Types of studies

As it was likely that the number of randomised controlled trials (RCTs) that could be included in this review were small given the authors' tacit knowledge of the subject, we included data from non‐randomised studies (NRS) for all outcome evaluations. We only included NRS which were prospective in design and included a rigorous definition of the interventions and outcomes measured as an attempt to assess as wide a data volume as possible. Such studies included quasi‐randomised, non‐RCTs and prospective cohort studies. We excluded retrospective and uncontrolled studies (e.g. case series) due to the potential for unacceptable bias. Despite the inferiority of study quality that most NRS offered, a review of eligible NRS might identify a need for further RCTs and prove informative to the design of such a trial.

Types of participants

We included studies conducted on people with haematological malignancies requiring a red blood cell (RBC) transfusion whilst receiving intensive chemotherapy or radiotherapy, or both, with or without HSCT. We Included studies on participants of all ages, with the exception of neonates (up to 28 days old). We accepted the individual study's definition of intensive chemotherapy or radiotherapy, sufficient to cause myelosuppression. The number of potential intensive chemotherapy or radiotherapy regimens, or both was broad. We listed some examples, but these were not limited to, chemotherapeutic regimes listed on public oncology web sites such as:

• Public Health England, Systemic anti‐cancer therapy Chemotherapy Dataset;

• Cancer Care Ontario Drug Formulary;

• Macmillan Cancer Support Combination Chemotherapy Regimen;

• BC Cancer Agency Lymphoma and Myeloma Chemotherapy Protocols;

• Royal Surrey County Hospital NHS Foundation Trust Haematology Chemotherapy.

We included studies that involved participants at all stages of treatment. We included studies on people with high risk myelodysplasia (refractory anaemia with excess of blasts (RAEB), and RAEB in transformation into acute leukaemia, (RAEB‐t)) if intensive chemotherapy or radiotherapy, or both was received. We excluded studies on people with a haematological malignancy requiring transfusion support only.

Where a study included mixed populations of participants, we planned to only use data relevant to haematological malignancies. There were no studies with mixed populations of participants with haematological and solid tumours or non‐intensive and intensive chemotherapy. In future updates of this review, when a study includes mixed populations of participants with haematological and solid tumours, we will only use data relevant to haematological malignancies. If the data are not available within the published literature or through direct author contact, we will exclude the study if fewer than 80% of the study population had haematological malignancies or high‐risk myelodysplasia. In future updates of this review, when studies include participants receiving intensive and non‐intensive chemotherapy or radiotherapy for haematological malignancies, we will only include data from the former group. If these data are not available, we will exclude the study if fewer than 80% of the study population were receiving intensive chemotherapy or radiotherapy for haematological malignancies.

Types of interventions

We included allogeneic red blood cell (RBC) transfusion strategies defined as 'restrictive' and 'liberal'. We expected that such definitions varied between studies, however, we accepted the individual study definitions of these strategies such that the restrictive (intervention) group in all included studies should receive a transfusion of allogeneic RBC units below a certain 'trigger' or 'threshold' haemoglobin or haematocrit level. The control group should receive a transfusion of RBC in accordance with a more liberal transfusion policy such as when a higher haemoglobin or haematocrit threshold was reached. For the purposes of this Cochrane review, we applied a standard published equation to convert haematocrit to haemoglobin: Haemoglobin (g/dL) = Haematocrit (%)/3 (Quintó 2006).

Types of outcome measures

We reported the outcomes from RCTs and NRS separately (Reeves 2011).

Electronic searches

Bibliographic databases

We searched for RCTs and NRS in the following databases:

• Cochrane Central Register of Controlled Trials (CENTRAL, the Cochrane Library 2016, Issue 6 ‐Appendix 1);

• MEDLINE (1946 to 15 June 2016‐ Appendix 2);

• Embase (1974 to 15 June 2016 ‐ Appendix 3);

• CINAHL (1982 to 15 June 2016 ‐ Appendix 4);

• PubMed (epublications only ‐ Appendix 5);

• Transfusion Evidence Library (www.transfusionevidencelibrary.com) (1980 to 15 June 2016 ‐ Appendix 6);

• LILACS (1982 to 15 June 2016 ‐ Appendix 7);

• IndMed (1986 to 15 June 2016 ‐ Appendix 8);

• KoreaMed (1995 to 15 June 2016 ‐ Appendix 9);

• Web of Science: Conference Proceedings Citation Index‐Science (CPCI‐S) (Thomson Reuters, 1990 to 15 June 2016 ‐ Appendix 10).

We combined searches in MEDLINE with the Cochrane RCT highly sensitive search filter, as detailed in Lefebvre 2011 and with the SIGN observational studies filter for MEDLINE. Also, we merged searches in Embase and CINAHL with the relevant SIGN RCT and observational studies filters (www.sign.ac.uk/methodology/filters.html).

Ongoing studies databases

We also checked for ongoing RCTs and NRS in the following databases to 15 June 2016:

• ClinicalTrials.gov (Appendix 11);

• ISRCTN Register (Appendix 12);

• WHO International Clinical Trials Registry (ICTRP) (Appendix 13);

• EU Clinical Trials Register (EUDRACT) (Appendix 14).

Searches were not limited by date, language or publication status.

Searching other resources

We augmented database searching with the following.

Handsearching reference lists

We checked the reference lists of all included studies, relevant reviews, conference abstracts and current treatment guidelines to identify additional studies that were not retrieved through databases searches.

Personal contact

We contacted the lead authors of relevant studies, study groups and international experts working in this field to identify any unpublished material, or to gather information regarding ongoing studies.    

Selection of studies

We selected studies according to Chapter 7 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). The Systematic Review Initiative’s Information Specialist (CD) initially screened all search hits for relevance against the eligibility criteria and discarded all those that were clearly irrelevant. Thereafter, two review authors (LE, RM) independently screened all the remaining references for relevance against the full eligibility criteria. We retrieved full‐text papers for all references for which a decision on eligibility could not be made from title and abstract alone. We assessed study design features against the review inclusion criteria. Additional information was requested from study authors as necessary to assess the eligibility for inclusion of individual studies. Two review authors (LE, RM) discussed the results of study selection and resolved all disagreements by discussion without the need to consult a third review author (MM). We recorded the reasons why potentially relevant studies failed to meet the eligibility criteria. We reported the results of study selection using a PRISMA flow diagram (Moher 2009) (Figure 1).

Figure 1

---

---

Study flow diagram for study selection

We initially assessed the RCTs. As there were not two or more well‐conducted eligible RCTs including at least 1000 patients, we extended the assessment and selection of studies to NRS.

Data extraction and management

As recommended in Higgins 2011a, two review authors (RM, LE) independently extracted data onto standardised forms and performed a cross‐check. The two review authors piloted the data extraction form for included RCTs and NRS. Any disagreements between the review authors were resolved by consensus. We were not blinded to the names of the study authors, institutions, journals or the study outcomes. If a trial had multiple publications we used one form only to extracted relevant data. When these sources provided insufficient information, we contacted the authors and study groups for additional details. One review author (RM) performed data entry into software, which a second review author (LE) checked for accuracy.

We extracted the following information for each study:
1. Source: Study ID; report ID; review author ID; date of extraction; ID of author checking extracted data; citation of paper; contact authors details;
2. Eligibility: Fate of study; reason for exclusion (as appropriate);
3. General study information: Publication type; study objectives; funding source; conflict of interest declared; other relevant study publication reviewed;
4. Study design: Was there a comparison; how were participants allocated to groups?; which part of the study were prospective?; on which variables was comparability between groups assessed?; type of study;
5. Study details and methods: Location; country; setting; number of centres; total study duration; recruitment dates; length of follow‐up; power calculation; primary analysis (and definition); stopping rules; method of sequence generation; allocation concealment; blinding (of clinicians, participants and outcome assessors); concerns regarding bias; inclusion & exclusion criteria; primary outcome(s); secondary outcomes;
6. Characteristics of interventions: Number of study arms; description of experimental arm; description of control arm; duration of red cell storage; frequency of minor ABO mismatched transfusions; other treatment (e.g. gamma irradiation);
7. Characteristics of participants: Age; gender; ethnicity; body surface area; primary diagnosis; stage of disease; category of intensive chemotherapy received; details of radiotherapy received; type of HSCT received; additional therapy received; risk of alloimmunisation; baseline haematology laboratory; subgroups evaluated; cofounders reported;
8. Participant flow: Total number screened for inclusion; total number recruited; total number excluded; total number allocated to each study arm; total number analysed (for primary outcome); number of allocated patients who received planned treatment; number of drop‐outs with reasons (percentage in each arm); protocol violations; missing data;
9. Outcomes: All cause mortality (undefined and within short, medium and long term periods); mortality due to infection, bleeding, TRALI, TACO, ABO incompatibility, TTI, other (with details); number and severity of bleeding episodes, adverse transfusion reactions, serious infections, toxicity score and quality of life scores; number and volume of red cell transfusion units received per patient; interval between red cell transfusions, number and volume of platelet doses received per patient; interval between platelet transfusion; duration of hospital admission; duration of intensive care admission; readmission to hospital during follow‐up period.

10. In the case of the NRS, we additionally extracted information on confounding factors, results from adjusted models and variables adjusted for.

Assessment of risk of bias in included studies

We assessed the quality of all included trials (RCTs) using the Cochrane risk of bias tool as described in Higgins 2011b. Two review authors (RM, LE) independently assessed each element of potential bias listed below as either 'high', 'low' or 'unclear risk of bias'. We also provided a brief description in the Characteristics of included studies table of the judgement statements upon which the review authors assessed potential bias. We reached a consensus on the degree of risk of bias through comparison of the review authors' statements and, where necessary, through consultation with a third author (SH).

To assess risk of bias, we addressed the following questions in the ’Risk of bias’ table for each included study.

• Selection bias: Was the allocation sequence randomly generated? Was allocation adequately concealed prior to assignment?

• Performance bias: Where possible, were the study participants and personnel adequately blinded?

• Detection bias: Was blinding of the outcome assessors effective in preventing systematic differences in the way in which the outcomes were determined?

• Attrition bias: Were incomplete outcome data adequately addressed for every outcome?

• Reporting bias: Are reports of the study free of selective outcome reporting?

• Other issues: Was the study apparently free of other problems that could put it at risk?

We assessed the quality of the NRS using an adapted Newcastle‐Ottawa scale (Wells 2013). We included an expansion of the comparability of cohorts section with the inclusion of a list of important confounding factors we felt should be addressed in the studies methods or analysis. Also, we included an additional question concerning equal follow‐up durations for mortality, the primary and secondary outcomes of this review. We scored the studies with reference to the Ottawa Hospital Research Institute's coding manual and scale guidelines. We provided judgement statements for each score in tabular form as for the 'Risk of bias' assessment of interventional trials. The score was reported out of a maximum of 10 stars using the following categories.

Selection (one star each, maximum four stars).

• Representiveness of the exposed cohort

• Selection of the non‐exposed cohort

• Ascertainment of exposure

• Demonstration that outcome of interest was not present at start of study

Comparability (maximum of two stars).

• Recognition of at least 75% of the main potential confounding factors as listed below within study design or adjustment in analysis (two stars).

• • Primary diagnosis, separated by: haematological malignancy (acute leukaemia, high risk myelodysplasia, chronic lymphocytic leukaemia (CLL), myeloma, lymphoma) or natural history of primary diagnosis e.g. a) 'Indolent' (myeloma, low grade lymphoproliferative disorders (e.g. CLL, low grade lymphoma), acute lymphoblastic leukaemia/acute myeloid leukaemia in complete remission (CR1), chronic myeloid leukaemia in chronic phase, high risk myelodysplastic syndrome) or b) 'aggressive' (high grade lymphomas (e.g. diffuse large B‐cell lymphoma (DLBCL), Burkitt), all other stages of acute leukaemia, chronic myeloid leukaemia in accelerated or blast phase).

  • Age: variability in the age of patients included, e.g. paediatric (< 18 years) versus adult (> 18 years).

  • Gender: male to female ratio.

  • Previous severe bleeding (e.g. WHO/CTCAE grade > 3 or equivalent).

  • Use of anticoagulation during study.

  • Previous alloimmunisation.

  • Co‐existing cardiovascular disease.

  • Performance Status (e.g. ECOG, KPS).

  • HSCT: autograft versus allograft (allograft source: sibling, matched unrelated, cord donation; conditioning type: myeloablative including total body irradiation versus reduced intensity conditioning); harvest type: peripheral blood stem cell versus bone marrow harvest); ABO compatibility; cytomegalovirus (CMV) compatibility, stem cell dose received.

  • Radiation use in addition to intensive chemotherapy.

• Recognition of 50% to 75% of the main potential confounding factors within study design or adjustment in analysis (one star).

Outcome (one star each, maximum of four stars).

• Assessment of outcome.

• Was follow‐up long enough for outcomes to occur?

• Adequacy of follow‐up of cohorts.

• Follow‐up equal between groups for primary and secondary outcomes?

We requested additional information from study authors as necessary to address the quality of individual studies. Two review authors discussed the results of the study quality assessment and try to resolve any discrepancies between themselves.

We planned to use the 'Risk of bias' assessment to explore statistical heterogeneity in each included study and to perform sensitivity analyses, but no meta‐analyses were performed. For both RCTs and NRS, we planned to exclude monocentric studies from a sensitivity analysis due to their higher likelihood of bias, but no meta‐analyses were performed.

Measures of treatment effect

We did not combine data retrieved from RCTs and NRS in a single meta‐analysis, but reported data separately (Reeves 2011).

We undertook quantitative assessments using Review Manager (RevMan) 2014.

Unit of analysis issues

There were no cluster‐randomised trials or cross‐over studies eligible for inclusion in this review.

We treated other unit of analysis issues in accordance with the advice given in Higgins 2011c. There was a unit of analysis issue for this review for the number of RBC or platelet transfusions reported in Webert 2008. Data were reported per participant day rather than per participant. However, the authors used a recurrent events analysis to take into account the repeated events data (Cook 1997).

Dealing with missing data

Where we identified data as missing or unclear in published literature, we contacted study authors directly. We recorded the number of patients lost to follow‐up for each study. Where possible, we planned to analyse data by intention‐to‐treat (ITT) but if insufficient data were available, we planned to present per protocol (PP) analyses (Higgins 2011c).

Assessment of heterogeneity

We only performed one meta‐analysis due to the small number of eligible studies and the lack of studies reporting similar outcomes.

We had planned to perform separate meta‐analyses for data extracted from RCTs and observational studies, if the clinical and methodological characteristics of studies were sufficiently homogeneous. We had also planned to assess the statistical heterogeneity of treatment effects between pooled studies using a Chi² test with a significance level at P < 0.1. We would have used the I² statistic to quantify the degree of potential heterogeneity and classify it as moderate if I² > 30% or considerable if I² > 75%.

We also anticipated that there would be at least moderate clinical and methodological heterogeneity within the included studies and the random‐effects model would be appropriate. If the heterogeneity was still considerable, we also planned to not report the overall summary statistic. In addition, we planned to assess potential causes of heterogeneity by sensitivity and subgroup analyses (Deeks 2011).

Assessment of reporting biases

We did not perform a formal assessment of potential publication bias (small‐trial bias) (Sterne 2011), because the review included fewer than 10 trials.

Data synthesis

We performed analyses according to the recommendations of Deeks 2011 using aggregated data for analysis. For statistical analysis, we entered data into the Cochrane statistical package of Review Manager (RevMan) 2014. One review author (RM) entered the data into the software. A second author (LE) checked data for accuracy. We were only able to perform a limited number of meta‐analyses due to the small number of eligible studies and the perceived clinical and methodological heterogeneity between these studies. Aditionally, we provided a narrative quantitative summary for relevant outcome data retrieved from RCTs and NRS separately. If the qualitative assessment was not feasible, we therefore described outcome data extracted from each study individually and we also commented on any apparent trends.

We used the GRADE system to build a 'Summary of Findings' table, as suggested in Schünemann 2011a and Schünemann 2011b. In the review protocol we considered the following most relevant outcomes.

• All‐cause mortality

• Quality of life

• Bleeding episodes

• Serious infections

• Length of hospital admission

• Hospital readmission rate

Subgroup analysis and investigation of heterogeneity

No subgroup analyses were performed because we only performed a limited number of meta‐analyses. We had planned to perform subgroup analysis for each outcome if data were available for the following.

• Intensive chemotherapy versus radiotherapy.

• Paediatric (less than 18 years) versus adult (18 years or older).

• HSCT versus no HSCT.

• For those that received HSCT:

  • autologous HSCT versus allogeneic HSCT;

  • reduced intensity transplant versus myeloablative HSCT.

• For those that received intensive chemotherapy without HSCT:

  • induction versus consolidation chemotherapy (at least for acute leukaemia);

  • acute leukaemia versus non acute leukaemia.

We had also planned to undertake a subgroup analysis of results from studies including participants with pre‐existing cardiovascular diseases, if the relevant data were given separately in the study.

Sensitivity analysis

This was not feasible as we only performed a limited number of meta‐analyses and one of which was where data from the three included RCTs were combined. However, we originally planned to assess the robustness of our findings by performing the following sensitivity analyses where appropriate:

• including only those studies with a 'low risk of bias' (e.g. RCTs with methods assessed as low risk for random sequence generation and concealment of treatment allocation);

• including only those studies with less than a 20% dropout rate;

• including only multicentric studies.