Types of studies

We will include RCTs. If the process of randomisation is unclear, we will contact the trial authors to obtain further information. If we are unable to contact the authors, we will include the trial in the review and consider it to be at unclear risk of bias. To be eligible, trials must compare at least one of our interventions of interest (placebo versus active treatment, or active treatment versus another active treatment). We will use both abstracts and full‐text publications if they report adequate information about study design, participant characteristics, and interventions. We will only include trials that have been prospectively registered, unless the final trial report was published before 2010 (Roberts 2015).

Types of participants

We will include participants who have undergone treatment in an emergency department for bleeding caused by a blunt or penetrating injury. There will be no restrictions on age, gender, or ethnicity. If an eligible trial contains a mixed age group which includes a paediatric population, then we will perform subgroup analysis of the data.

Mechanisms causing blunt force trauma or penetrating injury will include but are not limited to the following.

• Fall from height

• Fall from horse or other moving object

• Injury from a falling object

• Assault with a blunt object

• Motor vehicle collisions

• Bicycle accidents

• Accidental falls

• Sporting injuries

• Punches and kicks

• Blast injury

• Crush injury

• Stabbing

• Gun shot

• Nail gun wounds

• Shrapnel injury

• Sharp‐glass injury

If an eligible trial contains a mixed population of people (e.g. bleeding due to trauma and other causes), then we will only use data contributed by our population of interest. If no subgroup data are given, and we are unable to contact the corresponding author to provide this information, we will include the trial if at least 80% of the sample size is from our population of interest. We will include participants if they were taking anticoagulant medication or antiplatelet therapy at the time of injury. We will exclude participants with known bleeding disorders, such as haemophilia.

Types of interventions

Eligible trials will have compared one or more of the following interventions:

• antifibrinolytics:

  • tranexamic acid;

  • aprotinin;

  • epsilon‐aminocaproic acid;

• desmopressin;

• recombinant factor VIIa and factor XIII;

• fibrinogen‐containing products;

• fibrin sealants;

• non‐fibrin sealants.

We will not combine different interventions and treatments in the NMA other than those listed above. Trials must compare an intervention of interest versus placebo, or an intervention of interest versus another intervention of interest. We will include trials that use interventions of interest combined with another agent or blood product in each arm (e.g. tranexamic acid plus platelets versus placebo plus platelets), as we consider that the effect of the additional agent in both arms will cancel out.

To explore the optimal treatment pathway, we will consider interventions administered over a range of doses, as both single or multiple doses via intravenous, subcutaneous, intranasal, oral, or topical routes, and at different timings.

The variations in dose, route, and times for interventions may differ greatly. We expect that tranexamic acid will be the intervention most commonly assessed in the included trials, therefore we anticipate this to be the focus of our NMA.

Types of outcome measures

We will use the outcome measures below to assess the relative hierarchy of our interventions.

Electronic searches

Searching other resources

To supplement the database searches, we will handsearch the bibliographies of all included trials, relevant review articles, and other current evidence in order to identify additional trials missed by the electronic searches. We will also contact the authors of ongoing trials to obtain any unpublished data. We will contact authors a maximum of three times.

We will check the reference lists of included trials and relevant systematic reviews to identify any trials not found by the electronic searches. We will also check for relevant retraction statements and errata for the included trials.

Selection of studies

Three review authors (VE, AN and AH) will independently screen the titles and abstracts of citations identified by the electronic searches for potential eligibility. If the title and abstract of the citation are found to be irrelevant, we will exclude it at this stage. The same review authors will then independently screen the full‐text articles of the citations considered to be potentially eligible against the criteria set out in this protocol. If the information needed to make a decision regarding eligibility is insufficient, we will request additional information from the corresponding author of the trial. We will contact the author up to three times within six weeks. If there is no response after six weeks of our initial attempted contact, we will exclude the study. We will keep records of the study selection process and use the information to generate a PRISMA flowchart to illustrate the study flow (Moher 2009). We will record reasons for the exclusion of trials at the full‐text review stage in the 'Characteristics of excluded studies' table. If we encounter translation needs, we will use colleagues or Cochrane resources such as Task Exchange.

Data extraction and management

Three review authors (VE, JS  and AH) will independently use a standardised, piloted form to undertake data extraction of the included trials. The form will be designed following the methods described in chapter 5 of the Cochrane Handbook for Systematic Reviews of Interventions (Li 2020). The three  review authors (VE, JS and AH) will not be blinded to authors, institutions, or outcomes of the trials they are extracting. Colleagues providing translation support will be expected to help extract data. We will pilot the data extraction form on two included trials selected at random (equally split between the review authors). Following this process, if necessary, we will make amendments to the data extraction form.

We will contact corresponding authors up to three times to request further trial data. We will classify the data as unobtainable if there is no response from the authors within six weeks of the initial email request. If a conflict arises over data sources, we will give preference to published data over unpublished data, as published work is more likely to have undergone a rigorous peer‐review process.

The potential dose, route, and timing combinations for each intervention are illustrated in Table 1. There are numerous treatment pathway combinations for each intervention, which will makes it difficult to decide on an approach for data synthesis prior to data extraction. To overcome this, we will use a staged approach to carry out data extraction. In the first stage, three review authors (VE, JS and AH ) will extract trial, participant, and intervention characteristics. Using these data, we will convene an independent blinded external expert panel to categorise the clinically meaningful interventions by dose, route, and timing, in order to decide which to compare. The external expert panel will also help us to create clinically meaningful groups ready for data analysis. Once this process has been completed, the three  review authors (VE, JS and AH) will independently extract the outcome data.

We will extract data for the following items and list these and the outcomes from each trial in the 'Characteristics of included studies' table.

• General information: name of review author carrying out data extraction, date of data extraction, study identifier, surname and contact address of first author, language of trial.

• Trial information: RCT trial design – location of where the trial was run, setting, sample size, duration of trial, power calculation, treatment arms, randomisation, inclusion and exclusion criteria, comparability of groups, length of study.

• Characteristics of participants: age, sex, breakdown of total numbers for those randomised and analysed,  site of injury, severity of injury (as and if reported by trialist), dropouts (percentage in each arm) with reasons and protocol violations, participants on anticoagulants or antiplatelet therapy at the time of injury.

• Characteristics of interventions: number of treatment arms, description of experimental arm(s), description of control arm(s), timing, dose, and route of administration of intervention, and other differences between intervention arms.

• Outcomes (all within 30 days of injury unless otherwise specified): mortality within 48 hours of admission to the emergency department, allogenic blood transfusion in emergency department, mortality due to any cause, mortality within six hours of admission to the emergency department, allogenic blood transfusion within 24 hours of injury, mean number of units of red blood cells transfused, length of ICU and hospital stay, and adverse effects (thromboembolism, transfusion reactions (within 24 hours), and adverse drug reactions). We will use the ICH‐GCP definition of serious adverse events (ICH‐GCP 2018). Where that definition is not used in the included studies, we will extract information about how 'adverse effect' and 'serious adverse effect' were defined in each study.

• Risk of bias assessment: allocation concealment, blinding (participants, personnel, outcome assessors), incomplete outcome data, selective outcome reporting, other sources of bias.

We will extract arm‐level data, rather than study‐level data, from both abstracts and full‐text papers. If there are multiple publications from the same trial, we will use one extraction form and obtain maximal data through extracting data from all available publications. If insufficient information is given in the full text, we will contact the corresponding author, study group, or sponsor to obtain additional data. Where studies have multiple publications, we will collate the reports of the same study so that each study, rather than each report, is the unit of interest for the review, and such studies will have a single identifier with multiple references. We will conduct the review according to this published protocol and report any deviations from it in the 'Differences between protocol and review' section of the full review.

Whilst extracting trial data, we will collect any data about cost, resource usage, and quality of life provided in the included studies. Whilst this will not form a formal economic evaluation, it will offer useful information that may guide decision making. Should any quality of life data be available, we will comment on these outcomes descriptively in our discussion.

We will list all treatment arms in each study in the 'Characteristics of included studies' table.

Two review authors (VE and JS) will independently enter data into Review Manager 5 or Review Manager Web (Review Manager 2020, RevMan Web 2019). Each review author will cross‐check the other review author's entries for accuracy.

Assessment of risk of bias in included studies

We will assess the quality of the included trials using the Cochrane risk of bias tool (ROB1) described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will test the tool on a random sample of trials. Three  review authors (VE, JS and AH) ) will independently assess the risk of bias within each trial and classify it as low, high, or unclear. The judgements of the two review authors will be compared and consensus reached; if there is discrepancy, a third review author (SB) will be consulted. We will report our judgements in the risk of bias table for each trial.

Using the information generated, we will look for statistical heterogeneity in each trial and perform sensitivity analyses accordingly. We will assess risk of bias for the following domains:

• selection bias (random sequence generation and allocation concealment);

• performance bias (blinding of participants and personnel);

• detection bias (blinding of outcome assessment);

• attrition bias (incomplete outcome data);

• reporting bias (selective reporting); and

• other forms of bias.

We will classify each of the domains listed above as follows:

• low risk: if the criterion has been adequately fulfilled in the study;

• high risk: if the criterion has not been fulfilled in the study;

• unclear risk: if insufficient information is provided to permit a definitive judgement.

Measures of treatment effect

When extracting data for dichotomous outcomes (proportion of participants needing an allogenic blood transfusion, mortality, adverse events), we will record the number of participants and events in both the intervention and control arms. We will express the results as odds ratios with 95% confidence intervals (CIs).

However, when extracting arm‐level data for continuous outcomes (e.g. mean number of allogenic blood transfusions per participant), we will record means, standard deviations (or medians with interquartile ranges), and the total number of participants in both the intervention and control arms. If only study‐level data are available, we will note the reported effect size and standard errors. If the data allow, we will use Stata to do the quantitative analyses (Stata 2017).

For continuous outcome data measured using the same scale, we will use mean difference with a 95% CI. However, if this outcome is measured using different scales, we will use standardised mean difference with 95% CI.

Unit of analysis issues

When performing pair‐wise meta‐analyses, we will treat trials with multiple treatment comparisons as individual, independent two‐arm studies. However, this will not be the case in the NMA, where we will include all comparisons, if and when there are adequate data to do so. These trials will be analysed by taking into account the respective treatment effects. The NMA method correctly accounts for correlations in relative effects from trials with more than two arms. We will analyse data using the participant as the unit of analysis.

Dealing with missing data

We will handle missing data using the approach described in Chapter 10 of the Cochrane Handbook for Systematic Reviewsof Interventions (Deeks 2020). We will contact corresponding authors of studies to obtain missing data. We will keep a record of the number of participants lost to follow‐up in each trial. If an included trial has a mixed population, we will only extract data from the relevant population. If this is not possible, we will contact the authors up to three times to request additional information. If we are still unable to obtain the information, and the missing data are thought to lead to serious bias, we will perform a sensitivity analysis to assess the impact of the missing outcome data.

Assessment of heterogeneity

Assessment of reporting biases

We will investigate the presence of small‐study effects in the pair‐wise meta‐analyses through funnel plots and linear regression, if there are at least 10 studies. We will consider a P value of 0.10 or below as statistically significant. Several factors can contribute to the association between study effect size and funnel plot asymmetry. We will differentiate between funnel plot asymmetry caused by publication bias using contour‐enhanced funnel plots (Peters 2008). The contour lines in the plot demonstrate levels of statistical significance. We will assume that a lack of studies in areas of non‐significance will show signs of publication bias.

Data synthesis

We will use Stata to undertake a multivariate NMA which will treat each comparison as a different outcome. We will perform the analyses using the network package in Stata (Stata 2017). We will provide the estimated treatment effect for each comparison with a 95% CI. For pair‐wise meta‐analyses, we will perform direct treatment comparisons using the methods described in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2020). Where data are sufficiently homogeneous, we will perform meta‐analyses in Review Manager 5 or Review Manager Web (Review Manager 2020; RevMan Web 2019). Forest plots illustrating these results will be shown with 95% CIs for all analyses. Both review authors (VE and AH) will enter data into Review Manager 5 or Review Manager Web, and will cross‐check each other's data entry for accuracy.

Where appropriate, we will categorise interventions into clinically meaningful groups during the first stage of data extraction. Each group will act as a single node within the network. We will run sensitivity analyses using different groupings. Each group will contain one type of pharmacological intervention, for example only tranexamic acid, but may include a narrow dose range, route and timing variables, so as to have a pharmacologically similar predicted effect.

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

We will examine the strength of the overall results by performing sensitivity analyses, where appropriate, with and without the trials thought to be at high risk of bias. A sensitivity analysis is used to determine the robustness of an assessment by examining the extent to which results are affected by changing the methods or models of analysis, values of unmeasured variables or assumptions.

We will perform our main analyses using studies deemed as being at low risk of bias, and then undertake a sensitivity analysis that incorporates all the included studies. We will look at the effect of participant dropout, and will categorise the trials into groupings of:

• less than 20% dropout;

• 20% to 50% dropout; and

• more than 50% dropout.

We will analyse each group separately. We will explore heterogeneity using a fixed‐effect model to assess sensitivity.

Summary of findings and assessment of the certainty of the evidence

We will create a 'Summary of findings' table for each intervention for the following outcomes: need for allogenic blood transfusion in the emergency department or within 24 hours of injury, all‐cause mortality (deaths occurring within 30 days after the injury), mean number of red blood cell transfusion units per person (within 30 days), length of ICU and hospital stay, and adverse events (within 30 days).

We will use the GRADE approach, which involves consideration of study limitations, consistency of effect, imprecision, indirectness, and publication bias, to assess the certainty of the evidence contributing data to the analyses for the prespecified outcomes. We will use the methods described in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2020), employing GRADEpro GDT software (GRADEpro GDT).

Two review authors (VE and AH) will independently judge the certainty of the evidence. Any disagreements will be resolved by discussion or by consulting a third review author (SB). We will provide justifications in the 'Summary of findings' table for downgrading of the certainty of the evidence in order to aid the reader's understanding of the review, and we will incorporate these judgements into the review for each outcome.

For the NMA, we will evaluate the confidence of the evidence using the CINeMA framework (Confidence in Network Meta‐Analysis) (Salanti 2014). We will use the online CINeMA tool, which assesses confidence for each comparison within the network and is based on within‐study bias, across‐studies bias, indirectness, imprecision, heterogeneity, and incoherence (CINeMA 2017).

We will create a summary of findings table before writing the 'Results' and 'Authors' conclusions' sections of the review.