Design
This protocol is based on The Cochrane Handbook for Systematic Reviews of Interventions and specific recommendations for conducting systematic reviews and meta-analysis of preclinical studies [25, 26]. We will report our results according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) [27]. The PRISMA-P checklist for reporting standards of systematic review protocols is also available as Additional file 1. Our systematic review protocol is registered online in PROSPERO (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42018086662) CRD42018086662.
Eligibility criteria
We will include studies using in vivo animal models with acute cerebral lesions limited to traumatic brain injury (all types) or stroke (ischemic or hemorrhagic) comparing outcome in injured animals treated with a specific RBC transfusion strategy (restrictive or liberal) with outcome in injured animals treated with a different RBC transfusion strategy or no treatment or any other intervention. In our study, the hemorrhagic stroke group refers to animal models of intracerebral hemorrhage (ICH) or subarachnoid hemorrhage (SAH). Both animal models of focal and global brain ischemia will be included. There will be no restrictions in terms of species and comorbidities of models. We will consider studies regardless of their primary outcomes of interest. There will be no restriction applied to date or language of publication. We will consult translators for publications not written English or French. Our structured research question and our study eligibility criteria are presented in Tables 1 and 2, respectively.
Table 1
Population | Animal models of: - Traumatic brain injury - Ischemic stroke - Hemorrhagic stroke (ICH, SAH) |
Intervention | Red blood cell transfusion |
Comparator | An alternative red blood cell transfusion strategy (hemoglobin threshold/target) or No transfusion or Any other intervention |
Primary outcomes | Neurobehavioral |
Secondary outcomes | All-cause mortality Infarct size Intracranial pressure Cerebral perfusion pressure (CPP) Cerebral blood flow (CBF) Brain tissue oxygen tension (PbtO2) |
Study design | Comparative preclinical studies |
ICH : Intracerebral hemorrhage, SAH: Subarachnoid hemorrhage |
Table 2
Inclusion criteria | Exclusion criteria |
1. Studies comparing at least two interventional groups | 1. Human studies |
2. In vivo animal models of TBI or ischemic stroke or hemorrhagic stroke (ICH or SAH) | 2. In vitro studies |
3. RBC transfusion in at least one interventional group | 3. Non-mammals used as animal models |
4. Different transfusion strategy or no transfusion or any other intervention as a comparator | 4. Transfusion before cerebral injury is induced |
5. Any study outcome | |
TBI: Traumatic brain injury, ICH: Intracerebral hemorrhage, SAH: Subarachnoid hemorrhage, RBC: Red blood cell |
Table 3
Construct and external validity
Domains | Description |
Age | All same age vs different ages |
Sex | All male vs all female vs mixed |
Comorbidities | All healthy vs models with comorbidities vs mixed |
Breeding | Inbred only vs outbred (wild-type) only vs mixed |
Number of participating centers | Single vs multicentered |
Species | Rats, mice, monkeys, cats, dogs, others |
TBI models | Fluid percussion injury (FPI), controlled cortical impact (CCI), weight-drop model, acceleration model |
Acute stroke models | Spontaneous vs induced, global vs focal ischemia, endovascular vs open surgery with ligature, embolic, photothrombosis |
ICH models | Blood vs collagenase injection in cerebral parenchyma |
SAH models | Endovascular perforation of internal carotid artery, blood injection in basal cisterns |
Blood transfusion type | Whole blood vs packed red blood cells |
Anemia | Hemorrhage or hemodilution as part of model preparation |
Co-interventions | Crystalloids, colloids, medications, mechanical ventilation, others |
Timing of treatment | Immediate vs delayed treatment measures after brain injury |
Death | Animals found dead or sacrificed when met criteria for important deterioration (imminent death) |
Study identification
We will search MEDLINE (PubMed), EMBASE and Web of Science using a structured search strategy. We have developed a strategy for MEDLINE using a combination of keywords related to “anemia”, “red blood cell transfusion”, “traumatic brain injury”, “stroke”, “intracranial hemorrhage”. We have also included the Medical Subject Headings (MeSH) terms linked to the keywords mentioned previously and a search filter for animal studies [28]. This search strategy has been pre-tested to obtain a high sensitivity and acceptable specificity. It was also reviewed by an information specialist with expertise in health sciences for additional robustness. The strategy was then modified for EMBASE (with Emtree and specific filter for animal studies) and Web of Science [29]. The search strategy used for MEDLINE is presented in Additional file 2. We will also review references of the included studies identified through databases searches to identify additional studies. We will search databases from their inception and we will update the search prior to submission of the systematic review in order to include the most recent eligible studies.
Study selection
Records resulting from electronic database searches will be imported in Endnote (version X8, New York City: Thomson Reuters, 2016) where duplicates will be identified and removed. Titles and abstracts of retrieved records will then be transferred to a Microsoft Excel (version 15.29, Redmond, WA: Microsoft, 2016) spreadsheet and screened independently by two reviewers. Full-text articles of potentially eligible studies will be retrieved and reviewed to assess full eligibility. Disagreement will be resolved by consensus. If consensus is not possible, a senior team member will arbitrate.
Data collection
Data from included studies will be abstracted independently by two reviewers using a standardized and pre-tested form. In case of discrepancy, consensus will be reached through discussion or the involvement of a third reviewer. We will collect data on study characteristics (design, length of follow-up), sample characteristics (species, number, age, weight, gender, hemoglobin level), model preparation (anesthesia, type and severity of injury, hemorrhage, hemodilution), intervention (target, threshold, units, volume, timing with injury), co-interventions (crystalloids, colloids, medications, surgical interventions, etc.) and outcomes. We will use graph analysis tools to extract data of interest if results are not reported in the text or tables. If necessary, we will contact authors for additional information.
Outcomes
Our primary outcome is neurological function (neurobehavioral performance). Some examples of neurobehavioral tests that we expect to encounter include, but are not limited to: cylinder test, beam balance, beam walking, rotarod test, open field test, Morris water maze, elevated T-maze, passive avoidance. Our secondary outcomes include mortality, infarct size, intracranial pressure, cerebral perfusion pressure, cerebral blood flow and brain tissue oxygen tension. We will consider all neurobehavioral tests reported in the included studies. We will collect information on the definition of mortality used in included studies; whether animals had died spontaneously or if they received euthanasia when they reached a state of imminent death according to specific criteria of deterioration of their condition.
Risk of bias
We will use an adaptation of the Collaborative Approach to Meta Analysis and Review of Animal Data from Experimental Studies (CAMARADES) tool (See Additional file 3) for risk of bias assessment [30]. The same two reviewers who abstracted data will independently assess the risk of bias. The domains evaluated are: i) selection bias (inclusion and exclusion criteria, randomization), ii) information bias (blinding of outcome assessment), iii) confounding (blinding of induction to cerebral lesions and care during the follow-up, comorbidities, temperature control, anesthesia agents), iv) other bias or considerations (selective reporting, conflict of interest, sample size, peer-reviewed publications, animal welfare). The assessment of criteria for animals to enter the study and the assessment of selective reporting are the two elements that we added to the original tool.
Quality of reporting
We will evaluate the quality of reporting in individual publications with a list of questions according to the NIH Principles and Guidelines for Reporting Preclinical Research [31]. These principles were elaborated with editors from more than 30 preclinical and basic science journals. Many journals have agreed to endorse these guidelines.
Construct and external validity
Two reviewers will independently assess each included study for the potential applicability of the results to the clinical setting defined as construct validity. We will evaluate baseline characteristics of the animals (species, age, sex, comorbidities) as well as methods for model preparation (TBI, acute ischemic stroke, ICH, SAH, anemia), type of blood transfusion, definition of death and co-interventions. Setting of the studies (single vs multicentered) and heterogeneity of the animal population (age, sex, comorbidities, breeding) will help evaluate external validity (generalizability of the results).
Data analysis and synthesis
A descriptive synthesis of our results will be presented. We plan to report data on neurobehavioral performance, and any other continuous data, with normalised mean difference (NMD) as the effect-size measure with 95%CIs. This is considering that data from sham and uninjured animal models are available or can be easily inferred. If there are no data available on sham animals or it cannot be inferred, we will report neurobehavioral results with standardised mean difference (SMD). In a single study, different tests can be conducted to assess neurobehavioral performance in the same cohort of animals. To synthesize the overall neurobehavioral performance, we will combine data from neurobehavioral tests into a single “nested’ outcome, when possible, using a method previously described by members of our research team (Vesterinen 2014) [32]. Mortality data, as well as any other dichotomous data, will be reported with risk ratios (RR) as the effect-size measure and 95% confidence intervals (95%CI).
If appropriate, we will pool data from our primary and secondary outcomes using the inverse of the variance with random-effects models. We will need at least 3 studies using similar animal models, transfusion strategy and comparator to conduct a meta-analysis. Cochrane Review Manager (RevMan) version 5.3 (The Cochrane Collaboration, Copenhagen, Denmark, 2014) will be used. Statistical heterogeneity will be evaluated using the I2 index and classified as very low (0–25%), low (25–50%), moderate (50–75%) or high (> 75%) [33]. Funnel plots will be used for the evaluation of potential publication bias for neurobehavioral outcome and mortality results with additional analysis for each type of brain injury. Subgroup analyses will be conducted when possible using the following: mammal class orders (primates vs. rodents vs. all others), transfusion thresholds (restrictive vs. liberal strategies), the presence of co-interventions, induced anemia during preparation (hemorrhage vs. hemodilution vs none), type of neurocritical condition (TBI vs. ischemic stroke vs. ICH vs. SAH) and risk of bias (high/unclear risk of bias vs. low risk of bias).