dose hydroxychloroquine was associated with reduced mortality in COVID-19 patients, as seen in observational studies but not in randomised clinical trials, which used high doses of hydroxychloroquine. These findings can help disentangling the debate on hydroxychloroquine use in COVID-19.
Background Hydroxychloroquine (HCQ) was proposed as potential treatment for COVID-19, but its association with mortality is not well characterized. We conducted two meta-analyses to evaluate the association between HCQ (with or without azithromycin (AZM)) and total mortality in COVID-19 patients.
Methods Articles were retrieved until October 20th, 2020 by searching in seven databases. Data were combined using the general variance-based method on relative risk estimates.
Results A total of 26 articles were found (N=44,521 COVID-19 patients, including N=7,324 from 4 randomized clinical trials (RCTs)); 10 studies were valuable for analysing the association of HCQ+AZM. Overall, the use of HCQ was associated with 21% lower mortality risk (pooled risk ratio: 0.79, 95%CI: 0.67 to 0.93; high level of heterogeneity: I2=82%, random effects). This association vanished (1.10, 95%CI: 0.99 to 1.23 and 1.10, 95%CI: 0.99 to 1.23) when daily dose >400 mg or total dose >4,400 mg were used, respectively). HCQ+AZM was also associated with 25% lower mortality risk, but uncertainty was large (95%CI: 0.50 to 1.13; P=0.17). No association was apparent when only pooling the 4 RCTs (13.8% of the overall weight; pooled risk ratio: 1.11, 95%CI: 0.99 to 1.24).
Conclusions HCQ use was not associated with either increased or decreased mortality in COVID-19 patients when 4 RCTs only were evaluated, while a 7% to 33% reduced mortality was observed when observational studies were also included. The association was mainly apparent when pooling studies using lower doses of HCQ. These findings can help disentangling the debate on HCQ use in COVID-19.
The aminoquinoline hydroxychloroquine (HCQ) is an anti-malaria drug, with immunomodulatory and anti-thrombotic properties, currently used in the treatment of autoimmune diseases like rheumatoid arthritis, systemic lupus erythematosus and anti phospholipid syndrome [1, 2]. At the beginning of the pandemic, it was proposed as a possible therapy in COVID-19 patients since it could
directly inhibit viral entry and spread in several in vitro and in vivo models .
Indeed, HCQ has been used in Ebola virus disease , human immunodeficiency virus (HIV) infection , SARS-CoV-1 infection and the Middle East Respiratory Syndrome . Despite the lack of evidence of efficacy from few randomized clinical trials, HCQ became very popular and widely used by many clinicians. In Italy over 70% of COVID-19 hospitalised patients were treated with HCQ . On the other hand, this drug was given an inexplicable political connotation that shifted the focus more to a political battle than to a discussion based on scientific evidence. The publication of a very questionable study  by one of the most reliable scientific journals showing that the use of HCQ was associated to an increased risk of death, lead to the stop of several clinical trials in their tracks and of the HCQ arm in the Recovery trial . This Lancet study was retracted 13 days after publication , because its data turned out to be fabricated.
Anyway, the principal drug Agencies decided to suspend the authorization to use HCQ for COVID-19. Consequently, the review process of papers on HCQ became increasingly difficult, creating a publication bias that affected all meta-analyses published until now. However, a number of questions remain open on the relationship between HCQ treatments in COVID-19 patients: is there a dose issue? Does mortality rate of a population or the severity of the disease affect HCQ efficacy? Is there any interaction with other anti-COVID19 drugs? In the last few months, at least three large, well-conducted observational studies have been published showing a protective effect of HCQ on mortality risk in hospitalized COBID-19 patients [7,11,12]. All three studies have not been included in previous meta-analysis , and used HCQ doses lower than those administered in randomized clinical trial (RCT), as the Recovery or the Solidarity trials [9,14].
Therefore, we decided to conduct an updated meta-analysis on observational and RCT studies on HCQ use and the mortality outcome in patients hospitalized for COVID-19. We also performed
subgroup analyses to dissect whether treatment effects differ according to methodological or clinical characteristics of the primary studies.
This study was conducted according to the recommendations outlined in the Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0, and reported in line with the PRISMA statement. Institutional review board approval was not required as the study did not directly involve human participants.
Flow diagram for study selection is reported in Figure 1. Articles published in English were retrieved from inception to October 20th
, 2020 by searching in Medline, Embase, PubMed, Web of Science, Cochrane Central Database, MedRvix and Preprints.org, with the search terms: “(COVID19 OR Cov-Sars-2) AND (hydroxychloroquine OR chloroquine)”. In addition, the reference lists of relevant articles for potential studies were also manually reviewed. After initial search, the duplicate results were removed. The remaining articles were screened for relevance by their titles and abstracts by two of us independently (SC and ADC).
All selected potential articles were then reviewed by the remaining investigators to ensure their eligibility for inclusion. Disagreements about eligibility of the literature were resolved by consensus based on the agreements of all investigators. To be included in this meta-analysis, the study had to meet the following criteria: (1) clinical trials or cross-sectional studies or cohort studies; (2) quantitatively investigating the difference in mortality risk in unselected COVID-19 patients according to use or not of HCQ.
Twenty-seven articles were identified [7,9,11,12,14-36]. For ten of them [11,15,17,19,21,23,25,31-33] it was possible to extract data necessary for comparing HCQ+AZM versus no HCQ+AZM. For the other studies, it was not possible to systematically distinguish if HCQ therapy was
complemented or not with AZM. The RCT by Mitjà et al.  was not included since the Authors found zero deaths both in HCQ and control group. The study of Mehra et al.  was excluded since the Authors retracted it owing to several concerns on the veracity of the data .
Data Extraction and data analysis
For each study, odds ratio (OR) or hazard ratio (HR) and/or number of events (number of deaths and number of total COVID-19 patients) in both the HCQ (or HCQ+AZM) and respective control groups were extracted. If available, measure of association adjusted for covariates were retrieved.
Number of events were used to calculate relative risk and 95% confidence intervals (CIs) when other measures of association were not available from the primary study. The following information was also extracted: study design, if the article was not peer-reviewed, region, level of adjustment, sample size, mortality rate in the entire cohort, percentage of patients treated with HCQ, mean duration of the treatment, mean daily dose after the first day and mean total dose of HCQ used.
The total dose of HCQ was calculated as the sum of the amount of drug used in the first day plus daily dose multiplied by number of days of treatment after the first. Pre-specified subgroup analyses have been conducted for all the additional characteristics retrieved. All analyses were performed using standard statistical procedures provided in RevMan5.1 (The Cochrane Collaboration, Oxford, United Kingdom). Data were combined using the general variance-based method, that requires information on the relative risk (or OR or HR) estimate and their 95% CI for each study. 95% CI were used to assess the variance and the relative weight of each study.
Heterogeneity was assessed using the Higgin’s I2 metric. Fixed and random effects were considered, but due to the large heterogeneity observed, findings from random effects were considered as primary analysis. The hypothesis that publication bias might have affected the validity of the estimates was visually tested by a funnel plot–based approach.
Characteristics of the studies
The workflow of the process of study selection is reported in Figure 1. A total of 27 articles were found in the search. Twenty-six of these eligible studies were enrolled for analysing the association with mortality of HCQ use in patients with COVID-19, and 10 of them were valuable for analysing the association of HCQ+AZM. The main characteristics of the 26 studies included in the meta-analysis are shown in Table 1.
We found 4 RCT studies [9,14,17,34] and 22 observational studies; 5 articles were not published in peer reviewed journals; 3 studies reported not adjusted measure of association between HCQ and mortality; 13 studies have been conducted in Europe, 9 in North America and 3 in other countries. The outcome considered was the mortality for any cause; for the large majority of the studies, the mortality was intra-hospital.
In all studies the control group was formed by patients without HCQ exposure (HCQ or HCQ+AZM). All studies included adult men and women COVID-19 patients, with the exception of two RCT [10,25], that included a portion of individuals with uncertain positivity to Sars-CoV-2. A total of N=44,521 COVID-19 patients (including N=7,324 from the 4 RCTs) were counted in the meta-analysis of HCQ, and N=12,153 in the meta-analysis of HCQ+AZM.
Pooled analysis, HCQ as exposure
Forest plot on the association between HCQ and mortality is reported in Figure 2. Pooling of data from 22 observational studies, which accounted for 86.2% of the total weight, the use of HCQ has
been associated with 25% lower mortality risk (pooled risk ratio: 0.75, 95%CI: 0.63 to 0.89; high level of heterogeneity: I2=80%, random effects). Conversely, the association was not evident in the 4 RCTs (13.8% of the weight; pooled risk ratio: 1.11, 95%CI: 0.99 to 1.24; I2=0%).
Grand total by pooling of both RCTs and observational studies (weight 100%) pointed to a relative reduction of mortality in support of HCQ by 7% to 33% (figure 2). Subgroup analyses according to main features of primary studies are presented in Table 2. The association of HCQ with lower mortality was observed with very low differences in all subgroups, with the exception of dose grouping. The reduced mortality was in fact confined to studies that used a daily dose ≤400 mg (as estimated in days of treatment after the first, in which a higher (double for most) dose of drug was administered); pooling n=9 studies which used more than 400 mg of HCQ daily resulted in an overall measure of association equal to 1.10 (95%CI: 0.99 to 1.23; table 2 and supplementary figure 1).
Also, the magnitude of the association was higher in studies which used HCQ for 5 or less days (table 2 and supplementary figure 2). Pooling studies which used more than 4,400 mg of HCQ
during the entire phase of treatment set an overall measure of association equal to 1.10 (95%CI: 0.99 to 1.23; table 2 and supplementary figure 3). These findings were confirmed in subgroup
analyses restricted to observational studies: the pooled measure of association was equal to 0.98 (95%CI: 0.54 to 1.77) and 1.05 (95%CI: 0.73 to 1.53) in studies with a total HCQ dose >4,400 mg
(n=3) or a daily HCQ dose >400 mg (n=5), respectively.
Pooled analysis, HCQ+AZM as exposure
Figure 3 reported random forest for 10 studies comparing HCQ+AZM. Use of the combination HCQ+AZM was associated with 25% lower mortality risk, with very large uncertainty (pooled risk ratio: 0.75, 95%CI: 0.50 to 1.13; P for testing of overall effect=0.17; high level of heterogeneity: I 2=91%, random effects). By visual inspection of funnel plots (Supplementary Figure 4), we failed to observe any selection bias for both meta-analyses
Pooled analysis of adverse events
Pooled analyses of RCTs on the relationship between HCQ and incidence of adverse effects are reported in supplementary figure 5. HCQ use was associated with an increased risk of adverse effects of any type (panel A). On the contrary, patients treated with HCQ in RCTs showed a similar rate of serious adverse events (panel B), as that non-treated with HCQ (pooled risk ratio: 1.16, 95%CI: 0.86 to 1.55; P for testing of overall effect=0.33).
In a meta-analysis of 26 studies (4 RCTs) involving 44,814 COVID-19 patients, the use of HCQ was associated with a 21% lower risk of total mortality. The association was apparent by pooling 22 observational studies and was more evident in studies which used lower HCQ doses.
No association was found pooling 4 RCTs. Use of HCQ was not associated with severe adverse events. The potential for selection bias in observational studies is not negligible. The decision from the clinician to utilize or not a drug may depend on comorbidities and baseline risk of the patient. However, in the pandemic, and in absence of guidelines and specific anti COVID-19 drugs, allocation of HCQ in observational studies was not associated systematically with a lower or higher baseline risk profile.
For example, in the CORIST study  patients receiving HCQ were more likely younger and less likely had ischemic heart disease, cancer or chronic kidney disease, but, on the contrary, they were more likely men and had higher levels of C-reactive protein. As a consequence, it is not clear if in that particular study HCQ patients were potentially at higher or lower risk of a negative prognosis. In attempting to account for baseline differences between patients who received HCQ and those who did not, we used the results for adjusted measure of association for each study, and this was possible for 22 out of 26 studies included in the metaanalysis. After the exclusion of 3 unadjusted studies [16,21,24], the strength of the overall association of HCQ with mortality was merely reduced from 0.79 to 0.80.
Although we attempted to control for potential confounding factors inherent to patient and clinical characteristics, it is possible that unmeasurable confounding still remains, and this may explain the different finding between observational and RCT studies. However, it is hard to determine which are, if any, the unmeasured characteristics that have confused so strongly the association between HCQ and mortality in COVID-19 that was observed in observational studies. In fact, these features must be a) unmeasured in observational studies; b) associated with mortality in COVID-19 and c) associated with HCQ use, in a way that when the risky conditions are present the clinicians tend systematically to avoid using HCQ. For example, HCQ is contraindicated in patients with cardiomyopathy but this condition has been mostly measured in observational studies and was not recognised as a risk factor for mortality in COVID-19 patients.
The dissimilar findings between observational and RCTs we found might also be explained by differences in HCQ dosage . Interestingly, we observed that the reduced mortality associated
with HCQ treatment is actually confined to studies that used a daily dose ≤400 mg, or a total dose ≤4,400 mg or which used HCQ for 5 or less days. Obviously these three conditions largely overlapped in studies, that we can now designate as “at low HCQ dose”. Remarkably, none of the 4 RCTs are in this category. In detail, the RECOVERY  and the SOLIDARITY study  used 800 mg/day for 9 or 10 days (after the first), respectively and a total dose of 9200 or 10000 mg of HCQ (including the dose at first day) respectively, a very high dose regimen as confronted to the rest of studies, particularly of the observational ones.
The possibility that HCQ reduced the risk of negative prognosis in COVID-19 patients when only administered at “low dose” cannot be here undoubtedly proven starting from our findings, but it is a plausible hypothesis that may explain the different result between observational and RCT studies and, more importantly, might be useful in disentangling the debate on HCQ use in COVID-19. High levels of HCQ administration were used in RCTs to maximise the antiviral activity of the drug that was considered to be the main mechanism of action of HCQ in this context. In some studies, the inverse association of HCQ with inpatient mortality was more evident in elderly, in patients who experienced a higher degree of COVID-19 severity or having elevated C-reactive protein levels , suggesting that the anti-inflammatory potential of HCQ may have had a more important role than its antiviral properties.
HCQ, indeed, beside an antiviral activity, may have both anti-inflammatory and anti-thrombotic effects . This can justify its effect in reducing mortality risk, since Sars-Cov-2 can induce pulmonary microthrombi and coagulopathy, that are a possible cause of its severity [38, 39] and the lack in preventing SARS-CoV-2 infection after exposure . On the other end, National guidelines suggested to use HCQ 200 mg twice daily for 5-7 days probably to maintain a better risk benefit profile hypothesizing that low doses could be more effective and safer. Indeed, non-sigmoidal, bell-shaped dose-response curves are possible with drugs having complex biological effects, multiple-binding sites or cellular and organ targets.
On the other hand, anti SARS-2-CoV-2 activity of HCQ has been confirmed in Vero cells . HCQ is also reported to reduce secretion of IFN-γ and IL-17 in activated Th1 and Th17 cells, respectively . The concomitant use of azithromycin seems to not neither increase nor decrease the effect, if any, of the HCQ since the combination of the two drugs was associated with a lower mortality risk at very similar extent to that observed for HCQ alone, but the assumption is inconclusive because of the very large uncertainty in the findings.
A main concern with HCQ treatment have been its side effects, in particular a severe cardiovascular toxicity. Indeed, HCQ can cause prolongation of the QT interval on electrocardiogram , which could be exacerbated by coadministration with azithromycin, widely prescribed as co-treatment in Covid-19 treatment. Our meta-analysis of data from RCTs, that allowed a proper evaluation of side effects, shows that use of HCQ was associated with an increase in side effects of any type, but not of major type, including cardiovascular events. This despite the high prevalence of cardiovascular disease in patients with COVID-19 or the high dose used in RTCs.
This meta-analysis has the strength of including all available data recently published and that have not been included in previous meta-analyses [13,44], and of considering modification of effect by
dosing of HCQ. As major limitations, we recognised that majority of the primary studies were observational, the pooled findings suffer of a high degree of heterogeneity and that results in observational and RCT studies were different.
In conclusion, HCQ was not associated with increased or decreased mortality in COVID-19 patients when only RCTs were pooled, but it was associated with 8% to 35% reduced mortality when observational studies were also included. The association was mainly apparent by pooling studies using lower doses of HCQ. Use of HCQ was not associated with severe adverse events.
These results should be considered with caution, because the majority of the studies included were observational and retrospective and the possibility of confounding could not be fully excluded. However, at present, this is the largest comprehensive quantitative overview on the association of HCQ with mortality in COVID-19 patients, and our findings underscoring HCQ dosage effects can help disentangling the debate on HCQ use and encourage the planning of RCTs using low doses of HCQ in hospitalised COVID-19 patients.
SC was the recipient of a Fondazione Umberto Veronesi Travel Grant.
ADC and LI contributed to the conception and design of the work and interpretation of data; SC and ADC managed study selection and data extraction and critically reviewed the results; SC analysed
the data; ADC and LI wrote the paper; LI, AC, RC and GdG originally inspired the research and critically reviewed the manuscript.
All Authors approved the final version of the manuscript.
Source of support: none.
Conflicts of interest: The Authors report no conflict of interest related to the current work