Prophylactic anticoagulants for people hospitalised with COVID‐19

Summary of findings 1. Anticoagulants (all types) compared to no treatment for people hospitalised with COVID‐19

Outcomes	Impact	№ of participants (studies)	Certainty of the evidence (GRADE)
Anticoagulants (all types) compared to no treatment for people hospitalised with COVID‐19
Patient or population: people hospitalised with COVID‐19 Setting: hospital (ICU and ward) Intervention: anticoagulants (all types) Comparison: no treatment
All‐cause mortality Follow‐up: range 8 to 28 days	One study reported reduction of mortality by OR adjusted for confounding (reduction of 58% on chance of death; 2075 participants). One study reported reduction of mortality only in a subgroup of severely ill participants (HR 0.86, 95% CI 0.82 to 0.89; 395 participants). Three studies reported no differences by adjusted OR (1.64, 95% CI 0.92 to 2.92; 449 participants), unadjusted OR (1.66, 95% CI 0.76 to 3.64; 154 participants) or adjusted RR (1.15, 95% CI 0.29 to 2.57; 192 participants). One study reported zero events in both intervention groups.	5685 (6 retrospective NRS)	⊕⊝⊝⊝ Very low^a,b,c
Necessity for additional respiratory support	No study measured this outcome
Mortality related to COVID‐19	No study measured this outcome
Deep vein thrombosis	No study measured this outcome
Pulmonary embolism	No study measured this outcome
Major bleeding Follow‐up: not reported	One study reported 24 (3%) bleeding events in the intervention group and 38 (1.9%) bleeding events in the control group (OR 1.62, 95% CI 0.96 to 2.71).	2773 (1 retrospective NRS)	⊕⊕⊝⊝ Low^c,d
CI: confidence interval; HR: hazard ratio; ICU: intensive care unit; NRS: non‐randomised studies; OR: odds ratio; 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 a 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.
^aDowngraded one level due to study limitations. Overall critical/serious risk of bias across studies, especially related to confounding. ^bDowngraded one level due to inconsistency. We decided not to pool data due to the heterogeneity of studies (especially due to differences in interventions). ^cDowngraded one level due to imprecision. Narrative synthesis was conducted with imprecise estimates. ^dDowngraded one level due to study limitations. Overall serious risk of bias, especially related to confounding.

Summary of findings 2. Anticoagulants (therapeutic dose) compared to anticoagulants (prophylactic dose) for people hospitalised with COVID‐19

Outcomes	Impact	№ of participants (Studies)	Certainty of the evidence (GRADE)
Anticoagulants (therapeutic dose) compared to anticoagulants (prophylactic dose) for people hospitalised with COVID‐19
Patient or population: people hospitalised with COVID‐19 Setting: hospital (ICU and ward) Intervention: anticoagulants (therapeutic dose) Comparison: anticoagulants (prophylactic dose)
All‐cause mortality Follow‐up: 35 days	One study reported an absolute rate of death lower in the therapeutic group (34.2% versus 53%) and an HR adjusted for confounding of 0.21 (95% CI 0.10 to 0.46).	244 (1 retrospective NRS)	⊕⊕⊝⊝ Low^a,b
Necessity for additional respiratory support	No study measured this outcome
Mortality related to COVID‐19	No study measured this outcome
Deep vein thrombosis	No study measured this outcome
Pulmonary embolism	No study measured this outcome
Major bleeding Follow‐up: 35 days	One study reported 51 (31.7%) bleeding events in the intervention group and 17 (20.5%) bleeding events in the control group (OR 1.80, 95% CI 0.96 to 3.37).	244 (1 retrospective NRS)	⊕⊕⊝⊝ Low^a,b
CI: confidence interval; HR: hazard ratio; ICU: intensive care unit; NRS: non‐randomised studies; OR: odds 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 a 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.
^aDowngraded one level due to study limitations. Overall serious risk of bias, especially related to selection bias. ^bDowngraded one level due to imprecision. Narrative synthesis was conducted with imprecise estimates based on fewer than 400 participants.

Background

See Table 1 for a glossary of terms.

Table 1. Glossary of terms

Term	Definition
Anticoagulants	Drugs that suppress, delay or prevent blood clots
Antiplatelet agents	Drugs that prevent blood clots by inhibiting platelet function
Arterial thrombosis	An interruption of blood flow to an organ or body part due to a blood clot blocking the flow of blood
Body mass index (BMI)	Body mass divided by the square of the body height, universally expressed in units of kg/m²
Catheters	Medical devices (tubes) that can be inserted in the body for a broad range of functions, such as to treat diseases, to perform a surgical procedure, and to provide medicine, fluids and food.
COVID‐19	An infectious disease caused by SARS‐CoV‐2 virus
Deep vein thrombosis (DVT)	Coagulation or clotting of the blood in a deep vein, i.e. far beneath the surface of the skin
Disseminated intravascular coagulopathy	A severe condition in which blood clots form throughout the body, blocking small blood vessels and that may lead to organ failure. As clotting factors and platelets are used up, bleeding may occur, throughout the body (e.g. in the urine, in the stool, or bleeding into the skin)
Duplex ultrasound	Non‐invasive evaluation of blood flow through the arteries and veins by ultrasound devices
Heparin (also known as unfractionated heparin (UFH))	A drug used to prevent blood clotting (anticoagulant, blood thinner)
Hypercoagulability	An abnormality of blood coagulation that increases the risk of blood clot formation in blood vessels (thrombosis)
Low molecular weight heparin	A drug used to prevent blood clotting (anticoagulant)
Obesity	Amount of body fat beyond healthy conditions (BMI > 30 kg/m²)
Placebo	Substance or treatment with no active effect, like a sugar pill
Platelet	Colourless blood cells that help blood clot by clumping together
Pulmonary embolism (PE)	Blood clot in the lung or blood vessel leading to the lung. The clot originates in a vein (e.g. deep vein thrombosis) and travels to the lung
Quasi‐randomised controlled trial (Quasi‐RCT)	A study in which participants are divided by date of birth or by hospital register number, i.e. not truly randomly divided into separate groups to compare different treatments
Randomised controlled trial (RCT)	A study in which participants are divided randomly into separate groups to compare different treatments
Respiratory failure	An abnormality that results from inadequate gas exchange by the respiratory system
SARS‐CoV‐2	The virus (coronavirus 2) that causes COVID‐19
Thrombosis	Local coagulation of blood (clot) in a part of the circulatory system
Vascular	Relating to blood vessels (arteries and veins)
Venous	Relating to a vein
Venous thromboembolism (VTE)	A condition that involves a blood clot that forms in a vein and may migrate to another location (e.g. the lung)

Description of the condition

The novel coronavirus disease strain, coronavirus disease 2019 (COVID‐19), is caused by severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2). COVID‐19 emerged in Wuhan, China and rapidly spread worldwide (Lai 2020). SARS‐CoV‐2 is a highly transmittable virus, and up to 16% of people hospitalised may develop a severe form of the disease (Giannis 2020). Pulmonary effects are typical, but due to high inflammation, hypoxia, immobilisation and diffuse intravascular coagulation, COVID‐19 may predispose patients to both arterial and venous thromboembolism (Ackermann 2020; Dolhnikoff 2020; Fox 2020; Long 2020). Venous and arterial thromboembolic complications affect 16% of people hospitalised with COVID‐19 and 31% to 49% of people with COVID‐19 in intensive care units (ICUs), with 90% of such cases being venous thromboembolism (Bilaloglu 2020; Klok 2020a; Klok 2020b). Viral infections induce an imbalance between anticoagulant and procoagulant mechanisms and raise the systemic inflammatory response. Indeed, people with COVID‐19 commonly present with both elevated D‐dimer (fibrin degradation product) and reductions of factors related to clot formation (Giannis 2020). Excessive activation of the coagulation cascade and platelets can explain these haematological findings (Giannis 2020). Coagulopathy and vascular endothelial dysfunction have been proposed as complications of COVID‐19. Emerging data support that asymptomatic people with COVID‐19 are at risk of developing pathologic thrombosis. The association between large‐vessel stroke and COVID‐19 in young asymptomatic people requires further investigation (Oxley 2020), but Li 2020 found the incidence of stroke among people hospitalised with COVID‐19 was approximately 5% in a retrospective cohort. Activation of the coagulation system seems to be important in the development of acute respiratory distress syndrome, one of the most typical complications of COVID‐19 infection and it can be related to pulmonary microthrombosis (Ackermann 2020; Dolhnikoff 2020; Fox 2020; Marini 2020).

Description of the intervention

Anticoagulants are pharmacological interventions used in reducing hypercoagulability (Amaral 2020). The decision to use, or not use, thromboprophylaxis, depends on the risk stratification of each patient (NHS 2020).

Anticoagulants are medications used in the prevention and treatment of venous or arterial thromboembolic events (Amaral 2020; Biagioni 2020; Clezar 2020). When used for a prophylactic purpose, the dose of anticoagulants is usually half or significantly lower than that given for therapeutic purposes (Alquwaizani 2013). Even so, adverse events, such as bleeding may occur, and can have a significant impact on patient care (Amaral 2020; AVF 2020; Biagioni 2020; Clezar 2020).

How the intervention might work

D‐dimers are a reflection of the pathophysiology in COVID‐19, which is highly associated with increased mortality in people with COVID‐19 infection (Becker 2020). The elevated D‐dimer levels seen are most likely a reflection of the overall clot burden and critically ill people with COVID‐19 have lower levels of fibrinolytic system activation than the reference population (Panigada 2020). Tang 2020 reported decreased mortality after use of heparin in people with COVID‐19 (40.0% versus 64.2%, P = 0.029). Long 2020 reported that anticoagulation (mainly low molecular weight heparin), may reduce mortality in people with severe COVID‐19 infection or those with higher levels of D‐dimer (e.g. greater than six times the upper limit).

Some authors had also correlated this effect with the anti‐inflammatory effect of heparinoids, for instance, binding and neutralising a wide variety of mediators released from inflammatory cells, reducing IL‐6 and as potent inhibitors of the complement system, which may have effects on the clinical evolution of people with COVID‐19 (Liu 2019; Shi 2020; Tang 2020; Young 2008). It can attenuate ongoing tissue damage (Liu 2019; Young 2008). Practical guidelines and specialist consensus are addressing the management of thromboprophylaxis and anticoagulation in people with COVID‐19 infection (Bikdeli 2020; NHS 2020; Obe 2020; Ramacciotti 2020). However, the effects of anticoagulants on people with COVID‐19 is still under debate.

Objectives

To assess the effects of prophylactic anticoagulants versus active comparator, placebo or no intervention, on mortality and the need for respiratory support in people hospitalised with COVID‐19.

Methods

Criteria for considering studies for this review

Types of studies

The protocol for this review was prospectively registered with the Open Science Framework on 7 August 2020 (Flumignan 2020).

We considered parallel or cluster‐randomised controlled trials (RCTs), quasi‐RCTs, and cohort studies. Cohort studies may be useful for rare adverse events and clinical decisions if there is a lack of controlled studies. We did not consider studies without a comparator group. Although cohort studies (non‐randomised) were considered, we planned to limit our primary analyses to specific studies, that is, RCTs and quasi‐RCTs. We did not perform a meta‐analysis of non‐randomised studies (NRS), and we analysed their data narratively. In future updates of this review, when at least 400 participants are included from RCTs, we will no longer consider NRS for inclusion. We considered all other types of studies irrelevant for this review. Please find further explanations in Appendix 1.

In order to minimise selection bias for NRS, we planned to include only studies that used statistical adjustment for baseline factors using multivariate analyses for at least these confounding factors:

participants already using anticoagulants (e.g. atrial fibrillation)
participants who underwent surgery during the hospitalisation
active cancer treatment
concomitant antiplatelet use
history of venous thromboembolism

We considered only studies with a minimum duration of two weeks.

Types of participants

We included all participants eligible for prophylactic anticoagulation, both male and female of all ages, hospitalised with the diagnosis of COVID‐19. Any hospitalised participants with confirmed COVID‐19 infection were eligible, independently of the disease severity (e.g. patients hospitalised in ICUs or wards). We had also considered participants with the previous history of venous thromboembolism for inclusion in this review. However, the participants with COVID‐19 treated out of the hospital, i.e. those who were not hospitalised were not eligible for our review.

In future updates of this review, if we find studies with mixed populations, that is, hospitalised and non‐hospitalised participants, and only a subset of the participants meets our inclusion criteria, we will attempt to obtain data for the subgroup of interest from the study authors in order to include the study. For studies with mixed populations for which we cannot get the subgroup of interest's data but at least 50% of the study population are of interest, we will include all participants in our analysis. Moreover, we will explore the effect of this decision in a sensitivity analysis. Studies in which less than 50% of the population are of interest and the subgroup of interest data are not available will be excluded.

Types of interventions

We considered the following pharmacological interventions.

Heparinoids, that is, both unfractionated heparin and low molecular weight heparin, and pentasaccharides (synthetic and selective anticoagulant drugs similar to low molecular weight heparin)
Vitamin K antagonists
Direct anticoagulants, both factor Xa inhibitors and direct thrombin inhibitors, that is, direct oral anticoagulants and non‐oral direct anticoagulants (e.g. bivalirudin).

We considered studies comparing different formulations, doses and schedules of the same intervention (e.g. heparinoids).

Some commonly applicable prophylactic doses of the interventions of interest are low molecular weight heparin 30 mg twice a day or 40 mg daily, and unfractionated heparin 5000 IU three times a day. However, we considered all doses of anticoagulants, when used for primary or secondary prophylaxis of thromboembolism, eligible for our review.

Types of comparisons

We included studies that compared one pharmacological intervention (agent or drug) versus another active comparator, or placebo or no treatment with any combination of interventions, provided that co‐treatments were balanced between the treatment and control arms. We allowed other potential interventions (e.g. antiplatelet agents, elastic stockings, intermittent pneumatic compression) as comparators or additional interventions. We also included studies that compared different doses of drugs. We pooled the studies that addressed the same comparisons.

Anticoagulant versus placebo or no treatment (we planned to pool all anticoagulants together – heparinoids, vitamin K antagonists, direct anticoagulants, etc. – if possible)
Anticoagulant versus a different anticoagulant
Anticoagulant versus a different dose, formulation, or schedule of the same anticoagulant
Anticoagulant versus other pharmacological interventions such as antiplatelet agents
Anticoagulant versus non‐pharmacological interventions

Types of outcome measures

We evaluated core outcomes as pre‐defined by the Core Outcome Measures in Efectiveness Trials Initiative for people with COVID‐19 (COMET 2020). We also considered the outcomes after hospital discharge. We intended to present the outcomes at two different time points following the start of the intervention if data were available:

early outcomes (at hospital discharge or before);
long‐term outcomes (after hospital discharge).

Our time point of primary interest is early; we, therefore, intended to produce related 'Summary of findings' tables only for this time point but we also planned to report the long‐term outcomes at the longest possible time of follow‐up.

Primary

All‐cause mortality
Necessity for additional respiratory support:
- oxygen by non‐invasive ventilators or high flow
- intubation and mechanical ventilation
- extracorporeal membrane oxygenation

Secondary

Mortality related to COVID‐19
Deep vein thrombosis (DVT), symptomatic or asymptomatic, first episode or recurrent confirmed by ultrasonography or angiography (e.g. by computed tomography (CT), magnetic resonance imaging (MRI) or by digital subtraction) from any site (e.g. lower limbs, upper limbs, abdominal).
Pulmonary embolism (symptomatic or asymptomatic, first episode or recurrent, fatal or non‐fatal): a diagnosis had to be confirmed by angiography (e.g. by CT, MRI or digital subtraction) and ventilation‐perfusion scan, or both. We also considered post mortem examination as an objective confirmation of DVT and pulmonary embolism.
Major bleeding: defined by a haemoglobin concentration decrease of 2 g/dL or more, a retroperitoneal or intracranial bleed, a transfusion of two or more units of blood, or fatal haemorrhagic events, as defined by International Society on Thrombosis and Haemostasis (Schulman 2010).
Adverse events. We will consider all possible adverse events separately, as individual outcomes, such as minor bleeding, gastrointestinal adverse effects (e.g. nausea, vomiting, diarrhoea, abdominal pain), allergic reactions, renal failure and amputations
Hospitalisation time in days
Quality of life: participant's subjective perception of improvement (yes or no) as reported by the study authors or using any validated scoring system such as the Short Form‐36 Health Survey (SF‐36) (Ware 1992).

We planned to include studies in the review irrespective of whether measured outcome data were reported in a ‘usable’ way.

Search methods for identification of studies

An information specialist (LLA) designed and conducted all searches on 20 June 2020, which were informed and verified by a content expert (RLGF) and independently peer reviewed.

Electronic searches

We identified eligible study references through systematic searches of the following bibliographic databases.

Cochrane Central Register of Controlled Trials (CENTRAL; 2020, Issue 6) in the Cochrane Library (searched 20 June 2020; Appendix 2)
MEDLINE PubMed (1946 to 20 June 2020; Appendix 3)
Embase Wiley (1974 to 20 June 2020; Appendix 4)
LILACS Virtual Health Library (Latin American and Caribbean Health Sciences Literature database; 1982 to 20 June 2020; Appendix 5)
IBECS Virtual Health Library (Indice Bibliográfico Español de Ciencias de la Salud; 2015 to 20 June 2020; Appendix 5)

We adapted the preliminary search strategy for MEDLINE (PubMed; Appendix 3) for use in the other databases. We did not apply any RCT filters for any databases, but we selected the study design manually because we also considered NRS for inclusion in this review.
We searched all databases from their inception to the present, and we did not restrict the language of publication or publication status. We considered the adverse effects described in the included studies only.

Searching other resources

We also conducted a search of the Cochrane COVID-19 Study Register (Appendix 6), and medRxiv (Appendix 7), for ongoing or unpublished studies (both searched 20 June 2020).
We checked reference lists of all included studies and any relevant systematic reviews identified for additional references to studies. We examined any relevant retraction statements and errata for included studies. We contacted the authors of the included studies for any possible unpublished data. Furthermore, we contacted field specialists to enquire about relevant ongoing or unpublished studies.

Data collection and analysis

Inclusion of non‐English language studies

We considered abstracts and full texts in all languages for inclusion. All potentially eligible non‐English language abstracts progressed to full‐text review, with methods translated for eligibility, and full text translated for data extraction.

Selection of studies

Two review authors (JDST, LCUN) independently screened titles and abstracts of all the potential studies we identified as a result of the search and coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve', using the Covidence tool. If there were any disagreements, we asked a third review author to arbitrate (RLGF). We retrieved the full‐text study reports/publications, and two review authors (JDST, LCUN) independently screened the full text and identified studies for inclusion, and identified and recorded reasons for exclusion of the ineligible studies. We resolved any disagreement through discussion or, if required, we consulted a third person (RLGF). We identified and excluded duplicates and collated multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram and ‘Characteristics of excluded studies’ table (Liberati 2009). We considered studies reported as full text, those published as abstract only, and unpublished data. We considered abstracts and conference proceedings if they were eligible and had usable data.

Data extraction and management

We managed and synthesised the available data using Review Manager 5 (Review Manager 2020). If there was a conflict between data reported across multiple sources for a single study (e.g. between a published article and a trial registry record), we planned to use the article published for numerical analysis, and we planned to report the differences and consider it on the certainty of evidence (GRADE approach; Schünemann 2013).

We planned to use a data collection form, which we piloted on at least one study in the review, for study characteristics and outcome data. We planned that one review author (RLGF) would extract study characteristics from included studies. We planned to extract the following study characteristics.

Methods: study design, total duration of the study, number of study centres and location, study setting, and date of the study
Participants: comorbidities, ventilation support, pregnancy, number randomised, number lost to follow‐up/withdrawn, number analysed, number of interest, mean age, age range, gender, the severity of the condition, inclusion criteria, and exclusion criteria
Interventions: intervention and comparison characteristics (e.g. manufacture, dosage, additional procedures, method of administration), concomitant medications, and excluded medications
Outcomes: primary and secondary outcomes specified and collected (e.g. how outcomes are measured), and time points reported. For NRS: confounding factors controlled for each relevant analysis presented
Notes: funding for the trial, and notable conflicts of interest of study authors

We planned for one review author (RLGF) to extract outcome data from included studies independently, which would be verified by the other two review authors (CM, BT). We planned to resolve disagreements by discussion. We planned for one review author (RLGF) to transfer data into Review Manager 5 (RevMan 5; Review Manager 2020). We planned to double‐check that data were entered correctly by comparing the data presented in the systematic review with the data extraction form. We planned for two review authors (CM, BT) to spot‐check study characteristics for accuracy against the study report.

Assessment of risk of bias in included studies

For data from RCTs we planned to use the 'Risk of bias' 1.0 tool to analyse the risk of bias in the underlying study results (Higgins 2017). For data from quasi‐RCTs or prospective NRS, we planned to use the Risk Of Bias in Non‐randomised Studies of Interventions (ROBINS‐I) tool (Sterne 2016). We also planned to use ROBINS‐I to assess the risk of bias in retrospective NRS. Please refer to Appendix 1 for detailed information regarding how we planned to assess the risk of bias of RCTs, quasi‐RCTs, and NRS.

We considered the following confounders for the assessment of ROBINS‐I domain on 'confounding' and used the Robvis tool to create the 'risk of bias' graphs for NRS (McGuinness 2020).

Participants already using anticoagulants (e.g. atrial fibrillation)
Participants who underwent surgery during hospitalisation
Active cancer treatment
Concomitant antiplatelet use
History of venous thromboembolism

Measures of treatment effect

Please refer to Appendix 1 for information regarding how we had planned to measure the treatment effects of RCTs, quasi‐RCTs and NRS.

Unit of analysis issues

As we included NRS only, meta‐analysis was not appropriate. Instead, we narratively described and presented results per study also using tables.

Please refer to Appendix 1 for information regarding how we had planned to combine studies with multiple treatment groups.

Dealing with missing data

We planned to contact investigators or study sponsors in order to verify key study characteristics and obtain missing numerical outcome data where possible (e.g. when a study is identified as abstract only). Where possible, we planned to use the RevMan 5 calculator to calculate missing standard deviations using other data from the trial, such as confidence intervals. Where this was not possible, and the missing data were thought to introduce serious bias, we planned to explore the impact of including such studies in the overall assessment of results by a sensitivity analysis. For all outcomes, we planned to follow intention‐to‐treat (ITT) principles to the highest degree possible: that is, we planned to analyse participants in their randomised group regardless of what intervention they received. We planned to use available‐case data for the denominator if ITT data were not available. We estimated the mean difference (MD) using the method reported by Wan 2014 to convert median and interquartile range (IQR) into MD and confidence intervals (CI). When it was not possible, we narratively described skewed data reported as medians and IQRs.

Dealing with sparse data

We planned to adjust comparisons (e.g. grouping broader categories of participants (all ages), grouping broader of variations of intervention (all types of anticoagulants) accordingly, regardless of sparse data.

Assessment of heterogeneity

As we identified NRS only, meta‐analysis was not appropriate. Instead, we narratively described and presented results per study in tables.

Please refer to Appendix 1 for information regarding how we had planned to assess heterogeneity.

Assessment of reporting biases

If we were able to pool more than 10 studies, we planned to create and examine a funnel plot to explore possible small‐study biases for the primary outcomes.

Data synthesis

Please refer to Appendix 1 for information regarding how we had planned to synthesise data from RCTs, quasi‐RCTs and NRS. We did not meta‐analyse data from NRS. We reported outcome data of each included study narratively and using tables.

Synthesis without meta‐analysis

We planned to synthesise the data using RevMan 5 (Review Manager 2020). We planned to report data narratively if it was not appropriate to combine in a meta‐analysis. We planned to undertake meta‐analyses only where this was meaningful, that is, if the treatments, participants and the underlying clinical question were similar enough for pooling to make sense.
We planned to analyse data from NRS separately in a spreadsheet with the exposure of the sample number and the quantitative and qualitative variables relevant to the review.
We planned to describe skewed data reported as medians and interquartile ranges narratively.

If a meta‐analysis was not possible, we explored the possibilities above to show data of all relevant outcomes considered in this review. Where there was substantial clinical, methodological, or statistical heterogeneity across studies that prevented the pooling of data, we used a narrative approach to data synthesis. We planned to describe narratively skewed data reported as medians and interquartile ranges.

Subgroup analysis and investigation of heterogeneity

We planned to explore the following subgroups related to participants or interventions, if heterogeneity was substantial.

Different doses of drugs
Duration of prophylaxis (e.g. until 30 days after the start of intervention or more)
Age (e.g. children (up to 18 years), adults (18 years to 64 years) and seniors (65 years and over))
Gender
Comorbidities
Type of ventilator support:
- oxygen by non‐invasive ventilators or high flow
- intubation and mechanical ventilation
- extracorporeal membrane oxygenation

Sensitivity analysis

We planned to carry out the following sensitivity analyses to test whether critical methodological factors or decisions have affected the main result. We planned to group according to study design (RCTs or cluster‐RCTs, quasi‐RCTs, NRS).

Only including studies with a low risk of bias, as previously specified ('Assessment of risk of bias in included studies').
We planned to examine both the fixed‐effect model and random‐effects model meta‐analyses, and we planned to explore the differences between the two estimates.
We planned to explore the decision to include all participants when at least 50% were of interest in a study with a mixed population.
We planned to explore the impact of missing data. If we identified studies with missing data that were unobtainable, we planned to repeat analyses excluding these studies to determine their impact on the primary analyses.

We also planned to carry out sensitivity analyses considering cluster‐RCTs. We planned to investigate the effect of variation in the intracluster correlation coefficient (ICC), and we also planned to acknowledge heterogeneity in the randomisation unit and perform a sensitivity analysis to investigate the effects of the randomisation unit. We planned to present these results and compare them with the overall findings. We planned to justify any post hoc sensitivity analyses that arose during the review process in the final report.

Summary of findings and assessment of the certainty of the evidence

We created a 'Summary of findings' table for the early time point using the following outcomes.

All‐cause mortality
Necessity for additional respiratory support
Mortality related to COVID‐19
DVT
Pulmonary embolism
Major bleeding

We used the five GRADE considerations (study limitations; consistency of effect; imprecision; indirectness; and publication bias) to assess the certainty of a body of evidence as it relates to the studies that contribute data to the analyses for the prespecified outcomes. We used methods and recommendations described in Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2019) using GRADEpro software (GRADEpro GDT 2015). We made a separate 'Summary of findings' table for each of the following comparisons with available data.

Anticoagulant (all types) versus no treatment
Anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)

We justified all decisions to downgrade the certainty of studies using footnotes, and we made comments to aid the reader's understanding of the review where necessary.

Two review authors (RLGF, LCUN) made judgements about the certainty of the evidence, with disagreements resolved by discussion or by involving a third review author (CM, BT). We justified, documented and incorporated judgements into reporting of results for each outcome.
We plan to extract study data, format our comparisons in data tables and prepare a 'Summary of findings' table with meta‐analysis before writing the results and conclusions of future updates of our review.

Results

Results of the search

We retrieved a total of 1148 records from our searches. After excluding 103 duplicate records, we screened 1045 unique records. We considered a total of 991 records not relevant at this stage and selected 54 for full‐text reading. We excluded 12 studies (11 reports) (see Characteristics of excluded studies). Twenty‐two studies are ongoing (see Characteristics of ongoing studies). We considered another 13 studies not relevant after a full‐text analysis. For this review, we found seven non‐randomised studies (NRS) with available data for inclusion. See Figure 1 for the study flow diagram (Liberati 2009).

Figure 1

Study flow diagram
RCTs: randomised controlled trials; NRS: non‐randomised studies

Included studies

See Table 2 for the summarised characteristics of included studies.

Table 2. Summary of characteristics of included studies

Study (design)	Country	Participant age (mean)	Setting	Intervention type (dose)	Comparator	All‐cause mortality	Necessity for additional respiratory support	Follow‐up time (mean days)	Total participants allocated	Intervention group participants (anticoagulant)
Ayerbe 2020 (Retrospective cohort)	Spain	67	Hospital^a	Heparin (NR)	NA	OR 0.42 (95% CI 0.26 to 0.66) P < 0.001, in favour of intervention group	NR	8	2075	1734
Liu 2020 (Retrospective cohort)	China	72	ICU (intervention) vs hospital ward (comparator)	Heparin (NR)	NA	Unadjusted OR 1.66, 95% CI 0.76 to 3.64	NR	NR	154	61
Paranjpe 2020 (Retrospective cohort)	USA	NR	Hospital^a	Treatment dose anticoagulation	NA	In‐hospital mortality: intervention 22.5% versus comparator 22.8% In subgroup who required mechanical ventilation: intervention 29.1% versus comparator 62.7% (adjusted HR 0.86, 95% CI 0.82 to 0.89; 395 participants, P < 0.001)	NR	NR	2773	786
Russo 2020 (Retrospective cohort)	Italy	67	Hospital^a	DOACS (NR) in 18 participants and VKA (NR) in 8 participants	NA	RR 1.15 (95% CI 0.29 to 2.57), P = 0.995	NR	NR	192	26
Shi 2020 (Retrospective cohort)	China	69	Hospital^a	LMWH	NA	Reported no deaths in both groups	NR	NR	42	21
Tang 2020 (Retrospective cohort)	China	65	Hospital^a	UFH (10,000 to 15,000 IU/d in 5 participants and LMWH (40 mg/d to 60 mg/d) in 94 participants	NA	No difference (general mortality): (adjusted OR 1.64, 95% CI 0.92 to 2.92; 449 participants) Subgroup analysis: participants with SIC score of ≥ 4(unadjusted OR 0.37, 95% CI 0.15 to 0.90; 97 participants) Participants with D‐dimer > 6 times the ULN (unadjusted OR 0.44, 95% CI 0.22 to 0.86; 161 participants)	NR	28	449	99
Trinh 2020 (Retrospective cohort)	USA	59	ICU	UFH 15 IU/kg/h; or enoxaparin 1 mg/kg twice or once daily; or apixaban 10 mg (if no prior anticoagulation) or 5 mg (if prior anticoagulation) twice daily^b	UFH 5000 IU two to three times daily; or enoxaparin 40 mg twice or once daily; or apixaban 2.5 mg or 5 mg twice daily^b	Reduction in all‐cause mortality (adjusted HR 0.21, 95% CI 0.10 to 0.46) and a lower absolute rate of death in the therapeutic group (34.2% versus 53%)	NR	35	244	161
Total	China: 3 Italy: 1 Spain: 1 USA: 2	‐	‐	‐	‐	6 studies considered mortality; 1 study did not report mortality data	No study considered additional respiratory support	8 to 35 (3 studies)	5929	2888
CI: confidence interval; DOACS: direct oral anticoagulants; GFR: glomerular filtration rate;HR: hazard ratio; ICU: intensive care units;LMWH: low molecular weight heparin; NA: no anticoagulation; NR: not reported; NRS: non‐randomised study;OR: odds ratio; RR: risk ratio; SIC: sepsis‐induced coagulopathy; UFH: unfractionated heparin; VKA: vitamin K antagonist

^aHospital: includes intensive care unit, hospital wards or emergency department.
^bAnticoagulation used twice daily if glomerular filtration rate (GFR) was greater than 30 mL/min, or once daily if GFR was 30 mL/min or less.

We included seven studies describing 5929 participants in this review, of whom at least 2888 received anticoagulants (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020; Trinh 2020). The seven included studies were all non‐randomised studies (NRS) of interventions, with a comparator group. Of the seven included studies, four originated from China (Liu 2020; Shi 2020; Tang 2020; Trinh 2020), one from Italy (Russo 2020), one from Spain (Ayerbe 2020), and one from the USA (Paranjpe 2020).

Trinh 2020 compared different doses of anticoagulant (prophylactic versus therapeutic) and the six other included studies compared anticoagulation versus no anticoagulation (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). Only three included studies reported the follow‐up period that varied from 8 to 35 days (Ayerbe 2020; Tang 2020; Trinh 2020). Liu 2020 compared participants from the ICU (intervention group) with participants in hospital wards (comparator group). Trinh 2020 included only participants from the ICU in both groups. The five other studies considered participants from all settings (ICU, hospital wards and emergency departments; Ayerbe 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). Paranjpe 2020 did not report data regarding age of participants. The mean age of the other six studies' participants varied from 59 to 72 years (Ayerbe 2020; Liu 2020; Russo 2020; Shi 2020; Tang 2020; Trinh 2020). Six studies reported data on mortality (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Tang 2020; Trinh 2020), and none reported data for necessity for additional respiratory support.

Paranjpe 2020 did not describe the type or dose of anticoagulation. Ayerbe 2020 and Liu 2020 used heparin in the intervention group, but they did not report details about the type of heparin or dose. Shi 2020 used low molecular weight heparin and Russo 2020 used direct oral anticoagulants in 18 participants and vitamin K antagonist in eight other participants, but neither reported more details. Tang 2020 used unfractionated heparin 10,000 IU/day to 15,000 IU/day in five participants and low molecular weight heparin (enoxaparin) 40 mg/day to 60 mg/day in 94 participants. Trinh 2020 used unfractionated heparin 15 IU/kg/hour or enoxaparin 1 mg/kg twice daily if glomerular function rate (GFR) was greater than 30 mL a minute, or once daily if GFR was 30 mL a minute or less. In addition to these anticoagulants, the comparator group in Trinh 2020 also used apixaban 2.5 mg or 5 mg twice daily.

Please refer to the Characteristics of included studies for detailed information.

Excluded studies

We excluded 12 studies for at least one reason (Characteristics of excluded studies). Eleven of the studies had an irrelevant study design because of at least one of the following reasons (Al‐Samkari 2020; Artifoni 2020; EudraCT2020‐001823‐15; Helms 2020; Khider 2020; NCT04354155; NCT04359212; NCT04368377; NCT04394000; NCT04427098; Zhang 2020):

retrospective cases series without a consistent comparator group;
prospective cohort study without a comparator group (single‐arm study);
prospective cohort study without an intervention purpose;
prospective before‐after cohort study without a parallel comparator group;
prospective cohort study without a parallel comparator group of intervention.

One study had an irrelevant intervention, that is, it is a RCT of aspirin for COVID‐19, and there was no difference between the intervention groups regarding anticoagulants (NCT04365309).

Ongoing studies

Twenty‐two ongoing studies met our inclusion criteria, which plan to evaluate 15,727 participants. We tried to contact study authors; we also searched by study registration number and by title of the study on all databases of interest for this review. However, there are no additional data for all these ongoing studies. See the Characteristics of ongoing studies table for further details.

Four of the ongoing studies plan to include 1000 participants or more (NCT04333407; NCT04359277; NCT04366960; NCT04372589). NCT04333407 plans to compare aspirin, clopidogrel, rivaroxaban, atorvastatin, and omeprazole with no treatment in 3170 participants to assess mortality at 30 days. NCT04359277 plans to compare higher‐dose versus low‐dose prophylactic heparin to assess composite outcomes that include mortality in 1000 participants. NCT04366960 plans to compare 40 mg subcutaneous enoxaparin twice daily versus 40 mg subcutaneous enoxaparin once daily to assess venous thromboembolism in 2712 participants. NCT04372589 plans to compare therapeutic anticoagulation using heparin for 14 days with prophylactic anticoagulation to assess intubation and mortality in 3000 participants. See Table 3 for a summary of the characteristics of ongoing studies.

Table 3. Summary of characteristics of ongoing studies

Study	Country	Design	Primary outcomes	Estimated number of participants	Estimated primary completion date
ACTRN12620000517976	Australia	RCT	Time to separation from invasive ventilation	172	25 July 2021
ChiCTR2000030700	China	RCT	Time to virus eradication	60	30 September 2020
ChiCTR2000030701	China	RCT	Time to virus eradication	60	30 September 2020
ChiCTR2000030946	China	Prospective cohort	Biochemical indicators	120	24 April 2020
Marietta 2020	Italy	RCT	Clinical worsening (includes death and necessity for additional respiratory support)	300	June 2021
NCT04333407	UK	RCT	All‐cause mortality at 30 days after admission	3170	30 March 2021
NCT04344756	France	RCT	Survival without ventilation	808	31 July 2020
NCT04345848	Switzerland	RCT	Composite outcome of arterial or venous thrombosis, disseminated intravascular coagulation and all‐cause mortality	200	30 November 2020
NCT04352400	Italy	RCT	Time to clinical improvement	256	December 2021
NCT04359277	USA	RCT	Composite incidence of: all‐cause mortality, cardiac arrest, symptomatic deep venous thrombosis, PE, arterial thromboembolism, myocardial infarction, stroke, or shock	1000	21 April 2021
NCT04360824	USA	RCT	Risk of all‐cause mortality	170	16 April 2021
NCT04362085	Canada	RCT	Composite outcome of ICU admission (yes/no), non‐invasive positive pressure ventilation (yes/no), invasive mechanical ventilation (yes/no), or all‐cause death (yes/no) up to 28 days	462	November 2020
NCT04366960	Italy	RCT	Incidence of VTE detected by imaging	2712	August 2020
NCT04367831	USA	RCT	Total number of patients with clinically relevant venous or arterial thrombotic events in ICU	100	November 2020
NCT04372589	Canada	RCT	Intubation and mortality	3000	January 2021
NCT04373707	France	RCT	VTE	602	September 2020
NCT04377997	USA	RCT	Risk of composite efficacy endpoint of death, cardiac arrest, symptomatic deep venous thrombosis, PE, arterial thromboembolism, myocardial infarction, or haemodynamic shock Risk of major bleeding event according to the ISTH definition	300	1 January 2021
NCT04393805	Italy	Retrospective cohort	Bleeding Thrombosis Mortality	877	December 2020
NCT04394377	Brazil	RCT	Hierarchical composite endpoint composed of mortality, number of days alive, number of days in the hospital and number of days with oxygen therapy at the end of 30 days	600	December 2020
NCT04397510	USA	RCT	Mean daily PaO₂:FiO₂	50	31 December 2020
NCT04401293	USA	RCT	Composite outcome of arterial thromboembolic events, venous thromboembolic events and all‐cause mortality at day 30 ± 2 days	308	22 October 2020
NCT04416048	Germany	RCT	Composite endpoint of VTE (DVT and/or fatal or non‐fatal PE), arterial thromboembolism, new myocardial infarction, non‐haemorrhagic stroke, all‐cause mortality or progression to intubation and invasive ventilation	400	30 April 2021
Total number of studies	Australia: 1 Brazil: 1 Canada: 2 China: 3 France: 2 Germany: 1 Italy: 4 Switzerland: 1 UK: 1 USA: 6	Prospective cohort: 1 RCT: 20 Retrospective cohort: 1	12 studies considered mortality Six studies considered additional respiratory support	15,727 participants 997 from NRS 14,730 from RCTs	13 studies to December 2020 8 studies to July 2021 1 study to December 2021
DVT: deep vein thrombosis; FiO₂: fraction of inspired oxygen; ISTH: International Society on Thrombosis and Haemostasis; NRS: non‐randomised studies; PaO₂: arterial oxygen pressure;PE: pulmonary embolism; RCT: randomised controlled trial; VTE: venous thromboembolism

Risk of bias in included studies

We assessed the risk of bias at the result level, in each comparison, using ROBINS‐I tool (Sterne 2016). The specific judgements ('critical risk', 'serious risk', 'moderate risk', 'low risk', or 'no information') by available outcomes, in each comparison, are presented in Figure 2, Figure 3, Figure 4, Figure 5, Figure 6 and Figure 7. The support for judgement is explained in the related 'Risk of bias' tables (Table 4, Table 5, Table 6, Table 7, Table 8 and Table 9). The overall risk of bias for all‐cause mortality and for hospitalisation in the comparison 'anticoagulants (all types) versus no treatment' was critical and in the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' was serious. The overall risk of bias for major bleeding was serious for both comparisons.

Figure 2

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (all‐cause mortality)

Figure 3

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (major bleeding)

Figure 4

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (hospitalisation)

Figure 5

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (all‐cause mortality)

Figure 6

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (majorbleeding)

Figure 7

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (hospitalisation)

Table 4. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (all‐cause mortality)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Ayerbe 2020	Critical risk	No information	Serious risk	Low risk	Critical risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is no baseline characteristics table comparing the two groups. Essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism, were not considered.	Participants included in both groups were selected from 17 hospitals, and the study was retrospective, therefore it is not possible to know whether the selection was free from bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. The type and doses of heparin in the intervention group were not described.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	There were missing outcome data for 56 participants with no specific information or appropriate analyses. These missing data could cause a critical impact on the estimates.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.
Liu 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Serious risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is no baseline characteristics table comparing the two groups. Essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism, were not considered.	Participants included in both groups were selected from a single hospital, and the study was retrospective, therefore it is not possible to know whether the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We cold not adjust the analyses for this selection bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. The type and doses of heparin in the intervention group were not described.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data was reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias in selection of reported effect estimate, based on the results, from different subgroups analyses.	The study is too problematic to provide useful evidence.
Paranjpe 2020	Serious risk	Moderate risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, an adjusted analysis with propensity scores was performed considering confounding demographic, clinical, and medication use. However, the confounding factors 'participants who underwent surgery during the hospitalisation', 'active cancer treatment', 'concomitant antiplatelet use' and 'history of venous thromboembolism' were not considered.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome, but the study authors used appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems
Russo 2020	Serious risk	Moderate risk	Serious risk	No information	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores, considering confounding demographic and clinical factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'active cancer treatment' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome, but the study authors used appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	Insufficient information to judge. No information is reported on whether there was deviation from the intended intervention.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems
Shi 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is a baseline characteristics table comparing the two groups with limited items. However, the study did not compare essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism.	The participants of the two groups (intervention and comparator) were selected from the same hospital, but as the study was retrospective, it is not possible to know if the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	There is a risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.
Tang 2020	Critical risk	Critical risk	Serious risk	No information	Low risk	Low risk	Critical risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced among the compared groups. There was no table comparing the characteristics of the two groups at baseline. The comparator group included participants who used heparin for less time or did not use heparin. These participants may be less severely ill than those in the intervention group.	Participants included in both groups were selected from the same hospital, but as the study was retrospective, it is not possible to know whether the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. Besides, the comparator group also included participants who used heparin for less than seven days. This proximity to the case definition for the intervention group increases the risk of error in the classification of participants. Also, the comparator group considered two very different types of intervention.	Insufficient information to judge. No information is reported on whether there was deviation from the intended intervention.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both, and it is not possible to exclude bias in selection of reported effect estimate, based on the results, from multiple measurements within the outcome domain, multiple analyses of the intervention‐outcome relationship, and different subgroups analyses.	The study is too problematic to provide useful evidence.

Table 5. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (major bleeding)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Paranjpe 2020	Serious risk	Serious risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an adjusted analysis with propensity scores considering confounding demographic and clinical factors, and medication use. However, the study did not consider confounding factors 'participants who underwent surgery during the hospitalisation', 'active cancer treatment', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome. For this outcome, the authors did not use appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems

Table 6. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (hospitalisation)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Shi 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is a baseline characteristics table comparing the two groups with limited items. However, the study did not compare essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism.	The participants of the two groups (intervention and comparator) were selected from the same hospital, but as the study was retrospective, it is not possible to know if the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	There is a risk that the interventions received by participants in the same group have not been standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.

Table 7. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (all‐cause mortality)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Table 8. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (major bleeding)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Table 9. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (hospitalisation)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Bias due to confounding

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment'. We rated four of them as critical risk because one or more prognostic variables are likely to be unbalanced between the compared groups (Ayerbe 2020; Liu 2020; Shi 2020; Tang 2020). There is not a baseline characteristics table comparing the two groups in Ayerbe 2020, Liu 2020 and Tang 2020. There is a baseline characteristics table, with limited items, comparing the two groups in Shi 2020. However, they did not consider essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism. In Tang 2020, the comparator group included participants who used heparin for less time or did not use heparin. These participants may be less severely ill than those in the intervention group.

We rated the other two studies as serious risk because, to minimise the impact of the absence of randomisation, the studies authors performed an adjusted analysis with propensity scores, considering confounding demographic and clinical factors, and medication use. However, neither Paranjpe 2020 nor Russo 2020 considered the confounding factors 'participants who underwent surgery during hospitalisation', 'active cancer treatment', and 'history of venous thromboembolism'. Paranjpe 2020 also did not consider 'concomitant antiplatelet use' as a confounder. See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as serious risk because, to minimise the impact of the absence of randomisation, we performed an analysis with propensity scores, considering confounding demographic, clinical, and laboratory factors, and medication use. However, Trinh 2020 did not consider the confounding factors 'participants who underwent surgery during hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as serious risk because, to minimise the impact of the absence of randomisation, we performed an adjusted analysis with propensity scores, considering confounding demographic and clinical factors, and medication use. However, Paranjpe 2020 did not consider the confounding factors 'participants who underwent surgery during hospitalisation', 'active cancer treatment', 'concomitant antiplatelet use' and 'history of venous thromboembolism'. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as serious risk because, to minimise the impact of the absence of randomisation, we performed an analysis with propensity scores, considering confounding demographic, clinical, laboratory factors and medication use. However, Trinh 2020 did not consider the confounding factors 'participants who underwent surgery during hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as critical risk because one or more prognostic variables are likely to be unbalanced between the compared groups. There is a baseline characteristics table, with limited items, comparing the two groups. However, Shi 2020 did not compare essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as serious risk because, to minimise the impact of the absence of randomisation, we performed an analysis with propensity scores, considering confounding demographic, clinical, laboratory factors and medication use. However, Trinh 2020 did not consider the confounding factors 'participants who underwent surgery during hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'. See Figure 7 and Table 9.

Bias in selection of participants into the study

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment'. We rated three of them as critical risk (Liu 2020; Shi 2020; Tang 2020). These studies selected participants included in both groups (intervention and comparator) from a single hospital, and the studies were retrospective, so it is not possible to know whether the selection was free from bias. The selection for the studies was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias (Liu 2020; Shi 2020; Tang 2020).

We rated Paranjpe 2020 and Russo 2020 as moderate risk because they selected the included participants in both groups from the same hospital, and selection may have been related to intervention and outcome, but the study authors used appropriate methods to adjust for selection bias.

We rated Ayerbe 2020 as 'no information' because they selected participants included in both groups from 17 hospitals, and the study was retrospective, therefore it is not possible to know whether the selection was free from bias. See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as moderate risk because they selected the included participants in both groups from the same hospital. Trinh 2020 considered for inclusion all patients who met the inclusion criteria, and who were treated in each period. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as serious risk because they selected the included participants in both groups from the same hospital, and selection may have been related to intervention and outcome. For this outcome, the study authors did not use appropriate methods to adjust for selection bias. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as moderate risk because they selected the included participants in both groups from the same hospital. Trinh 2020 considered for inclusion all patients who met the inclusion criteria, and who were treated in each period. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as critical risk because they selected the participants of the two groups (intervention and comparator) from the same hospital, but as the study was retrospective, it is not possible to know if the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as moderate risk because they selected the included participants in both groups from the same hospital. Trinh 2020 considered for inclusion all patients who met the inclusion criteria, and who were treated in each period. See Figure 7 and Table 9.

Bias in classification of interventions

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment' (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). We rated them as serious risk because there is a high risk that these studies did not standardise interventions received by participants in the same group. Ayerbe 2020 and Liu 2020 did not describe the type and doses of heparin in the intervention group. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively (Paranjpe 2020; Russo 2020; Shi 2020). Besides, in Tang 2020, the comparator group also included participants who used heparin for under seven days. This proximity to the case definition for the intervention group increases the risk of error in the classification of participants. Also, the comparator group in Tang 2020 considered two very different types of intervention. See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because intervention status was well defined based on information collected at the time of intervention. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as serious risk because there is a high risk that they did not standardise the interventions received by participants in the same group. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as low risk because intervention status was well defined based on information collected at the time of intervention. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as serious risk because there is a risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because intervention status was well defined based on information collected at the time of intervention. See Figure 7 and Table 9.

Bias due to deviations from the intended intervention

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment'. We rated four of them as low risk because they did not report any deviations from the intended intervention, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome (Ayerbe 2020; Liu 2020; Paranjpe 2020; Shi 2020). We rated Russo 2020 and Tang 2020 as 'no information' because there is insufficient information to judge. They did not report any information on whether there was deviation from the intended intervention. See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because they did not report any deviations from the intended intervention in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because they did not report any deviations from the intended intervention in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as low risk because they did not report any deviations from the intended intervention in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because they did not report any deviations from the intended intervention in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because they did not report any deviations from the intended intervention in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome. See Figure 7 and Table 9.

Bias due to missing data

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment' (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). We rated Ayerbe 2020 as critical risk because there were missing outcome data for 56 participants with no specific information or appropriate analyses. These missing data could cause a critical impact on the estimates. We rated the other five studies as low because there were no missing data for this outcome (Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). See Figure 2 and Table 4.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because there were no missing data for this outcome. See Figure 3 and Table 5.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because there were no missing data for this outcome. See Figure 4 and Table 6.

Bias in measurement of outcomes

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment'. We rated them as low risk because it is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because it is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because it is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as low risk because it is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because it is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as low risk because it is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants. See Figure 7 and Table 9.

Bias in selection of the reported result

All‐cause mortality

Six studies reported mortality for the comparison 'anticoagulants (all types) versus no treatment' (Ayerbe 2020; Liu 2020; Paranjpe 2020; Russo 2020; Shi 2020; Tang 2020). We rated Tang 2020 as critical risk because we did not identify the study protocol or it was not available, and it is not possible to exclude bias in selection of reported effect estimates, based on the results, from multiple measurements within the outcome domain, multiple analyses of the intervention‐outcome relationship, and analyses of different subgroups. We rated Liu 2020 as serious risk because we did not identify the study protocol or it was not available (only a preprint was available), and it is not possible to exclude bias in selection of reported effect estimates, based on the results, from analyses of different subgroups. We rated the other four studies as low because we did not identify the study protocol but all reported results corresponded to the intended outcome (Ayerbe 2020; Paranjpe 2020; Russo 2020; Shi 2020). See Figure 2 and Table 4.

Trinh 2020 reported mortality for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as serious risk because we did not identify the study protocol or it was not available (only a preprint was available), and it is not possible to exclude bias. See Figure 5 and Table 7.

Major bleeding

Paranjpe 2020 reported major bleeding for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because we did not identify the study protocol but all reported results corresponded to the intended outcome. See Figure 3 and Table 5.

Trinh 2020 reported major bleeding for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)'. We rated this study as serious risk because we did not identify the study protocol or it was not available (only a preprint was available), and it is not possible to exclude bias. See Figure 6 and Table 8.

Hospitalisation

Shi 2020 reported hospitalisation for the comparison 'anticoagulants (all types) versus no treatment'. We rated this study as low risk because we did not identify the study protocol but all reported results corresponded to the intended outcome. See Figure 4 and Table 6.

Trinh 2020 reported hospitalisation for the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' and we rated this study as serious risk because we did not identify the study protocol or it was not available (only a preprint was available), and it is not possible to exclude bias. See Figure 7 and Table 9.

Effects of interventions

Since we included seven NRS of interventions and no RCTs, or quasi‐RCTs, we did not perform any quantitative data analysis (meta‐analysis). Otherwise, we restricted our analysis on the qualitative aspects of the results reported by the study authors.

1. Anticoagulants (all types) versus no treatment

Four studies compared heparins (Ayerbe 2020; Liu 2020; Shi 2020; Tang 2020), and Russo 2020 compared oral anticoagulants (direct oral anticoagulants or vitamin K antagonists) to no treatment. Paranjpe 2020 compared 'therapeutic anticoagulation' (including oral, subcutaneous, or intravenous forms) to no treatment, but did not describe the type or dose of the pharmacological intervention. See summary of findings Table 1.

Primary outcomes

All‐cause mortality

Ayerbe 2020 reported all‐cause mortality as the proportion of participants and as odds ratio (OR) after adjusting for some covariates (e.g. age, gender, saturation of oxygen < 90% and temperature > 37 °C). They found 242 (13.9%) deaths in the intervention group and 44 (15.4%) deaths in the comparator group (adjusted OR 0.42, 95% confidence interval (CI) 0.26 to 0.66; P < 0.001; 2075 participants), in favour of the intervention group after all adjustments.

Liu 2020 evaluated all‐cause mortality in the context of the use or not of substitutive dialysis therapy, but not comparing the use or not of heparin. In this setting, mortality in the intervention group was 22.5% and in the comparator group was 22.8%. Extracting data from the reported tables, irrespective of setting (ICU or ward), we found 35 deaths in 106 participants with anticoagulants and 11 deaths in 48 participants without anticoagulants (unadjusted OR 1.66, 95% CI 0.76 to 3.64; 154 participants).

Paranjpe 2020 reported 22.5% in‐hospital mortality for the intervention group and 22.8% for the comparator group. However, in participants who required mechanical ventilation (n = 395), in‐hospital mortality was 29.1% in the intervention group and 62.7% in the comparator group. In this subgroup, after a multivariate adjustment, the hazard ratio (HR) was 0.86 (95% CI 0.82 to 0.89; P < 0.001; 395 participants).

Russo 2020 reported all‐cause mortality, after regression adjustment, as: risk ratio (RR) 1.15 (95% CI 0.29 to 2.57; P = 0.995; 192 participants).

Shi 2020 did not foresee this outcome but reported that no deaths occurred during the follow‐up period.

Tang 2020 reported no difference in all‐cause mortality between the intervention group (30.3%) and the comparator group (29.7%; P = 0.910) in general (adjusted OR 1.64, 95% CI 0.92 to 2.92; 449 participants). Among participants with a sepsis‐induced coagulopathy (SIC) score of 4 or more (n = 97), mortality was 40% in the intervention group and 64.2% in the comparator group (P = 0.029). The unadjusted OR was 0.37 (95% CI 0.15 to 0.90; 97 participants). Besides, mortality among participants with high levels of D‐dimer (e.g. greater than 6 times the upper limit) was 32.8% in the intervention group and 52.4% in the comparator group (P = 0.017). The unadjusted OR was 0.44 (95% CI 0.22 to 0.86; 161 participants).

It is very uncertain whether anticoagulants (all types) reduce all‐cause mortality compared with no treatment because the certainty of evidence is very low.

Necessity for additional respiratory support

There were no available data for this outcome.

Secondary outcomes

Mortality related to COVID‐19

There were no available data for this outcome.

Deep vein thrombosis (DVT)

There were no available data for this outcome.

Pulmonary embolism

There were no available data for this outcome.

Major bleeding

Liu 2020 did not define their bleeding criteria and reported bleeding in lung tissues of one participant. They did not clarify if this diagnosis was made on necropsies or clinically. Therefore, we did not consider this information as an available datum.

Paranjpe 2020 defined 'major bleeding' as:

haemoglobin less than 7 g/dL and any red blood cell transfusion;
at least 2 units of red blood cell transfusion within 48 hours; or
a diagnosis code for major bleeding including intracranial haemorrhage, haematemesis, melena, peptic ulcer with haemorrhage, colon, rectal, or anal haemorrhage, haematuria, ocular haemorrhage, and acute haemorrhagic gastritis.

They reported 24 (3%) events in the intervention group and 38 (1.9%) events in the comparator group (P = 0.2). Of the 24 participants who had bleeding events in the intervention group, 15 (63%) had bleeding events after starting anticoagulation and 9 (37%) had bleeding events before starting anticoagulation. Bleeding events were more common among intubated participants (30 of 395; 7.5%) than among non‐intubated participants (32 of 2378; 1.35%).

Ayerbe 2020, Russo 2020, Shi 2020 and Tang 2020 did not report data for this outcome.

Anticoagulants (all types) may make no difference in major bleeding compared with no treatment, but the certainty of evidence is low.

Adverse events (minor bleeding, gastrointestinal adverse effects (e.g. nausea, vomiting, diarrhoea, abdominal pain), allergic reactions, renal failure and amputations)

Tang 2020 reported that "the prophylactic dose of low molecular weight heparin was used in most of our heparin users, bleeding complications were unusual and commonly mild, and it is not known if higher doses would have been better." However, the trial authors did not report any related number of events or comparison between the groups. Therefore, we did not consider this information as an available datum.

Ayerbe 2020, Liu 2020, Paranjpe 2020, Russo 2020 and Shi 2020 did not report data for this outcome.

Hospitalisation time in days

Paranjpe 2020 reported 5 days (interquartile range (IQR) 3 to 8) as their median hospitalisation time, but they did not compare this outcome among the intervention and comparator groups. Therefore, we did not consider this information as an available datum.

Shi 2020 reported a median of 29 days (IQR 17 to 42) as hospitalisation time in the intervention group and 27 days (IQR 24 to 31) in the comparator group (P = 0.41). We estimated the mean difference (MD) 2 days (95% CI −0.80 to 4.80) using the method reported by Wan 2014 to convert median and IQR into MD and CI.

Tang 2020 used hospital stay of less than seven days as an exclusion criterion, but they did not report data for analysis.

Ayerbe 2020, Liu 2020 and Russo 2020 did not report data for this outcome.

It is very uncertain whether anticoagulants (all types) have any effect on hospitalisation time compared with no treatment because the certainty of evidence is very low (42 participants, 1 retrospective NRS). We downgraded the certainty of evidence by one level due to study limitations because the overall risk of bias was critical, especially related to confounding. We downgraded the certainty of evidence by two levels due to imprecision because the narrative synthesis was conducted with imprecise estimates based on few participants.

Quality of life

There were no available data for this outcome.

2. Anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)

Trinh 2020 compared heparins (unfractionated heparin or low molecular weight heparin) or direct oral anticoagulants (apixaban) in therapeutic doses (161 participants) versus heparins (unfractionated heparin or low molecular weight heparin) or direct oral anticoagulants (apixaban) in prophylactic doses (83 participants). See summary of findings Table 2.

Primary outcomes

All‐cause mortality

Trinh 2020 reported in‐hospital mortality with a follow‐up of 35 days in 43.5% of the intervention group and 74.8% of the control group (P = 0.0003). Participants in the intervention group experienced reduced mortality: HR 0.43, (95% CI 0.23 to 0.78). In the subgroup of intubated participants with mechanical ventilation, mortality was 34.2% in the intervention group and 53% in the comparative group: adjusted HR 0.21, 95% CI 0.10 to 0.46.

Anticoagulants (therapeutic dose) may reduce all‐cause mortality compared with anticoagulants (prophylactic dose), but the certainty of evidence is low.

Necessity for additional respiratory support

There were no available data for this outcome.

Secondary outcomes

Mortality related to COVID‐19

There were no available data for this outcome.

Deep vein thrombosis (DVT)

There were no available data for this outcome.

Pulmonary embolism

There were no available data for this outcome.

Major bleeding

Trinh 2020 did not define major and minor bleeding, therefore, we considered their reported bleeding as major bleeding. They reported 51 (31.7%) events of bleeding in the intervention group and 17 (20.5%) in the comparator group (OR 1.80, 95% CI 0.96 to 3.37, P = 0.07).

Anticoagulants (therapeutic dose) may lead to no difference in major bleeding compared with anticoagulants (prophylactic dose), but the certainty of evidence is low.

Adverse events (minor bleeding, gastrointestinal adverse effects (e.g. nausea, vomiting, diarrhoea, abdominal pain), allergic reactions, renal failure and amputations)

Trinh 2020 reported stroke events (intervention: 6 (3.7%) and comparator: 5 (6%), P = 0.41); renal failure requiring dialysis (intervention: 67 (42.7%) and comparator: 25 (30.9%), P = 0.08); and liver failure (intervention: 3 (1.9%) and comparator: 2 (2.4%), P = 1.00). However, they did not report any of the adverse events of interest for this review.

Hospitalisation time in days

Trinh 2020 reported hospitalisation time as mean ± SD for the intervention (23.3 ± 7.7 days) and comparator (15.7 ± 8.9 days) groups (MD 7.6 days, 95% CI 5.35 to 9.85; P < 0.001).

There was low‐certainty evidence (244 participants, one retrospective NRS) that anticoagulants (therapeutic dose) may increase hospitalisation time compared with anticoagulants (prophylactic dose). We downgraded the certainty of evidence by one level due to study limitations because the overall risk of bias was serious, especially related to selection bias. We downgraded the certainty of evidence by one level due to imprecision. Narrative synthesis was conducted with imprecise estimates based on fewer than 400 participants.

Quality of life

There were no available data for this outcome.

Discussion

This review aimed to assess the effects of prophylactic anticoagulants versus active comparator, placebo or no intervention on mortality and need for additional respiratory support for people hospitalised with COVID‐19.

Summary of main results

We found no RCTs, no quasi‐RCTs, and no prospective NRS with available data assessing the effects of prophylactic anticoagulants compared to active comparator, placebo or no intervention on mortality and need for additional respiratory support for people hospitalised with COVID‐19.

We found 22 ongoing studies (from Australia (1), Brazil (1), Canada (2), China (3), France (2), Germany (1), Italy (4), Switzerland (1), UK (1), and USA (6)) that plan to evaluate 15,727 participants in this setting, of whom 14,730 are from 20 RCTs, and 997 are from one prospective NRS (120 estimated participants) and one retrospective NRS (877 estimated participants). See Table 3.

Twelve ongoing studies plan to report data for mortality. Six ongoing studies plan to report data for necessity for additional respiratory support. Thirteen ongoing studies are expected to be completed in December 2020 (6959 estimated participants), eight in July 2021 (8512 estimated participants), and one in December 2021 (256 estimated participants). Four of these ongoing studies plan to include 1000 participants or more.

One of the studies plans to compare aspirin, clopidogrel, rivaroxaban, atorvastatin, and omeprazole with no treatment in 3170 participants to assess mortality at 30 days of follow‐up. One study plans to compare a higher dose versus lower dose of prophylactic heparin to assess composite outcomes that include mortality in 1000 participants. One study plans to compare different doses of enoxaparin to assess venous thromboembolism in 2712 participants. Another study plans to compare therapeutic anticoagulation using heparin for 14 days with prophylactic anticoagulation to assess intubation and mortality in 3000 participants.

We found six retrospective NRS (5685 participants) with limited evidence of anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (Table 2). The overall risk of bias for all‐cause mortality and for hospitalisation was critical and for major bleeding was serious in this comparison. Two studies reported reduction of mortality by odds ratio (reduction of 58% on chance of death) or hazard ratio (HR 0.86, 95% CI 0.82 to 0.89; 395 participants), both adjusted for confounding. Another study reported reduction of mortality only in a subgroup of severely ill participants, two studies reported no differences by unadjusted OR 1.66 (95% confidence interval (CI) 0.76 to 3.64) or adjusted risk ratio (RR) 1.15 (95% CI 0.29 to 2.57), and another study reported zero events in both intervention groups. One study reported 3% of bleeding events in the intervention group and 1.9% in the control group (OR 1.62, 95% CI 0.96 to 2.71). One study reported a median of 29 days of hospitalisation (IQR 17 to 42) in the intervention group and 27 days (IQR 24 to 31) in the control group (MD 2 days, 95% CI −0.80 to 4.80). See summary of findings Table 1.

We found one retrospective NRS (244 participants) with limited evidence about anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19. One study reported an absolute rate of death lower in the intervention group (34.2% versus 53%) and an adjusted HR 0.21 (95% CI 0.10 to 0.46) for confounding. One study reported 31.7% of bleeding events in the intervention group and 20.5% in the control group (OR 1.8, 95% CI 0.96 to 3.37). One study reported a mean increase of 7.6 days of hospitalisation (95% CI 5.35 to 9.85) in length of hospital stay. See summary of findings Table 2.

Overall completeness and applicability of the evidence

While most of the studies reported our primary outcome of all‐cause mortality, we identified very little evidence relating to adverse effects of anticoagulants. It is also noteworthy that none of the studies measured our other primary outcome (necessity for additional respiratory support) or our secondary outcomes such as mortality related to COVID‐19, DVT, pulmonary embolism and quality of life.

There was substantial heterogeneity in the methods of the included studies and many of them did not provide complete and clear information about their data. This hindered the qualitative analyses and the assessment of the risk of bias of many outcomes in many studies.

The number of studies for each of the possible comparisons was small, ranging from one to six studies. Moreover, the included studies had small primary sample sizes, except for only two included studies that evaluated more than 2000 participants. The largest study involved 2773 participants treated with anticoagulation, but did not provide details about the type or dose of the pharmacological interventions. Another issue is the poor reporting quality of most of these studies, which directly affects data extraction and judgement of risk of bias.

There was considerable variation in the use of the same intervention (e.g. dosages, type, method of application). The variation of assessment for the confounding factor in NRS also impaired the results.

It is noteworthy that the studies included in this review were conducted in four different countries, most of which (75%) were high‐income countries. Social and cultural aspects of the evaluated interventions can also interfere with their acceptability and effectiveness for the treatment of people hospitalised with COVID‐19. Therefore, the external validity of the overall evidence presented in this review should be considered with caution.

We acknowledge that designing and conducting an appropriate study with available data for this topic is difficult. The new approach regarding prophylactic anticoagulants for people hospitalised with COVID‐19 has been used to provide high levels of anticoagulants for these people, although there is no available evidence based on RCTs or quasi‐RCTs to support their use. This reinforces the importance of this review and serves as an incentive for further investigation.

Certainty of the evidence

We found no RCTs, quasi‐RCTs or prospective NRS with available data that were eligible for this review, and we included only seven retrospective NRS.

Despite the increasing number of studies on prophylactic anticoagulants for people hospitalised with COVID‐19 in the past months, the overall risk of bias for all‐cause mortality and for hospitalisation in the comparison 'anticoagulants (all types) versus no treatment' was critical and in the comparison 'anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose)' was serious. The overall risk of bias for major bleeding was serious for the both comparisons with available data. We judged the bias domains due to confounding, selection of participants into the study, classification of interventions, deviations from the intended intervention, measurement of outcomes, and selection of the reported results from low to critical risk of bias. There was no information from three included studies for the all‐cause mortality assessment in the comparison 'anticoagulants (all types) versus no treatment'.

The certainty of evidence is low to very low. We downgraded the certainty of evidence due to risk of bias, particularly with regard to overall critical/serious risk of bias across studies, especially related to confounding or selection bias. We downgraded the certainty of evidence due to inconsistency and we decided not to pool data due to heterogeneity of studies (especially due to differences in the interventions). We also downgraded the certainty of evidence by one or two levels due to imprecision because the narrative synthesis was conducted with an imprecise estimate based on fewer than 400 participants (in some cases in very few participants).

It is very uncertain if anticoagulants (all types) compared with no treatment, reduce all‐cause mortality at 28 days after the intervention (5685 participants, 6 retrospective NRS), or have any effect on hospitalisation time (42 participants, 1 retrospective NRS, follow‐up not reported) because the certainty of evidence is very low for both outcomes. Anticoagulants (all types) may make no difference in major bleeding compared with no treatment, but the certainty of evidence is low (2773 participants, 1 retrospective NRS, follow‐up not reported). See summary of findings Table 1.

Anticoagulants (therapeutic dose), compared with anticoagulants (prophylactic dose), may reduce all‐cause mortality, may make no difference in major bleeding or may increase hospitalisation time, but the certainty of evidence is low (244 participants, 1 retrospective NRS, follow‐up 35 days) for all these outcomes. See summary of findings Table 2.

Potential biases in the review process

We performed a comprehensive search of the literature, and we performed study selection according to the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2020). We believe that we identified all the relevant studies that met our inclusion criteria. However, the possibility remains that we may have missed some studies, particularly in the grey literature. We adhered to the inclusion and exclusion criteria prespecified in the protocol in order to limit subjectivity (Flumignan 2020). We made efforts to obtain additional relevant data from study authors but were unable to do so. If we can source supplementary data, we will consider them in future updates. Two review authors selected studies in duplicate, independently, to reduce potential bias of the review process. One review author extracted data and assessed risk of bias of the included studies, and another review author checked the data extraction and 'Risk of bias' judgements, to accelerate the process and also to reduce potential bias of the review process.

Agreements and disagreements with other studies or reviews

A systematic review of 'potential rapid diagnostics, vaccine and therapeutics for COVID‐19' searched for published articles in PubMed, Embase and Cochrane Library, and found 27 studies for inclusion, but none of them regarding anticoagulants (Pang 2020).

A systematic review of 'therapeutic management of patients with COVID‐19' searched for studies published in English in Embase, MEDLINE and Google Scholar between 1 December 2019 and 31 March 2020, without any criteria regarding study design. Tobaiqy 2020 included 41 studies (three clinical studies, seven case reports, 10 case series, and 11 retrospective and 10 prospective observational studies) in their review. However, none of their included studies evaluated anticoagulants.

A systematic review of 'hypercoagulation and antithrombotic treatment in coronavirus 2019' searched for studies published in English in PubMed, ISI Web of Science, SCOPUS, and Cochrane Library on 28 March 2020, without any restrictions on publication date or publication status (Violi 2020). They excluded studies without a control group, animal studies, case reports, editorials, commentaries, letters, review articles, and guidelines from their analysis. No additional criteria for the included studies were described. Violi 2020 included nine NRS, which reported measures of clotting activation and their relationship with COVID‐19 clinical severity. However, no included study evaluated prophylactic anticoagulants for people hospitalised with COVID‐19.

Two narrative reviews regarding 'pharmacologic treatments for COVID‐19' and 'management of critically ill adults with COVID‐19' analysed several pharmacological interventions for the management of these people, but neither addressed prophylactic anticoagulants directly (Poston 2020; Sanders 2020).

In order to prevent microvascular thrombosis, some clinicians use higher‐dose anticoagulation rather than prophylactic dosing for inpatients with COVID‐19 (AVF 2020; Bikdeli 2020; Obe 2020). However, this practice is not supported by robust evidence. Although some practical guidelines address the management of prophylactic anticoagulation in people with COVID‐19, all of these recommendations are based on non‐COVID‐19 populations or low‐quality COVID‐19‐related evidence (AVF 2020; Bikdeli 2020; NHS 2020; Obe 2020; Ramacciotti 2020).

Figure 1

Study flow diagram
RCTs: randomised controlled trials; NRS: non‐randomised studies

Figure 2

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (all‐cause mortality)

Figure 3

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (major bleeding)

Figure 4

ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (hospitalisation)

Figure 5

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (all‐cause mortality)

Figure 6

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (majorbleeding)

Figure 7

ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (hospitalisation)

Summary of findings 1. Anticoagulants (all types) compared to no treatment for people hospitalised with COVID‐19

Outcomes	Impact	№ of participants (studies)	Certainty of the evidence (GRADE)
Anticoagulants (all types) compared to no treatment for people hospitalised with COVID‐19
Patient or population: people hospitalised with COVID‐19 Setting: hospital (ICU and ward) Intervention: anticoagulants (all types) Comparison: no treatment
All‐cause mortality Follow‐up: range 8 to 28 days	One study reported reduction of mortality by OR adjusted for confounding (reduction of 58% on chance of death; 2075 participants). One study reported reduction of mortality only in a subgroup of severely ill participants (HR 0.86, 95% CI 0.82 to 0.89; 395 participants). Three studies reported no differences by adjusted OR (1.64, 95% CI 0.92 to 2.92; 449 participants), unadjusted OR (1.66, 95% CI 0.76 to 3.64; 154 participants) or adjusted RR (1.15, 95% CI 0.29 to 2.57; 192 participants). One study reported zero events in both intervention groups.	5685 (6 retrospective NRS)	⊕⊝⊝⊝ Very low^a,b,c
Necessity for additional respiratory support	No study measured this outcome
Mortality related to COVID‐19	No study measured this outcome
Deep vein thrombosis	No study measured this outcome
Pulmonary embolism	No study measured this outcome
Major bleeding Follow‐up: not reported	One study reported 24 (3%) bleeding events in the intervention group and 38 (1.9%) bleeding events in the control group (OR 1.62, 95% CI 0.96 to 2.71).	2773 (1 retrospective NRS)	⊕⊕⊝⊝ Low^c,d
CI: confidence interval; HR: hazard ratio; ICU: intensive care unit; NRS: non‐randomised studies; OR: odds ratio; 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 a 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.
^aDowngraded one level due to study limitations. Overall critical/serious risk of bias across studies, especially related to confounding. ^bDowngraded one level due to inconsistency. We decided not to pool data due to the heterogeneity of studies (especially due to differences in interventions). ^cDowngraded one level due to imprecision. Narrative synthesis was conducted with imprecise estimates. ^dDowngraded one level due to study limitations. Overall serious risk of bias, especially related to confounding.

Summary of findings 1. Anticoagulants (all types) compared to no treatment for people hospitalised with COVID‐19

Summary of findings 2. Anticoagulants (therapeutic dose) compared to anticoagulants (prophylactic dose) for people hospitalised with COVID‐19

Outcomes	Impact	№ of participants (Studies)	Certainty of the evidence (GRADE)
Anticoagulants (therapeutic dose) compared to anticoagulants (prophylactic dose) for people hospitalised with COVID‐19
Patient or population: people hospitalised with COVID‐19 Setting: hospital (ICU and ward) Intervention: anticoagulants (therapeutic dose) Comparison: anticoagulants (prophylactic dose)
All‐cause mortality Follow‐up: 35 days	One study reported an absolute rate of death lower in the therapeutic group (34.2% versus 53%) and an HR adjusted for confounding of 0.21 (95% CI 0.10 to 0.46).	244 (1 retrospective NRS)	⊕⊕⊝⊝ Low^a,b
Necessity for additional respiratory support	No study measured this outcome
Mortality related to COVID‐19	No study measured this outcome
Deep vein thrombosis	No study measured this outcome
Pulmonary embolism	No study measured this outcome
Major bleeding Follow‐up: 35 days	One study reported 51 (31.7%) bleeding events in the intervention group and 17 (20.5%) bleeding events in the control group (OR 1.80, 95% CI 0.96 to 3.37).	244 (1 retrospective NRS)	⊕⊕⊝⊝ Low^a,b
CI: confidence interval; HR: hazard ratio; ICU: intensive care unit; NRS: non‐randomised studies; OR: odds 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 a 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.
^aDowngraded one level due to study limitations. Overall serious risk of bias, especially related to selection bias. ^bDowngraded one level due to imprecision. Narrative synthesis was conducted with imprecise estimates based on fewer than 400 participants.

Summary of findings 2. Anticoagulants (therapeutic dose) compared to anticoagulants (prophylactic dose) for people hospitalised with COVID‐19

Table 1. Glossary of terms

Term	Definition
Anticoagulants	Drugs that suppress, delay or prevent blood clots
Antiplatelet agents	Drugs that prevent blood clots by inhibiting platelet function
Arterial thrombosis	An interruption of blood flow to an organ or body part due to a blood clot blocking the flow of blood
Body mass index (BMI)	Body mass divided by the square of the body height, universally expressed in units of kg/m²
Catheters	Medical devices (tubes) that can be inserted in the body for a broad range of functions, such as to treat diseases, to perform a surgical procedure, and to provide medicine, fluids and food.
COVID‐19	An infectious disease caused by SARS‐CoV‐2 virus
Deep vein thrombosis (DVT)	Coagulation or clotting of the blood in a deep vein, i.e. far beneath the surface of the skin
Disseminated intravascular coagulopathy	A severe condition in which blood clots form throughout the body, blocking small blood vessels and that may lead to organ failure. As clotting factors and platelets are used up, bleeding may occur, throughout the body (e.g. in the urine, in the stool, or bleeding into the skin)
Duplex ultrasound	Non‐invasive evaluation of blood flow through the arteries and veins by ultrasound devices
Heparin (also known as unfractionated heparin (UFH))	A drug used to prevent blood clotting (anticoagulant, blood thinner)
Hypercoagulability	An abnormality of blood coagulation that increases the risk of blood clot formation in blood vessels (thrombosis)
Low molecular weight heparin	A drug used to prevent blood clotting (anticoagulant)
Obesity	Amount of body fat beyond healthy conditions (BMI > 30 kg/m²)
Placebo	Substance or treatment with no active effect, like a sugar pill
Platelet	Colourless blood cells that help blood clot by clumping together
Pulmonary embolism (PE)	Blood clot in the lung or blood vessel leading to the lung. The clot originates in a vein (e.g. deep vein thrombosis) and travels to the lung
Quasi‐randomised controlled trial (Quasi‐RCT)	A study in which participants are divided by date of birth or by hospital register number, i.e. not truly randomly divided into separate groups to compare different treatments
Randomised controlled trial (RCT)	A study in which participants are divided randomly into separate groups to compare different treatments
Respiratory failure	An abnormality that results from inadequate gas exchange by the respiratory system
SARS‐CoV‐2	The virus (coronavirus 2) that causes COVID‐19
Thrombosis	Local coagulation of blood (clot) in a part of the circulatory system
Vascular	Relating to blood vessels (arteries and veins)
Venous	Relating to a vein
Venous thromboembolism (VTE)	A condition that involves a blood clot that forms in a vein and may migrate to another location (e.g. the lung)

Table 1. Glossary of terms

Table 2. Summary of characteristics of included studies

Study (design)	Country	Participant age (mean)	Setting	Intervention type (dose)	Comparator	All‐cause mortality	Necessity for additional respiratory support	Follow‐up time (mean days)	Total participants allocated	Intervention group participants (anticoagulant)
Ayerbe 2020 (Retrospective cohort)	Spain	67	Hospital^a	Heparin (NR)	NA	OR 0.42 (95% CI 0.26 to 0.66) P < 0.001, in favour of intervention group	NR	8	2075	1734
Liu 2020 (Retrospective cohort)	China	72	ICU (intervention) vs hospital ward (comparator)	Heparin (NR)	NA	Unadjusted OR 1.66, 95% CI 0.76 to 3.64	NR	NR	154	61
Paranjpe 2020 (Retrospective cohort)	USA	NR	Hospital^a	Treatment dose anticoagulation	NA	In‐hospital mortality: intervention 22.5% versus comparator 22.8% In subgroup who required mechanical ventilation: intervention 29.1% versus comparator 62.7% (adjusted HR 0.86, 95% CI 0.82 to 0.89; 395 participants, P < 0.001)	NR	NR	2773	786
Russo 2020 (Retrospective cohort)	Italy	67	Hospital^a	DOACS (NR) in 18 participants and VKA (NR) in 8 participants	NA	RR 1.15 (95% CI 0.29 to 2.57), P = 0.995	NR	NR	192	26
Shi 2020 (Retrospective cohort)	China	69	Hospital^a	LMWH	NA	Reported no deaths in both groups	NR	NR	42	21
Tang 2020 (Retrospective cohort)	China	65	Hospital^a	UFH (10,000 to 15,000 IU/d in 5 participants and LMWH (40 mg/d to 60 mg/d) in 94 participants	NA	No difference (general mortality): (adjusted OR 1.64, 95% CI 0.92 to 2.92; 449 participants) Subgroup analysis: participants with SIC score of ≥ 4(unadjusted OR 0.37, 95% CI 0.15 to 0.90; 97 participants) Participants with D‐dimer > 6 times the ULN (unadjusted OR 0.44, 95% CI 0.22 to 0.86; 161 participants)	NR	28	449	99
Trinh 2020 (Retrospective cohort)	USA	59	ICU	UFH 15 IU/kg/h; or enoxaparin 1 mg/kg twice or once daily; or apixaban 10 mg (if no prior anticoagulation) or 5 mg (if prior anticoagulation) twice daily^b	UFH 5000 IU two to three times daily; or enoxaparin 40 mg twice or once daily; or apixaban 2.5 mg or 5 mg twice daily^b	Reduction in all‐cause mortality (adjusted HR 0.21, 95% CI 0.10 to 0.46) and a lower absolute rate of death in the therapeutic group (34.2% versus 53%)	NR	35	244	161
Total	China: 3 Italy: 1 Spain: 1 USA: 2	‐	‐	‐	‐	6 studies considered mortality; 1 study did not report mortality data	No study considered additional respiratory support	8 to 35 (3 studies)	5929	2888
CI: confidence interval; DOACS: direct oral anticoagulants; GFR: glomerular filtration rate;HR: hazard ratio; ICU: intensive care units;LMWH: low molecular weight heparin; NA: no anticoagulation; NR: not reported; NRS: non‐randomised study;OR: odds ratio; RR: risk ratio; SIC: sepsis‐induced coagulopathy; UFH: unfractionated heparin; VKA: vitamin K antagonist
^aHospital: includes intensive care unit, hospital wards or emergency department. ^bAnticoagulation used twice daily if glomerular filtration rate (GFR) was greater than 30 mL/min, or once daily if GFR was 30 mL/min or less.

Table 2. Summary of characteristics of included studies

Table 3. Summary of characteristics of ongoing studies

Study	Country	Design	Primary outcomes	Estimated number of participants	Estimated primary completion date
ACTRN12620000517976	Australia	RCT	Time to separation from invasive ventilation	172	25 July 2021
ChiCTR2000030700	China	RCT	Time to virus eradication	60	30 September 2020
ChiCTR2000030701	China	RCT	Time to virus eradication	60	30 September 2020
ChiCTR2000030946	China	Prospective cohort	Biochemical indicators	120	24 April 2020
Marietta 2020	Italy	RCT	Clinical worsening (includes death and necessity for additional respiratory support)	300	June 2021
NCT04333407	UK	RCT	All‐cause mortality at 30 days after admission	3170	30 March 2021
NCT04344756	France	RCT	Survival without ventilation	808	31 July 2020
NCT04345848	Switzerland	RCT	Composite outcome of arterial or venous thrombosis, disseminated intravascular coagulation and all‐cause mortality	200	30 November 2020
NCT04352400	Italy	RCT	Time to clinical improvement	256	December 2021
NCT04359277	USA	RCT	Composite incidence of: all‐cause mortality, cardiac arrest, symptomatic deep venous thrombosis, PE, arterial thromboembolism, myocardial infarction, stroke, or shock	1000	21 April 2021
NCT04360824	USA	RCT	Risk of all‐cause mortality	170	16 April 2021
NCT04362085	Canada	RCT	Composite outcome of ICU admission (yes/no), non‐invasive positive pressure ventilation (yes/no), invasive mechanical ventilation (yes/no), or all‐cause death (yes/no) up to 28 days	462	November 2020
NCT04366960	Italy	RCT	Incidence of VTE detected by imaging	2712	August 2020
NCT04367831	USA	RCT	Total number of patients with clinically relevant venous or arterial thrombotic events in ICU	100	November 2020
NCT04372589	Canada	RCT	Intubation and mortality	3000	January 2021
NCT04373707	France	RCT	VTE	602	September 2020
NCT04377997	USA	RCT	Risk of composite efficacy endpoint of death, cardiac arrest, symptomatic deep venous thrombosis, PE, arterial thromboembolism, myocardial infarction, or haemodynamic shock Risk of major bleeding event according to the ISTH definition	300	1 January 2021
NCT04393805	Italy	Retrospective cohort	Bleeding Thrombosis Mortality	877	December 2020
NCT04394377	Brazil	RCT	Hierarchical composite endpoint composed of mortality, number of days alive, number of days in the hospital and number of days with oxygen therapy at the end of 30 days	600	December 2020
NCT04397510	USA	RCT	Mean daily PaO₂:FiO₂	50	31 December 2020
NCT04401293	USA	RCT	Composite outcome of arterial thromboembolic events, venous thromboembolic events and all‐cause mortality at day 30 ± 2 days	308	22 October 2020
NCT04416048	Germany	RCT	Composite endpoint of VTE (DVT and/or fatal or non‐fatal PE), arterial thromboembolism, new myocardial infarction, non‐haemorrhagic stroke, all‐cause mortality or progression to intubation and invasive ventilation	400	30 April 2021
Total number of studies	Australia: 1 Brazil: 1 Canada: 2 China: 3 France: 2 Germany: 1 Italy: 4 Switzerland: 1 UK: 1 USA: 6	Prospective cohort: 1 RCT: 20 Retrospective cohort: 1	12 studies considered mortality Six studies considered additional respiratory support	15,727 participants 997 from NRS 14,730 from RCTs	13 studies to December 2020 8 studies to July 2021 1 study to December 2021
DVT: deep vein thrombosis; FiO₂: fraction of inspired oxygen; ISTH: International Society on Thrombosis and Haemostasis; NRS: non‐randomised studies; PaO₂: arterial oxygen pressure;PE: pulmonary embolism; RCT: randomised controlled trial; VTE: venous thromboembolism

Table 3. Summary of characteristics of ongoing studies

Table 4. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (all‐cause mortality)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Ayerbe 2020	Critical risk	No information	Serious risk	Low risk	Critical risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is no baseline characteristics table comparing the two groups. Essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism, were not considered.	Participants included in both groups were selected from 17 hospitals, and the study was retrospective, therefore it is not possible to know whether the selection was free from bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. The type and doses of heparin in the intervention group were not described.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	There were missing outcome data for 56 participants with no specific information or appropriate analyses. These missing data could cause a critical impact on the estimates.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.
Liu 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Serious risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is no baseline characteristics table comparing the two groups. Essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism, were not considered.	Participants included in both groups were selected from a single hospital, and the study was retrospective, therefore it is not possible to know whether the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We cold not adjust the analyses for this selection bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. The type and doses of heparin in the intervention group were not described.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data was reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias in selection of reported effect estimate, based on the results, from different subgroups analyses.	The study is too problematic to provide useful evidence.
Paranjpe 2020	Serious risk	Moderate risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, an adjusted analysis with propensity scores was performed considering confounding demographic, clinical, and medication use. However, the confounding factors 'participants who underwent surgery during the hospitalisation', 'active cancer treatment', 'concomitant antiplatelet use' and 'history of venous thromboembolism' were not considered.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome, but the study authors used appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems
Russo 2020	Serious risk	Moderate risk	Serious risk	No information	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores, considering confounding demographic and clinical factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'active cancer treatment' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome, but the study authors used appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	Insufficient information to judge. No information is reported on whether there was deviation from the intended intervention.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems
Shi 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is a baseline characteristics table comparing the two groups with limited items. However, the study did not compare essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism.	The participants of the two groups (intervention and comparator) were selected from the same hospital, but as the study was retrospective, it is not possible to know if the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	There is a risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.
Tang 2020	Critical risk	Critical risk	Serious risk	No information	Low risk	Low risk	Critical risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced among the compared groups. There was no table comparing the characteristics of the two groups at baseline. The comparator group included participants who used heparin for less time or did not use heparin. These participants may be less severely ill than those in the intervention group.	Participants included in both groups were selected from the same hospital, but as the study was retrospective, it is not possible to know whether the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	As this was a retrospective study, there is a high risk that the interventions received by participants in the same group were not standardised. Besides, the comparator group also included participants who used heparin for less than seven days. This proximity to the case definition for the intervention group increases the risk of error in the classification of participants. Also, the comparator group considered two very different types of intervention.	Insufficient information to judge. No information is reported on whether there was deviation from the intended intervention.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both, and it is not possible to exclude bias in selection of reported effect estimate, based on the results, from multiple measurements within the outcome domain, multiple analyses of the intervention‐outcome relationship, and different subgroups analyses.	The study is too problematic to provide useful evidence.

Table 4. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (all‐cause mortality)

Table 5. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (major bleeding)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Paranjpe 2020	Serious risk	Serious risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an adjusted analysis with propensity scores considering confounding demographic and clinical factors, and medication use. However, the study did not consider confounding factors 'participants who underwent surgery during the hospitalisation', 'active cancer treatment', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital, and selection may have been related to intervention and outcome. For this outcome, the authors did not use appropriate methods to adjust for selection bias.	There is a high risk that the interventions received by participants in the same group were not standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study has some important problems

Table 5. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (major bleeding)

Table 6. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (hospitalisation)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Shi 2020	Critical risk	Critical risk	Serious risk	Low risk	Low risk	Low risk	Low risk	Critical risk
Judgement	One or more prognostic variables are likely to be unbalanced between the compared groups. There is a baseline characteristics table comparing the two groups with limited items. However, the study did not compare essential characteristics, such as participants already using anticoagulants, participants who underwent surgery during the hospitalisation, concomitant antiplatelet use, and history of venous thromboembolism.	The participants of the two groups (intervention and comparator) were selected from the same hospital, but as the study was retrospective, it is not possible to know if the selection was free from bias. The selection for the study was strongly related to both the intervention and the outcome of interest. We could not adjust the analyses for this selection bias.	There is a risk that the interventions received by participants in the same group have not been standardised. There is a high risk of differential classification errors because the information on the status of the interventions was obtained retrospectively.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified but all reported results corresponded to the intended outcome.	The study is too problematic to provide useful evidence.

Table 6. ROBINS‐I assessments: anticoagulants (all types) versus no treatment for people hospitalised with COVID‐19 (hospitalisation)

Table 7. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (all‐cause mortality)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (death) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Table 7. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (all‐cause mortality)

Table 8. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (major bleeding)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (major bleeding) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Table 8. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (major bleeding)

Table 9. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (hospitalisation)

Study	Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from the intended intervention	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall risk of bias
Trinh 2020	Serious risk	Moderate risk	Low risk	Low risk	Low risk	Low risk	Serious risk	Serious risk
Judgement	To minimise the impact of the absence of randomisation, we performed an analysis with propensity scores considering confounding demographic, clinical and laboratory factors, and medication use. However, the study did not consider confounding factors 'participants who underwent a surgery during the hospitalisation', 'concomitant antiplatelet use' and 'history of venous thromboembolism'.	The included participants in both groups were selected from the same hospital. The study authors considered for inclusion all patients who met the inclusion criteria and who were treated in each period.	Intervention status was well defined based on information collected at the time of intervention.	No deviations from the intended intervention were reported in the study, and if any deviation occurred from usual practice, it was unlikely to impact on the outcome.	No missing data were reported for this outcome.	It is unlikely that the outcome assessment (length of hospital stay) was influenced by the knowledge of the intervention received by the study participants.	The study protocol was not identified or was not available or both (only a preprint was available), and it is not possible to exclude bias.	The study has some important problems

Table 9. ROBINS‐I assessments: anticoagulants (therapeutic dose) versus anticoagulants (prophylactic dose) for people hospitalised with COVID‐19 (hospitalisation)