Safety of hydroxychloroquine for treatment or prevention of SARS‐CoV‐2 infection: A rapid systematic review and meta‐analysis of randomized clinical trials

Abstract Introduction Hydroxycloroquine (HCQ) has been extensively studied for treatment and prevention of coronavirus diseases 2019 (COVID‐19) from the start of the pandemic. Conflicting evidence about its usefulness has begun to accrue. Methods In the face of controversial results about clinical efficacy of HCQ, we performed a rapid systematic review to assess its safety in the framework of COVID‐19 randomized clinical trials. Results Five studies investigating 2291 subjects were included. The use of HCQ was associated with higher risk of adverse event compared with placebo or standard of care: odds ratio 4.57, 95% confidence interval 2.14–9.45. Conclusion Safety profile of HCQ appears to be unsatisfactory when used to treat or prevent COVID‐19, especially in the light of unproved clinical benefit.


| INTRODUCTION
During the early phases of coronavirus diseases 2019 (COVID-19) pandemic, the repurposing of drugs used for other indications represented one of the few available options to fight severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of the new clinical entity. Chloroquine (CQ) and its more soluble and less toxic metabolitehydroxychloroquine (HCQ) are two antimalarial drugs derivative from 4-aminoquinoline employed in rheumatological setting due to their immunomodulatory properties. 1 In the setting of COVID-19 pandemic, they have drawn much attention for their potential antiviral activities, combined with other interesting features such as wide availability, inexpensive profile, manageabilityderived from long historical use. 1 Nevertheless, in spite of promising results from observational studies, evidence synthesis from randomized clinical trials (RCTs) indicates no efficacy of HCQ/CQ on relevant clinical outcomes, except time to symptom resolution, but with low certainty. 2 In the light of this very modest beneficial effect, we tried to gauge the safety profile of HCQ/CQ in the context of COVID-19 pandemic on the basis of data from RCTs published so far, to highlight the trade-off between pros and cons of HCQ/CQ repurposing to treat or prevent SARS-CoV-2 infection.

| METHODS
We performed a rapid systematic review and metaanalysis according to the Preferred Reporting Items for SystematicReviews and Meta-analyses reporting guidelines (Supporting Information Material S1). 3 The study protocol was preregistered and made publicly available on osf.io/qckf5.
We searched Medline, Scopus, and Embase for allinvestigations published as peer-review publicationany time from inception up to August 11, 2020, without language restriction. The details of the search strategy for each database were provided in Supporting Information Material S2.
Inclusion criteria relied on the following PICO strategy: subjects with or without infection by SARS-CoV-2 (P-participants), undergoing a course of HCQ/HQ for treatment or prevention, respectively, of SARS-CoV-2 infection (I-intervention); comparison according to the study, namely placebo or other comparator such standard of care without antimalarials (C-comparison); safety profile in terms of total adverse events (AEs) as main endpoint (O-outcome). Secondary outcomes werespecific types of AEs (e.g., gastrointestinal, cardiovascular and so on). We performed pooled analyses regarding secondary outcomes if consistent definitions across the RCTs were available in at least two RCTs. Studies were excluded if the intervention (HCQ/HQ) was not administered in an RCT or if the RCT enrolled ≤100 patients to avoid highly implausibleand hugely imprecise effect estimates as described elsewhere. 2 Information of interest extracted from included studies were: study authors, country, setting (hospitalized or community subjects), time span, type of drug (HCQ or HQ with dosages and potential concurrent agents), type of comparator, indications (prevention or treatment), main characteristics of enrolled patients, sample size, safety populations (the group of individuals receiving at least one dose of intervention or comparator drug and that was evaluated for any safety-related analysis), number of total AEs in each group and specific AE according to study's definition. Eventually, data regarding the main outcome of each RCT were extracted as well.
The quality of included studies was graded by means of the Rob2 tool (Revised tool for Risk of Bias in randomized trials). 4 Odds ratios (OR) with 95% confidence intervals (CI) were calculated as summary statistics. The pooled OR was computed with the Mantel-Haenszelmethod in the framework of a random-effects model for dichotomous data. The number needed to harm (NNH) was calculated for each study as well, as inverse of the absolute risk increase.
Statistical heterogeneity was measured though I 2 statistic and was quantified as low, moderate, and high, with upper limits of 25%, 50%, and 75% for I 2 , respectively. To evaluatepotential publication bias, Egger's test was carried out to estimate asymmetry of the funnel plot in case at least 10 studies were retrieved. Sensitivity analyses were performed to assess robustness of findings, in particular resorting to leave-one-out method to gauge the impact of each study. A 95% prediction interval was also computed, to define the range of values where observations from future studies will fall. 5 Results were deemed as statistically significant at p < .05. Analyses were performed through Review Manager 5.4 (Cochrane Collaboration).

| RESULTS
A total of five articles, 6-10 all investigating only HCQ and not CQ, were included in the quantitative analysis from the 417 records screened (the selection process is illustrated in Supporting Information Material S3). The number of examined patients was 2291. Study features are summarized in Table 1. Main outcomes were clinical or virologic efficacy of HCQ and in no case it was superior than comparator.
Overall, the proportion of AE was 43.2% (535/1238) in the HCQ group versus 16.8% (177/1053) in the comparator group. The use of HCQ was associated with higher risk of AE compared with placebo or standard of care: OR: 4.57, 95% CI: 2.14-9.45, with relevant heterogeneity (I 2 = 92%), as described in Figure 1. Funnel plot was not drawn due to the low number of included studies. The type of most common AE varied but the ones involving the gastrointestinal tract were the most frequent. When singled out specific AE, risk was always higher among subject receiving HCQ, but statistical significance was met only for nausea and skin reactions ( Table 2). In detail, OR for experience nausea was 3.31 (95% CI: 2.38-4.61), skin reaction was 3.71 (95% CI: 1.11-12.43). Concerning serious AEs, OR was 1.11 (95% CI: 0.45-2.79). NNH was quite low, ranging from 1.6 to 6.
Quality assessment was described in Supporting Information Material S4: the risk of bias was low in most studies. Of note, three of them were open-label, 6,8,10 being the other two double-blind. 7,9 Sensitivity analysis confirmed the robustness of the main finding: the omission of whichever study, including the largest one 7 or the one with the greatest OR, 8

| DISCUSSION
This is the first systematic review focusing on safety profile of HCQ/CQ in the context of COVID-19 summarizing data only from high-quality RCTs. The main finding was the higher risk of AE (more than fourfold) in individuals receiving HCQ compared with placebo or standard of care. Although the magnitude of the effect across the studies varied, the direction of effect in all RCTs pointed to a disadvantageous safety profile of the 4-aminoquinoline derivative unambiguously. Gastrointestinal AEs were the most frequently reported but among the most clinically relevant change in vision, arrythmia, and neurological manifestation stood out. The biological plausibility of these events is underpinned by sound data from both historical cohorts and worldwide pharmacovigilance. 11 Results are coherent with a previous evidence synthesis collecting data only from RCTs, but not limited to COVID-19, being focused on other settings such as rheumatological diseases characterized by different HCQ schedules and including fewer studies in the framework of treatment or prevention of SARS-CoV-2 infection. 12 Actually, the hypothesis that HCQ couldalter viral growth of SARS-CoV-2 was based on weak evidence and inconsistent results from other viral infections in vitro, in animals, and in clinical studies. 13 Additionally, the likelihood of benefit, on the basis of poor mechanistic reasoning, is quite limited, also taking into account the high doses used so far in COVID-19, increasing the risk of AEs and of hazardous interactions with commonly used drugs such as macrolide antibiotics. 14 Indeed, a very recent metaanalysis of both RCTs and observational studies demonstrated that HCQ alone was not associated with reduced death risk in hospitalized COVID-19 patients but its combination with azithromycin significantly increased mortality (relative risk 1.2, 95% CI: 1.04-1.54). 13 The hype generated by very first observational studies on HCQ was quite soon contested by the emergence of serious flaws in design and analysis of these works, calling for high-quality RCTs to properly assess the clinical usefulness of the drug. 15 The RCTs included in this meta-analysis included nonsevere COVID-19 patients (or healthy subjects who were at risk of being close contact of confirmed cases), thus it is reasonable to affirm that the risk of AEs could be even amplified in more fragile individuals, the ones with serious forms of disease.
Notwithstanding, some limitations of the present study need to be underlined. For starters, studies investigating CQ were not retrieved, therefore, all results and findings are related to HCQ only. Then, populations were heterogeneous: COVID-19 patients (although the most suffering from mild or moderate disease) and subjects at risk of SARS-CoV-2 infection, having different baseline likelihood to experience AE. Heterogeneous were doses and durations of HCQ as well. Comparators were different and not rarely HCQ was used in combination with other agents such as azithromycin in one study, 10 and, in that case, it was impossible to disentangle the effects of HCQ from the one of companion drugs. Furthermore, not all information could be pooled due to inconsistent definition of AE, limiting the number of feasible analyses and so potentially explaining why most results did not reach statistically significance. Moreover, the profile of AEs was very broad, also including minor occurrences usually not interfering with medicine adherence, drug efficacy, daily activities and not seriously threatening patients' health. Open-label studies were potentially prone to ascertainment bias at least with F I G U R E 1 Odds ratio of adverse events between subjects receiving HCQ and the ones receiving the comparator. HCQ, hydroxycloroquine regard to subjective AEs. Eventually, the prediction interval was very wide, even including values conferring a protective role of HCQ toward AEs versus comparator and so lowering the confidence in summary estimates, but this can mirror the clinical heterogeneity of the studies.
Nevertheless, this kind of heterogeneity, regarding patients' features and protocols, involve neither methodological nor statistical aspects: all works were RCTs and magnitude as well as direction of the effect were consistent across studies.
In conclusion, given the limited and controversial, if any, benefit linked with HCQ when used to treat or prevent SARS-CoV-2 infection, the burden of AE overwhelms the potential advantages. HCQ should be used solely in the framework of clinical trials for COVID-19. Routine use in clinical practice currently should be avoided on the basis of an unfavourable trade-off between benefits and harms.