Antibiotics for the treatment of COVID‐19

Summary of findings 1. Azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
Patient or population: people with moderate to severe disease (WHO scale 4 to 9) Setting: inpatient Intervention: azithromycin Comparison: placebo or standard of care
Outcomes	Risk with placebo or standard of care	Risk with azithromycin	Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
All‐cause mortality at day 28	223 per 1000	219 per 1000 (201 to 236)	RR 0.98 (0.90 to 1.06)	8600 (4 RCTs)	⊕⊕⊕⊕ High^a	Azithromycin has little or no effect on all‐cause mortality at day 28
Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28	261 per 1000	248 per 1000 (227 to 269)	RR 0.95 (0.87 to 1.03)	7311 (1 RCT)	⊕⊕⊕⊖ Moderate^b	Azithromycin probably has little or no effect on worsening of clinical status or death at day 28
Improvement of clinical status: participants discharged alive at day 28	672 per 1000	645 per 1000 (564 to 746)	RR 0.96 (0.84 to 1.11)	8172 (3 RCTs)	⊕⊕⊕⊖ Moderate^c	Azithromycin probably has little or no effect on improvement of clinical status at day 28
Quality of life at longest follow‐up available	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at quality of life.
Serious adverse events during the study period	214 per 1000	238 per 1000 (190 to 300)	RR 1.11 (0.89 to 1.40)	794 (4 RCTs)	⊕⊕⊕⊖ Moderate^d	Azithromycin probably has little or no effect on serious adverse events during the study period
Any adverse events during the study period	333 per 1000	400 per 1000 (306 to 523)	RR 1.20 (0.92 to 1.57)	355 (3 RCTs)	⊕⊕⊖⊖ Low^e	Azithromycin may increase any adverse events slightly during the study period
Cardiac arrhythmias during the study period	45 per 1000	41 per 1000 (33 to 52)	RR 0.92 (0.73 to 1.15)	7865 (4 RCTs)	⊕⊕⊕⊖ Moderate^f	Azithromycin probably has little or no effect on cardiac arrhythmias during the study period
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk on the comparison group and the relative effect of the intervention (and its 95% confidence interval). CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is the possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aNo change of result in sensitivity analysis (RR 1.04; 95% CI 0.82 to 1.31; 837 participants; 3 studies) excluding studies with mixed populations (negative or unknown RT‐PCR). Therefore, no downgrading of evidence for indirectness. ^bDowngrade by one level for serious indirectness due to the effect estimate based on only one study with a mixed population (8.5% participants with negative or unknown RT‐PCR). ^cDowngrade by one level for serious heterogeneity (I² = 49%). No change of result in sensitivity analysis (RR 0.89; 95% CI 0.65 to 1.21; 402 participants; 2 studies) excluding studies with mixed populations (negative or unknown RT‐PCR). Therefore, no downgrading of evidence for indirectness. ^dDowngrade by one level for serious risk of bias due to an as‐treated analysis in the study with the highest weight (97.0%). The certainty of evidence was not downgraded due to indirectness, even if the imprecision of the effect estimate had increased in the sensitivity analysis excluding the study with a mixed population (RR 0.98; 95% CI 0.26 to 3.60; 355 participants; 3 studies), as the same study was already the reason for downgrading due to risk of bias. ^eDowngrade by one level for serious risk of bias due to non‐blinded outcome assessment in the study with the highest weight (87.7%), and one level for serious imprecision due to few small studies. ^fDowngrade by one level for serious indirectness due to the effect estimate based mainly (weight 99.1%) on two studies with mixed populations (8.5% and 11% participants with negative or unknown RT‐PCR).

Summary of findings 2. Azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
Patient or population: people with mild disease (WHO scale 1 to 3) Setting: outpatient Intervention: azithromycin Comparison: placebo or standard of care
Outcomes	Risk with placebo or standard of care	Risk with azithromycin	Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
All‐cause mortality at day 28	2 per 1000	2 per 1000 (0 to 31)	RR 1.00 (0.06 to 15.69)	876 (3 RCTs)	⊕⊕⊖⊖ Low^a	Azithromycin may have little or no effect on all‐cause mortality at day 28
Admission to hospital or death within 28 days	67 per 1000	63 per 1000 (38 to 105)	RR 0.94 (0.57 to 1.56)	876 (3 RCTs)	⊕⊕⊖⊖ Low^b	Azithromycin may have little or no effect on admission to hospital or death within 28 days
All initial symptoms resolved (asymptomatic) at day 14	927 per 1000	955 per 1000 (881 to 1038)	RR 1.03 (0.95 to 1.12)	138 (1 RCT)	⊕⊕⊖⊖ Low^c	Azithromycin may have little or no effect on symptom resolution (all initial symptoms resolved) at day 14
Duration to symptom resolution	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at duration to symptom resolution.
Quality of life at longest follow‐up available	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at quality of life.
Any adverse events during the study period	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at any adverse events during the study period.
Serious adverse events during the study period	Two studies assessed serious adverse events during the study period, but none of the participants in either group were affected		Not estimable	454 (2 RCTs)	⊕⊖⊖⊖ Very low^d	We are uncertain whether azithromycin increases or reduces serious adverse events.
Cardiac arrhythmias during the study period	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at cardiac arrhythmias during the study period
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk on the comparison group and the relative effect of the intervention (and its 95% confidence interval). CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is the possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in that study contributing events to the analysis; and by one level for serious imprecision due to a wide confidence interval and few events. ^bDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in one study and lack of registering the outcome in another study; and by one level for serious imprecision due to a wide confidence interval. ^cDowngrade by two levels for very serious imprecision due to the effect estimate based on one small study. ^dDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in one study and lack of registering the outcome in the other study; and by two levels for very serious imprecision due to zero events in both studies.

Background

This work is part of a series of Cochrane Reviews investigating treatments and therapies for coronavirus disease 2019 (COVID‐19). Reviews of this series share information in the background section and methodology based on the first published reviews about monoclonal antibodies (Kreuzberger 2021b), convalescent plasma (Chai 2020), and ivermectin (Popp 2021) from the German research project “CEOsys” (COVID‐19 Evidence‐Ecosystem).

Description of the condition

COVID‐19 is a rapidly spreading infectious disease caused by severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2; WHO 2020a). On March 22, 2020 the World Health Organization (WHO) declared the current COVID‐19 outbreak a pandemic. Severity of this disease can be classified according to a standardised scale established by the WHO which is also used throughout this review (Marshall 2020). COVID‐19 is unprecedented in comparison to previous coronavirus outbreaks, such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS), with 813 and 858 deaths, respectively (WHO 2003; WHO 2019a). Despite intensive international efforts to contain its spread, as of July 2021, COVID‐19 has resulted in more than 180 million confirmed cases and over 3.9 million deaths worldwide (WHO 2021a). In the same time, the emergence of variants of SARS‐CoV‐2, with potential for altered transmission or disease characteristics and vaccine and therapeutics effectiveness could impact established and proven disease control methods (WHO 2021b).

Vaccination has been shown to be highly effective at reducing severe illness and death from COVID‐19. More than 2.95 billion doses of COVID‐19 vaccines have been administered at the global level, as of July 2021 (WHO 2021a). Additional vaccine candidates are in development (WHO 2020c). However, the majority of vaccines have been administered ina few high‐income countries. The majority of the world's population still remains susceptible to SARS‐CoV‐2 infection and is at risk of developing COVID‐19. Moreover, the duration and degree of protection against the disease, but also against infection and transmission is still not well‐defined, and vaccine hesitancy poses direct and indirect threats to health (Grubaugh 2020). Besides unequal access to vaccines, evidence is available indicating a significant impact of certain circulating variants of SARS‐CoV‐2 on immunity that is likely to have an impact on the epidemiological situation.

Specific risk factors for severe disease, hospitalisation and mortality have been identified: older age, certain underlying medical conditions, such as cancer, chronic kidney disease, chronic lung disease including chronic obstructive pulmonary disease (COPD), moderate‐to‐severe asthma, interstitial lung disease, cystic fibrosis, and pulmonary hypertension, dementia or other neurological conditions, diabetes mellitus, Down syndrome, heart conditions, HIV infection, immunocompromised state, liver disease, overweight and obesity, Sickle cell disease or thalassemia, smoking, solid organ or blood stem cell transplant, stroke or cerebrovascular disease, substance use disorders, and pregnancy can increase the risk for severe disease (Huang 2020; Liang 2020; WHO 2020a; Williamson 2020). Reported COVID‐19 case fatality ratios varied widely between countries and reporting periods (from 0% to more than 25%, Johns Hopkins University & Medicine). However, these numbers may be misleading as they tend to overestimate the infection fatality ratio due to varying testing frequency, lag in reporting dates, incomplete capturing of all cases, and variations in case definitions since the beginning of the pandemic (WHO 2020b).

The median incubation period is estimated to be between five and six days and 97.5% of symptomatic cases develop symptoms within 11.5 days of exposure (Lauer 2020). Sore throat, cough, fever, headache, fatigue, and myalgia (muscle pain)or arthralgia (joint pain) are the most commonly reported symptoms (Struyf 2020). Other symptoms include dyspnoea, chills, nausea or vomiting, diarrhoea and nasal congestion (WHO 2020a). The majority of infected people develops mild symptoms, not requiring hospitalisation or remains completely asymptomatic (approximately 80% to 90%), depending on the time of the investigation, the cohort investigated and the virus variant (Chen 2020; Funk 2021; Pan 2020; Wu 2020). The reported frequency of asymptomatic courses also varies greatly and ranges between 6% and 96% (Buitrago‐Garcia 2020; Funk 2021; Oran 2020). A smaller proportion is affected by severe (approximately 11% to 20%) or critical (approximately 1% to 5%) disease with hospitalisation and intensive care unit (ICU) admittance due to respiratory failure, septic shock or multiple organ dysfunction syndrome (Funk 2021; Wu 2020). In light of the extent of the COVID‐19 pandemic with its pressure on health systems, especially in the face of evolving virus variants with potential increased transmissibility and altered disease characteristics, the ongoing scarcity of effective treatments, and the low global vaccination coverage, there is an urgent need for effective and safe therapies to save lives and to reduce the burden on healthcare systems.

Description of the intervention

During the current COVID‐19 pandemic, several antimicrobials have been investigated as treatment options for COVID‐19. International clinical trials are evaluating the effect of repurposed drugs including certain antibiotics. The effect of antibiotics with potential antiviral and anti‐inflammatory properties such as azithromycin are being investigated as treatment for a viral, and not treatment or prophylaxis of primary bacterial superinfections.

Most randomised trials and observational studies which evaluated the efficacy and safety of antibiotics studied macrolides such as azithromycin and clarithromycin. Macrolides are widely available and their safety profile is well‐established. The antibacterial mechanism of action involves binding to the bacterial ribosome leading to inhibition of bacterial protein synthesis. The most common adverse reactions encountered with macrolides include gastrointestinal discomfort such as nausea, diarrhoea and abdominal pain. A potential adverse reaction is a clinically relevant QT‐prolongation predisposing to cardiac arrhythmias. Especially when azithromycin is combined with other drugs known to prolong QT, like hydroxychloroquine.
. Established dosing regimens range from 250 mg to 500 mg per day, administered orally or intravenously (FDA 2021). Dosing of clarithromycin in bacterial infections is 500 mg to 1000 mg per day, administered orally or intravenously. Besides their frequent use in acute bacterial infections, including community‐acquired respiratory tract infections, macrolides are sometimes used for their anti‐inflammatory, immunomodulatory effects in chronic inflammatory diseases (Hansen 2019). Macrolides belong to antibiotic classes that have higher resistance potential than others. The WHO advice is to prioritise these medicines as key targets of stewardship program and monitoring to reduce antimicrobial resistance (AMR) (WHO 2019b). Macrolide use during the pandemic is being discussed as a relevant contributor to the spread of AMR which is an ongoing problem that has implications for global health and the world economy (WHO 2020d).

Another antibiotic studied as treatment for COVID‐19 is the broad‐spectrum antibiotic doxycycline, a long‐acting tetracycline. Tetracyclines exhibit bacteriostatic properties and also inhibit the bacterial protein synthesis. The wide range of activity includes gram‐positive bacteria, gram‐negative bacteria, intracellular organisms, and protozoan parasites. Doxycycline is an important antimalarial agent (Krakauer 2003). Tetracyclines are generally well‐tolerated. The most common adverse reactions include gastrointestinal side effects and increased photosensitivity occurring in sun‐exposed areas. Established dosing regimens range from 100 mg to 200 mg per day orally or intravenously (FDA 2021). It is commonly used in many areas of the world.

Finally, lincosamides are also being studied in the context of COVID‐19. Lincomycin is pharmaceutically similar to clindamycin, but has no therapeutic advantages over clindamycin, and is rarely used in infections caused by susceptible bacterial strains in penicillin‐allergic patients. Dose regimens of 600 mg two to three times daily by intravenous or intramuscular route, or lower oral dosing are considered acceptable (DrugBank Online; FDA 2021). Lincosamides have a similar mechanism of action as macrolides. Lincomycin has largely been superseded by its semisynthetic derivative clindamycin due to higher efficacy. In the EU lincomycin is approved for veterinary use only. However, it is still used in human medicine in the US, Canada and Russia amongst others (DrugBank Online; EMA 1998).

How the intervention might work

There are several rationales for using antibiotics in the context of COVID‐19. The most intuitive is using antibiotics to treat secondary bacterial infections or bacterial co‐infections which antibiotics could effectively treat. However, this review focuses on using antibiotics as repurposed treatment options with direct activity against the SARS‐CoV‐2 virus and a reduction of the pro‐inflammatory immune response through anti‐inflammatory properties.

Macrolides have been shown to have in‐vitro antiviral activity against a range of viruses and have been reported to inhibit SARS‐CoV‐2 replication in vitro. Their anti‐inflammatory and immunomodulatory activity triggered their use in patients with COVID‐19. Especially azithromycin is being used in patients with chronic pulmonary inflammatory conditions to decrease inflammation and exacerbation frequency, e.g. panbronchiolitis, chronic obstructive pulmonary disease or cystic fibrosis. The beneficial effect is proposed to be based on down‐regulating inflammatory mediators like interleukin 6 (IL‐6), interleukin 1beta (IL‐1ß) and tumour necrosis factor (TNF) (Pani 2020; Shinkai 2008). Further, in vitro activity inhibiting virus replication was observed for influenza H1N1, Ebola and Zika virus (Al‐Horani 2020). The precise mechanism is unknown, but theories include the drug's capacity of creating an acidic environment that blocks endocytosis, release of viral content after endocytosis or interfering with the SARS‐CoV‐2 spike protein mediated cell entry, all mechanisms would ultimately limit virus replication (Al‐Horani 2020). Similarly, clarithromycin is suggested to potentially inhibit virus replication and correct the immune dysregulation in critically ill patients (Al‐Horani 2020; Shinkai 2008).

Doxycycline is being tested for its benefit in COVID‐19 patients as well. On the one hand, it has also been suggested to suppress pro‐inflammatory cytokine production, especially IL‐6. On the other hand, it inhibits certain enzymes' activity that normally help the coronavirus entering and replicating in host cells (Al‐Horani 2020; Krakauer 2003).

For lincosamides, especially lincomycin, less research is available on activity outside its antibacterial effect. However, it is suggested impeding hyperinflammation by blocking neutrophil chemotaxis and down‐regulating cytokines (Spížek 2004).

Why it is important to do this review

There is a need for evidence‐based information to guide clinical decision‐making for COVID‐19 patients. Current treatment consists of supportive care with oxygen therapy in cases with moderate disease, and with respiratory support as mechanical ventilation, and extracorporeal membrane oxygenation in cases with severe disease (CDC 2020; WHO 2020b). Overall, data from randomised controlled trials (RCTs) do not demonstrate a clear, major clinical benefit with most drugs evaluated so far. Data from RCTs at this stage support the role of corticosteroids for severe COVID‐19 and clinical guidelines recommend their use (National COVID‐19 Clinical Evidence Taskforce 2021; Siemieniuk 2020). Further, tocilizumab is recommended for certain patient groups, while other drugs, such as hydroxychloroquine, are not recommended for the treatment of COVID‐19 (National COVID‐19 Clinical Evidence Taskforce 2021; Siemieniuk 2020). Furthermore, evidence on the efficacy of anti‐SARS‐CoV‐2 monoclonal antibodies as treatments for COVID‐19 patients is being generated (Kreuzberger 2021b). Management strategies for most patient subgroups are still to be identified.

The idea of drug repurposing follows the desire and need to rapidly find an effective treatment against COVID‐19 as the drug candidates are mostly well‐investigated in other clinical scenarios and thus adverse effects are well‐known. However, it has to be kept in mind that every drug has side effects and possible negative consequences. Safety and potential collateral damage have to be elucidated just as much as the potential efficacy. Antibiotic use was already excessive before 2020 resulting in increased antimicrobial resistance (AMR) rates worldwide (Russell 2021). The WHO has declared that AMR is one of the top 10 global public health threats facing humanity (WHO 2020d). As long as the current body of evidence regarding the efficacy of antibiotics as antiviral treatment option for COVID‐19 is not completely mapped, the uncontrolled use of antibiotics in COVID‐19 patients could enhance this problem even more.

Extensive work in the field of systematic reviews for interventions to treat COVID‐19 has already been undertaken, including treatment with antibiotics (Kamel 2021; Mangkuliguna 2021; Sharma 2021). However, none fulfilled all quality standards for evidence syntheses such as risk of bias assessment and the GRADE approach.

Therefore, we aim to provide a complete evidence profile based on current Cochrane standards for antibiotics with regard to efficacy and safety as treatment of COVID‐19.

Objectives

To assess the efficacy and safety of antibiotics compared to each other, standard of care alone, placebo, or any other active intervention with proven efficacy for treatment of COVID‐19 outpatients and inpatients.

Methods

Criteria for considering studies for this review

Types of studies

The main description of methods is based on the standard template of the Cochrane Haematology review group and is in line with a series of Cochrane Reviews investigating treatments and therapies for COVID‐19 (e.g. Kreuzberger 2021b; Stroehlein 2021; Popp 2021). The original review protocol for this review was registered with PROSPERO on January 25, 2021 (CRD42021233062). As this review and the other reviews of the Cochrane Review series are living systematic reviews during the COVID‐19 pandemic, specific adaptions related to the research question, including participants, interventions, comparators, outcomes, and methods were necessary. Relevant protocol changes are transparently reported in the section Differences between protocol and review.

To assess the efficacy and safety of antibiotics for the treatment of people with COVID‐19, we included randomised controlled trials (RCTs), as this study design, if performed appropriately, provides the best evidence for experimental therapies in highly controlled therapeutic settings. We used the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2020a). Non‐standard RCT designs, such as cluster‐randomised and cross‐over trials, were not eligible for the review. These designs are not appropriate in this context, since the underlying cause of COVID‐19 is an infection with the SARS‐CoV‐2 virus and the medical condition evolves over time.

We excluded controlled non‐randomised studies of intervention and observational studies. We also excluded animal studies, pharmacokinetic studies, and in vitro studies.

We included the following formats, if sufficient information was available on study design, characteristics of participants, interventions and outcomes.

Full‐text journal publications
Preprint articles
Abstract publications
Results published in trial registries
Additional personal communication with investigators, if results were available in any of the above listed format

We included preprints and conference abstracts to have a complete overview of the ongoing research activity, especially for tracking newly emerging studies about antibiotics as antiviral and anti‐inflammatory treatments for COVID‐19. We did not apply any limitation with respect to the length of follow‐up or language of the publication.

Types of participants

We included studies investigating adults with a confirmed diagnosis of COVID‐19 (reverse transcription polymerase chain reaction (RT‐PCR) or antigen testing) irrespective of age, gender, ethnicity, disease severity or setting. If studies included participants with a confirmed or suspected COVID‐19 diagnosis, we used only the data for the patient population with confirmed COVID‐19 diagnosis. In cases, where data were not reported separately for patients with confirmed or suspected COVID‐19 diagnosis, we included the mixed population. The participants' status of included studies including the type of COVID‐19 diagnosis is reported in the Characteristics of included studies. If mixed population studies contributed data to meta‐analyses, we excluded these studies in sensitivity analyses (Sensitivity analysis) to test the robustness of the results.

We excluded studies that evaluated antibiotic treatment for other coronavirus diseases such as SARS or MERS, or other viral diseases, such as influenza. If studies enrolled populations with or exposed to mixed viral diseases, we had planned to only include studies if trial authors provided subgroup data for SARS‐CoV‐2 infection.

Types of interventions

Any antibiotic given with antiviral and anti‐inflammatory intent was eligible as an intervention for this review, e.g. azithromycin, lincomycin, doxycycline, and clarithromycin. All doses and regimens were eligible. We categorised dose categories into usual dose as recommended for other infectious diseases (FDA 2021), low dose (< usual dose), and high dose (> usual dose) for the used antibiotics as following.

Azithromycin usual dose:
- oral route: 250 to 500 mg once daily; low dose < 250 mg; and high dose: > 500 mg;
- intravenous route: 500 mg daily; low dose < 500 mg; and high dose > 500 mg.
Clarithromycin usual dose: 250 mg twice daily by the oral or intravenous route; low dose < 500 mg daily; and high dose > 500 mg daily.
Doxycycline usual dose: 200 mg daily followed by 100 mg daily by the oral route or intravenous route; low dose < 200 mg and 100 mg daily, respectively; and high dose > 200 mg and 100 mg daily, respectively;
Lincomycin usual dose: 600 mg twice daily by the intravenous or intramuscular route; low dose < 1200 mg; and high dose > 1200 mg. Oral use is not recommended by the FDA any more.

We compared antibiotics with each other, standard of care alone, or placebo. Placebo and standard of care alone were treated as the same control group, as well as standard of care at different institutions and time points during the pandemic. Standard of care (co‐interventions) had to be comparable between intervention and control groups. We have planned to compare any antibiotic with any other active pharmacological comparator with proven efficacy for the treatment of COVID‐19. While for this review version this considers mainly dexamethasone and remdesivir, more interventions with proven efficacy might become available for future updates. For dexamethasone, it has been shown that mortality from COVID‐19 was lower among people who were randomised to receive dexamethasone than among those who received the usual standard of care (Siemieniuk 2020; RECOVERY 2021). Remdesivir showed some benefit for people hospitalised with COVID‐19, though to a lesser extent (Beigel 2020). For patients that qualify for e.g. dexamethasone therapy or for another intervention that proves to be beneficial in the future, it would be unethical to further conduct trials that use placebo only. In contrary, studies using comparators without proven efficacy, e.g. hydroxychloroquine, may confound the assessment of the efficacy or safety of antibiotics and were excluded. Although those kinds of interventions were possibly used at a certain point of time during the pandemic with the best intentions, their use was never supported by actual evidence, and they have potential side effects as well (Singh 2021). From those comparisons no reliable evidence can be obtained, therefore, active comparators without proven efficacy were not eligible for this review. For the current review, we found no study using an active comparator with proven efficacy.

We excluded studies evaluating antibiotics in combination with other active treatments, if the same treatment was not used in the control group. We also excluded studies investigating the efficacy and safety to prevent COVID‐19.

We created the following comparisons.

Any antibiotic versus placebo/standard of care alone.
Any antibiotic versus other antibiotic.
Any antibiotic vs active pharmacological intervention for the treatment of COVID‐19 with proven efficacy (no studies were available for the current review version).

Types of outcome measures

We evaluated core outcomes in accordance with the Core Outcome Measures in Effectiveness Trials (COMET) Initiative for COVID‐19 patients (COMET 2020; Marshall 2020), and additional outcomes that have been prioritised by consumer representatives and the German guideline panel for inpatient therapy of people with COVID‐19. The current outcome set differed between previous protocols and reviews and the current review. Changes to the outcomes were necessary due to the risk of competing events associated with the original outcome set. All differences are reported in the section Differences between protocol and review.

We defined outcome sets with primary and secondary outcomes for two populations\;

hospitalised individuals with moderate to severe COVID‐19; and
ambulatory managed individuals with asymptomatic or mild COVID‐19.

Primary outcomes were used to inform the summary of findings tables.

Inpatients with moderate to severe COVID‐19

All‐cause mortality at day 28, day 60, time‐to‐event, and at hospital discharge.
Clinical status at day 28, day 60, and up to the longest follow‐up, including:
- Worsening of clinical status:
  - participants with clinical deterioration (new need for invasive mechanical ventilation) or death.
- Improvement of clinical status:
  - participants discharged alive. Participants should be discharged without clinical deterioration or death.
Quality of life, including fatigue and neurological status, assessed with standardised scales (e.g. WHOQOL‐100) at up to 7 days; up to 28 days, and longest follow‐up available.
Serious adverse events during the study period, defined as number of participants with any event.
Adverse events (any grade) during the study period, defined as number of participants with any event.
Cardiac arrhythmias during the study period; cardiac arrhythmias included any event that was classified as such by the individual study definition. We reported on the incidence of QT prolongation narratively.

Outpatients with asymptomatic or mild COVID‐19

All‐cause mortality at day 28, day 60, time‐to‐event, and up to the longest follow‐up
Admission to hospital or death within 28 days
Symptom resolution:
- all initial symptoms resolved (asymptomatic) at day 14, day 28, and up to the longest follow‐up;
- duration to symptom resolution.
Quality of life, including fatigue and neurological status, assessed with standardised scales (e.g. WHOQOL‐100) at up to 7 days, up to 28 days, and longest follow‐up available.
Serious adverse events during the study period, defined as number of participants with any event.
Adverse events (any grade) during the study period, defined as number of participants with any event.
Cardiac arrhythmias during the study period; cardiac arrhythmias included any event that was classified as such by the individual study definition. We reported on the incidence of QT prolongation narratively.

Timing of outcome measurement

We collected information on outcomes from all time points reported in the publications. If only a few studies contributed data to an outcome, we pooled different time points, provided the studies had produced valid data and pooling was clinically reasonable.

In case of time‐to‐event analysis, e.g. for time to death, we included the outcome measure based on the longest follow‐up time and measured from randomisation.

We included serious adverse events and adverse events occurring during the study period, including adverse events during active treatment and long‐term adverse events as well. If sufficient data were available, we grouped the measurement time points of eligible outcomes, for example, adverse events and serious adverse events, into those measured directly after treatment (up to 7 days after treatment), medium‐term outcomes (up to 14 days after treatment) and longer‐term outcomes (more than 28 days after treatment).

Inpatients with moderate to severe COVID‐19

Additional outcomes for efficacy of antibiotics (not included in summary of findings table)

Clinical status at day 28, day 60, and up to the longest follow‐up, including:
- Worsening of clinical status:
  - new need for invasive mechanical ventilation;
  - new need for non‐invasive mechanical ventilation or high flow;
  - new need for oxygen by mask or nasal prongs.
- Improvement of clinical status:
  - weaning or liberation from invasive mechanical ventilation in surviving patients;
  - Ventilator‐free days;
  - duration to liberation from invasive mechanical ventilation;
  - liberation from supplemental oxygen in surviving patients;
  - duration to liberation from supplemental oxygen.
Need for dialysis at up to 28 days.
Admission to the intensive care unit (ICU) at day 28.
Duration of hospitalisation.
Viral clearance, assessed with reverse transcription polymerase chain reaction (RT‐PCR) test for SARS‐CoV‐2 at baseline, up to 3, 7, and 14 days
Hospital‐acquired infections at up to 28 days.

Outpatients with asymptomatic or mild COVID‐19 (outpatients)

Additional outcomes for efficacy of antibiotics (not included in summary of findings table)

Clinical status at day 28 and up to the longest follow‐up:
- worsening of clinical status (moderate to severe COVID‐19 symptoms):
  - need for invasive mechanical ventilation;
  - need for non‐invasive mechanical ventilation or high flow;
  - need for hospitalisation with need for oxygen by mask or nasal prongs;
  - need for hospitalisation without oxygen therapy.
Viral clearance, assessed with RT‐PCR for SARS‐CoV‐2 at baseline, up to 3, 7, and 14 days.
Hospital‐acquired infections at up to 28 days.

Search methods for identification of studies

Electronic searches

On 14 June 2021 our Information Specialist (MIM) conducted systematic searches in the following sources from the inception of each database to 14 June 2021 (date of last search for all databases) and did not place restrictions on the language of publication:

Cochrane COVID‐19 Study Register (CCSR) (www.covid-19.cochrane.org), comprising:
- MEDLINE (PubMed), daily updates;
- Embase, weekly updates;
- ClinicalTrials.gov (www.clinicaltrials.gov), daily updates,
- World Health Organization International Clinical Trials Registry Platform (ICTRP) (www.who.int/trialsearch), weekly updates,
- medRxiv (www.medrxiv.org), weekly updates;
- Cochrane Central Register of Controlled Trials (CENTRAL), monthly updates.
Web of Science Core Collection:
- Science Citation Index Expanded;
- Emerging Sources Citation Index.
WHO COVID‐19 Global literature on coronavirus disease (https://search.bvsalud.org/global-literature-on-novel-coronavirus-2019-ncov/).

For detailed search strategies, see Appendix 1.

Searching other resources

We searched for other potentially eligible studies or ancillary publications by searching the reference lists of included studies, systematic reviews and meta‐analyses. In addition, we contacted the investigators of included studies to obtain additional information on the retrieved studies.

We searched for grey literature, which we defined as searching study registries such as ClinicalTrials.gov and WHO ICTRP contained in the CCSR, as well as searching preprint servers and grey literature indexes contained in CCSR and WHO COVID‐10 Global Literature database. Once we established our set of included studies, we searched for preprints via Europe PMC, to check if any preprints for included studies were published since our database search.

Data collection and analysis

Selection of studies

In accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2020), two review authors (MP, MR) independently screened the results of the search strategies for eligibility for this review by reading the titles and abstracts using Covidence (Covidence). We then retrieved full‐text articles and assessed eligibility of the remaining records against pre‐defined eligibility criteria in duplicate. Discrepancies were resolved by discussion within the group of review authors. We included studies in the review irrespective of whether measured outcome data were reported in a ‘usable’ way. We collated multiple reports of the same study, so that the study, rather than the report, was the unit of interest in the review.

We documented the study selection process in a flow diagram, as recommended in the PRISMA statement (Moher 2009), and show the total numbers of retrieved references and the numbers of included, ongoing, excluded studies as well as awaiting classification. We listed all studies that we excluded after full‐text assessment and the reasons for their exclusion in the 'Characteristics of excluded studies' section, same procedure was conducted for the studies awaiting classification.

Data extraction and management

We conducted data extraction according to the guidelines proposed by Cochrane (Li 2020). Two out of three review authors (MP, MR, SW) extracted data independently and in duplicate, using a customised data extraction form developed in Microsoft Excel (Microsoft 2018). We solved disagreements by discussion. If no agreement was obtained, a third review author was involved to resolve the disagreement.

Two review authors (MP, SW) independently assessed eligible studies in the process of study selection (as described above) for methodological quality and risk of bias. If the review authors were unable to reach a consensus, a third review author was consulted.

We extracted the following information, if reported.

General information: author, title, source, country, language, type of publication, publication date.
Study characteristics: setting and dates, inclusion/exclusion criteria, number of study arms, comparability of groups, treatment cross‐overs, length of follow‐up, funding.
Participant characteristics: number of participants randomized/received intervention/analysed, COVID‐19 diagnostics, severity of disease, age, gender, co‐morbidities (e.g. diabetes, immunosuppression), concurrent interventions, time since symptom onset.
Intervention: type of antibiotic, dose, frequency, duration and route of administration.
Control intervention: type of control, frequency, duration and route of administration.
Outcomes: as specified under Types of outcome measures.

Assessment of risk of bias in included studies

We used the risk of bias 2.0 (RoB 2) tool to analyse the risk of bias of study results contributing information to our primary outcomes (Sterne 2019). Of interest for this review was the effect of the assignment to the intervention (the intention‐to‐treat (ITT) effect), thus, we performed all assessments with RoB 2 on this effect. The outcomes that we assessed are the primary outcomes specified for inclusion in the summary of findings tables.

Two review authors (MP, SW) independently assessed the risk of bias for each outcome. In case of discrepancies among their judgements and inability to reach consensus, we consulted the third review author to reach a final decision. We assessed the following types of bias as outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2020b).

Bias arising from the randomisation process.
Bias due to deviations from the intended interventions.
Bias due to missing outcome data.
Bias in measurement of the outcome.
Bias in selection of the reported result.

To address these types of bias we used the signalling questions recommended in RoB 2 and make a judgement using the following options.

'Yes': if there is firm evidence that the question is fulfilled in the study (i.e. the study is at low or high risk of bias for the given the direction of the question).
'Probably yes': a judgement has been made that the question is fulfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).
'No': if there is firm evidence that the question is unfilled in the study (i.e. the study is at low or high risk of bias for the given the direction of the question).
'Probably no': a judgement has been made that the question is unfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).
'No information': if the study report does not provide sufficient information to allow any judgement.

We used the algorithms proposed by RoB 2 to assign each domain one of the following levels of bias.

Low risk of bias
Some concerns
High risk of bias

Subsequently, we derived an overall risk of bias rating for each pre‐specified outcome in each study in accordance with the following suggestions.

'Low risk of bias': we judge the trial to be at low risk of bias for all domains for this result.
'Some concerns': we judge the trial to raise some concerns in at least one domain for this result, but not to be at high risk of bias for any domain.
'High risk of bias': we judge the trial to be at high risk of bias in at least one domain for the result, or we judge the trial to have some concerns for multiple domains in a way that substantially lowers confidence in the results.

We used the RoB 2 version 2019. We used the RoB 2 Excel tool to implement RoB 2 (available on the website www.riskofbias.info/welcome/rob-2-0-tool/current-version-of-rob-2), and stored and presented our detailed RoB 2 assessments in the analyses section.

Measures of treatment effect

For dichotomous outcomes, we recorded the number of events and total number of participants in both treatment and control groups. We reported the pooled risk ratio (RR) with a 95% confidence intervals (CIs) (Deeks 2020).

For continuous outcomes, we recorded the mean, standard deviation and total number of participants in both treatment and control groups. Where continuous outcomes used the same scale, we performed analyses using the mean difference (MD) with 95% CIs (Deeks 2020). For continuous outcomes measured with different scales, we planned to perform analyses using the standardised mean difference (SMD) (Deeks 2020). For interpreting SMDs, we planned to re‐express SMDs in the original units of a particular scale with the most clinical relevance and impact. For the current review, all outcomes were measured on comparable scales.

If available, we extracted and report hazard ratios (HRs) for time‐to‐event outcomes (e.g. time to hospital discharge). If HRs were not available, we would have made every effort to estimate the HR as accurately as possible from available data using the methods proposed by Parmar and Tierney (Parmar 1998; Tierney 2007). Had sufficient studies provided HRs, we planned to use HRs rather than RRs or MDs in a meta‐analysis, as they provide more information.

Unit of analysis issues

The unit of analysis for this review is the individually‐randomised‐participant.

In studies with multiple intervention groups, we followed the recommendations in Chapter 6 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2020c). Studies with multiple treatment groups of the same intervention (i.e. dose, route of administration) were combined if study arms were sufficiently homogeneous. If arms could not be pooled, which was not the case for the current review, we had planned to compare each arm with the common comparator separately. For pairwise meta‐analysis, we had planned to split the ‘shared’ group into two or more groups with smaller sample size, and include two or more (reasonably independent) comparisons. For this purpose, in case of dichotomous outcomes, both the number of events and the total number of participants would have been divided, and for continuous outcomes, the total number of participants would have been divided with unchanged means and standard deviations (SDs).

Dealing with missing data

There are many potential sources of missing data in a systematic review or meta‐analysis, which can affect the level of studies, outcomes, summary data, individuals, or study‐level characteristics (Deeks 2020). Incomplete data can introduce bias into the meta‐analysis, if they are not missing at random. We addressed all sources of missing data. Missing studies may be the result of reporting bias, and we addressed this as described in the Assessment of reporting biases section. Missing outcomes and summary data may be the result of selective reporting bias; missing individuals may be the result of attrition from the study or lack of intention‐to‐treat analysis. We addressed these sources of missing data using the RoB 2 tool (Assessment of risk of bias in included studies). If data were incompletely reported, we contacted the study authors to request additional information.

Assessment of heterogeneity

We used the descriptive statistics reported in the Characteristics of included studies to assess whether the studies within each pairwise comparison are homogenous enough, with respect to study and intervention details and population baseline characteristics, that the assumption of homogeneity might be plausible. In case of excessive clinical heterogeneity, we did not pool the findings of included studies.
We measured statistical heterogeneity using the Chi² test and the I² statistic (Deeks 2020), and the 95% prediction interval (PI) for random‐effects meta‐analysis (IntHout 2016). The prediction interval helps in the clinical interpretation of heterogeneity by estimating what true treatment effects can be expected in future settings (IntHout 2016). We restricted calculation of a 95% PI to meta‐analyses with four or more studies (≥ 200 participants), since the interval would be imprecise when a summary estimate is based on only a few small studies. We used the open‐source statistical software R package meta to calculate 95% PIs (Meta). We declared statistical heterogeneity if the P value was less than 0.1 for the Chi² statistic, or the I² statistic was equal to or greater than 40% (40% to 60%: moderate heterogeneity; 50% to 90%: substantial heterogeneity; 75% to 100%: considerable heterogeneity, Deeks 2020), or the range of the 95% PI revealed a different clinical interpretation of the effect estimate compared to the 95% CI.

Assessment of reporting biases

We sought to identify all research that meets our predefined eligibility criteria. Missing studies can introduce bias to the analysis. We searched for completed non‐published trials in trials registers, contacted authors to seek assurance that the results will be made available, and classified them as 'awaiting classification' until the results are reported. We reported the number of completed non‐published trials.
We planned to investigate risk of reporting bias (publication bias) in pairwise meta‐analyses using contour‐enhanced funnel plots, when there were 10 or more relevant studies pooled in a meta‐analysis. In the current review, there are no meta‐analyses including 10 or more studies. For future review updates, if funnel plot asymmetry is suggested by a visual assessment, we plan to perform exploratory analyses (e.g. Rücker’s arcsine test for dichotomous data and Egger’s linear regression test for continuous data) to further investigate funnel plot asymmetry. A P value of less than 0.1 will be considered as the level of statistical significance. In future review updates, we will analyse reporting bias using the open‐source statistical software R package meta (Meta).

Data synthesis

We analysed trials including different severities of disease separately, grouping them into asymptomatic to mild, and moderate to severely ill, as these are different populations in different settings, resulting in different outcome sets (see Types of outcome measures). We analysed trials with the following participant populations separately, as follows.

Inpatients moderate to severe COVID‐19.
Outpatients with asymptomatic or mild COVID‐19.

For these populations, we created the following comparisons.

Any antibiotic versus placebo/standard of care alone.
Any antibiotic versus other antibiotic.
Any antibiotic vs active pharmacological intervention for the treatment of COVID‐19 with proven efficacy (no studies available for the current review version).

Placebo and standard of care alone were treated as the same intervention, as well as standard of care at different institutions and time points during the pandemic.

We performed meta‐analyses according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2020). If clinical and methodological characteristics of individual studies were sufficiently homogeneous, we pooled the data in meta‐analysis. We used the RevMan Web software for meta‐analyses (RevMan Web 2020). One review author entered the data into the software, and a second review author checked the data for accuracy. When meta‐analysis was feasible, we used the random‐effects model as we assume that the intervention effects are related but are not the same for the included studies. For dichotomous outcomes, we performed meta‐analyses using the Mantel‐Haenszel method under a random‐effects model to calculate the summary (combined) intervention effect estimate as a weighted average of the intervention effects estimated in the individual studies. For continuous outcomes, we used the inverse‐variance method.

We planned to present descriptive statistics only, if we deemed meta‐analysis inappropriate for a certain outcome because of heterogeneity or because of serious study limitations leading to considerably high risk of bias (e.g competing risk of death not taken into account in outcome measurement). This was not the case for the current review version.

If meta‐analysis was possible, we assessed the effects of potential biases in sensitivity analyses (see Sensitivity analysis) and considered investigating heterogeneity in subgroup analyses (see Subgroup analysis and investigation of heterogeneity). Subgroup analyses were not possible due to the low number of studies per outcome. For future review updates, if we could not find a cause for the heterogeneity, we had planned to not perform a meta‐analysis, but to comment on the results as a narrative with the results from all studies presented in tables.

Forest plots were used to visualise for meta‐analyses of primary outcomes only, including risk of bias assessment. Secondary outcomes without risk of bias assessment were reported in additional tables.

Subgroup analysis and investigation of heterogeneity

Subgroup analyses were restricted to the primary outcomes.

For hospitalised individuals with moderate to severe COVID‐19, we performed a subgroup analysis independent of heterogeneity and number of studies for the following characteristic, because we considered it essential to test the intervention for its impact in different stages of the disease.

Severity of condition at baseline: moderate (WHO 4 to 5) versus severe disease (WHO 6 to 9) as defined by the WHO Clinical Progression Scale. Studies providing data only for a mixed population including moderate and severe participants where included in the subgroup 'moderate to severe disease (WHO 4 to 9)'. Details of the patients' status at baseline were provided in the footnotes of the forest plots.

We investigated heterogeneity by visual inspection of the forest plot. Details of the intervention, comparator, and the population were reported for each study in the footnotes of the forest plot.

In future review updates, we will perform subgroup analyses to investigate heterogeneity for the following characteristics.

Dose of antibiotic (usual dose versus low dose versus high dose);
Route of administration (oral versus intravenous versus inhalational)
Age (children versus adults).

We had planned to explore heterogeneity by subgroup analysis to calculate RR or MD in conjunction with the corresponding CI for each subgroup, if sufficient studies had been available (at least 10 studies per outcome). In the current review, there are not enough studies available. In future review updates, we will perform subgroup analyses, if statistical heterogeneity is present (P < 0.1 for the Chi² test of heterogeneity, I² ≥ 50%, or a different clinical conclusion of 95% CI versus 95% PI). We planned to use the tests for interaction to test for differences between subgroup results.

Sensitivity analysis

We performed sensitivity analyses of the following characteristics for our primary outcomes.

Risk of bias assessment (only studies with a low risk of bias or some concerns).
Comparison of preprint articles versus peer‐reviewed articles (only studies published as journal articles).
Confirmed versus mixed (suspected and confirmed) COVID‐19 diagnosis (only studies/participants with confirmed COVID‐19 diagnosis).

Summary of findings and assessment of the certainty of the evidence

Summary of findings

We created separate summary of findings tables for the different patient populations and for the different comparisons with regard to the intervention and comparator, respectively. For the current review version, there was no study with active comparator available. We evaluated the certainty of the evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for interventions evaluated in RCTs.

We used the MAGICapp software to create summary of findings tables (MAGICapp). For time‐to‐event outcomes, we planned to calculate absolute effects at specific time points, as recommended in the GRADE guidance 27 (Skoetz 2020). For the current review there were no time‐to‐event data available. According to Chapter 14 of the updated Cochrane Handbook for Systematic Reviews of Interventions, the “most critical and/or important health outcomes, both desirable and undesirable, limited to seven or fewer outcomes” should be included in the summary of findings table(s) (Schünemann 2020). We included outcomes prioritised according to the Core Outcome Set for intervention studies (COMET 2020), and patient‐relevance. Those are the primary outcomes of this review.

Inpatients with moderate to severe COVID‐19

All‐cause mortality; all‐cause mortality at hospital discharge most favourable; if not reported all‐cause mortality day 60, followed by day 28, or time‐to‐event estimate, will be included in summary of findings table.
Worsening of clinical status at day 28:
- participants with clinical deterioration (new need for invasive mechanical ventilation) or death.
Improvement of clinical status at day 28:
- participants discharged alive.
Quality of life at longest follow‐up available.
Serious adverse events during the study period.
Any adverse events during the study period.
Cardiac arrhythmias during the study period.

Outpatients with asymptomatic or mild COVID‐19

All‐cause mortality; all‐cause mortality at longest follow‐up and > 60 days most favourable; if not reported all‐cause mortality day 60, followed by day 28, or time‐to‐event estimate, will be included in the summary of findings table.
Admission to hospital or death within 28 days
Symptom resolution:
- all initial symptoms resolved (asymptomatic) at day 14;
- duration to symptom resolution.
Quality of life at longest follow‐up available.
Serious adverse events during the study period.
Any adverse events during the study period.
Cardiac arrhythmias during the study period.

Assessment of the certainty in the evidence

We used the GRADE approach to assess the certainty in the evidence for the outcomes listed in the previous section.

The GRADE approach uses five domains (risk of bias, consistency of effect, imprecision, indirectness and publication bias) to assess the certainty in the body of evidence for each prioritised outcome.

We downgraded our certainty of evidence for:

serious (‐1) or very serious (‐ 2) risk of bias;
serious (‐1) or very serious (‐ 2) inconsistency;
serious (‐1) or very serious (‐ 2) uncertainty about directness;
serious (‐1) or very serious (‐ 2) imprecise or sparse data;
serious (‐1) or very serious (‐ 2) probability of reporting bias.

The GRADE system used the following criteria for assigning grade of evidence.

High: we are very confident that the true effect lies close to that of the estimate of the effect.
Moderate: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of effect, but there is a possibility that it is substantially different.
Low: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect.
Very low: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

We followed the current GRADE guidance for these assessments in its entirety as recommended in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 14 (Schünemann 2020).

We used the overall risk of bias judgement, derived from the RoB 2 Excel tool, to inform our decision on downgrading for risk of bias. We phrased the findings and certainty in the evidence as suggested in the informative statement guidance (Santesso 2020).

Methods for future updates

Living systematic review considerations

Our information specialist (MIM) will provide us with new search records each week, which two review authors will screen, extract, evaluate, and integrate following the guidance for Cochrane living systematic reviews (Cochrane LSR).

We will manually check platform trials that were previously identified and listed as 'studies awaiting classification' for additional treatment arms.

We will wait until the accumulating evidence changes our conclusions of the implications of research and practice before republishing the review. We will consider one or more of the following components to inform this decision.

The findings of one or more prioritised outcomes
The credibility (e.g. GRADE rating) of one or more prioritised outcomes
New settings, populations, interventions, comparisons or outcomes studied.

In case of emerging policy relevance because of global controversies around the intervention, we will consider republishing an updated review even though our conclusions remain unchanged. We will review the review scope and methods approximately monthly, or more frequently if appropriate, in light of potential changes in COVID‐19 research (for example, when additional comparisons, interventions, subgroups or outcomes, or new review methods become available).

Results

Description of studies

Results of the search

The literature search resulted in 3733 records. Three records were additionally identified via han search of reference lists, resulting in overall 3736 records. After removing duplicates, 1781 records remained. During title and abstract screening 1633 records were judged as irrelevant. We proceeded to full text screening with 148 records, considering published full texts or if these were not available, trial register entries. 77 records related to 64 studies were excluded after full‐text assessment. Three studies were cancelled before starting patient recruitment. 25 studies investigated a combined treatment including an antibiotic plus another intervention, nine studies used an active comparator without proven efficacy, and one study analysed a wrong study population including patients in recovery phase after surviving COVID‐19. Furthermore, 10 studies were not RCTs and 16 studies turned out to focus on another intervention than antibiotics. We identified 19 ongoing studies with 24 records and 15 studies with 21 records awaiting assessment. Finally, we included 11 studies with 26 records in our qualitative synthesis, with 10 studies being eligible for meta‐analyses, of which nine studies contributed data to the primary outcomes of this review. The search process is visualised in Figure 1.

Figure 1

Designs of the studies and publication status

The characteristics of including studies are available in the Characteristics of included studies section.

We included 11 studies describing 11,281 adult participants randomised to study arms relevant for the review question (Cavalcanti 2020; CJWT629A12301; Furtado 2020; Guvenmez 2020; Hinks 2020; NCT04335552; Omrani 2020; PRINCIPLE 2021; Rashad 2021; RECOVERY 2021; Sekhavati 2020). Eight studies had an open‐label design (Cavalcanti 2020; Furtado 2020; Guvenmez 2020; Hinks 2020; NCT04335552; PRINCIPLE 2021; RECOVERY 2021; Sekhavati 2020), two studies were placebo‐controlled and double‐ (Omrani 2020) or quadruple‐blinded (CJWT629A12301), and one study did not sufficiently report on its blinding method (Rashad 2021). Seven studies were performed as multi‐centre studies in Brazil (Cavalcanti 2020; Furtado 2020), UK (Hinks 2020; PRINCIPLE 2021; RECOVERY 2021), USA (CJWT629A12301), and Qatar (Omrani 2020).
The remaining four studies were performed as single‐centre studies in Egypt (Rashad 2021), Iran (Sekhavati 2020), Turkey (Guvenmez 2020), and the USA (NCT04335552). Two studies were prematurely terminated due to poor recruitment and enrolled 11 and 20 participants, respectively (CJWT629A12301; NCT04335552). The smallest regularly completed study was Guvenmez 2020 with 24 randomised participants, and RECOVERY 2021 was the largest trial with a sample size of 7763 participants. The remaining studies ranged in sizes between 111 (Sekhavati 2020) and 1415 participants (PRINCIPLE 2021).

Two studies were published as preprint articles by the end of July 2021 (Hinks 2020; Rashad 2021), the two prematurely terminated trials posted their results in the trial registry (CJWT629A12301; NCT04335552). The remaining seven studies were available as publications in indexed journals.

Except for one study with a retrospective trial registry entry (Rashad 2021) and another one with neither any publicly available protocol nor registry entry (Guvenmez 2020), all included studies registered a study protocol prospectively.

All studies reported information on the responsible ethics committee. However, five publications did not provide any approval reference number (Cavalcanti 2020; CJWT629A12301; Furtado 2020; NCT04335552; Omrani 2020).

Four studies were partially or fully funded by pharmaceutical companies producing or distributing azithromycin, including EMS Pharma (Cavalcanti 2020; Furtado 2020), Novartis Pharmaceuticals (CJWT629A12301), Pfizer Inc. (Hinks 2020) and six studies were solely funded by departmental or governmental resources as well as non‐profit organisations (NCT04335552; Omrani 2020; PRINCIPLE 2021; Rashad 2021; RECOVERY 2021; Sekhavati 2020). One study did not report its funding source (Guvenmez 2020).

Participants

Four of the 11 eligible studies used antibiotics for treating mild COVID‐19 cases in outpatient settings (Hinks 2020; Omrani 2020; PRINCIPLE 2021; Rashad 2021). Of those outpatient trials, only one also included asymptomatic participants (Omrani 2020), while the other three investigated ambulatory treatment exclusively in participants presenting COVID‐19 symptoms.

The majority of the seven studies conducted in the inpatient settings included moderately ill participants according to WHO 4 to 5 (Cavalcanti 2020; CJWT629A12301; Guvenmez 2020; NCT04335552; Sekhavati 2020). Only Furtado 2020 and RECOVERY 2021 included participants with moderate to severe COVID‐19 according to WHO scale 4 to 7.

Five trials included participants with suspected or test‐confirmed SARS‐CoV‐2 infection (Cavalcanti 2020; Furtado 2020; Hinks 2020; PRINCIPLE 2021; RECOVERY 2021). The exact rates of negative or unknown RT‐PCR test results are recorded in the Characteristics of included studies section for each study. For Cavalcanti 2020, Hinks 2020 and PRINCIPLE 2021, data for all outcomes of interest for our meta‐analyses were reported separately for RT‐PCR confirmed SARS‐CoV‐2 infections, either in the publication or via personal communication with the study authors. Furtado 2020 reported efficacy outcomes for the SARS‐CoV‐2 positive subgroup, however the study's safety outcomes were only reported for the overall population which included 11% of participants with negative RT‐PCR result. The largest eligible study reported data for all outcomes for the overall population only, which included 8.5% of participants with a negative or unknown RT‐PCR result (RECOVERY 2021). The authors of Furtado 2020 nor RECOVERY 2021 did not provide a subset of outcome data for confirmed diagnosis upon request.

Except for Guvenmez 2020 who gave very little information on eligibility criteria, all included studies listed pre‐existing QT‐prolongation or electrocardiogram (ECG) abnormalities observed prior to the trial initiation as exclusion criteria to secure patient safety in the face of macrolides’ known proarrhythmic potential.

The overall population mean age of the included studies was 54 years. Omrani 2020 included the youngest participants with a population mean age of 41 years. RECOVERY 2021 included the oldest participants with a population mean age of 65 years. The mean proportion of males in all included studies was 64%. The lowest proportion of males was included in PRINCIPLE 2021 with 43% males, while Omrani 2020 included the highest proportion with 98% males.
Co‐morbidities and relevant risk factors such as obesity, diabetes, respiratory diseases and hypertension were partially reported by the included studies and listed in the Characteristics of included studies. In one study, existing co‐morbidities were excluded and partially specified in the inclusion and exclusion criteria (Omrani 2020). Five studies did not report data on risk factors in their publications or study reports (CJWT629A12301; Guvenmez 2020; NCT04335552; Rashad 2021; Sekhavati 2020).

Interventions and comparators

All 11 included studies investigated azithromycin. All outpatient studies administered azithromycin orally, while four of the seven inpatient studies additionally allowed nasogastric (NCT04335552) or nasogastric and intravenous application (Cavalcanti 2020; Furtado 2020; RECOVERY 2021). The first day's dosage was consistently defined as 500 mg. Further doses and duration of treatment varied between studies. After day one, most studies exceeded usual dosage of azithromycin for bacterial infections which is defined as 250 mg per day on days 2 to 5. CJWT629A12301, Guvenmez 2020, NCT04335552 and Omrani 2020 used exactly that regimen, while the remaining seven studies proceeded with 500 mg daily. The shortest duration was 500 mg per day for three days in PRINCIPLE 2021, whereas the longest time period was 500 mg per day for 14 days (Hinks 2020). With five days Sekhavati 2020 ranged in between those time periods, as well as Cavalcanti 2020 and Rashad 2021 with seven days and Furtado 2020 and RECOVERY 2021 with 10 days of treatment duration.

Apart from two studies that used azithromycin‐placebo tablets (CJWT629A12301; Omrani 2020), the antibiotic was compared with standard of care alone. Hydroxychloroquine was used as part of the standard of care treatment in all study arms of six included studies (Cavalcanti 2020; CJWT629A12301; Furtado 2020; NCT04335552; Omrani 2020; Sekhavati 2020). Three studies did not include hydroxychloroquine, but other active treatments such as corticosteroids (Hinks 2020; RECOVERY 2021); although PRINCIPLE 2021 indicated similar proceedings (‘standard of care according to the NICE guideline’), no details of utilised substances were provided. Standard of care varied between the included studies, but was comparable between the study arms of the individual studies.

Guvenmez 2020 compared azithromycin with lincomycin, another antibiotic, which was administered orally 600 mg twice daily for 5 days. The second study investigating another antibiotic than azithromycin was Rashad 2021. This study had three arms investigating azithromycin or clarithromycin (500 mg twice daily for 7 days) compared to supportive care only.

No studies were identified that compared antibiotics with an active comparator with proven efficacy.

Outcome measures

Neither of the studies comparing azithromycin to another antibiotic reported our defined primary outcomes. Rashad 2021 focused on duration of individual symptoms and time to PCR negativity, but did not report any relevant outcomes for this review. Guvenmez 2020 provided data on PCR‐negativity at day 6 only, which was eligible for our secondary outcome viral clearance up to seven days. Ten studies where included in the meta‐analysis, from which only nine studies contributed data to the primary outcomes of this review.

All‐cause mortality at 28 days was reported by four studies in inpatients settings (Cavalcanti 2020; Furtado 2020; RECOVERY 2021; Sekhavati 2020) and three studies in outpatient settings (Hinks 2020; Omrani 2020; PRINCIPLE 2021).

One study in the inpatient setting reported data assessing clinical worsening by participants with new need for invasive mechanical ventilation or death at day 28 (RECOVERY 2021). Three studies in outpatient settings reported data for ‘admission to hospital or death within 28 days’, the equivalent outcome for ambulant patients (Hinks 2020; Omrani 2020; PRINCIPLE 2021).

Improvement of clinical status was reported by three studies in the inpatient setting as ‘patients discharged without respiratory deterioration or death at 28 days’ (CJWT629A12301; Furtado 2020; RECOVERY 2021). In the outpatient setting, two studies reported improvement of clinical status as participants with all initial symptoms resolved at 28 days (Omrani 2020; PRINCIPLE 2021), additionally Omrani 2020 reported this outcome for day 14 as well.

Four studies in the inpatient setting reported the number of participants with serious adverse events during the study period, which ranged from 15 to 60 days (Cavalcanti 2020; CJWT629A12301; Furtado 2020; NCT04335552). Two studies reported this safety outcome in the outpatient setting including study durations of 21 and 28 days (Hinks 2020; Omrani 2020). Regarding adverse events of any grade, three studies provided data for inpatients during the study period, which was followed up for a maximum of 60 days overall for this outcome (Cavalcanti 2020; CJWT629A12301; NCT04335552). No study reported usable data this outcome for outpatients.

Five studies investigating hospitalised patients reported the number of participants with ‘cardiac arrhythmias during the study period’ employing slightly varying though equally relevant definitions (Cavalcanti 2020; CJWT629A12301; Furtado 2020; RECOVERY 2021; Sekhavati 2020).

None of the included studies reported data for the primary outcome ‘duration to symptom resolution’ and ‘cardiac arrhythmias during the study period’ in the outpatient setting. Neither of the eligible inpatient nor outpatient studies reported data on quality of life or any efficacy outcome at day 60.

We defined a set of secondary outcomes for both healthcare settings which are not presented in the summary of findings tables. If data were available for those outcomes, we report the effect estimates in the Effects of interventions section and list further details of the meta‐analyses in Table 1 and Table 2.

Table 1. Secondary outcomes for azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Outcomes	Relative effect (95% CI)	N° of participants (studies)	Measures of heterogeneity
New need for invasive mechanical ventilation at day 28	RR 0.65 (0.16 to 2.71)	7422 (2 RCTs)	Tau² = 0.64; Chi² = 1.57, df = 1 (P = 0.21); I² = 36%
Ventilator‐free days	MD ‐0.10 (‐1.16 to 0.96)	331 (1 RCT)	NA
New need for non‐invasive mechanical ventilation or high flow at day 28	RR 0.91 (0.78 to 1.05)	4073 (1 RCT)	NA
Weaning or liberation from invasive mechanical ventilation in surviving patients at day 28	RR 1.11 (0.85 to 1.45)	452 (1 RCT)	NA
Liberation from supplemental oxygen in surviving patients at day 28	RR 0.85 (0.67 to 1.08)	395 (1 RCT)	NA
Need for dialysis at up to 28 days	RR 1.03 (0.85 to 1.25)	8052 (2 RCTs)	Tau² = 0.00; Chi² = 1.27, df = 1 (P = 0.26); I² = 21%
Admission to the intensive care unit (ICU) at day 28	RR 0.28 (0.06 to 1.29	111 (1 RCT)	NA
Duration of hospitalisation	MD ‐0.61 (‐2.58 to 1.37)	448 (3 RCTs)	Tau² = 1.61; Chi² = 5.37, df = 2 (P = 0.07); I² = 63%
Viral clearance up to 7 days	RR 1.50 (0.35 to 6.40)	14 (1 RCT)	NA
Hospital‐acquired infections at up to 28 days	RR 1.16 (0.90 to 1.49)	395 (1 RCT)	NA
CI: confidence interval; NA: not applicable; MD: mean difference RCT: randomised controlled trial; RR: risk ratio

See: Summary of findings 1 Azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19; Summary of findings 2 Azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Table 2. Secondary outcomes for azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Outcomes	Relative effect (95% CI)	N° of participants (studies)	Measures of heterogeneity
Need for invasive mechanical ventilation at 28 days	RR 0.61 (0.11 to 3.27)	405 (1 RCT)	NA
Need for hospitalisation with need for oxygen by mask or nasal prongs at 28 days	RR 0.80 (0.29 to 2.22)	406 (1 RCT)	NA
Viral clearance up to 7 days	RR 0.83 (0.44 to 1.54)	301 (1 RCT)	NA
CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio

Ongoing studies

The characteristics of ongoing studies are available in the Characteristics of ongoing studies section.

We classified a total of 19 studies as ongoing. 12 studies which are still active, either currently recruiting or not yet recruiting, and compare azithromycin with standard of care alone (Akram 2020; CTRI/2020/04/024904; EUCTR‐2020‐001527‐14/NL; NCT04328272; NCT04336332; NCT04344444; NCT04359316; NCT04359953; NCT04363060; NCT04371107; NCT04405921), except for one trial using placebo (NCT04332107). Those studies evaluate primarily inpatients with between 40 and 2271 participants, completion date according to the trial registries lies in the past (2020) for five studies, another five studies should be finished during the course of 2021, and another two studies did not give information on a planned completion date.

The large platform trial REMAP‐CAP, evaluating 71,00 participants, plans to investigate a range of different antibiotics and indicates its overall completion date for 2023 (Angus 2020).

Four trials evaluating between 94 and 330 participants investigate other antibiotics than azithromycin with the last one to be finished in 2022. Two trials in the outpatient setting compare doxycycline with placebo or standard of care alone, one study is not yet recruiting (NCT04371952), and the other study is currently still recruiting (NCT04729140). Another two trials use cotrimoxazol as antibiotic intervention in an inpatient setting, one recruiting (CTRI/2020/10/028297) and the other not yet recruiting (NCT04470531).

Two trials evaluating azithromycin in 300 and 500 participants have been suspended, from the information in their registry entry it remains unclear if they will be terminated or resume patient recruitment (NCT04363203; NCT04341727).

Studies awaiting classification

The characteristics of studies awaiting classification are available in the Characteristics of studies awaiting classification section.

Fifteen studies are awaiting classification until either publication of results, a protocol update or clarification of details by the study authors. If eligible eventually, they will be considered in the next review update (Duska 2020; EUCTR‐2020‐001156‐18/ES; EUCTR‐2020‐001606‐33/ES; Feeney 2020; Gyselinck 2021; IRCT20080901001165N50; IRCT20191211045691N2; IRCT20200428047228N2; IRCT20200608047686N1; IRCT20200718048129N1; NCT04345861; NCT04370782; Purwati 2021; Sayed 2020; Yamamoto 2021). Two studies had already been published their results in journal articles. However, despite using the term 'randomised controlled trial', we could not find sufficient details on fulfilment of a genuine randomisation process nor any study protocol that allowed us to classify those studies as RCTs (Purwati 2021; Sayed 2020). Authors were given a chance to clarify and adjust this assessment through personal communication, though we have not received any response at the time of review publication. We identified six completed and three prematurely terminated, but potentially eligible RCTs from trial register entries, but no results are available or have been published yet. Four of those studies investigated oral azithromycin versus placebo or standard of care alone in a total of 317 participants in an inpatient setting (Duska 2020; Feeney 2020; IRCT20200428047228N2; NCT04345861), one study including 80 patients applied azithromycin via an intranasal spray (IRCT20080901001165N50). Another three completed studies investigated various other antibiotics like clarithromycin (IRCT20200718048129N1; 45 outpatients), doxycycline (IRCT20191211045691N2; 60 inpatients) or metronidazole (IRCT20200608047686N1; 39 inpatients) versus standard of care alone. One study with completed recruitment compared azithromycin to doxycycline in 18 outpatients (NCT04370782). Another four studies are not explicit enough in their protocol to make a final decision on eligibility. Those studies have not reported a clear description of the type of control intervention used as comparator or in case of standard of care as control it remained unclear if this was administered in all study arms equally (EUCTR‐2020‐001156‐18/ES; EUCTR‐2020‐001606‐33/ES; Gyselinck 2021; Yamamoto 2021).

Excluded studies

The characteristics of excluded studies are available in the Characteristics of excluded studies section.

We excluded 64 studies that did not match our inclusion criteria. Three studies were withdrawn before enrolling patients, therefore, we did not expect any interim results to be published in the future (NCT04365582; NCT04433078; NCT04575558). Twenty‐five studies evaluated a combination of antibiotics with other treatments that were different between groups (Ahmed 2020; Chowdhury 2021; Hashim 2020; IRCT20200418047121N1; jRCT2071200109; Mahmud 2021; NCT04334512; NCT04341870; NCT04354597; NCT04358068; NCT04365231; NCT04371406; NCT04374552; NCT04392128; NCT04399746; NCT04407130; NCT04482686; NCT04551755; NCT04715295; PACTR202005622389003; RBR‐3k4wxb; RBR‐7nzwkpg; RBR‐95yjmq; Sivapalan 2020; Spoorthi 2020). Nine studies investigated active comparators without proven efficacy (Brown 2020; CTRI/2020/10/028234; EUCTR‐2020‐001605‐23/ES; Johnston 2021; NCT04286503; NCT04329832; NCT04334382; NCT04355052; NCT04374903). Sixteen studies investigated a wrong intervention (CTRI/2020/08/027033; EUCTR‐2020‐001366‐11/ES; EUCTR‐2020‐001971‐33/ES; NCT04332094; NCT04345419; NCT04359095; NCT04373733; NCT04374019; NCT04382846; NCT04390594; NCT04395768; NCT04403555; NCT04459702; NCT04492501; NCT04621461; RBR‐3rywwg), one study analysed a wrong study population including patients post COVID‐19 (CTRI/2020/11/029305), and 10 studies were non‐RCTs (Abbas 2021; Bernardini 2021; Cadegiani 2020; EUCTR‐2020‐001466‐11/GR; Moschini 2021; NCT04339426;NCT04458948; NCT04699097; Okour 2020; Tsiakos 2020) .

Risk of bias in included studies

We assessed methodological quality and risk of bias for nine RCTs contributing results to our primary outcomes using the RoB 2 tool (Cavalcanti 2020; CJWT629A12301; Furtado 2020; Hinks 2020; NCT04335552; Omrani 2020; PRINCIPLE 2021; RECOVERY 2021; Sekhavati 2020). We did not judge risk of bias for two studies (Guvenmez 2020; Rashad 2021). Guvenmez 2020 reported results only eligible for our secondary outcomes, and Rashad 2021 has not reported any outcome of interest for this review. In total, the nine studies contributed 30 study results to 11 outcomes, six outcomes for hospitalised COVID‐19 individuals and five outcomes for outpatients, that were finally assessed using RoB 2. The RoB 2 judgements for all study results per outcomes and for all domains are available in an interactive risk of bias table (Supplementary File_Antibiotics_Risk of Bias) and are briefly summarised below. The complete set of data is available in the supplementary file (Supplementary File_Antibiotics_Risk of Bias).

Overall risk of bias by study

About half of the 30 study results (56.7%) were assessed as overall low risk of bias and the remaining study results (43.3%) were assessed as some concerns for the overall risk of bias. None of the 30 study results was assessed as overall high risk of bias.

Overall risk of bias by outcome

The following section summarises the risk of bias per outcome for all primary outcomes included in the summary of findings tables (summary of findings Table 1; summary of findings Table 2).

Azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

We have no concerns regarding risk of bias across studies for the outcomes ‘all‐cause mortality at day 28’, ‘participants with clinical deterioration or death at day 28’, ‘participants discharged alive at day 28’, and ‘cardiac arrhythmias during the study period’. For these outcomes ‘low risk of bias studies’ dominated the effect estimates with more than 90% of the weight in the meta‐analyses. Key concerns across studies and per outcome were identified for the following outcomes: ‘serious adverse events during the study period’ due to an as‐treated analysis in one study with 97.0% of the weight in the meta‐analysis (Furtado 2020); ‘adverse events (any grade) during the study period’ due to non‐blinded outcome assessment in the study with the 87.7% weight (Cavalcanti 2020).

Azithromycin compared to placebo or standard of care alone for outpatients with asymptomatic or mild COVID‐19

We have no concerns regarding risk of bias across studies for the outcome ‘all initial symptoms resolved at day 14’. For this outcome, one ‘low risk of bias study’ contributed data to the result. Key concerns across studies and per outcome were identified for the following outcomes: ‘all‐cause mortality at day 28’ due to possible trial context related deviations from the intended interventions in that study contributing events to the analysis; ‘admission to hospital or death within 28 days’ due to possible trial context related deviations from the intended interventions in one study (Hinks 2020) and lack of registering the outcome in another study (Omrani 2020) contributing together more than 50% weight to the meta‐analysis; and ‘serious adverse events during the study period’ due to possible trial context related deviations from the intended interventions in one of two studies (Hinks 2020) and lack of registering the outcome in the other study (Omrani 2020).

Effects of interventions

We included11 studies in the qualitative synthesis of this review (Cavalcanti 2020; CJWT629A12301; Furtado 2020; Guvenmez 2020; Hinks 2020; NCT04335552; Omrani 2020; PRINCIPLE 2021; Rashad 2021; RECOVERY 2021; Sekhavati 2020). Ten studies were included in the meta‐analyses (quantitative synthesis), as one outpatient study did not report any relevant outcome for our meta‐analyses (Rashad 2021). Apart from two studies that used an azithromycin‐placebo (CJWT629A12301; Omrani 2020), or another antibiotic (Guvenmez 2020), respectively, azithromycin was compared with standard of care alone in the remaining studies. No studies were identified that compared antibiotics with an active comparator with proven efficacy.

Five studies investigated azithromycin for treating COVID‐19 in the inpatient setting and contributed data to primary outcomes of our meta‐analyses. Two of these studies investigated participants with a mixed population regarding disease severity, corresponding to WHO scale 4 to 7 (Furtado 2020; RECOVERY 2021). All other studies investigated participants with COVID‐19 severity according to WHO scale 4 to 5. Therefore, planned subgroup analyses for severity at baseline were performed, considering the subgroups moderate COVID‐19 and moderate to severe COVID‐19, and were reported in the Data and analyses. Since for none of the outcomes any clinically relevant subgroup difference was detectable, we do not report results of the subgroup analyses repeatedly in the following. We performed sensitivity analyses excluding non peer‐reviewed articles (e.g. preprint articles, results in trial registry) and of studies with confirmed COVID‐19 diagnosis only. Sensitivity analyses regarding risk of bias were not required since none of the eligible outcomes were assessed as high risk of bias. The main findings for the inpatient setting are summarised in the summary of findings Table 1.

Three studies investigated azithromycin for treating COVID‐19 in the outpatient setting and contributed data to meta‐analyses (Hinks 2020; Omrani 2020; PRINCIPLE 2021). We performed sensitivity analyses including peer‐reviewed articles only. The main findings for the outpatient setting are summarised in the summary of findings Table 2.

For both settings, we assessed heterogeneity by calculating 95% prediction interval (PIs) for each outcome if at least four studies contributed to the result. In the current review, there are not enough studies available to perform the planned subgroup analyses to investigate heterogeneity for study characteristics regarding different dosage, route of administration and participants' age in either setting.

Azithromycin compared to placebo or standard of care alone for inpatients individuals with moderate to severe COVID‐19

Primary outcomes

All‐cause mortality at day 28

Four studies with peer‐reviewed publications reported on all‐cause mortality at day 28 comparing azithromycin with standard of care alone (Cavalcanti 2020; Furtado 2020; RECOVERY 2021; Sekhavati 2020). Azithromycin has little or no effect on all‐cause mortality at day 28 (RR 0.98; 95% CI 0.90 to 1.06; 8600 participants; 4 studies; Analysis 1.1). The 95% prediction interval ranges from 0.85 to 1.12. Certainty of the evidence for this outcome was high. We did not downgrade for risk of bias, although one study had some concerns, since the weight of the respective study was only 0.1% of the overall effect estimate (Sekhavati 2020). While Cavalcanti 2020 and Sekhavati 2020 included RT‐PCR confirmed moderate COVID‐19, Furtado 2020 and RECOVERY 2021 investigated moderately to severely ill participants, with the latter also including 8.5% participants with negative RT‐PCR. Excluding those patients in a sensitivity analysis did not result in a different conclusion (RR 1.04; 95% CI 0.82 to 1.31; 837 participants; 3 studies).

Two more studies reported on all‐cause mortality at time points that were not eligible for our meta‐analysis. CJWT629A12301 completed follow up after 15 days which was too short and NCT04335552 with a follow‐up of 46 days was considered clinically not useful for pooling in the 28‐day analysis of mortality. This study on moderately ill RT‐PCR confirmed COVID‐19 patients, recorded three and one death in the azithromycin and standard of care group, respectively, with a total of five participants per group. The result was posted as registry entry only.

Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28

RECOVERY 2021 provided data for worsening of clinical status assessed as new need for invasive mechanical ventilation or death at day 28 in an inpatient population with mixed RT‐PCR test results (8.5% participants with negative or unknown RT‐PCR) and moderate to severe disease. Azithromycin probably has little or no effect on worsening of clinical status or death at day 28 (RR 0.95; 95% CI 0.87 to 1.03; 7311 participants; 1 study; Analysis 1.2). We downgraded the certainty of the evidence by one level for serious indirectness due to the effect estimate based on only one study with a mixed population regarding COVID‐19 diagnosis. The study was published in a peer‐reviewed journal.

Other studies included in this review reported related endpoints that were not eligible for pooling.

Furtado 2020 reported the number of participants on invasive mechanical ventilation at day 29. However, the study included participants on all kinds of respiratory support at baseline. Therefore, these data cannot be classified as ‘new need for respiratory support’ or generally worsening of clinical status. Cavalcanti 2020 reported this outcome for 15 days, which was too short and not clinically useful to pool with the eligible time point. Sekhavati 2020 reported new need for invasive mechanical ventilation and death within 30 days separately, therefore we could not judge if patients would be counted twice when combining the data. NCT04335552 reported outcome data similarly which in addition to a follow‐up period of 46 days was not eligible for our meta‐analysis.

Improvement of clinical status: participants discharged alive at day 28

Three studies reported improvement of clinical status as participants being discharged alive at day 28 comparing azithromycin to standard of care alone (Furtado 2020; RECOVERY 2021) or placebo (CJWT629A12301) in RT‐PCR confirmed moderate to severe COVID‐19 participants. Azithromycin probably has little or no effect on improvement of clinical status at day 28 (RR 0.96; 95% CI 0.84 to 1.11; 8172 participants; 3 studies; Analysis 1.3). We downgraded the certainty of the evidence by one level for serious heterogeneity (I² = 49%). The sensitivity analysis excluding one study with a mixed population (RECOVERY 2021; 8.5% participants with negative or unknown RT‐PCR) did not change the conclusion (RR 0.89; 95% CI 0.65 to 1.21; 409 participants; 2 studies). The sensitivity analysis excluding the results posted as a registry entry (CJWT629A12301) and only including peer‐reviewed publications was almost identical to the original analysis (RR 0.92; 95% CI 0.72 to 1.18; 8158 participants; 2 studies) (Furtado 2020; RECOVERY 2021).

Although CJWT629A12301 reported this outcome for 15 days only, we considered it eligible and clinically reasonable for pooling since all study participants were discharged alive until day 15. In contrast, Cavalcanti 2020 with the same follow‐up for this outcome, was not eligible for meta‐analysis, since several patients were still in hospital at 15 days.

Quality of life at longest follow‐up available

No study reported data for this outcome.

Serious adverse events during the study period, defined as number of participants with any event

Four studies with moderately to severely ill inpatients recorded any serious adverse events during the study period comparing azithromycin to standard of care alone (Cavalcanti 2020; Furtado 2020; NCT04335552) or placebo (CJWT629A12301). Furtado 2020 included severely ill participants and the remaining three studies investigated moderate COVID‐19. Azithromycin probably has little or no effect on serious adverse events during the study period (RR 1.11; 95% CI 0.89 to 1.40; 794 participants; 4 studies; Analysis 1.4). The 95% prediction interval ranges from 0.67 to 1.83. We downgraded the certainty of the evidence one level for serious risk of bias. We performed two sensitivity analysis. Firstly, we excluded the only study with a mixed population regarding RT‐PCR results (Furtado 2020; 11% participants with negative RT‐PCR), which increased the imprecision of the estimated effect (RR 0.98; 95% CI 0.26 to 3.60; 355 participants; 3 studies). The certainty of evidence was not downgraded due to indirectness, as the same study was already the reason for the downgrading due to risk of bias. The second sensitivity analysis included peer‐reviewed publications only with a comparable result (RR 1.12; 95% CI 0.89 to 1.41; 770 participants; 2 studies) (Cavalcanti 2020; Furtado 2020).

RECOVERY 2021 reported data on serious adverse reaction to the study drug only, which did not fulfil the definition and criteria for assessment of any serious adverse events. Similarly, Furtado 2020 reported the additional outcome ‘serious adverse events suspected to be related to study drug’ (azithromycin group: 12/241 versus control group: 8/198), which we did not pool in the meta‐analysis for the same reason.

Adverse events (any grade) during the study period, defined as number of participants with any event

Three studies with RT‐PCR confirmed moderate COVID‐19 inpatients comparing azithromycin to standard of care alone (Cavalcanti 2020; NCT04335552) or placebo (CJWT629A12301) recorded any adverse events during the study period. Azithromycin may increase any adverse events slightly during the study period (RR 1.20; 95% CI 0.92 to 1.57; 355 participants; 3 studies; Analysis 1.5). The certainty of the evidence for this outcome was low due to serious risk of bias and serious imprecision because the effect estimate is based on few studies with small sample sizes. Only the largest of the three studies was published in a peer‐reviewed journal, resulting in a marginally deviating sensitivity analysis (RR 1.19; 95% CI 0.89 to 1.57; 331 participants; 1 study) (Cavalcanti 2020).

Cardiac arrhythmias during the study period

Four studies including moderately to severely ill patients provided data for this outcome, with varying outcome definitions between studies (Cavalcanti 2020; Furtado 2020; RECOVERY 2021; Sekhavati 2020). We considered all definitions as clinically relevant and overall similar. Meta‐analysis indicated that azithromycin probably has little or no effect on cardiac arrhythmias (RR 0.92; 95% CI 0.73 to 1.15; 7865 participants; 4 studies; Analysis 1.6). The 95% prediction interval ranges from 0.22 to 3.92. We downgraded the level of evidence for serious indirectness due to the effect estimate based mainly (weight 99.1%) on two studies with mixed populations (8.5% and 11% participants with unknown or negative RT‐PCR) (Furtado 2020; RECOVERY 2021). Sensitivity analysis including only RT‐PCR confirmed COVID‐19 diagnosis increased the imprecision of the effect estimate (RR 0.46; 95% CI 0.04 to 5.05; 442 participants; 2 studies) (Cavalcanti 2020; Sekhavati 2020). All studies were published in peer‐reviewed journals and used standard of care alone as comparator.

QT‐prolongation during the study period was reported by several studies. CJWT629A12301 did not report on overall cardiac arrhythmias, but on the incidence of prolonged QT interval, in RT‐PCR positive and moderately ill participants. This outcome was reported for one of seven participants in the azithromycin group and none of seven participants in the placebo control group within 15 days. Three studies comparing azithromycin to standard of care alone reported a prolonged QT interval in addition to any cardiac arrhythmias. Furtado 2020 reported 47 of 241 (intervention group) versus 42 of 198 participants (control group) with prolonged QT interval within 29 days. Cavalcanti 2020 recorded this outcome up to seven days only, registering 14 of 85 (intervention group) versus 11 of 77 (control group) affected participants, while Sekhavati 2020 registered zero events in either group during hospital stay.

Secondary outcomes

For inpatients, we defined a set of secondary outcomes of inferior clinical relevance compared to our primary outcome set. Therefore, we did not assess risk of bias or quality of the evidence for that data. For transparency and completeness, we extracted the available data from the eligible study pool and presented their overall effect estimates in the following and further details are listed in Table 1.

No studies were found that reported any or conclusive data on our secondary outcomes: clinical worsening at day 28 assessed as ‘new need for oxygen by mask or nasal prongs’, clinical improvement assessed as ‘duration to liberation from invasive mechanical ventilation’ or ‘duration to liberation from supplemental oxygen’, and ‘viral clearance at up to 3 or 14 days’.

New need for invasive mechanical ventilation at day 28

Two studies reported this outcome including hospitalised participants with RT‐PCR confirmed moderate disease (Sekhavati 2020) or with mixed RT‐PCR test results and moderate to severe disease (RECOVERY 2021; 8.5% participants with negative or unknown RT‐PCR). Due to a wide confidence interval, the effect of azithromycin compared to standard of care alone for the new need of invasive mechanical ventilation remained unclear (RR 0.65; 95% CI 0.16 to 2.71; 7422 participants; 2 studies; Table 1), statistical heterogeneity was minor (I² = 36%). Both studies were published in peer‐reviewed journals.

Furtado 2020 reported the number of participants on invasive mechanical ventilation at day 29, however the study included intubated participants at baseline. Therefore, these data cannot be classified as ‘new need’ or generally worsening of clinical status. Cavalcanti 2020 reported this outcome for 15 days, which was too short, while NCT04335552 followed participants up for 46 days which was not clinically useful for pooling in the analysis of this outcome.

Ventilator‐free days

Cavalcanti 2020, a journal publication, reported this outcome investigating RT‐PCR confirmed moderate disease and did not show any relevant differences between the azithromycin and the standard of care group in the days free from ventilator (MD ‐0.10; 95% CI ‐1.16 to 0.96; 331 participants; 1 study; Table 1). Furtado 2020 provided asymmetrical median data, that were not eligible for meta‐analysis.

New need for non‐invasive mechanical ventilation or high flow at day 28

RECOVERY 2021 provided data for this outcome in an inpatient population with mixed RT‐PCR test results (8.5% participants with negative or unknown RT‐PCR) and moderate to severe disease. Azithromycin showed no clinically relevant effect compared to standard of care alone (RR 0.91; 95% CI 0.78 to 1.05; 4073 participants; 1 study; Table 1). The study was published in a peer‐reviewed journal.

Furtado 2020 reported the number of participants on non‐invasive mechanical ventilation or high flow at day 29, however the study included participants on all kinds of respiratory support at baseline. Therefore, these data cannot be classified as ‘new need’ or generally worsening of clinical status. Cavalcanti 2020 reported this outcome for 15 days, which was too short as follow‐up and not clinically useful to pool with the eligible time points.

Weaning or liberation from invasive mechanical ventilation in surviving patients at day 28

RECOVERY 2021 provided data for this outcome in an inpatient population with mixed RT‐PCR test results (8.5% participants with negative or unknown RT‐PCR). The outcome was consequentially reported for intubated patients at baseline only. Azithromycin showed no clinically relevant effect compared to standard of care alone (RR 1.11; 95% CI 0.85 to 1.45; 452 participants; 1 study; Table 1). The study was published in a peer‐reviewed journal.

Furtado 2020 reported the number of participants on less extensive respiratory support than invasive mechanical ventilation at day 29, however the study included participants at equal or milder stages of the disease at baseline. Therefore, these data cannot be classified as either improvement or worsening of clinical status.

Liberation from supplemental oxygen in surviving patients at day 28

This outcome was reported by Furtado 2020 for moderately to severely ill inpatients with RT‐PCR confirmed diagnosis (RR 0.85; 95% CI 0.67 to 1.08; 395; 1 study; Table 1). Azithromycin showed no clinically relevant effect compared to standard of care alone. The study was published in a peer‐reviewed journal.

Cavalcanti 2020 reported number of patients without oxygen support, but the study also included participants without any oxygen need at baseline. Therefore, these data cannot be classified as either improvement or worsening of clinical status. Further it was reported for 15 days only, which was not an eligible time point for the inpatient setting.

Need for dialysis at up to 28 days

The two largest studies, published in a journal, and investigated moderate to severely ill patients reported results regarding need for dialysis at up to 28 days (Furtado 2020; RECOVERY 2021). The meta‐analysis revealed no difference between azithromycin and standard of care alone (RR 1.03; 95% CI 0.85 to 1.25; 8052 participants; 2 studies; Table 1; RECOVERY 2021 included 8.5% participants with negative or unknown RT‐PCR), with minor statistical heterogeneity (I² = 21%).

Admission to the intensive care unit (ICU) at day 28

One small study reported this outcome for moderately ill patients with RT‐PCR confirmed diagnosis in a journal publication (RR 0.28; 95% CI 0.06 to 1.29; 111 participants; 1 study; Table 1) (Sekhavati 2020). Due to the wide confidence interval, the effect of azithromycin compared to standard of care alone remained unclear.

Duration of hospitalisation

Three studies provided data for duration of hospitalisation for moderately ill inpatients with confirmed diagnosis (Cavalcanti 2020; NCT04335552; Sekhavati 2020). The meta‐analysis revealed no clinically relevant difference between the azithromycin and the standard of care group (MD 0.61; 95% ‐2.58 to 1.37; 448 participants; 3 studies; Table 1) with substantial statistical heterogeneity (I² = 63%). Cavalcanti 2020 and Sekhavati 2020 published their results in peer‐reviewed journals, while NCT04335552 posted its results in a trial registry. Furtado 2020 and RECOVERY 2021 provided asymmetrical median values with interquartile range, that were not eligible for meta‐analysis.

Viral clearance up to seven days

One study on moderately ill inpatients with confirmed COVID‐19 diagnosis reported on viral clearance at day 6 (CJWT629A12301). The effect of azithromycin compared to placebo remained unclear due to a wide confidence interval (RR 1.50; 95% CI 0.35 to 6.40; 14 participants; 1 study; Table 1).

Hospital‐acquired infections at up to 28 days

This outcome was reported by Furtado 2020 for moderately to severely ill inpatients with RT‐PCR confirmed diagnosis (RR 1.16; 95% CI 0.90 to 1.49; 395 participants; 3 studies). Azithromycin showed no clinically relevant effect compared to standard of care alone. The study was published in a peer‐reviewed journal.

Azithromycin compared to placebo or standard of care alone for outpatients with asymptomatic or mild COVID‐19

Primary outcomes

All‐cause mortality at day 28

Azithromycin may have little or no effect on all‐cause mortality at day 28 based on three studies comparing azithromycin with placebo (Omrani 2020) or standard of care alone (Hinks 2020; PRINCIPLE 2021) that reported outcome data in the outpatient setting including only RT‐PCR confirmed and mild COVID‐19 diagnosis (RR 1.00 ; 95% CI 0.06 to 15.59; 876 participants; 3 studies; Analysis 2.1). Omrani 2020 included asymptomatic participants for this outcome. We downgraded certainty of the evidence one level due for serious risk of bias and one level for serious imprecision caused by the wide confidence interval and few events (only one death in either group). Sensitivity analysis with only peer‐reviewed journal publications resulted in a non‐estimable effect estimate due to zero events in all arms of the remaining studies (RR not estimable; 726 participants; 2 studies) (Omrani 2020; PRINCIPLE 2021).

Admission to hospital or death within 28 days

Three studies comparing azithromycin with placebo (Omrani 2020) or standard of care alone (Hinks 2020; PRINCIPLE 2021) reported data on admission to hospital or death within 28 days in participants with mild disease confirmed by RT‐PCR. Omrani 2020 included asymptomatic participants for this outcome. Azithromycin may have little or no effect on admission to hospital or death within 28 days (RR 0.94 ; 95% CI 0.57 to 1.56; 876 participants; 3 studies; Analysis 2.2). We downgraded certainty of the evidence two levels for serious risk of bias and serious imprecision due to the wide confidence interval. Sensitivity analysis with only peer‐reviewed journal publications does not result in a different judgement (RR 0.90; 95% CI 0.47 to 1.74; 726 participants; 2 studies) (Omrani 2020; PRINCIPLE 2021).

All initial symptoms resolved (asymptomatic) at day 14

One study comparing azithromycin to placebo reported data on resolution of all initial symptoms at day 14 in participants with mild disease confirmed by RT‐PCR (Omrani 2020). Azithromycin may have little or no effect on symptom resolution at day 14 (RR 1.03; 95% CI 0.95 to 1.12; 138 participants; 1 study; Analysis 2.3). The certainty of the evidence is low due to very serious imprecision because the effect estimate is based on one small study. The study was published as a peer‐reviewed journal article.

All initial symptoms resolved (asymptomatic) at day 28

Two studies comparing azithromycin to placebo (Omrani 2020) or standard of care alone (PRINCIPLE 2021) provided data for resolution of all initial symptoms at day 28 in the outpatient setting. The effect estimate of the meta‐analysis showed no clinically relevant difference in this outcome between groups (RR 0.95; 95% CI 0.79 to 1.15; 549 participants; 2 studies; Analysis 2.4). Both studies included RT‐PCR positive, mildly ill patients only and were published in a peer‐reviewed journal.

Duration to symptom resolution

No study reported eligible data for this outcome. PRINCIPLE 2021 reported data as median with interquartile range which were not eligible for meta‐analysis.

Quality of life at longest follow‐up available

No study reported data for this outcome.

Serious adverse events during the study period, defined as number of participants with any event

Two studies comparing azithromycinwith standard of care alone (Hinks 2020) or placebo (Omrani 2020) assessed serious adverse events during the study period in the outpatient setting, but none of the participants in either study were affected. We downgraded certainty of the evidence by one level for serious risk of bias and two levels for very serious imprecision due to zero events in all arms of both studies. Hence, there were too few participants who experienced serious adverse events to determine whether azithromycin made a difference (RR not estimable; 454 participants; 2 studies; Analysis 2.5). Sensitivity analysis with exclusion of the preprint‐article (Hinks 2020) was not meaningful (RR not estimable; 304 participants; 1 study). All included participants were diagnosed by an RT‐PCR test and were mildly ill or even asymptomatic (see Omrani 2020).

Adverse events (any grade) during the study period, defined as number of participants with any event

No study reported data for this outcome. PRINCIPLE 2021 and Hinks 2020 reported on drug specific side effects and specific adverse events in their publications, therefore did not fulfil the definition and criteria for assessment of any adverse events.

Cardiac arrhythmias during the study period

No study provided data for this outcome. However, QT‐prolongation during the study period was reported in one study with 5 of 152 affected participants in the azithromycin group versus 4 of 152 in the placebo group within 21 days (Omrani 2020).

Secondary outcomes

For outpatients, we defined a set of secondary outcomes of inferior clinical relevance compared to our primary outcome set. Therefore, we did not assess risk of bias or quality of the evidence for that data. For transparency and completeness, we extracted the available data from the eligible study pool and present their overall effect estimates in the following, further details are listed in Table 2.

No studies were found that reported any or conclusive data on our secondary outcomes: clinical worsening at day 28 assessed as ‘need for non‐invasive mechanical ventilation or high flow’ or ‘need for hospitalisation without oxygen therapy’, ‘hospital‐acquired infections at up to 28 days’, and ‘viral clearance at up to 3 or 14 days’.

Need for invasive mechanical ventilation at day 28

One study comparing azithromycin to standard of care alone reported the need for invasive mechanical ventilation in RT‐PCR positive, mildly ill outpatients at day 28 (PRINCIPLE 2021). The study was published in a peer‐reviewed journal. The effect of azithromycin compared to standard of care alone remained unclear due to a wide confidence interval (RR 0.61, 95% CI 0.11 to 3.27; 405 participants; 1 study; Table 2).

Need for hospitalisation with need for oxygen by mask or nasal prongs at day 28

One study reported the number of RT‐PCR positive outpatients with need for hospitalisation and with need for oxygen by mask or nasal prongs at day 28 (PRINCIPLE 2021). The effect of azithromycin compared to standard of care alone remained unclear due to a wide confidence interval (RR 0.80, 95% CI 0.29 to 2.22; 406 participants; 1 study; Table 2).

Viral clearance up to day 7

One outpatient study reported viral clearance at up to 7 days for previously RT‐PCR positive participants with mild or no symptoms and was published in a peer‐reviewed journal (Omrani 2020). The effect of azithromycin compared to standard of care alone remained unclear due to a wide confidence interval (RR 0.83, 95% CI 0.44 to 1.54; 301 participants; 1 study; Table 2).

For other time points, there was no conclusive data available within the study pool. Although Omrani 2020 reported data for this outcome at day 14, we could not determine whether patients with initial viral clearance at day 6 were counted twice.

Azithromycin compared to other antibiotics for inpatients with moderate to severe COVID‐19

Primary outcomes

No study reported data for any primary outcome of this comparison.

Secondary outcomes

Viral clearance up to day 7

One peer‐reviewed publication compared azithromycin to lincomycin, another antibiotic, in moderately ill inpatients with confirmed COVID‐19 diagnosis and reported on viral clearance at day 6 (Guvenmez 2020). The study revealed a reduced viral clearance for azithromycin compared to lincomycin (RR 0.40; 95% CI 0.17 to 0.93; 24 participants; 1 study; Table 3).

Table 3. Secondary outcomes for azithromycin compared to other antibiotics for inpatients with confirmed moderate to severe COVID‐19

Outcomes

Relative effect (95% CI)

N° of participants (studies)

Measures of heterogeneity

Viral clearance up to 7 days

RR 0.40
(0.17 to 0.93)

(1 RCT)

Azithromycin compared to other antibiotics for outpatients COVID‐19 with asymptomatic or mild COVID‐19

One preprint article was included that compared azithromycin to clarithromycin in outpatients with confirmed COVID‐19 diagnosis comparing azithromycin to clarithromycin, another antibiotic (Rashad 2021). The study did not report any outcome relevant for this review.

Discussion

Summary of main results

This review included 11 studies with 11281 participants investigating azithromycin compared to placebo, standard of care alone or another antibiotic. The main findings of this review are summarised in the summary of findings Table 1 (azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19) and summary of findings Table 2 (azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19).

Azithromycin has no effect on all‐cause mortality at 28 days, the most important patient‐relevant outcome during this pandemic, neither in the inpatient (4 studies with 8600 participants) nor the outpatient setting (3 studies with 876 participants). The certainty for this finding was high for inpatients, and low for outpatients due to concerns of risk of bias and very few events of deaths in this setting.

For the inpatient setting, azithromycin probably has little or no effect on worsening of clinical status assessed by new need for invasive mechanical ventilation or death at day 28 or on improvement of clinical status assessed as participants discharged alive at day 28. Regarding safety outcomes, azithromycin probably has little or no effect on cardiac arrhythmias and serious adverse events during the study period, while azithromycin may increase any adverse events slightly during the study period.

For the outpatient setting, azithromycin may have little or no effect on admission to hospital or death within 28 days and on resolution of all initial symptoms at day 14. The certainty of the evidence for these findings was low. Regarding safety outcomes, we are uncertain whether azithromycin increases or reduces serious adverse events compared to placebo or standard of care alone, since no participant experienced serious adverse events in the studies. No study provided data for any adverse events or cardiac arrhythmias.

Overall completeness and applicability of evidence

The evidence summarised in this review applies to azithromycin for COVID‐19, but as yet no other antibiotic. Only two included studies investigated another antibiotic than azithromycin. However, those trials did not provide relevant data for this review nor for clinical practice (Guvenmez 2020; Rashad 2021). There are four ongoing trials with between 94 and 330 participants evaluating other antibiotics such as doxycycline (NCT04371952; NCT04729140) and cotrimoxazol (CTRI/2020/10/028297; NCT04470531). Four completed studies with results publication pending investigate clarithromycin (IRCT20200718048129N1), doxycycline (IRCT20191211045691N2; NCT04370782) and metronidazol (IRCT20200608047686N1). Except for the trial on doxycycline which evaluates overall 1490 patients, those studies are very small with trials sizes well below 100 participants. Therefore, we do not expect that publication of those trials alone will lead to reliable evidence. The ongoing platform trial REMAP‐CAP, evaluating 7100 participants, plans to investigate macrolides as well as flouroquinolones and ß‐lactam antibiotics, but it is unclear when those results will be available (Angus 2020).

With seven out of 11 included studies, the majority of studies were conducted in inpatient settings. Those studies included participants with moderate as well as severe disease (WHO scale 4 to 7). Therefore, the findings of this review are transferable to inpatients with COVID‐19 at any stage.

Four of the 11 included studies were conducted in outpatient settings with no, or mild COVID‐19 symptoms (WHO 1 to 3), of which three studies reported relevant outcomes. Due to the relatively small number of participants with confirmed COVID‐19 diagnosis, i.e. RT‐PCR test positives, and partially low number of events, certainty of the evidence is low for this particular population. However, as the inpatient studies included in this review show strong evidence for azithromycin’s futility for COVID‐19 treatment in inpatients who do not generally require medical support, applicability for the outpatient setting should be considered.

The overall mean age of participants in all studies was 54 years with a mean range from 41 to 65 years and included patients affected by co‐morbidities, e.g. obesity, hypertension or diabetes. Considering age and pre‐existing conditions as the most important risk factors for developing severe COVID‐19, the current evidence is applicable to patients which are at most risk of death from COVID‐19.

Three studies were conducted in the Middle East, three in Europe, two in the USA, two in South America, and one in Africa, covering states with high‐ and low‐healthcare expenditure.

All outpatient studies administered azithromycin per mouth, while in the inpatient setting oral, intravenous and nasogastric application was used. With 500 mg daily for three to 14 days, seven of the 11 included studies exceeded usual dosage of azithromycin for bacterial infections which is described as 500 mg loading dose followed by 250 mg daily for four subsequent days. Too few studies were available to investigate dose effects by subgroup analyses. Overall, regimens were very similar. Considering the strong evidence on azithromycin's lack of benefit for COVID‐19 patients, a future subgroup analysis investigating different doses is unlikely to reveal relevant new information. Evidence on other antibiotics is lacking.

A total of 19 studies are ongoing and 15 studies are awaiting classification pending publication of results or clarification of inconsistencies. It is questionable if ongoing studies investigating azithromycin will be completed due to the results of larger trials already available. Indeed, two of the 11 studies were already prematurely terminated due to results from large trials indicating no evidence of effect for azithromycin (CJWT629A12301; NCT04335552). It is not to be expected that including future publications of smaller or lower quality randomised controlled trials(RCTs) would change this review's conclusion on azithromycin. However, we will monitor preprints, ongoing studies and those awaiting classification for potentially relevant results that require updating this review.

For other antibiotics than azithromycin, there is currently a complete lack of evidence. Well‐designed RCTs are required that may allow for a judgement of the effectiveness and the safety of other antibiotics for treatment of COVID‐19. However, as the emergence and spread of antimicrobial resistance has been considered as a global threat already before the pandemic, prudent use of antibiotics is of utmost importance.

This review did not investigate antibiotic prophylaxis or preemptive therapy of bacterial infections in patients also infected with SARS‐CoV‐2, but focused on antibiotics used as COVID‐19 treatment only. Therefore, any conclusions drawn form our results are only valid for the latter rationale. There are no RCTs available on how to guide preemptive or prophylactic antibiotic prescription in COVID‐19 patients, which is an evidence gap to be filled urgently in the light of the global emergence of antimicrobial resistance as a major public health and development threat.

Quality of the evidence

The certainty of evidence for prioritised outcomes presented in the Summary of findings tables (summary of findings Table 1; summary of findings Table 2) ranged from low to high for the inpatient setting and from very low to low for the outpatient setting.

For the Summary of findings and assessment of the certainty of the evidence according to Schünemann 2020, we used the results from analysis of our primary outcomes. About half of the study results were assessed as overall low risk of bias. No respective study outcome was judged as high risk of bias, therefore, our meta‐analyses included only studies with low risk or some concerns of bias. Details of the risk of bias assessments per outcome are reported in Risk of bias in included studies.

Overall certainty of the evidence was higher for inpatients than for outpatients due to studies with larger size and low risk of bias. We are very certain that azithromycin has little or no effect on all‐cause mortality at day 28 in inpatients compared to standard of care alone. Most of the other findings for inpatients were assessed as moderate certainty of evidence. The certainty of evidence for the outcomes ‘worsening of clinical status or death at day 28’ and ‘cardiac arrhythmias during the study period’ was downgraded due to indirectness. The effect estimates of both outcomes were mainly based on one study that included about 10% of participants with negative or unknown RT‐PCR results. We did not downgrade ‘all‐cause mortality at day 28’ and ‘improvement of clinical status at day 28’ for indirectness, because the result did not change in a sensitivity analysis excluding studies with mixed populations. Inconsistency was the reason for downgrading the certainty of evidence for ‘improvement of clinical status at day 28’, because three studies contributed heterogenous results (I² = 49%). One study favoured standard of care alone and the other two showed no difference between study arms. Adverse events was the only outcome in the inpatient setting with low certainty of the evidence. The certainty of evidence was downgraded due to serious risk of bias because study results that contributed to the outcome were assessed as some concerns of bias and serious imprecision due to inclusion of few small studies.

Small studies with few events causing serious imprecision were the main reason for downgrading certainty of the evidence in the outpatient setting. Certainty of the evidence was downgraded one level if studies with some concerns for risk of bias contributed to the according outcome. Therefore, all but one result was not downgraded for risk of bias because there was always at least one study contributing results assessed as some concerns of bias.

In the current phase of the pandemic, it is impossible to reliably assess the risk of publication bias. Many of the registered studies are still ongoing or, in the case of a completed study status, their results have not yet been published. We will follow the publication and trial history of each ongoing study and study awaiting classification. Currently, we did not suspect publication bias for any outcome included in this review. However, this may change in future updates of this review.

Potential biases in the review process

To avoid potential bias in the review process, we were committed at all times to conduct a systematic review that followed published guidance provided by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2020a).

The review team is part of the German research project “CEOsys” (COVID‐19 Evidence‐Ecosystem). CEOsys is a consortium of clinical and methodologic experts supported by the German Federal Ministry of Education and Research to synthesise clinical evidence during this global pandemic. The involved medical information specialists of this consortium carried out a rigorous search of electronic databases including preprint servers and clinical trial registries to identify the complete extent of published and ongoing trials on this topic. The sensitive search included relevant electronic databases as well as clinical trial registries. As a supplement, we screened reference lists of included studies and compared our search results with those from the living network meta‐analysis (COVID‐NMA). Therefore, we are confident that we identified all relevant studies and will monitor ongoing studies as well as full publication of preprints closely after the publication of this review.

This review is part of a Cochrane Living Systematic Reviews Series on different interventions for treatment of COVID‐19 (Ansems 2021; Kreuzberger 2021b; Popp 2021; Stroehlein 2021; Wagner 2021) established and performed by the CEOsys consortium. Initially, a common core outcome set was established for all reviews in accordance with the Core Outcome Measures in Effectiveness Trials (COMET) Initiative for COVID‐19 patients (COMET 2020; Marshall 2020), and additional outcomes that have been prioritised by consumer representatives and the German guideline panel for inpatient therapy of people with COVID‐19. Prespecified outcomes in the protocol template and previous reviews (Ansems 2021; Kreuzberger 2021b; Popp 2021; Stroehlein 2021; Wagner 2021) have been changed to address issues of competing risks associated with certain outcomes describing the clinical status of COVID‐19 patients. These issues are extensively discussed in previous reviews (Ansems 2021; Wagner 2021). To overcome competing risks, we added outcomes for hospitalised and ambulatory managed patients that aim to simultaneously capture all participants of the population with clinical worsening and all participants with clinical improvement. This was possible by using composite outcomes, e.g. combining new need for invasive mechanical ventilation and death as clinical worsening for hospitalised individuals, and combining admission to hospital and death for outpatients. Clinical improvement for inpatients was represented by the number of participants discharged alive within the same time period used for clinical worsening, and for outpatients as complete resolution of initial symptoms. All previous outcomes defined in the protocol on clinical status assessed by need for respiratory support were analysed as secondary outcomes.

Two included studies were published as preprint articles and two other ones posted data in their registry entry. We are aware that articles may change following peer‐review. Nevertheless, we are convinced that including all eligible data in a highly dynamic situation such as the COVID‐19 pandemic is crucial to be up‐to‐date and to provide timely information on potentially promising treatment options. Journal publications and corresponding preprint articles were compared in terms of consistency and all study results were assessed for their risk of bias. Study authors were contacted if the publication included unclear or missing information. Unfortunately, not all attempts of gathering additional data were successful. Details of the communication with authors are provided in the Characteristics of included studies.

For two studies that had already published results, we could not finally judge eligibility due to insufficient description of their randomisation method and inconsistent information on analysed participants (Purwati 2021; Sayed 2020). We tried to outreach to the corresponding authors to clarify those questions, though we have not received any satisfying response at the time of review publication. Further, nine trials classified as awaiting classification have completed recruitment, but publication of results is still pending. Those trials will be closely monitored for publication in the near future.

None of the members of the review author team has any affiliation with any stakeholder group who favours or disapproves of any antibiotic or the comparators used in relevant studies.

Agreements and disagreements with other studies or reviews

The findings of our review on azithromycin as a treatment for COVID‐19, is similar to findings of the rapid guideline released by NICE 2020 and last updated in June 2021 (NICE 2021). The authors included four randomised controlled trials RCTs) on hospitalised patients, which were also identified in this review (Cavalcanti 2020; Furtado 2020; RECOVERY 2021; Sekhavati 2020). The additional two studies in our review were not included in the NICE guideline presumably due to the date of publication (CJWT629A12301; NCT04335552). For azithromycin in ambulatory settings, the evidence syntheses include the same study pool as this review (Hinks 2020; Omrani 2020; PRINCIPLE 2021). In agreement with our results, the guideline authors claimed that azithromycin does not reduce mortality in out‐ or inpatients, and there is a lack of benefit for azithromycin regarding discharge from hospital for inpatients or clinical recovery for outpatients, as well as adverse and serious adverse events in both settings. Certainty of the evidence for these outcomes ranged from very low to moderate.

Kamel 2021, a very up to date meta‐analysis investigating azithromycin as COVID‐19 treatment, included the same studies as NICE 2021, and draw the same conclusion on azithromycin to treat COVID‐19, correspondingly. Based on three studies identical to our review, they also report no relevant influence of azithromycin on cardiac arrhythmias, an additional safety outcome of high clinical relevance. Our work differs from this review in its methods as in Kamel 2021 different health care settings were pooled in some of their analyses, suspected COVID‐19 was accepted, and they did not use GRADE to rate the certainty of evidence.

To the best of our knowledge, there is currently no systematic review available that investigates antibiotics other than azithromycin as anti‐viral or anti‐inflammatory treatment of COVID‐19. Sharma 2021 includes non‐RCTs and RCTs in their meta‐analysis on different antibiotics, but focused on the prevalence of each antibiotic in COVID‐19 studies instead of their efficacy and safety.

Currently, there are few RCT‐ investigating the effectiveness and safety of antibiotics other than azithromycin used for COVID‐19 treatment. It is not well.

Figure 1

Analysis 1.1

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 1: All‐cause mortality at day 28

Analysis 1.2

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 2: Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28

Analysis 1.3

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 3: Improvement of clinical status: participants discharged alive at day 28

Analysis 1.4

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 4: Serious adverse events during the study period, defined as number of participants with any event

Analysis 1.5

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 5: Adverse events (any grade) during the study period, defined as number of participants with any event

Analysis 1.6

Comparison 1: Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19, Outcome 6: Cardiac arrhythmias during the study period

Analysis 2.1

Comparison 2: Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19, Outcome 1: All‐cause mortality at day 28

Analysis 2.2

Comparison 2: Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19, Outcome 2: Admission to hospital or death within 28 days

Analysis 2.3

Comparison 2: Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19, Outcome 3: All initial symptoms resolved (asymptomatic) at day 14

Analysis 2.4

Comparison 2: Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19, Outcome 4: All initial symptoms resolved (asymptomatic) at day 28

Analysis 2.5

Comparison 2: Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19, Outcome 5: Serious adverse events during the study period, defined as number of participants with any event

Summary of findings 1. Azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
Patient or population: people with moderate to severe disease (WHO scale 4 to 9) Setting: inpatient Intervention: azithromycin Comparison: placebo or standard of care
Outcomes	Risk with placebo or standard of care	Risk with azithromycin	Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
All‐cause mortality at day 28	223 per 1000	219 per 1000 (201 to 236)	RR 0.98 (0.90 to 1.06)	8600 (4 RCTs)	⊕⊕⊕⊕ High^a	Azithromycin has little or no effect on all‐cause mortality at day 28
Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28	261 per 1000	248 per 1000 (227 to 269)	RR 0.95 (0.87 to 1.03)	7311 (1 RCT)	⊕⊕⊕⊖ Moderate^b	Azithromycin probably has little or no effect on worsening of clinical status or death at day 28
Improvement of clinical status: participants discharged alive at day 28	672 per 1000	645 per 1000 (564 to 746)	RR 0.96 (0.84 to 1.11)	8172 (3 RCTs)	⊕⊕⊕⊖ Moderate^c	Azithromycin probably has little or no effect on improvement of clinical status at day 28
Quality of life at longest follow‐up available	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at quality of life.
Serious adverse events during the study period	214 per 1000	238 per 1000 (190 to 300)	RR 1.11 (0.89 to 1.40)	794 (4 RCTs)	⊕⊕⊕⊖ Moderate^d	Azithromycin probably has little or no effect on serious adverse events during the study period
Any adverse events during the study period	333 per 1000	400 per 1000 (306 to 523)	RR 1.20 (0.92 to 1.57)	355 (3 RCTs)	⊕⊕⊖⊖ Low^e	Azithromycin may increase any adverse events slightly during the study period
Cardiac arrhythmias during the study period	45 per 1000	41 per 1000 (33 to 52)	RR 0.92 (0.73 to 1.15)	7865 (4 RCTs)	⊕⊕⊕⊖ Moderate^f	Azithromycin probably has little or no effect on cardiac arrhythmias during the study period
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk on the comparison group and the relative effect of the intervention (and its 95% confidence interval). CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is the possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aNo change of result in sensitivity analysis (RR 1.04; 95% CI 0.82 to 1.31; 837 participants; 3 studies) excluding studies with mixed populations (negative or unknown RT‐PCR). Therefore, no downgrading of evidence for indirectness. ^bDowngrade by one level for serious indirectness due to the effect estimate based on only one study with a mixed population (8.5% participants with negative or unknown RT‐PCR). ^cDowngrade by one level for serious heterogeneity (I² = 49%). No change of result in sensitivity analysis (RR 0.89; 95% CI 0.65 to 1.21; 402 participants; 2 studies) excluding studies with mixed populations (negative or unknown RT‐PCR). Therefore, no downgrading of evidence for indirectness. ^dDowngrade by one level for serious risk of bias due to an as‐treated analysis in the study with the highest weight (97.0%). The certainty of evidence was not downgraded due to indirectness, even if the imprecision of the effect estimate had increased in the sensitivity analysis excluding the study with a mixed population (RR 0.98; 95% CI 0.26 to 3.60; 355 participants; 3 studies), as the same study was already the reason for downgrading due to risk of bias. ^eDowngrade by one level for serious risk of bias due to non‐blinded outcome assessment in the study with the highest weight (87.7%), and one level for serious imprecision due to few small studies. ^fDowngrade by one level for serious indirectness due to the effect estimate based mainly (weight 99.1%) on two studies with mixed populations (8.5% and 11% participants with negative or unknown RT‐PCR).

Summary of findings 1. Azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Summary of findings 2. Azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
Patient or population: people with mild disease (WHO scale 1 to 3) Setting: outpatient Intervention: azithromycin Comparison: placebo or standard of care
Outcomes	Risk with placebo or standard of care	Risk with azithromycin	Relative effect (95% CI)	N° of participants (studies)	Certainty in the evidence (GRADE)	Comment
All‐cause mortality at day 28	2 per 1000	2 per 1000 (0 to 31)	RR 1.00 (0.06 to 15.69)	876 (3 RCTs)	⊕⊕⊖⊖ Low^a	Azithromycin may have little or no effect on all‐cause mortality at day 28
Admission to hospital or death within 28 days	67 per 1000	63 per 1000 (38 to 105)	RR 0.94 (0.57 to 1.56)	876 (3 RCTs)	⊕⊕⊖⊖ Low^b	Azithromycin may have little or no effect on admission to hospital or death within 28 days
All initial symptoms resolved (asymptomatic) at day 14	927 per 1000	955 per 1000 (881 to 1038)	RR 1.03 (0.95 to 1.12)	138 (1 RCT)	⊕⊕⊖⊖ Low^c	Azithromycin may have little or no effect on symptom resolution (all initial symptoms resolved) at day 14
Duration to symptom resolution	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at duration to symptom resolution.
Quality of life at longest follow‐up available	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at quality of life.
Any adverse events during the study period	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at any adverse events during the study period.
Serious adverse events during the study period	Two studies assessed serious adverse events during the study period, but none of the participants in either group were affected		Not estimable	454 (2 RCTs)	⊕⊖⊖⊖ Very low^d	We are uncertain whether azithromycin increases or reduces serious adverse events.
Cardiac arrhythmias during the study period	NA	NA	NA	(0 RCTs)	NA	No study was found that looked at cardiac arrhythmias during the study period
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk on the comparison group and the relative effect of the intervention (and its 95% confidence interval). CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is the possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.
^aDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in that study contributing events to the analysis; and by one level for serious imprecision due to a wide confidence interval and few events. ^bDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in one study and lack of registering the outcome in another study; and by one level for serious imprecision due to a wide confidence interval. ^cDowngrade by two levels for very serious imprecision due to the effect estimate based on one small study. ^dDowngrade by one level for serious risk of bias due to possible trial context related deviations from the intended interventions in one study and lack of registering the outcome in the other study; and by two levels for very serious imprecision due to zero events in both studies.

Summary of findings 2. Azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Table 1. Secondary outcomes for azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Outcomes	Relative effect (95% CI)	N° of participants (studies)	Measures of heterogeneity
New need for invasive mechanical ventilation at day 28	RR 0.65 (0.16 to 2.71)	7422 (2 RCTs)	Tau² = 0.64; Chi² = 1.57, df = 1 (P = 0.21); I² = 36%
Ventilator‐free days	MD ‐0.10 (‐1.16 to 0.96)	331 (1 RCT)	NA
New need for non‐invasive mechanical ventilation or high flow at day 28	RR 0.91 (0.78 to 1.05)	4073 (1 RCT)	NA
Weaning or liberation from invasive mechanical ventilation in surviving patients at day 28	RR 1.11 (0.85 to 1.45)	452 (1 RCT)	NA
Liberation from supplemental oxygen in surviving patients at day 28	RR 0.85 (0.67 to 1.08)	395 (1 RCT)	NA
Need for dialysis at up to 28 days	RR 1.03 (0.85 to 1.25)	8052 (2 RCTs)	Tau² = 0.00; Chi² = 1.27, df = 1 (P = 0.26); I² = 21%
Admission to the intensive care unit (ICU) at day 28	RR 0.28 (0.06 to 1.29	111 (1 RCT)	NA
Duration of hospitalisation	MD ‐0.61 (‐2.58 to 1.37)	448 (3 RCTs)	Tau² = 1.61; Chi² = 5.37, df = 2 (P = 0.07); I² = 63%
Viral clearance up to 7 days	RR 1.50 (0.35 to 6.40)	14 (1 RCT)	NA
Hospital‐acquired infections at up to 28 days	RR 1.16 (0.90 to 1.49)	395 (1 RCT)	NA
CI: confidence interval; NA: not applicable; MD: mean difference RCT: randomised controlled trial; RR: risk ratio

Table 1. Secondary outcomes for azithromycin compared to placebo or standard of care alone for inpatients with confirmed moderate to severe COVID‐19

Table 2. Secondary outcomes for azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Outcomes	Relative effect (95% CI)	N° of participants (studies)	Measures of heterogeneity
Need for invasive mechanical ventilation at 28 days	RR 0.61 (0.11 to 3.27)	405 (1 RCT)	NA
Need for hospitalisation with need for oxygen by mask or nasal prongs at 28 days	RR 0.80 (0.29 to 2.22)	406 (1 RCT)	NA
Viral clearance up to 7 days	RR 0.83 (0.44 to 1.54)	301 (1 RCT)	NA
CI: confidence interval; NA: not applicable; RCT: randomised controlled trial; RR: risk ratio

Table 2. Secondary outcomes for azithromycin compared to placebo or standard of care alone for outpatients with confirmed asymptomatic or mild COVID‐19

Table 3. Secondary outcomes for azithromycin compared to other antibiotics for inpatients with confirmed moderate to severe COVID‐19

Outcomes

Relative effect (95% CI)

N° of participants (studies)

Measures of heterogeneity

Viral clearance up to 7 days

RR 0.40
(0.17 to 0.93)

(1 RCT)

Table 3. Secondary outcomes for azithromycin compared to other antibiotics for inpatients with confirmed moderate to severe COVID‐19

Comparison 1. Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 All‐cause mortality at day 28 Show forest plot	4	8600	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.90, 1.06]

1.1.1 Moderate disease (WHO 4 to 5)	2	442	Risk Ratio (M‐H, Random, 95% CI)	0.61 [0.21, 1.77]
1.1.2 Moderate to severe disease (WHO 4 to 9)	2	8158	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.90, 1.07]
1.2 Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28 Show forest plot	1	7311	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.87, 1.03]

1.2.1 Moderate to severe disease (WHO 4‐9)	1	7311	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.87, 1.03]
1.3 Improvement of clinical status: participants discharged alive at day 28 Show forest plot	3	8172	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.84, 1.11]

1.3.1 Moderate disease (WHO 4‐5)	1	14	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.78, 1.29]
1.3.2 Moderate to severe disease (WHO 4‐9)	2	8158	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.72, 1.18]
1.4 Serious adverse events during the study period, defined as number of participants with any event Show forest plot	4	794	Risk Ratio (M‐H, Random, 95% CI)	1.11 [0.89, 1.40]

1.4.1 Moderate disease (WHO 4‐5)	3	355	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.26, 3.60]
1.4.2 Moderate to severe disease (WHO 4‐9)	1	439	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.89, 1.41]
1.5 Adverse events (any grade) during the study period, defined as number of participants with any event Show forest plot	3	355	Risk Ratio (M‐H, Random, 95% CI)	1.20 [0.92, 1.57]

1.5.1 Moderate disease (WHO 4‐5)	3	355	Risk Ratio (M‐H, Random, 95% CI)	1.20 [0.92, 1.57]
1.6 Cardiac arrhythmias during the study period Show forest plot	4	7865	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.73, 1.15]

1.6.1 Moderate disease (WHO 4‐5)	2	442	Risk Ratio (M‐H, Random, 95% CI)	0.46 [0.04, 5.05]
1.6.2 Moderate to severe disease (WHO 4‐9)	2	7423	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.74, 1.16]

Comparison 1. Azithromycin compared to placebo or standard of care for inpatients with confirmed moderate to severe COVID‐19

Comparison 2. Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 All‐cause mortality at day 28 Show forest plot	3	876	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.06, 15.69]

2.2 Admission to hospital or death within 28 days Show forest plot	3	876	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.57, 1.56]

2.3 All initial symptoms resolved (asymptomatic) at day 14 Show forest plot	1	138	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.95, 1.12]

2.4 All initial symptoms resolved (asymptomatic) at day 28 Show forest plot	2	549	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.79, 1.15]

2.5 Serious adverse events during the study period, defined as number of participants with any event Show forest plot	2	454	Risk Ratio (M‐H, Random, 95% CI)	Not estimable

Comparison 2. Azithromycin compared to placebo or standard of care for outpatients with confirmed asymptomatic or mild COVID‐19

Risk of bias for analysis 1.1 All‐cause mortality at day 28

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.1.1 Moderate disease (WHO 4 to 5)
Sekhavati 2020

	Insufficient information on allocation concealment. Sparsely reported baseline details.

	Both participants and those delivering the intervention were aware of the intervention received. None of the participants withdrew or discontinued the assigned intervention.No information provided on relevant co‐interventions.The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered. The outcome was not registered, but clinically relevant and seems not to be selected based on the result.

	Sparsely reported baseline details. Insufficient information on allocation concealment and lack blinding of participants and health care providers.
Cavalcanti 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for all participants (mITT population).

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.
Subgroup 1.1.2 Moderate to severe disease (WHO 4 to 9)
Furtado 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for nearly all participants (mITT population).

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.
RECOVERY 2021

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 16 (< 1%) participants withdrew consent. 91% and 1% in the azithromycin and control group received azithromycin, respectively. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.

Risk of bias for analysis 1.1 All‐cause mortality at day 28

Risk of bias for analysis 1.2 Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.2.1 Moderate to severe disease (WHO 4‐9)
RECOVERY 2021

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 16 (< 1%) participants withdrew consent. 91% and 1% in the azithromycin and control group received azithromycin, respectively. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis excluding those on any form of ventilation at randomisation.

	Data were available for all randomized participants not on any form of invasive ventilation at randomisation.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was not reported as registered, but seems not to be selected based on the result.

	No concerns.

Risk of bias for analysis 1.2 Worsening of clinical status: participants with clinical deterioration (new need for invasive mechanical ventilation) or death at day 28

Risk of bias for analysis 1.3 Improvement of clinical status: participants discharged alive at day 28

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.3.1 Moderate disease (WHO 4‐5)
CJWT629A12301

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received.

	The protocol was prospectively registered and the outcome in the study registry was reported as registered.

	No concerns. The study was terminated early due to poor recruitment.
Subgroup 1.3.2 Moderate to severe disease (WHO 4‐9)
Furtado 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for nearly all participants (mITT population).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.
RECOVERY 2021

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 16 (< 1%) participants withdrew consent. 91% and 1% in the azithromycin and control group received azithromycin, respectively. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was not registered, but seems not to be selected based on the result.

	No concerns.

Risk of bias for analysis 1.3 Improvement of clinical status: participants discharged alive at day 28

Risk of bias for analysis 1.4 Serious adverse events during the study period, defined as number of participants with any event

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.4.1 Moderate disease (WHO 4‐5)
CJWT629A12301

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received.

	The protocol was prospectively registered and the outcome in the study registry was reported as registered.

	No concerns. The study was terminated early due to poor recruitment.
NCT04335552

	Sparsely reported baseline details.

	Both participants and those delivering the intervention were aware of the intervention received. One participant in the control group withdrew. None of the participants discontinued the assigned intervention. No information provided on relevant co‐interventions. The primary analysis was an intention‐to‐treat analysis (one withrawal was not included).

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered and the outcome in the study register was not reported as registered, but seem not to be selected based on the results.

	Due to insufficient information provided on baseline details and lack of blinding of participants and health care providers. The protocol was prospectively registered and the outcome in the study register was not reported as registered, but seem not to be selected based on the results. The trial was terminated early, strong evidence from larger trials of no therapeutic effect.
Cavalcanti 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for all participants (mITT population).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was prespecified in the study protocol.

	No concerns.
Subgroup 1.4.2 Moderate to severe disease (WHO 4‐9)
Furtado 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Safety analyses were done in all patients who received at least one dose of study treatment. 2% of participants were not analysed in the group to which they were randomized (as treated analysis).

	Data were available for all participants (safety population).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome in the journal publication was reported as prespecified.

	The outcome was analysed in an as‐treated analysis (safety population).

Risk of bias for analysis 1.4 Serious adverse events during the study period, defined as number of participants with any event

Risk of bias for analysis 1.5 Adverse events (any grade) during the study period, defined as number of participants with any event

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.5.1 Moderate disease (WHO 4‐5)
NCT04335552

	Sparsely reported baseline details.

	Both participants and those delivering the intervention were aware of the intervention received. One participant in the control group withdrew. None of the participants discontinued the assigned intervention. No information provided on relevant co‐interventions. The primary analysis was an intention‐to‐treat analysis (one withrawal was not included).

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of intervention status could have influenced outcome assessment but there is no reason to believe that it did.

	The protocol was prospectively registered and the outcome in the study register was not reported as registered, but seem not to be selected based on the results.

	Due to insufficient information provided on baseline details and lack of blinding of participants and health care providers. The protocol was prospectively registered and the outcome in the study register was not reported as registered, but seem not to be selected based on the results. The trial was terminated early, strong evidence from larger trials of no therapeutic effect.
CJWT629A12301

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received.

	The protocol was prospectively registered and the outcome in the study registry was reported as registered.

	No concerns. The study was terminated early due to poor recruitment.
Cavalcanti 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for all participants (mITT population).

	Outcome assessors were aware of the intervention received. Knowledge of intervention status could have influenced outcome assessment but there is no reason to believe that it did.

	The protocol was prospectively registered. The outcome was prespecified in the study protocol.

	Outcome assessors were not blinded and knowledge of intervention status could have influenced outcome assessment but there is no reason to believe that it did.

Risk of bias for analysis 1.5 Adverse events (any grade) during the study period, defined as number of participants with any event

Risk of bias for analysis 1.6 Cardiac arrhythmias during the study period

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Subgroup 1.6.1 Moderate disease (WHO 4‐5)
Sekhavati 2020

	Insufficient information on allocation concealment. Sparsely reported baseline details.

	Both participants and those delivering the intervention were aware of the intervention received. None of the participants withdrew or discontinued the assigned intervention.No information provided on relevant co‐interventions.The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was not registered, but clinically relevant and seems not to be selected based on the result.

	Sparsely reported baseline details. Insufficient information on allocation concealment and lack blinding of participants and health care providers.
Cavalcanti 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Few (< 1%) participants withdrew consent. 2% and 1% in the azithromycin and control group did not receive the assigned treatment. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an modified intention‐to‐treat analysis. RT‐PCR negative participants were excluded.

	Data were available for all participants (mITT population).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was prespecified in the study protocol.

	No concerns.
Subgroup 1.6.2 Moderate to severe disease (WHO 4‐9)
Furtado 2020

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. Safety analyses were done in all patients who received at least one dose of study treatment. 2% of participants were not analysed in the group to which they were randomized (as treated analysis).

	Data were available for all participants (safety population).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome in the journal publication was not registered.

	The outcome was analysed in an as‐treated analysis (safety population).
RECOVERY 2021

	Random sequence was generated using a computer‐based program. Allocation was concealed until after randomisation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 16 (< 1%) participants withdrew consent. 91% and 1% in the azithromycin and control group received azithromycin, respectively. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants with complete follow‐up forms (10% missing data).

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was not registered, but clinically relevant and seems not to be selected based on the result.

	No concerns.

Risk of bias for analysis 1.6 Cardiac arrhythmias during the study period

Risk of bias for analysis 2.1 All‐cause mortality at day 28

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Omrani 2020

	The study used a computer‐based randomization and a central allocation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received.

	The protocol was prospectively registered. The outcome was not registered, but clinically relevant and seems not to be selected based on the result.

	No concerns.
PRINCIPLE 2021

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 18 (1%) participants withdrew consent with comparable distribution between groups. 87% in the azithromycin received at least one dose azithromycin. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis. RT‐PCR negative participants were excluded in a mITT analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.
Hinks 2020

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of intervention received. The reason for complete withdrawals/withdrawal are consistent with routine care and equal between groups. Non‐compliance with experimental intervention (AZT) is major and reasons for non‐compliance not described, therefore it cannot be judged if those deviations were due to the trial context (open‐label). However, the analysis was appropriate in this context (mITT).

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of the intervention received could not have affected outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	Due to possible trial context related deviations from the intended interventions.

Risk of bias for analysis 2.1 All‐cause mortality at day 28

Risk of bias for analysis 2.2 Admission to hospital or death within 28 days

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Omrani 2020

	The study used a computer‐based randomization and a central allocation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received by the participants.

	The protocol was prospectively registered. The outcome in the journal publication was not registered.

	Due to lack of registering the outcome.
Hinks 2020

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of intervention received. The reason for complete withdrawals/withdrawal are consistent with routine care and equal between groups. Non‐compliance with experimental intervention (AZT) is major and reasons for non‐compliance not described, therefore it cannot be judged if those deviations were due to the trial context (open‐label). However, the analysis was appropriate in this context (mITT).

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered. The outcome was registered differently than reported, but results and reason for outcome modification were in reported full detail in the publication. Due to relevance of this outcome data in this context of this trial we did not assume that the results have have been selected.

	Due to possible trial context related deviations from the intended interventions.
PRINCIPLE 2021

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 18 (1%) participants withdrew consent with comparable distribution between groups. 87% in the azithromycin received at least one dose azithromycin. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis. RT‐PCR negative participants were excluded in a mITT analysis.

	Data were available for all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.

Risk of bias for analysis 2.2 Admission to hospital or death within 28 days

Risk of bias for analysis 2.3 All initial symptoms resolved (asymptomatic) at day 14

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Omrani 2020

	The study used a computer‐based randomization and a central allocation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The primary analysis was an intention‐to‐treat analysis.

	Data were available for > 90% of randomized participants. Missing data were equally distributed between groups and reasons were described.

	Outcome assessors were not aware of the intervention received by the participants.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.

Risk of bias for analysis 2.3 All initial symptoms resolved (asymptomatic) at day 14

Risk of bias for analysis 2.4 All initial symptoms resolved (asymptomatic) at day 28

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
PRINCIPLE 2021

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of the intervention received. 18 (1%) participants withdrew consent with comparable distribution between groups. 87% in the azithromycin received at least one dose azithromycin. Changes from assigned interventions are consistent with what would occur outside the trial context. The primary analysis was an intention‐to‐treat analysis. RT‐PCR negative participants were excluded in a mITT analysis.

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. Knowledge of intervention status could have influenced outcome assessment but there is no reason to believe that it did.

	The protocol was prospectively registered. The outcome on symptom recovery was added during the study due to lower hospitalistion rates than originally expected.

	Due to lack of blinding of outcome assessors.
Omrani 2020

	The study used a computer‐based randomization and a central allocation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The primary analysis was an intention‐to‐treat analysis.

	Data were available for > 90% of randomized participants. Missing data were equally distributed between groups and reasons were described.

	Outcome assessors were not aware of the intervention received by the participants.

	The protocol was prospectively registered and the outcome in the journal publication was reported as registered.

	No concerns.

Risk of bias for analysis 2.4 All initial symptoms resolved (asymptomatic) at day 28

Risk of bias for analysis 2.5 Serious adverse events during the study period, defined as number of participants with any event

Bias
Study	Randomisation process	Deviations from intended interventions	Missing outcome data	Measurement of the outcome	Selection of the reported results	Overall
Omrani 2020

	The study used a computer‐based randomization and a central allocation. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were not aware of the intervention received. The primary analysis was an intention‐to‐treat analysis.

	Data were available for all randomized participants.

	Outcome assessors were not aware of the intervention received by the participants.

	The protocol was prospectively registered. The outcome in the journal publication was not registered.

	Due to lack of registering the outcome.
Hinks 2020

	Random sequence was generated using a web‐based system. The trial team was masked to the randomisation ratios and allocation was concealed. There were no baseline imbalances that would suggest a problem with randomisation.

	Both participants and those delivering the intervention were aware of intervention received. The reason for complete withdrawals/withdrawal are consistent with routine care and equal between groups. Non‐compliance with experimental intervention (AZT) is major and reasons for non‐compliance not described, therefore it cannot be judged if those deviations were due to the trial context (open‐label). However, the analysis was appropriate in this context (mITT).

	Data were available for nearly all randomized participants.

	Outcome assessors were aware of the intervention received. By following a clinical protocol, knowledge of the intervention received could only minimally affect the outcome measurement.

	The protocol was prospectively registered and the outcome in the study register was reported as registered.

	Due to possible trial context related deviations from the intended interventions.

Risk of bias for analysis 2.5 Serious adverse events during the study period, defined as number of participants with any event