We first aim to evaluate current practices regarding red blood cell transfusion in critically ill patients with traumatic brain injuries by estimating the frequency of red blood cell transfusion in these patients. Secondly, we seek to evaluate transfusion thresholds, determinants and outcomes associated with transfusion strategies.

The protocol of the review is registered in PROSPERO (http://www.crd.york.ac.uk/prospero) CRD42014007402.

We will conduct a systematic review in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement 21 and The Cochrane Handbook for Systematic Reviews of Interventions 22 methodological recommendations.

Since we are interested in the frequency of red blood cell transfusion and its determinants, our systematic review will include prospective and retrospective cohort studies, and randomized controlled trials. Patients suffering from a traumatic brain lesion (any severity) will be considered. In case of a mixed population, at least 80% of patients included in a specific study have to respect this criterion for the study to be eligible. Since we expect that few studies will report precisely if patients were recruited at hospital admission or intensive care unit (ICU) admission, we will not restrict inclusion to critically ill patients but rather consider a population-based approach of acute care hospital admissions. Studies and trials will have to report data on red blood cell transfusion frequency. Studies who specifically studied patients with blood disorders and coagulopathies will be excluded. Table 1 and Table 2 present the structured study question and inclusion and exclusion criteria, respectively.

We will systematically search MEDLINE, Embase, BIOSIS and The Cochrane Library (from their inception up to a maximum of nine months before submission for publication) for eligible studies. References of included articles and abstracts of major conferences will be screened to identify additional potentially eligible studies. Experts in neurocritical care medicine, not members of our team, will also be contacted to identify additional ongoing studies. We will request available transfusion data from investigators of retrieved studies if deemed necessary.

Our search strategy will be based on keywords related to transfusion and anemia, as well as traumatic brain injury. Clinicians, investigators with expertise in transfusion or in neurocritical care, and information specialists will be consulted to verify the search strategy, identify synonyms and additional search terms. Relevant index terms (Medical Subject Headings and Emtree) will be added to the strategy. The search will be limited to human studies 22 . No language or date of publication restriction will be used. The search strategy will be first designed for Medline and Embase, and will be adapted for other electronic databases afterwards. The current version of our Medline search strategy is presented in Additional file 1. This preliminary strategy has been tested through an iterative process in order to achieve sufficient specificity while maintaining high sensitivity. Results will be imported in EndNote (version X7.0.1, New York City: Thomson Reuters, 2011) and duplicates will be removed. References will then be exported to a Microsoft Excel (version 14.1.0, Redmond, WA: Microsoft, 2011) spreadsheet in order to complete the selection process.

Two independent reviewers will screen all identified references to determine eligibility, first from titles and abstracts, and then based on full text evaluation for studies that could be potentially eligible. In case of disagreement on the inclusion of a study, a third reviewer will be consulted. In case the blood product transfused is not clearly specified, authors will be contacted to ensure that reported data pertains to red blood cell transfusion.

A translation of non-English or non-French articles will be obtained. Agreement on study selection will be evaluated with a kappa coefficient. Considering the high sensitivity of the search strategy, we expect the kappa will indicate moderate agreement. In case the agreement is too low, indicating an evasive interpretation of eligibility criteria, a third reviewer will review records’ titles and abstracts.

A preliminary version of the abstraction form will be pilot-tested and customized by two reviewers using four publications. Two independent reviewers will abstract data using the standardized form. In case of discrepancy, consensus will be reached with the involvement of a third reviewer. Authors will be contacted if relevant data is missing or clarification is needed.

Data pertaining to study characteristics (design, date of completion, funding sources, and so on), patients’ baseline characteristics (age, gender, type of neurological disease, severity of the lesion on admission, and so on), clinical management (surgical, medical), hemoglobin levels, blood products (type of products received, timing, quantity, repetition, thresholds, and so on), co-interventions (type, timing), clinical outcomes (mortality and withdrawal of life-sustaining therapies and their timing, length of hospital and ICU stay, neurological outcome (any scale; for example, Glasgow Outcome Score (GOS)), complications, and so on) [Additional file 2] will be extracted from published reports.

Risk of bias of included RCTs will be assessed using The Cochrane Collaboration tool for assessing the risk of bias 22 . Risk of bias in cohort studies will be assessed using a pilot version of a new tool for the assessment of the risk of bias in non-randomized studies, currently under development by Cochrane’s Non-Randomized Studies Methods Group.

We will report the cumulative incidence of transfusion in the course of hospital stay (primary outcome). Risk ratios of the association between red blood cell transfusion and potential determinants (categorical variables such as sex) or any relevant clinical outcomes, such as mortality, unfavorable neurological outcome, complications (any or specific complications, or categories of complications 23 ), will be reported. Mean differences of potential determinants or outcomes (continuous or ordinal variables such as age, hemoglobin levels, severity of the traumatic brain injury, length of stay, and so on) according to transfusion status will be reported. We will compute mean transfusion thresholds as well as mean differences in number of units transfused according to outcomes when available. A two-sided 5% type I error will be considered for all analyses.

If appropriate, results from cohort studies will be pooled with results from randomized controlled trials that did not randomize patients to specific transfusion strategies. Results from randomized controlled trials allocating patients to different transfusion strategies will be pooled separately, if deemed appropriate.

Variances of cumulative incidences of transfusion from all studies will be stabilized using a Freeman-Tukey transformation 24 and proportions will be pooled with DerSimonian and Laird random-effects approach 25 using R statistical software (version 2.15.1: R Core Team, R Foundation for Statistical Computing June 2012, Vienna, Austria). Means and mean differences will be pooled with inverse variance method with random effects. Risk ratio analyses will be conducted with Review Manager (RevMan) (version 5.1, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2012) using Mantel-Haenszel random-effect models. Pooled effect sizes and their 95% confidence limits will be reported.

Statistical heterogeneity will be measured using the Cochran’s Q-test and I² statistics 26 , the latter being interpreted as low from 0 to 40%, moderate from 30 to 60%, substantial from 50 to 90% and considerable from 75 to 100% 22 .

We will evaluate the risk of publication bias by visual exploration of funnel plots. We will also evaluate the risk of selective reporting of outcomes within studies by searching for previously published protocols on registration website (http://www.controlled-trials.com and clinicaltrials.gov).

The ability of a study to answer the review question will be evaluated in terms of applicability. Applicability concerns relate to deviation of a study from the ideal study designed to answer our research question (in relation to our primary outcome). For instance, a study recruiting patients with any severity of traumatic brain injury at hospital admission, therefore including patients with mild traumatic brain injury that might not be admitted to an ICU, will be considered as having high applicability concerns.

This project will allow the knowledge synthesis regarding transfusion practices in patients with traumatic brain injury. Considering the paucity of data and the equipoise on the optimal transfusion strategy in this population, it is of major importance to assess current practices. Therefore, this project, in addition to our previous systematic review of comparative studies, will systematically group original research data on the topic of red blood cell transfusion in traumatic brain injury.

Considering the observed lack of evidence from comparative studies and the risk of bias associated with observational studies, we expect results to be heterogeneous. Our results will provide information to inform the design of further studies in traumatic brain injury and red blood cells transfusion. Ultimately, we will obtain critical data regarding transfusion practices in patients with traumatic brain injury that will allow us to better design a high-quality non-inferiority trial. For example, information on the frequency of transfusion among patients with traumatic brain injury will facilitate the calculation of sample size estimates and recruitment rates. The knowledge of the usual hemoglobin thresholds observed in patients with traumatic brain injury and those associated with better outcomes will help setting acceptable, plausible and realistic comparative thresholds in a future trial.

Despite the use of rigorous methodology, we do expect high statistical and clinical heterogeneity in our analyses and few studies of low risk of bias. The strength of our conclusions may thus be limited by those factors. We may uncover only a limited number of comparative studies of transfusion strategies that included a small number of patients. This may potentially limit the planned sensitivity and subgroup analyses. In addition, we will conduct analysis of potential determinants and outcomes associated with transfusion through univariate analysis. In cohort studies, univariate analyses of associations are prone to confusion bias. If sufficient data are available, we will construct a meta-regression, which will help to, at least partially, control for potential confusion.

Little is known on optimal red blood cell transfusion strategies in patients with traumatic brain injury. In order to design studies to improve clinical practices, evidence-based information has to be gathered. We propose to conduct a systematic review that will synthesize the current knowledge from published clinical studies in the field. Our results will be used to optimize future prospective studies on this topic in order to conduct high-quality and rigorous studies with the aim of increasing the quality of care received by patients with traumatic brain injury.