Efficacy and Safety of Hydroxychloroquine for Hospitalized COVID-19 Patients: A Systematic Review and Meta-Analysis

We systematically reviewed the efficacy and safety of hydroxychloroquine as treatment for hospitalized COVID-19. Randomized controlled trials (RCTs) evaluating hydroxychloroquine as treatment for hospitalized COVID-19 patients were searched until 2nd of December 2020. Primary outcomes were all-cause mortality, need of mechanical ventilation, need of non-invasive ventilation, ICU admission and oxygen support at 14 and 30 days. Secondary outcomes were clinical recovery and worsening, discharge, radiological progression of pneumonia, virologic clearance, serious adverse events (SAE) and adverse events. Inverse variance random effects meta-analyses were performed. Thirteen RCTs (n=18,540) were included. Hydroxychloroquine total doses ranged between 2000 and 12,400 mg; treatment durations were from 5 to 16 days and follow up times between 5 and 30 days. Compared to controls, hydroxychloroquine non-significantly increased mortality at 14 days (RR 1.07, 95%CI 0.92–1.25) or 30 days (RR 1.08, 95%CI 1.00–1.16). Hydroxychloroquine did not affect other primary or secondary outcomes, except SAEs that were significantly higher than the control (RR 1.24, 95%CI 1.05–1.46). Eleven RCTs had high or some concerns of bias. Subgroup analyses were consistent with main analyses. Hydroxychloroquine was not efficacious for treating hospitalized COVID-19 patients and caused more severe adverse events. Hydroxychloroquine should not be recommended as treatment for hospitalized COVID-19 patients.


Introduction
As of the 22nd of February 2021, over 110 million people have been diagnosed with COVID-19, resulting in the death of~2.5 million individuals [1]. Treatment options for COVID-19 are currently limited, with few treatment recommendations from the National Institutes of Health (NIH). The current guidance includes the use of corticosteroids in hospitalized patients requiring supplemental oxygen or remdesivir in hospitalized patients requiring non-mechanical oxygen supplementation [2].
Hydroxychloroquine (HCQ), an oral drug used for severe forms of rheumatoid arthritis and lupus erythematosus, has in vitro antiviral efficacy against coronaviruses including SARS-CoV-2 and immune modulating effects [3][4][5][6][7] and was among the first COVID-19 treatments investigated in human studies. HCQ, however, has several known adverse events such as QTc prolongation, cardiomyopathy, hypoglycemia and retinal toxicity in the non-COVID-19 environment [8]. By late Spring 2020, most of the evidence assessing HCQ efficacy and safety in hospitalized patients were from observational studies and showed widely divergent findings [9]. Since that time, several randomized controlled trials (RCTs) have been completed and provided a much clearer picture of the role of hydroxychloroquine in the treatment of hospitalized patients with COVID- 19. We systematically assessed RCTs evaluating HCQ effects vs. controls on clinical and safety outcomes in hospitalized COVID-19 patients. We also assessed quality of evidence of all outcomes using GRADE methodology.

Data Sources and Searches
Two investigators (V.P. and A.V.H.) developed the search strategy, which was revised and approved by the other investigators. We searched the following databases until 2nd of December 2020: PubMed-MEDLINE, EMBASE-OVID, Scopus, Web of Science, the Cochrane Library, medRxiv.org (www.medrxiv.org, accessed on 2 December 2020) and Preprints (www.preprints.org, accessed on 2 December 2020). The PubMed search strategy is shown in the Supplementary Materials.

Study Selection
Randomized controlled trials (RCTs) in any language reporting benefit or harm outcomes from use of hydroxychloroquine vs. control (placebo or usual care) as treatment in hospitalized with reverse transcription-polymerase chain reaction (RT-PCR)-confirmed COVID- 19. We excluded studies in COVID-19 outpatients, studies of prophylaxis with hydroxychloroquine (i.e., in those without COVID-19 disease) and studies evaluating chloroquine. Three investigators (A.V.H., V.P., Y.M.R.) independently screened each record title and abstract for potential inclusion. Three investigators (V.P., J.J.B., Y.M.R.) then read the full text of the records whose abstracts had been selected by at least one investigator. Discrepancies were resolved through discussion or by a fourth investigator (A.V.H.).

Outcomes
Primary outcomes were all-cause mortality, need of mechanical ventilation, need of non-invasive ventilation, ICU admission and oxygen support at 14 and 28 or 30 days. Secondary outcomes were clinical recovery and worsening, discharge, radiological progression of pneumonia, virologic clearance, serious adverse events (SAE) and adverse events. Clinical recovery and worsening were extracted as defined by authors.

Data Extraction
Two investigators (A.P., J.J.B.) independently extracted the following variables from studies: study setting, country, COVID-19 diagnosis, hydroxychloroquine dose and duration, type of control and description, length of stay, primary and secondary outcomes and time of follow up. Discrepancies were resolved through discussion or by a third investigator (A.V.H.).

Risk of Bias Assessment
Three investigators (A.P., V.P., J.J.B.) independently assessed RoB by using the Cochrane Risk of Bias 2.0 tool for RCTs [10]; disagreements were resolved by discussion with a fourth investigator (A.V.H.). RoB per domain and study as low, some concerns and high for RCTs.

Statistical Analysis
We reported our systematic review according to 2009 PRISMA guidelines [11]. Inverse variance random effect meta-analyses were performed to evaluate effect of HCQ vs. control or SoC on outcomes when outcome data was available for at least two RCTs judged to have homogeneous study characteristics. Effects of meta-analyses were reported as relative risks (RR) for dichotomous outcomes; we also calculated their 95% confidence intervals (CIs). CIs of effects were adjusted with the Hartung-Knapp method [12], and the between study variance tau 2 was calculated with the Paule-Mandel method [13]. Heterogeneity of effects among studies was quantified with the I 2 statistic (an I 2 > 60% means high heterogeneity). We pre-specified subgroup analyses by type of control and RoB; the p for interaction test <0.05 indicated effect modification by subgroup. The meta package of R 3.5.1 (www.r-project.org, accessed on 10 December 2020) was used for meta-analyses. The quality of evidence was evaluated using the GRADE methodology, which covers 5 aspects: risk of bias, inconsistency, indirectness, imprecision and publication bias [14]. Quality of evidence was evaluated per outcome and described in summary of findings (SoF) tables; GRADEpro GDT was used to create SoF tables [15].

Characteristics of Included Studies
The general characteristics of the included RCTs are included in Table 1. In most RCTs, all participants had COVID-19 confirmation via RT-PCR. However, patients in Chen, L. et al. [19], Cavalcanti et al. [20] and Horby et al. [21] had baseline RT-PCR positivity rates of 63%, 76% and 90%, respectively, and Chen, J. et al. [16] did not report the percentage of patients with RT-PCR positivity. All RCTs included adult populations. Placebo was the comparator in three RCTs (Self et al. [24], Dubee et al. [25], Ulrich et al. [28]) while usual care was the comparator in the other RCTs. The total dose of hydroxychloroquine in the RCTs varied between 2000 and 12,400 mg with follow up times between 5 and 30 days.

Risk of Bias of Included RCTs
Four RCTs [18][19][20]23] had high risk of bias, one due to deviations from intended interventions, [18] one due to deviations from intended interventions and missing outcome data, [19] one due to measurement of the outcome, [20] and one due to selection of reported results [23]. Seven RCTs [16,17,22,24,[26][27][28] had some concerns of bias, three due to the randomization process, [16,24,28] one due to randomization and deviation from intended interventions [26], two due to deviations from intended target, [22,27] and one due to selection of the reported results. [17] Two RCTs had low risk of bias in all categories [21,25] ( Figure 2).

R PEER REVIEW 8 of 15
Four RCTs [18][19][20]23] had high risk of bias, one due to deviations from intended interventions, [18] one due to deviations from intended interventions and missing outcome data, [19] one due to measurement of the outcome, [20] and one due to selection of reported results [23]. Seven RCTs [16,17,22,24,[26][27][28] had some concerns of bias, three due to the randomization process, [16,24,28] one due to randomization and deviation from intended interventions [26], two due to deviations from intended target, [22,27] and one due to selection of the reported results. [17] Two RCTs had low risk of bias in all categories [21,25] (Figure 2).
x FOR PEER REVIEW 8 of 15 Four RCTs [18][19][20]23] had high risk of bias, one due to deviations from intended interventions, [18] one due to deviations from intended interventions and missing outcome data, [19] one due to measurement of the outcome, [20] and one due to selection of reported results [23]. Seven RCTs [16,17,22,24,[26][27][28] had some concerns of bias, three due to the randomization process, [16,24,28] one due to randomization and deviation from intended interventions [26], two due to deviations from intended target, [22,27] and one due to selection of the reported results. [17] Two RCTs had low risk of bias in all categories [21,25] (Figure 2).

Subgroup Analyses
Most of subgroup analyses by type of control (placebo or usual care) and RoB were consistent with main analyses (Supplementary Figure S20). No p for interaction tests were significant for clinical outcomes. HCQ non-significantly increased the composite endpoint of clinical recovery by 26% in two RCTs [23,25] but with high heterogeneity (RR 1.26, 95%CI 0.87-1.83, I 2 = 76%, Supplementary Figure S6) while HCQ also significantly increased the composite endpoint of clinical worsening by 14% in two RCTs [21,25] with no noted heterogeneity (RR 1.14, 95%CI 1.11-1.18, I 2 = 0%, Supplementary Figure S7).

Subgroup Analyses
Most of subgroup analyses by type of control (placebo or usual care) and RoB were consistent with main analyses (Supplementary Figure S20). No p for interaction tests were significant for clinical outcomes.

Subgroup Analyses
Most of subgroup analyses by type of control (placebo or usual care) and RoB were consistent with main analyses (Supplementary Figure S20). No p for interaction tests were significant for clinical outcomes.

Quality of Evidence
The quality of evidence using the GRADE tool was high for HCQ causing clinical worsening; moderate for HCQ causing all-cause mortality at 14 and 30 days, need for mechanical ventilation at 30 days and serious adverse events; low for HCQ impacting need for supplementary oxygen and discharge from hospital; and very low for HCQ impacting need for mechanical ventilation at 14 days, need for high-flow nasal cannula or non-invasive ventilation, need for ICU admission, clinical recovery and overall adverse events ( Table 2). Main drivers of low quality of evidence were individual studies with high or some concerns of bias, imprecision of effects and inconsistency of effects between RCTs.

Discussion
Since the time of our last systematic review update on HCQ in hospitalized patients, [9,29] the literature set is now robust enough to warrant meta-analysis. The dataset now includes 13 RCTs with only 30% deemed to have a high risk of bias. We can now say with moderate confidence that there was a trend towards HCQ use increasing all-cause mortality by 7-8%, with high confidence that patients treated with HCQ were 14% more likely to experience clinical worsening over their hospital course, and with moderate confidence that patients receiving HCQ were 24% more likely to experience serious adverse events. We also can conclude with low confidence that there was a trend that HCQ use can increase the need for oxygen support by 26%, and with very low confidence HCQ use had a trend towards increasing overall adverse events by 39%, ICU admission by 36%, noninvasive nasal cannula or non-invasive ventilation by 6% and mechanical ventilation at 14 days by 98%. Of all the aforementioned endpoints, no statistical heterogeneity was found, except for overall adverse events and mechanical ventilation at 14 days where it was high in both cases.
With moderate confidence, HCQ showed a trend towards reducing the need for mechanical ventilation at 30 days by 7%, with low confidence HCQ showed a trend towards speeding hospital discharge by 3%, and with very low confidence HCQ showed a trend towards improving clinical recovery by 26%. Unfortunately, HCQ did not reduce the need for mechanical ventilation at 14 days, the hospital discharge analysis had moderate statistical heterogeneity, and the clinical recovery analysis had very high statistical heterogeneity.
Taken together, HCQ should not be used in hospitalized patients with COVID-19 because the RCTs completed to date did not demonstrate a favorable balance of benefits to harm. We found no significant benefits with HCQ therapy and patients were significantly more likely to clinically worsen and have serious adverse events when given HCQ. The trend towards HCQ increasing overall mortality is especially troubling. It is unlikely that future studies will reveal positive benefits for hydroxychloroquine that outpace the potential for harms. This is strongly supported by the results of our subgroup analyses where only analyzing RCTs with a true placebo group or selectively analyzing RCTs with lower risk of bias were not different than what we found in our full dataset analyses.
Unlike previous systematic reviews, [30][31][32] including our previous review [9] and final update, [29] we limited this systematic review with meta-analyses to RCTs because this study type is inherently stronger and less prone to biases. Cohort studies assessing HCQ in COVID-19 have extensive clinical and methodological heterogeneity, especially those conducted earlier in the pandemic where the projects were rushed, and publications of lower quality projects were more likely. Likewise, the results of cohort studies were heterogenous and hard to interpret. In instances where the strength of evidence for an outcome was very low, if the direction of effect for an outcome was the same in RCT and cohort study analyses, they bolstered each other. However, it was impossible in most cases to reconcile areas where the different study types showed directions of effect moving in different directions. In those cases, the hierarchy of evidence would still suggest that the data from RCTs would be more likely to approximate the actual effects.
Among  [20]. For both comparisons, there was no effect of HCQ ± AZ vs. control on mortality. Bayesian secondary analysis gave similar results to primary analyses [32].
A recent systematic review of mortality outcomes by Axfors et al. [33] evaluated ongoing, completed or discontinued RCTs on HCQ or chloroquine treatment for any COVID-19 patients until October 16, 2020. This inclusion criteria were therefore broader than our study. We looked at a wider variety of outcomes than other systematic reviews to ensure that there were not unique benefits or harms that might not be identified in narrower assessments. We felt this was vital in truly understanding the balance of benefits to harms and as a result of our systematic review and meta-analysis saying the balance for HCQ use is unfavorable. A final advantage of our new systematic is that the literature search is updated to 2 December 2020 and included all major RCTs published on efficacy of hydroxychloroquine in hospitalized patients.
Chloroquine, an antimalarial drug, was proposed as therapeutic agent for COVID-19 because they were observed to inhibit SARS-CoV-2 viral replication in vitro in primate cells [5]; however, a later study found that chloroquine did not inhibit infection of human lung cells with SARS-CoV-2 [34]. HCQ, a less toxic derivative of chloroquine, inhibits trained immunity in vitro in peripheral blood mononuclear cells, which may not be beneficial for the antiviral innate immune response to SARS-CoV-2 infection in patients [35]. Chloroquine also has been investigated in vitro as potential treatment for other viruses such as human immunodeficiency virus [36], human coronavirus OC43, enterovirus EV-A71, Zika virus, influenza A H5N1, hepatitis C virus and chikungunya virus [37]; however, these studies did not show beneficial effects.
Our study had several limitations. First, most of outcomes had low or very low quality of evidence mainly driven by high or some concerns of bias, imprecision of effects and heterogeneity of effects. Second, there was heterogeneity of definitions of clinical worsening and clinical improving among RCTs; this situation was particularly prevalent in older studies. Third, we did not assess individual adverse events or serious adverse events due to scarcity of reporting across RCTs. Finally, we did not evaluate the effect of adding azithromycin to hydroxychloroquine as there was only one RCT [20] evaluating such combination.

Conclusions
Hydroxychloroquine was not efficacious for treating hospitalized COVID-19 patients and caused more severe adverse events. Hydroxychloroquine should not be recommended as treatment for hospitalized COVID-19 patients.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/jcm10112503/s1, Figure S1: Effect of hydroxychloroquine on need for mechanical ventilation at 14 days in hospitalized COVID-19 patients; Figure S2: Effect of hydroxychloroquine on need for mechanical ventilation at 30 days in hospitalized COVID-19 patients; Figure S3: Effect of hydroxychloroquine on need for high-flow nasal cannula or non-invasive ventilation at 14 days in hospitalized COVID-19 patients; Figure S4: Effect of hydroxychloroquine on need for ICU admission in hospitalized COVID-19 patients; Figure S5: Effect of hydroxychloroquine on need for oxygen support at 14 days in hospitalized COVID-19 patients; Figure S6: Effect of hydroxychloroquine on clinical recovery in hospitalized COVID-19 patients; Figure S7: Effect of hydroxychloroquine on clinical worsening (death or invasive mechanical ventilation) in hospitalized COVID-19 patients; Figure  S8: Effect of hydroxychloroquine on discharge in hospitalized COVID-19 patients; Figure S9: Effect of hydroxychloroquine on radiological progression of pneumonia in hospitalized COVID-19 patients; Figure S10: Effect of hydroxychloroquine on virologic clearance at 14 days in hospitalized COVID-19 patients; Figure S11: Effect of hydroxychloroquine on gastrointestinal adverse events (nausea, vomiting, abdominal pain) in hospitalized COVID-19 patients; Figure S12: Effect of hydroxychloroquine on abnormal liver function in hospitalized COVID-19 patients; Figure S13: Effect of hydroxychloroquine on rash in hospitalized COVID-19 patients; Figure S14: Effect of hydroxychloroquine on headache in hospitalized COVID-19 patients; Figure S15: Effect of hydroxychloroquine on QTc prolongation in hospitalized COVID-19 patients; Figure S16: Effect of hydroxychloroquine on anemia in hospitalized COVID-19 patients; Figure S17: Effect of hydroxychloroquine on ventricular tachycardia in hospitalized COVID-19 patients; Figure S18: Effect of hydroxychloroquine on leukopenia in hospitalized COVID-19 patients; Figure S19: Effect of hydroxychloroquine on lymphocytopenia in hospitalized COVID-19 patients; Figure S20: Subgroup analyses by type of control and risk of bias for all outcomes in hospitalized COVID-19 patients.