Red blood cell transfusion for people undergoing hip fracture surgery

Summary of findings for the main comparison. Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery

Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery
Patient or population: people undergoing hip fracture surgery¹ Settings: hospital Intervention: liberal threshold red blood cell transfusion² Comparison: restrictive threshold red blood cell transfusion³
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Restrictive threshold	Liberal threshold
30‐day mortality Follow‐up: mean 30 days	50 per 1000⁴	46 per 1000 (33 to 63)	RR 0.92 (0.67 to 1.26)	2683 (5 studies)	⊕⊕⊝⊝ low^5,6	‐
Inability to walk 10 feet (3 m; or across a room) without human assistance Follow‐up: mean 60 days	283 per 1000⁴	283 per 1000 (246 to 326)	RR 1.00 (0.87 to 1.15)	2083 (2 studies)	⊕⊕⊝⊝ low^7,8	‐
Thromboembolism (in hospital)	20 per 1000⁴	23 per 1000 (11 to 47)	RR 1.15 (0.56 to 2.37)	2416 (4 studies)	⊕⊕⊝⊝ low^6,9	‐
Stroke (in hospital)	2 per 1000⁴	5 per 1000 (2 to 14)	RR 2.4 (0.85 to 6.79)	2416 (4 studies)	⊕⊕⊝⊝ low^6,9	‐
Wound infection (in hospital)	8 per 1000⁴	13 per 1000 (6 to 27)	RR 1.61 (0.77 to 3.35)	2332 (3 studies)	⊕⊕⊝⊝ low^6,10	‐
Cardiovascular events ‐ myocardial infarction	24 per 1000⁴	14 per 1000 (9 to 23)	RR 0.59 (0.36 to 0.96)	2217 (3 studies)	⊕⊝⊝⊝ very low^6,11	‐
Respiratory infections (namely pneumonia)	18 per 1000⁴	24 per 1000 (17 to 35)	RR 1.35 (0.95 to 1.92)	2416 (4 studies)	⊕⊕⊝⊝ low^6,12	‐
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
1. Although we included evidence for pre‐operative, peri‐operative and postoperative transfusion, the majority of the evidence applied to postoperative transfusion. 2. The liberal transfusion threshold was a haemoglobin concentration of about 10 g/dL in four trials, and 11.3 g/dL in one trial. 3. The restrictive transfusion threshold in four trials was a haemoglobin concentration of about 8 g/dL or symptoms of anaemia, and 9.7 g/dL in one trial. 4. The assumed risk was the median control risk across studies. 5. We downgraded the evidence one level because of risk of bias: 57% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 6. We downgraded the evidence one level because of imprecision: generally because of the small number of events in the studies reporting data for this outcome has resulted in wide confidence intervals for these studies. 7. We downgraded the evidence one level because of risk of bias: 96% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 8. We further downgraded the evidence one level because of risk of bias: both participants and study personnel "were aware of study group assignment after randomisation" in the two studies reporting data for this subjective outcome. Given that the participants themselves are involved in assessing this outcome, knowledge of treatment allocation may influence outcome measurement. 9. We downgraded the evidence one level because of risk of bias: 60% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 10. We downgraded the evidence one level because of risk of bias: 70% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 11. We downgraded the evidence two levels because of risk of bias: 94% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups; and, although numbers were comparable between the two transfusion threshold groups in this trial, overall 265 (13%) participants had incomplete electrocardiographic results and in 355 (18%) participants there was no blood sample for troponin testing. Thus, attrition bias was a potential problem. 12. We downgraded the evidence one level because of bias: 93% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups.

Background

Description of the condition

Hip fracture is the common term for fractures of the proximal end of the femur (upper end of the thigh bone). These fractures are often subdivided into intracapsular fractures, which are fractures located within the hip joint capsule, and extracapsular fractures, which are outside the hip joint capsule. These main types have different degrees of mean blood loss and require different fixation techniques. Most hip fractures occur in old and frail people (mean age about 80 years) and are usually the result of a low impact injury or a fall. The incidence of hip fracture is increasing in line with increases of the mean age of populations in many countries: the number of hip fractures worldwide has been estimated to rise from 1.7 million in 1990 to 6.3 million in 2050 (Gullberg 1997).

There is general agreement among orthopaedic surgeons that hip fracture requires surgery either to repair the fracture or to replace part or all of the hip joint (Egol 2009). About 5% to 10% of people with hip fracture die within one month of their fracture (Parker 2006). Many people who have hip fracture surgery do not recover fully after their hip fracture, being less mobile and less independent than before their injury (Egol 2009).

Description of the intervention

Alongside the need for surgery, many people with hip fracture will receive red blood cell transfusions because of anaemia or bleeding. Anaemia in these people (resulting in a deficiency in the oxygen‐carrying red blood cells) may be present prior to their fracture (Penninx 2006). It is well recognised that the incidence of anaemia in the general population rises with increasing age, often reflecting concurrent illnesses and comorbidities. Anaemia may also occur as a consequence of blood loss at the time of the fracture or during and after surgery (Foss 2006). It is now recognised that people with hip fracture may have large blood loss that occurs after the fracture and before surgery (Smith 2011). Mean blood loss has been calculated at 1.5 units of red blood cells in intracapsular fractures and 2 units in extracapsular fractures. People with hip fracture are often frail and have less reserve than younger people to cope with the resulting haemodynamic changes. In the UK, many people wait more than 24 hours for surgery although early operation was incentivised by the Department of Health in April 2010 through their 'best practice tariffs' scheme (Department of Health Payment by Results team 2012). Red blood cell transfusion is then used to improve the oxygen‐carrying capacity of the blood and is a key part of supportive management of people undergoing hip fracture surgery. People receiving blood transfusions require active monitoring of all vital signs including pulse, blood pressure and temperature to detect any acute adverse reactions.

The use, timing and quantity of red blood cell transfusion may depend on several factors, including the severity of anaemia. One retrospective cohort study of over 3000 people operated on for hip fracture reported that nearly 30% received a perioperative allogeneic (blood donated from other people) blood transfusion (Johnston 2006). Various studies of people with hip fracture have demonstrated substantial variability in the use of red blood cell transfusion among both physicians and hospitals (e.g. Hutton 2005). The decision for red blood cell transfusion is often based on a threshold haemoglobin concentration with a red blood cell transfusion being triggered should the haemoglobin fall below this threshold value. For instance, Foss 2006 reported that people with hip fracture were given red blood cell transfusion if the haemoglobin concentration fell below 10 g/dL at any point during their hospitalisation. The decision for transfusion may also be influenced by other factors such as participant age and co‐existing medical morbidity such as coronary or respiratory disease (Dillon 2005).

Other strategies for preventing or correcting anaemia should be considered in people with hip fracture, in addition to transfusion needs. Methods aimed at reducing the need for red blood cell transfusion include perioperative cell salvage or treatment with oral or intravenous iron, erythropoietin, or iron plus erythropoietin. Erythropoietin is a hormone that promotes the formation of red blood cells by the bone marrow.

How the intervention might work

It is generally accepted that red blood cell transfusion corrects any pre‐existing anaemia and replaces lost blood in people with hip fracture and improves or maintains the oxygen‐carrying capacity in the circulation. Improved tissue oxygenation may then help recovery during and after surgery. However, limited data indicate improved functional outcomes after surgery with higher haemoglobin concentrations in transfused people with hip fracture (Lawrence 2003). There are also well‐recognised risks associated with red blood cell transfusion as for any blood component for transfusion. Adverse effects of transfusion include transfusion of the wrong blood products (due to errors in the pathways of processing or administration) resulting in acute haemolytic transfusion reactions, transfusion‐transmitted infections, other types of haemolytic reactions, respiratory complications and allergic reactions (SHOT 2010). A 2‐unit transfusion of red blood cells represents a volume of over 500 mL and may result in transfusion‐associated circulatory overload in older people who may poorly tolerate infusion of even moderate volumes of fluid because of co‐morbidities such as cardiac disease.

Why it is important to do this review

Red blood cell transfusion is a frequently used clinical intervention with around two million red blood cell units issued by UK transfusion services per year (SHOT 2010). It is a costly and scarce resource and is associated with risks (see How the intervention might work) (Carson 1999; SHOT 2010). Many red blood cell transfusions are given to stable and non‐bleeding people where the evidence from clinical studies suggests no clear benefit (Carless 2010). Surgery for hip fracture is common and many people receive red blood cell transfusions. In these people, important outcomes also include postoperative functional recovery, mobility and quality of life (QoL) (Adunsky 2008; Foss 2008). Given the rising incidence of hip fracture, the wide variation in transfusion practice and risks of red blood cell transfusion, it is important to identify and appraise the evidence for its safe and effective use in order to inform practice. This review includes evidence from randomised controlled trials on the use of red blood cell transfusions (or alternatives) in people with hip fracture.

Objectives

To assess the effects (benefits and harms) of red blood cell transfusion in people undergoing surgery for hip fracture.

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials or quasi‐randomised controlled trials (where the method of allocating participants to a treatment is not strictly random: e.g. by date of birth, hospital record number, alternation) assessing red blood cell transfusion for people undergoing hip fracture surgery.

Types of participants

People requiring or undergoing surgery for hip fracture (examples of surgery for hip fracture include sliding hip screw, intermedullary hip screw and arthroplasty).

Types of interventions

We set out to compare the following:

red blood cell transfusion versus no transfusion;
red blood cell transfusion versus alternative methods such as cell salvage, iron supplements or erythropoietin (we excluded studies evaluating use of tranexamic acid, as this is the subject of an another Cochrane review (Perel 2013));
red blood cell transfusion protocol A versus red blood cell transfusion protocol B;

- an example of this would be a comparison of a red blood cell transfusion given according to criteria detailed in one protocol (e.g. volume of transfusion, rate of transfusion) with a red blood cell transfusion given according to criteria detailed in another protocol (e.g. with a different volume of transfusion or different rate of transfusion);

red blood cell transfusion threshold A versus red blood cell transfusion threshold B. Trials using different measures, such as haematocrit (a measure of the percentage of red blood cells to the total blood) for setting thresholds were also eligible;

- an example of this would be a comparison of a liberal versus a restrictive haemoglobin concentration threshold, whereby a person would only be eligible to receive a red blood cell transfusion when their haemoglobin concentration fell below a given liberal or restrictive transfusion threshold.

Types of outcome measures

We included the following outcomes. We did not specify in advance the time points for the measurement of these outcomes, as we were interested in recording all measures that had been made per outcome. Overall, however, we were interested in outcomes reported in the immediate postoperative period through to outcomes reported during the follow‐up for the trial. We have reported the time points at which each outcome was measured alongside the analysis of data.

Primary outcomes

Mortality.
Mobility and functional recovery.
Postoperative morbidity, including medical complications (e.g. wound infections (in hospital), thromboses (in hospital), stroke, myocardial infarction and other cardiovascular events) and respiratory infections (including pneumonia).

Secondary outcomes

Postoperative or discharge haemoglobin.
Quality of life (QoL).
Length of stay in hospital.
Adverse effects of transfusion (including haemolytic transfusion reaction, inappropriate blood component transfusion, allergic reactions and transfusion‐transmitted infections).

Search methods for identification of studies

Electronic searches

The Systematic Review Initiative's Information Specialist (CD) formulated the search strategies in collaboration with the Cochrane Heart Group.

Bibliographic databases

We searched the following databases:

Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (31 October 2014);
Cochrane Central Register of Controlled Trials (CENTRAL, 2014, Issue 10);
MEDLINE (Ovid) (1946 to 20 November 2014);
EMBASE (Ovid) (1974 to 20 November 2014);
PubMed (searched for e‐publications ahead of print on 20 November 2014);
CINAHL (NHS Evidence) (1982 to 20 November 2014);
British Nursing Index Database (NHS Evidence) (1992 to 20 November 2014);
Transfusion Evidence Library (1980 to 20 November 2014);
LILACS (1982 to 20 November 2014);
IndMed (1985 to 20 November 2014);
KoreaMed (1997 to 20 November 2014);
PakMediNet (1995 to 20 November 2014);
Web of Science: Conference Proceedings Citation Index‐Science (CPCI‐S) (Thomson Reuters, 1990 to 20 November 2014).

Online databases of ongoing trials

ClinicalTrials.gov (20 November 2014).
ISRCTN Registry (20 November 2014).
World Health Organization International Clinical Trials Registry Search Platform (WHO ICTRP) (20 November 2014).

Appendix 1 shows all search strategies used. We combined searches in MEDLINE with the Cochrane Highly Sensitive RCT Search Filter as detailed in Chapter 6 of the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2011). We combined searches in EMBASE and CINAHL with adaptations of the relevant Scottish Intercollegiate Guidelines Network (SIGN) RCT filters. We applied no restrictions on language or publication status.

Searching other resources

We checked reference lists of relevant articles to identify any eligible studies missed through the electronic searching.

Data collection and analysis

Selection of studies

Three review authors (SB, SM and AS) screened all titles and abstracts of papers identified via the electronic searches. We excluded only clearly irrelevant references at this first stage. We retrieved the full‐texts of the remaining references and three review authors independently assessed them for inclusion using a review‐specific eligibility form. We resolved any screening disagreements by discussion.

Pairs of review authors from four authors (SB, SM, AS and SS) undertook the screening of the references identified in the three search periods ending 21 February 2012, 8 October 2013 and 20 November 2014.

Data extraction and management

Two review authors (AS and SM) independently undertook data extraction using a piloted study‐specific data extraction form. We extracted data on the setting of the trial, the methods and statistical assumptions made, the characteristics of participants and interventions, details of the outcomes measured and timing of the outcome assessments. We resolved disagreements by discussion.

Assessment of risk of bias in included studies

Two review authors (AS and SM) independently assessed risk of bias for each trial using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). We resolved disagreements by discussion and through consultation with a third review author (SB) as required. For each of the included trials, we assessed the risk of bias as low risk, high risk or unclear risk for the following domains.

Generation of random sequence (selection bias).
Concealment of treatment allocation schedule (selection bias).
Blinding of participants to treatment allocation (performance bias or detection bias).
Blinding of personnel (person(s) delivering the treatment) to treatment allocation (performance bias).
Blinding of outcome assessors to treatment allocation (detection bias).
Completeness of the outcome data (including checks for possible attrition bias through withdrawals, loss to follow‐up and protocol violations).
Selective reporting of outcome (reporting bias).
Other sources of bias (other bias). We assessed whether each trial was free of problems not identified via the above domains.

Measures of treatment effect

We calculated risk ratios (RR) for dichotomous outcomes. We expressed treatment effects for continuous data outcomes as mean differences (MD). We used 95% confidence intervals (CI) throughout. Where reported, we described non‐parametric measures such as medians and interquartile ranges in the text and tables.

Unit of analysis issues

We did not anticipate finding or find cross‐over or cluster randomised trials. Should either of these have been found, we would have made appropriate adjustments according to the advice in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b).

Dealing with missing data

Where possible, we sought missing data directly from the author(s) of the individual trial(s). For all included trials, we noted levels of attrition. If data had been available, we would have undertaken sensitivity analysis to examine the impact of losses on dichotomous outcomes.

Assessment of heterogeneity

Assessment of clinical heterogeneity included consideration of participant characteristics (e.g. underlying morbidity, type of fracture and surgery), trial design and risk of bias, care programmes provided to trial participants such as method of anaesthesia, and outcome definition and measurement.

We assessed statistical heterogeneity of treatment effects between trials using a Chi² test with a significant level at P value < 0.1. We used the I² statistic to quantify the percentage of variability that was due to heterogeneity (where I² > 40% indicated moderate heterogeneity and I² > 75% indicated considerable heterogeneity).

Assessment of reporting biases

We did not formally assess reporting biases in this review. We made every effort to identify unpublished studies through the search activities identified earlier in this report. If there had been sufficient numbers of studies (over 10) included in individual meta‐analysis, we would have used funnel plots to assess possible reporting biases.

Data synthesis

Meta‐analysis was undertaken using Review Manager 5 where there were sufficient data (RevMan 2011). We used a fixed‐effect model for combining data in all instances, as we observed no substantial heterogeneity. If substantial heterogeneity had been identified in a fixed‐effect meta‐analysis, we would have noted this and repeated the analysis using a random‐effects model.

Subgroup analysis and investigation of heterogeneity

If there had been sufficient data, we would have undertaken subgroup analyses of type of hip fracture (intracapsular vs. extracapsular), type of surgery (hip fracture fixation vs. hip replacement) and gender to examine for significant differences in treatment effects. We did not have appropriate data to substantiate this type of analysis within the review, but we intend to undertake these subgroup analyses in future updates of the review as data allow. We would use the test for subgroup differences provided in Review Manager 5 to establish whether the subgroups are statistically significantly different from one another (RevMan 2011).

Sensitivity analysis

We planned to undertake sensitivity analyses exploring aspects of trial and review methodology. These would have included exploring the effects of removing trials at high or unclear risk of selection bias (reflecting lack of confirmation of random sequence generation and allocation concealment); detection bias (reflecting lack of assessor blinding) or attrition bias, such as from high levels of missing data. There were not enough data to enable these analyses to be undertaken. In future updates of this review, we will perform these sensitivity analyses as data allow.

However, we undertook two post‐hoc sensitivity analyses. In the first, we removed the largest trial from all pooled analyses where it reported data to examine the result of pooling the other smaller trials. In the second, we removed the trial that randomised and transfused participants perioperatively to determine whether the timing of the study eligibility for transfusion (perioperative or postoperatively) affected the pooled estimates for those outcomes where it had reported data. We provided details of these sensitivity analyses in the Effects of interventions.

'Summary of findings' tables

We constructed a 'Summary of findings' table for this review; this included using the GRADE approach to assess the quality of evidence related to the primary outcomes listed in the Types of outcome measures (Schünemann 2011).

Results

Description of studies

See Characteristics of included studies and Characteristics of excluded studies.

Results of the search

The search strategy identified 1086 references of which we excluded 208 in the first screening for being either a duplicate or clearly irrelevant to the scope of this review. Of the remaining 878 references, we excluded 844: 537 did not meet the inclusion criteria for participants, interventions, or both and 307 were not randomised controlled trials.

We obtained the full text of the remaining 34 references. We contacted, by email, the authors of seven identified trials for further information about their trials (Carson 2011; Foss 2009; Gregersen 2015; Matot 2012; Nielsen 2012; Palmer 1998; Parker 2013). All the authors responded to our enquiries and we incorporated the information and data they provided into this review.

Of the 34 references subject to full‐text screening, we deemed six trials (reported in 22 references) to be eligible for inclusion (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Palmer 1998; Parker 2013), and one as a trial awaiting assessment (ChiCTR‐TRC‐10000822). We excluded 11 studies for not meeting the eligibility criteria of this review (Gampopoulou 2004; Izuel‐Rami 2005; Izuel‐Rami 2006; Jans 2011; Matot 2012; Moghaddam 2009; Muir 1995; Nielsen 2012; Prasad 2009; Serrano Trenas 2011; Zufferey 2010). We identified no ongoing trials. Full details are reported in the PRISMA flow diagram (Figure 1).

Figure 1

Study flow diagram

Included studies

Five of the six included studies were published in full (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013), whereas Palmer 1998 was published only in a conference abstract and presented no outcome data. All six studies compared red blood cell transfusion threshold A versus red blood cell transfusion threshold B. Thus, no included studies investigated our other three listed comparisons: red blood cell transfusion versus no transfusion, red blood cell transfusion versus an alternative method to red blood cell transfusion or red blood cell transfusion protocol A versus red blood cell transfusion protocol B.

Carson 1998 acted as a pilot trial for the Functional Outcomes in Cardiovascular Patients Undergoing Surgical Hip Fracture Repair (FOCUS) trial (Carson 2011), but otherwise these were conducted as separate trials with no overlap in recruitment. Some overlap occurred between Carson 1998 and Palmer 1998; four participants were reported as being enrolled at the Edinburgh site in one month "at the end of recruitment" in Carson 1998 and included in the analysis. A further account found in Table 25.2 of McClelland 2009 linked the two trials and stated that "two patients included in this series [Palmer 1998] are included among the 80 patients in the US study [Carson 1998]". The absence of results for Palmer 1998 means that any overlap (whether two or four participants) does not impact on the results of this review and we have, for convenience but at the risk of some very limited double counting, treated Carson 1998 and Palmer 1998 as separate trials.

Sample sizes

The six trials randomised 2722 participants. The numbers of participants randomised into each trial were 18 (Palmer 1998), 84 (Carson 1998), 120 (Foss 2009), 200 (Parker 2013), 284 (Gregersen 2015), and 2016 (Carson 2011). The total number of participants included per outcome is detailed in the Effects of interventions section and ranged from 107 to 2683 participants.

Setting

Two trials were multicentre with participants from across the USA (Carson 1998) or from both the USA and Canada (Carson 2011). As described above, Carson 1998 also performed the study in Edinburgh (UK) for one month only. The four single‐centre trials were from Denmark (Foss 2009; Gregersen 2015), and the UK (Palmer 1998; Parker 2013).

Participants

All trials included people requiring surgery for a hip fracture. Table 1 details the types of surgical procedures and the types of hip fracture in the trials. Palmer 1998 did not report any further demographic data for their randomised participants. Table 2 describes age, gender, cardiac and red blood cell transfusion history in the trials.

Table 1. Type of surgical procedure and type of fracture

Study ID	Type of surgical procedure	Type of hip fracture
Carson 1998	Multiple screws/plates (n = 59) Hemiarthroplasty (n = 12) Bipolar hemiarthroplasty (n = 13)	Femoral neck (n = 30)* Intertrochanteric (n = 55) Subtrochanteroc (n = 6)
Carson 2011	Not stated	Femoral neck (n = 854)^# Intertrochanteric (n = 1034) Subtrochanteroc (n = 183) Reverse oblique (n = 21)
Foss 2009	Screws/pins (n = 10) Arthroplasty (n = 46) Sliding hip screw (n = 46) Intermedullary hip screw (n = 18)	Not stated
Gregersen 2015	Internal fixation (n =221) Arthroplasty (n = 57) Other (n = 6)	Not stated
Palmer 1998	Not stated	Not stated
Parker 2013	Not stated Percutaneous screw fixation was an exclusion criterion	Not described in full Intracapsular (n = 68)

* Seven participants (five in the liberal and two in the restrictive transfusion threshold groups) in the trial had more than one type of hip fracture.

^# Several participants had more than one type of hip fracture: there was an excess of 41 fractures in the liberal group and 39 in the restricted group (data not available for four participants).

See: Summary of findings for the main comparison Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery

Table 2. Baseline characteristics of randomised participants

Study ID	Liberal transfusion threshold	Restrictive transfusion threshold
Carson 1998	Number of participants reporting these data: 42 Age: mean (SD): 81.3 (8.1) years; range: 50‐94 years Gender: number (%) men: 9 (21.4) Number (%) people withcardiac conditions: Any: 19 (45.2) Coronary artery disease: 12 (28.6) Congestive heart failure: 6 (14.3) Transfusion history: Number of red blood cell transfusions received before randomisation, mean (SD): 0.5 (1.0) Last preoperative haemoglobin concentration, mean (SD): 11.7 (1.6) g/dL Randomisation haemoglobin concentration, mean (SD): 9.1 (0.6) g/dL	Number of participants reporting these data: 42 Age: mean (SD): 83.3 (10.8) years; range: 32‐95 years Gender: number (%) men: 11 (26.2) Number (%) people withcardiac conditions: Any: 19 (45.2) Coronary artery disease: 12 (28.6) Congestive heart failure: 6 (14.3) Transfusion history: Number of red blood cell transfusions received before randomisation, mean (SD): 0.3 (0.6) Last preoperative haemoglobin concentration, mean (SD): 11.6 (1.0) g/dL Randomisation haemoglobin concentration, mean (SD): 9.1 (0.6) g/dL
Carson 2011	Number of participants reporting these data: 1007 Age: mean (SD): 81.8 (8.8) years Gender: number (%) men: 250 (24.8) Number (%) people withcardiac conditions: Any: 637 (63.3) Coronary artery disease: 402 (39.9) Congestive heart failure: 184 (18.3) Hypertension: 824/1003 (82.2) Transfusion history: transfusions before randomisation: number/total number (%) of participants receiving 0 red blood cell transfusions: 754/1006 (75.0) number/total number (%) of participants receiving ≥ 1 red blood cell units: 252/1006 (25.0) Haemoglobin concentration before surgery, mean (SD): 11.3 (1.5) g/dL Haemoglobin concentration during eligibility screening, mean (SD): 9.0 (0.8) g/dL	Number of participants reporting these data: 1009 Age: mean (SD): 81.5 (9.0) years Gender: number (%) men: 239 (23.7) Number (%) people withcardiac conditions: Any: 631 (62.5) Coronary artery disease: 403 (39.9) Congestive heart failure: 167 (16.6) Hypertension: 821/1005 (81.7) Transfusion history: transfusions before randomisation: number/total number (%) of participants receiving 0 red blood cell transfusions: 720/1008 (71.4) number/total number (%) of participants receiving ≥ 1 red blood cell units: 288/1008 (28.6) Haemoglobin concentration before surgery, mean (SD): 11.3 (1.5) g/dL Haemoglobin concentration during eligibility screening, mean (SD): 9.0 (0.8) g/dL
Foss 2009	Number of participants reporting these data: 60 Age: mean (SD): 81 (6.8) years Gender: number (%) men: 14 (23) Number (%) people withcardiac conditions: Hypertension: 14 (23) Atrial fibrillation: 3 (5) Congestive heart failure: 5 (8) Ischemic heart disease: 4 (7) Cardiovascular disease: 21 (35) Transfusion history: not stated Haemoglobin concentration on admission: reported graphically and not all the data were readily interpretable, but mean was clearly 13.0 g/dL	Number of participants reporting these data: 60 Age: mean (SD): 81 (7.3) years Gender: number (%) men: 14 (23) Number (%) people withcardiac conditions: Hypertension: 20 (33) Atrial fibrillation: 3 (5) Congestive heart failure: 3 (5) Ischemic heart disease: 10 (17) Cardiovascular disease: 28 (47) Transfusion history: not stated Haemoglobin concentration on admission: reported graphically and not all the data were readily interpretable, but mean was between 13.0 and 14.0 g/dL
Gregersen 2015	Number of participants reporting these data: 140 Age: mean (SD): 88 (6.9) years Gender: number (%) men: 34 (24) Number (%) people withcardiac conditions: Cardiovascular disease: 25 (18) Transfusion history: not stated Number (%) of people with anaemia at baseline: 68 (49) Postoperative haemoglobin levels between 9.7 g/dL and 11.3 g/dL during the first 6 postoperative days Repeated measurements of haemoglobin concentration levels showed maintained mean haemoglobin concentration levels of 12.2 g/dL (95% CI 12.2 to 12.3)	Number of participants reporting these data: 144 Age: mean (SD): 86 (6.8) years Gender: number (%) men: 36 (25) Number (%) people withcardiac conditions: Cardiovascular disease: 34 (24) Transfusion history: not stated Number (%) of people with anaemia at baseline: 70 (49) Postoperative haemoglobin levels between 9.7 g/dL and 11.3 g/dL during the first 6 postoperative days Repeated measurements of haemoglobin concentration levels showed maintained mean haemoglobin concentration levels of 11.3 g/dL (95% CI 11.3 to 11.4)
Palmer 1998	No data reported	No data reported
Parker 2013	Number of participants reporting these data: 100 Age: mean (range): 84.4 years (60‐104) Gender: number (%) men: 17 (17) Number (%) of people withcardiac conditions: Any cardiac disease: 37 (37) Hypertension: 42 (42) Angina: 7 (7) Previous myocardial infarction: 4 (4) Previous congestive cardiac failure: 5 (5) Other cardiac disease: 21 (21) Transfusion history: not stated Haemoglobin concentration (mean) on admission: 11.5 g/dL Haemoglobin concentration (mean) after surgery (time point post surgery not defined): 8.7 g/dL	Number of participants reporting these data: 100 Age: mean (range): 84.2 years (60‐97) Gender: number (%) men: 15 (15) Number (%) of people with cardiac conditions: Any cardiac disease: 50 (50) Hypertension: 46 (46) Angina: 9 (9) Previous myocardial infarction: 9 (9) Previous congestive cardiac failure: 3 (3) Other cardiac disease: 29 (29) Transfusion history: not stated Haemoglobin concentration (mean) on admission: 11.8 g/dL Haemoglobin concentration (mean) after surgery (time point post surgery not defined): 8.9 g/dL

SD: standard deviation.

In summary, the mean age of participants ranged from 81 to 87 years and the trials included far fewer men than women: overall 16% of the participants were men in Parker 2013, 23% were men in Foss 2009, 24% were men in both Carson 1998 and Carson 2011, and 25% were men in Gregersen 2015. Five trials included people with baseline cardiac conditions (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). The percentage of people with any cardiac condition at baseline was reported by four trials and was 21% in Gregersen 2015, 43.5% in Parker 2013, 45.2% in Carson 1998, and 62.9% in Carson 2011. Foss 2009 reported the number of people with any of five pre‐existing chronic cardiac conditions and was the only trial explicitly excluding participants with "acute cardiac or other acute severe medical conditions".

Timing of randomisation varied, with four trials randomising participants when their haemoglobin concentration dropped postoperatively: in Carson 1998 and Carson 2011, when haemoglobin concentration fell to less than 10 g/dL within the first three days post operation; in Palmer 1998 when haemoglobin concentration dropped to between 8 and 10 g/dL within the first two days post operation and in Parker 2013 when haemoglobin concentration measured between 8 and 9.5 g/dL on their first or second day postoperatively. Foss 2009 randomised participants preoperatively at admission to hospital. Gregersen 2015 randomised participants after surgery.

Two trials reported details of the number of red cell transfusions prior to study randomisation (Carson 1998; Carson 2011). Carson 1998 reported the mean (standard deviation (SD)) number of red blood cell transfusions received before randomisation, which were similar between groups at 0.5 (SD 1.0) in the liberal transfusion threshold group and 0.3 (SD 0.6) in the restrictive transfusion threshold group. Carson 2011 reported the number of participants receiving at least 1 unit of red blood cells: 252 (25%) in the liberal transfusion threshold group and 288 (28.6%) in the restrictive transfusion threshold group.

Interventions

In all trials, the intervention groups were a liberal haemoglobin concentration threshold for red blood cell transfusion versus a restrictive haemoglobin concentration threshold for red blood cell transfusion. The liberal transfusion threshold was receipt of 1 unit of packed red blood cells at the time of random assignment and as much blood as necessary to maintain the haemoglobin concentration greater than 10 g/dL (Carson 1998; Carson 2011; Palmer 1998; Parker 2013), receipt of a red blood cell transfusion when haemoglobin concentration fell to below 10.0 g/dL at any time between admittance "to the post‐anaesthesia care unit" and the fifth postoperative day (Foss 2009), and receipt of 1 or 2 units of red blood cells when the haemoglobin threshold was at or below 11.3 g/dL within the first three weeks following surgery (Gregersen 2015).

The restrictive transfusion threshold was receipt of a red blood cell transfusion if participants showed symptoms of anaemia or if their haemoglobin dropped to less than 8 g/dL in Carson 1998 and Carson 2011, when participants were symptomatic of anaemia in Parker 2013, when perceived necessary by the physicians (Palmer 1998), when the haemoglobin concentration was 8 g/dL or less (with transfusion not based on symptoms or presence of clinical anaemia) in Foss 2009, and receipt of 1 or 2 units of red blood cells when the haemoglobin threshold was at 9.7 g/dL or less within the first three weeks following surgery (Gregersen 2015).

Outcomes

The follow‐up periods for outcomes identified as primary outcomes by individual trials ranged from three days to one year post operation. Of our primary outcomes, mortality data were available at 30 days for five trials (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013), at 60 days for three trials (Carson 1998; Carson 2011; Parker 2013), at 90 days for two trials (Gregersen 2015; Parker 2013), and at 120 and 365 days for one trial (Parker 2013). Five trials reported on postoperative function or mobility. Two trials reported on inability to walk 10 feet (3 m; or across a room) without human assistance at 60 days (Carson 1998; Carson 2011). Foss 2009 reported on the gaining of functional independence during hospitalisation and the cumulated ambulation score (CAS) on the first three postoperative days. Parker 2013 assessed mobility at eight weeks from discharge using a commonly used mobility score. Gregersen 2015 measured physical ability before and at 90 days after surgery and reported the data as physical recovery. Four studies reported data on complications (Carson 1998; Carson 2011; Foss 2009; Parker 2013), as detailed in the Characteristics of included studies table. Gregersen 2015 only reported on complications as the main causes of death. Palmer 1998 did not report what outcomes were measured (primary or secondary) within their study.

Excluded studies

For the full list of excluded studies, see Characteristics of excluded studies.

We excluded 11 studies from this review following assessment of the full text or correspondence with the authors. We excluded six studies because of ineligible interventions (Izuel‐Rami 2005; Izuel‐Rami 2006; Moghaddam 2009; Prasad 2009; Serrano Trenas 2011; Zufferey 2010); two studies because their study population was ineligible (Gampopoulou 2004; Nielsen 2012); one study was not a randomised clinical trial (Muir 1995); and two studies, which were only reported in trial registrars, were stopped at the early stages due to poor recruitment (Jans 2011; Matot 2012).

Ongoing trials and studies awaiting classification

We assessed one study as awaiting assessment (ChiCTR‐TRC‐10000822; see Characteristics of studies awaiting classification). We identified no ongoing trials.

Risk of bias in included studies

Figure 2 presents a summary of the risk of bias assessments by risk of bias domain and by trial. In the following, we provide details only on the risk of bias in the five fully reported trials (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). Reflecting the lack of detailed information on methods and absence of results, we judged that Palmer 1998 was at unclear risk of bias for all domains.

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Random sequence generation

Four trials reported details of the randomisation sequence and used methods that we assessed as being at low risk of bias (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015). The methods used were automated telephone randomisation systems (Carson 1998; Carson 2011), a computer‐generated list by a person not affiliated with the project (Foss 2009), or a web‐based clinical trial support system (Gregersen 2015). As Parker 2013 did not report details of their generation of the randomisation sequence method, we assessed this as having an unclear risk of bias for this domain.

Concealment of treatment allocation

We deemed the method of randomisation (as described above) adequate (low risk of bias) to conceal treatment allocation in three trials (Carson 1998; Carson 2011; Gregersen 2015), with allocation being concealed using data co‐ordinating centres at central locations in Carson 1998 and Carson 2011, and use of a web‐based system in Gregersen 2015.

In two trials, we considered the methods used to conceal treatment allocation to be at unclear risk of bias (Foss 2009; Parker 2013). Neither trial used 'sequentially, numbered, opaque, sealed envelopes', which are believed to be the most robust way of concealing treatment allocation (Higgins 2011a): one trial used 'sealed envelopes' (Foss 2009); the other trial used 'numbered, opaque, sealed envelopes' (Parker 2013).

Blinding

We reported details of who was blinded to treatment allocation separately for participants, study personnel and outcome assessors.

Blinding of participants

Only Foss 2009 and Gregersen 2015 described the blinding of trial participants. We judged Foss 2009, Gregersen 2015, and Parker 2013 to be at low risk of bias as the participants were not involved in outcome assessment in any of these trials and so any participant knowledge of treatment allocation would have limited impact on the data measured by the study. There was no blinding in the other two trials, which we judged to be at high risk of bias as the participants were involved in assessing outcomes and knowledge of treatment allocation may have an influence on their outcome measurement (Carson 1998; Carson 2011). There was insufficient information reported in Gregersen 2015 to judge whether there was any blinding of participants to treatment allocation.

Blinding of study personnel

In these trials (all comparing a liberal with a restrictive haemoglobin concentration threshold for red blood cell transfusion), the blinding of clinicians would have been difficult as the clinicians themselves determined whether a participant met the requirements for a red blood cell transfusion. Thus, we classified five trials to be at high risk of performance bias relating to lack of blinding of study personnel (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013).

Blinding of outcome assessors

Five trials reported that there was blinding of outcome assessors to treatment allocation (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). In all these trials, the methods used to enact this blinding were described. In three trials, we deemed the methods used to be adequate (low risk of bias) as the manuscripts reported that outcome assessors were blinded to treatment allocation (Carson 1998; Carson 2011; Gregersen 2015).

In two trials, we deemed the methods used to blind outcome assessors to treatment allocation to be inadequate (high risk of bias) (Foss 2009; Parker 2013). In one trial, details of the participant's treatment allocation were not securely stored and could have been seen by outcome assessors (Foss 2009). In the other trial, the outcome assessor for one outcome (change in mobility score) was blinded to treatment allocation: while other outcome assessments were made by clinicians aware of treatment allocation, which may have influenced outcome measurement (Parker 2013).

Incomplete outcome data

We deemed four trials to have a low risk of attrition bias: two trials included all randomised participants in the analysis of outcome data and did not lose any participants during follow‐up (Carson 1998; Parker 2013), and two trials reported missing data for some outcome analyses but documented the level and reason of attrition per treatment group (Foss 2009; Gregersen 2015). Carson 2011 reported the numbers of participants not included in the 60‐day follow‐up and reasons why, but did not provide details of the reasons for attrition at the 30‐day follow‐up period and had 13% missing data for each group on one key outcome. Although this small variability may not have affected event rates (pooled or individual study), we downgraded the risk of bias to unclear for Carson 2011.

Selective reporting

In five trials, all pre‐specified outcomes were reported on in their results section and we deemed them to be at low risk of reporting bias (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013).

Other potential sources of bias

Other potential sources of bias (significant difference in protocol violations and baseline imbalance) were observed in three trials (Carson 2011; Foss 2009; Parker 2013). In Carson 2011, there was a statistical difference in the number of major protocol violations between the two transfusion threshold groups. In Foss 2009, there was baseline imbalance in the type of surgery received and the American Society of Anesthesiologists (ASA) classification between the liberal and liberal transfusion threshold groups, and, in Parker 2013, the baseline imbalance was in the proportion of participants presenting with cardiac disease between the interventions groups. We rated the impact that the protocol violations had as high in Carson 2011, and the impact of the baseline imbalances on the outcome measurements as unclear in Foss 2009.

Effects of interventions

The six trials investigated only one of the four comparisons outlined in the Types of interventions section.

Red blood cell transfusion threshold A (liberal) versus red blood cell transfusion threshold B (restrictive)

Five trials, randomising 2704 participants, presented outcome data on red blood cell transfusion threshold A (liberal) versus red blood cell transfusion threshold B (restrictive) (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). No outcome results were available for 18 participants of the sixth included study (Palmer 1998). Throughout, the denominator numbers we have used and reported were the number of participants included in each particular outcome analysis in the trial reports; this does not always correspond to the number of participants randomised into the respective trial groups.

One study randomised participants at hospital admission and participants' eligibility for the trial was assessed perioperatively (Foss 2009), while the other trials randomised and assessed eligibility post surgery (see Characteristics of included studies table). As there is a difference between assessing haemoglobin concentration (and thus study eligibility) perioperatively and postoperatively, we undertook a sensitivity analysis (by temporarily removing Foss 2009 from each meta‐analysis in which it was included) and if the pooled result of the sensitivity analysis did not differ from the overall pooled result, we have not reported the results of the sensitivity analysis in this section.

Carson 2011 was a large study, contributing 75% of the total number of participants randomised across the four trials that reported outcome data. We undertook sensitivity analyses for all outcomes that included Carson 2011 by removing these data from each pooled result. If the pooled result of the sensitivity analysis did not differ from the overall pooled result, we have not reported the results of the sensitivity analysis in this section.

Table 3 presents data on the number of participants receiving a red blood cell transfusion and on the quantity of red blood cells received in the two transfusion threshold groups. The percentage of participants receiving a red blood cell transfusion ranged from 74% to 100% in the liberal transfusion threshold group and from 11% to 45% in the restrictive transfusion threshold group. Gregersen 2015 did not report data on the percentage of participants receiving a red blood cell transfusion.

Table 3. Quantity of red blood cell units received (post randomisation)

Study ID	Liberal transfusion threshold		Restrictive transfusion threshold
Study ID	Number (%) of participants transfused	Quantity of red blood cell units transfused	Number (%) of participants transfused	Quantity of red blood cell units transfused
Carson 1998	41 (98)	Median number of red blood cell units transfused: 2, maximum of 4 (interquartile range 1 to 2)	19 (45)	Median number of red blood cell units transfused: 0, maximum of 6 (interquartile range 0 to 2) 45% (n = 19) participants received a red blood cell transfusion
Carson 2011	970 (97)	Median number of red blood cell units transfused: 2 (interquartile range 1 to 2) 96.6% (n = 970) participants received a red blood cell transfusion 3.3% (n = 33) participants received 0 units of red blood cells 41.9% (n = 420) participants received 1 unit of red blood cells 34.5% (n = 346) participants received 2 units of red blood cells 13.2% (n = 132) participants received 3 units of red blood cells 7.2% (n = 72) participants received ≥ 4 units of red blood cells	413 (41)	Median number of red blood cell units transfused: 0 (interquartile range 0 to 1) 41% (n = 413) participants received a red blood cell transfusion 59.0% (n = 594) participants received 0 units of red blood cells 24.4% (n = 246) participants received 1 unit of red blood cells 12.6% (n = 127) participants received 2 units of red blood cells 2.4% (n = 24) participants received 3 units of red blood cells 1.6% (n = 16) participants received ≥ 4 units of red blood cells
Foss 2009	44 (74)	Median number of red blood cell units transfused: 2 (interquartile range 1 to 2) 74% (n = 44) participants received a red blood cell transfusion	22 (37)	Median number of red blood cell units transfused: 1 (interquartile range 1 to 2) 37% (n = 22) participants received a red blood cell transfusion
Gregersen 2015	Not stated	Median number of red blood cell units per patient was 3.0 (interquartile range 2 to 5)	Not stated	Median number of red blood cell units per patient was 1.0 (interquartile range 1 to 2)
Palmer 1998	9 (100)**	Not reported	1 (11)**	Not reported
Parker 2013	100 (100)	A mean of 1.9 units of red blood cells transfused 16 participants received 1 unit of red blood cells 92 participants received 2 units of red blood cells 1 participant received 3 units of red blood cells 4 participants received 4 units of red blood cells	11 (11)	Participants received either 1 or 2 units of red blood cells: no further details reported

* 4 of these 42 participants received a transfusion in violation of the protocol (i.e. they did not have symptoms of anaemia or a haemoglobin of < 8 g/dL).

** Data taken from Table 25.2 (McClelland 2009).

Carson 1998 reported five protocol violations (6.2%): one participant in the liberal transfusion threshold group did not receive a transfusion and four in the restrictive transfusion threshold group received a transfusion but did not have symptoms of anaemia or haemoglobin less than 8 g/dL. Carson 2011 reported "major protocol violations" in 147 participants (7.3%): 91 participants (9.0%) in the liberal threshold group of whom 30 did not receive a transfusion and 61 were discharged with a haemoglobin level less than 10 g/dL; and 56 participants (5.6%) in the restricted threshold group who received transfusion despite not having symptoms or rapid bleeding. Gregersen 2015 reported eight deviations from protocol in each group, either not receiving or receiving transfusion outside the group threshold; and a further four drop‐outs in each group, seven of whom refused transfusion. There was no mention of protocol violations relating to transfusion in the other trials.

Primary outcomes

Mortality

All studies measured mortality as an outcome. Five trials reported mortality at 30 days (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). There was no evidence of a difference in mortality at 30 days post operation between the liberal and restrictive transfusion threshold groups (RR 0.92, 95% CI 0.67 to 1.26; I² = 38%; 2683 participants; event rate 68/1337 in the liberal transfusion threshold group vs. 75/1346 in the restrictive transfusion threshold group; Analysis 1.1).

Three studies also reported 60‐day mortality (Carson 1998; Carson 2011; Parker 2013). There was no evidence of a difference in mortality at 60 days between the liberal and restrictive transfusion threshold groups (RR 1.08, 95% CI 0.80 to 1.44; I² = 0%; 2283 participants; event rate 87/1140 in the liberal transfusion threshold group vs. 81/1143 in the restrictive transfusion threshold group; Analysis 1.2).

Two trials reported mortality at 90 days (Gregersen 2015; Parker 2013), and Parker 2013 reported 120‐day and 365‐day mortality. At all time points there was no evidence of a difference in mortality between the liberal and restrictive threshold groups, with an RR of 0.88 (95% CI 0.55 to 1.16) at 90 days (484 participants; event rate 40/240 in the liberal transfusion threshold group vs. 51/244 in the restrictive transfusion threshold group); an RR of 1.18 (95% CI 0.56 to 2.51) at 120 days (200 participants; event rate 13/100 in the liberal transfusion threshold group vs. 11/100 in the restrictive transfusion threshold group); and an RR of 1.04 (95% CI 0.65 to 1.65) at 365 days (200 participants; event rate 27/100 in the liberal transfusion threshold group vs. 26/100 in the restrictive transfusion threshold group; Analysis 1.3).

Our post‐hoc sensitivity analysis, which excluded Foss 2009, found similar results but heterogeneity was significantly reduced (I² = 0%). The second sensitivity analyses undertaken, which excluded Carson 2011, also found a similar lack of difference between the two groups. We have not presented these results in the review.

Mobility and functional recovery

All five studies reported on mobility and functional recovery. However, because the trials used a variety of measures to assess functional recovery, only limited meta‐analysis was possible.

Carson 2011 measured this outcome as the inability to walk 10 feet (3 m; or across a room) without human assistance at 30‐day follow‐up. There was little difference in this measurement of mobility between the liberal and restrictive transfusion threshold groups (RR 0.93, 95% CI 0.84 to 1.03; 1995 participants; event rate 407/995 in the liberal transfusion threshold group vs. 438/1000 in the restrictive transfusion threshold group; Analysis 1.4). Two trials measured the same outcome at 60‐day follow‐up (Carson 1998; Carson 2011). There was no difference in this measurement of mobility between the liberal and restrictive transfusion threshold groups (RR 1.00, 95% CI 0.87 to 1.15; I² = 26%; 2083 participants; event rate 292/1040 in the liberal transfusion threshold group vs. 292/1043 in the restrictive transfusion threshold group; Analysis 1.4).

Foss 2009 measured the number of participants who "regained functional independence during hospitalisation" and found very little difference between the liberal and restrictive transfusion threshold groups (RR 0.95, 95% CI 0.67 to 1.34; 107 participants; event rate 29/54 in the liberal transfusion threshold group vs. 30/53 in the restrictive transfusion threshold group; Analysis 1.5). Foss 2009 also reported CAS (CAS is a composite score evaluating independence in walking or getting up from the chair, with scores ranging from 0 to 18; higher scores indicate better mobility). Foss 2009 reported there was no statistically significant difference (reported P value = 0.46) between the liberal and restrictive transfusion threshold groups in the cumulative CAS values over three days (median CAS (interquartile range): 9 (9 to 15) in the liberal transfusion threshold group vs. 9 (9 to 13.5) in the restrictive transfusion threshold group).

Gregersen 2015 reported data using the New Mobility Score and the CAS and measured the independence/dependence of participants in transferring from bed to chair and their independence/dependence with regards to their walking ability. Gregersen 2015 reported there was no statistically significant difference (reported P value = 0.49) between the liberal and restrictive transfusion threshold groups in the New Mobility Score 10 days after hip fracture surgery (median (interquartile range): 1 (0 to 1) in the liberal transfusion threshold group vs. 1 (0 to 1) in the restrictive transfusion threshold group). Gregersen 2015 reported the numbers of participants who were able to walk, perform sit‐to‐stand‐to‐sit or were bedridden based on CAS categories at 10 days after hip fracture surgery. They found no evidence of a difference between the liberal and restrictive transfusion threshold groups in people unable to walk (RR 1.05, 95% CI 0.96 to 1.16; 284 participants; event rate 124/140 in the liberal transfusion threshold group vs. 121/144 in the restrictive transfusion threshold group) or people who were bedridden (RR 0.92, 95% CI 0.68 to 1.24; 284 participants; event rate 50/140 in the liberal transfusion threshold group vs. 56/144 in the restrictive transfusion threshold group) (Analysis 1.6).

Parker 2013 assessed mobility at eight weeks from discharge using a self developed but commonly used mobility score (where 0 represented a bed‐bound person and 9 represented full mobility indoors and outdoors without walking aids). They found very little difference between the liberal and restrictive transfusion threshold groups (MD 0.40, 95% CI ‐0.43 to 1.23; 106 participants; Analysis 1.7).

Postoperative morbidity

The four trials assessed postoperative morbidity using different events; we reported data for each event separately.

Four studies reported thromboembolism (inpatient) (Carson 1998; Carson 2011; Foss 2009; Parker 2013). There was little difference in the incidence of thromboembolism between the liberal and restrictive transfusion threshold groups (RR 1.15, 95% CI 0.56 to 2.37; I² = 0%; 2416 participants; event rate 15/1207 in the liberal transfusion threshold group vs. 13/1209 in the restrictive transfusion threshold group; Analysis 1.8).

Four studies reported stroke (inpatient) (Carson 1998; Carson 2011; Foss 2009; Parker 2013). There was little difference in the incidence of stroke between the liberal and restrictive transfusion threshold groups (RR 2.40, 95% CI 0.85 to 6.79; I² = 0%; 2416 participants; event rate 11/1207 in the liberal transfusion threshold group vs. 4/1209 in the restrictive transfusion threshold group; Analysis 1.9).

Three studies reported wound infection (both superficial and deep wound infections) (Carson 2011; Foss 2009; Parker 2013). There was little difference in the incidence of wound infection between the liberal and restrictive transfusion threshold groups (RR 1.61, 95% CI 0.77 to 3.35; I² = 31%; 2332 participants; event rate 18/1165 in the liberal transfusion threshold group vs. 11/1167 in the restrictive transfusion threshold group; Analysis 1.10).

Five studies reported on cardiovascular events (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013), but often using different terminology. We selected two types: myocardial infarction and congestive heart failure. The incidence of myocardial infarction was reduced in the liberal transfusion threshold group in comparison with the restrictive transfusion threshold group (RR 0.59, 95% CI 0.36 to 0.96; I² = 0%; three trials; 2217 participants; event rate 23/1107 in the liberal transfusion threshold group vs. 40/1110 in the restrictive transfusion threshold group; Analysis 1.11). This result was dominated by data from Carson 2011; where, although numbers were comparable in both groups, 13% (135 vs. 130) of participants had incomplete electrocardiographic results and in 18% (180 vs. 175), there was no blood sample for troponin testing. A sensitivity analysis removing Carson 2011 resulted in little difference between the two groups in the incidence of myocardial infarction for the remaining two trials (0/102 in the liberal transfusion group vs. 2/102 in the restrictive transfusion group; RR 0.33, 95% CI 0.04 to 3.15; I² = 0%; 204 participants).
There was little difference in the number of new‐onset congestive heart failure events between the two transfusion threshold groups (RR 0.77, 95% CI 0.48 to 1.23; I² = 0%; three trials; 2332 participants; event rates: +29/1165 in the liberal transfusion threshold group vs. 38/1167 in the restrictive transfusion threshold group; Analysis 1.11). Once again, the findings of Carson 2011 dominated these results.

Five studies reported respiratory infections (Carson 1998; Carson 2011; Foss 2009; Gregersen 2015; Parker 2013). Pooled data showed a slightly increased incidence of respiratory infection (namely pneumonia) in the liberal transfusion threshold group compared with the restrictive transfusion threshold group (RR 1.35, 95% CI 0.95 to 1.92; I² = 0%; 2416 participants; event rate 69/1207 in the liberal transfusion threshold group vs. 51/1209 in the restrictive transfusion threshold group; Analysis 1.12).

Aside from myocardial infarction, our other sensitivity analyses excluding Carson 2011 did not show important differences in effect. The same applied for sensitivity analyses in which Foss 2009 was excluded for all five postoperative morbidity events.

Gregersen 2015 reported only on complications as the main cause of death of the 70 participants who had died during 90 days post operation (Table 4). Although these are incomplete data in terms of complications for the whole trial population, we observed that the greater incidence of stroke and lower incidence of heart failure in the liberal transfusion group compared with the restrictive threshold group were consistent with the patterns found in Analysis 1.9 for stroke and Analysis 1.11 for heart failure, whereas the opposite was the case for pneumonia (Analysis 1.12).

Table 4. Main cause of death (Gregersen 2014)

Cause of death	Liberal threshold	Restrictive threshold
	30 deaths	40 deaths
Stroke	8 (27%)	2 (5%)
Heart failure	3 (10%)	11 (28%)
Pneumonia	8 (27%)	18 (45%)
Sepsis	3 (10%)	5 (12%)
Dementia	3 (10%)	4 (10%)
Liver failure	5 (17%)	0 (0%)

Percentages do not add up to 100 in the liberal threshold column because of rounding errors

Secondary outcomes

Postoperative or discharge haemoglobin

Postoperative haemoglobin level data, split by treatment group, at days one, two, four and seven post operation were presented graphically in Carson 2011; at days one, two, three and seven post operation in Foss 2009; and at days three, 10, 17, 24 and 30 post operation in Gregersen 2015. Exact figures (means and SDs) were not available for these trials. It was clear that the haemoglobin levels were higher in the liberal transfusion threshold groups in all three trials. Carson 2011 reported the mean haemoglobin level before transfusion was 1.3 g/dL higher in the liberal transfusion threshold group.

Quality of life

Carson 2011, Foss 2009, and Gregersen 2015 reported QoL data using very different outcome measures. Foss 2009 provided no data for pooling.

Carson 2011 used three measurement scales to assess QoL at 30 and 60 days post randomisation: the Lower‐extremity Physical Activities of Daily Living scale, the Instrumental Activities of Daily Living scale and the Functional Assessment of Chronic Illness Therapy‐Fatigue (FACIT‐Fatigue) scale. Appendix 2 presents details of the components of each scale and how the scale is scored. There was a substantial drop in the numbers of participants contributing these data. There was minimal or no difference in the mean change in score between the liberal and restrictive transfusion threshold groups for any scale at any measurement time point: Lower‐extremity Physical Activities of Daily Living scale at 30 days post randomisation (MD ‐0.10, 95% CI ‐0.60 to 0.40; 979 participants; Analysis 1.13) and at 60 days post randomisation (MD 0.00, 95% CI ‐0.51 to 0.51; 1076 participants; Analysis 1.13); the Instrumental Activities of Daily Living scale at 30 days post randomisation (MD 0.00, 95% CI ‐0.06 to 0.06; 887 participants; Analysis 1.14) and at 60 days post randomisation (MD 0.00, 95% CI ‐0.12 to 0.12; 800 participants; Analysis 1.14); and the FACIT‐Fatigue scale at 30 days post randomisation (MD 0.10, 95% CI ‐0.89 to 1.09; 915 participants; Analysis 1.15) and at 60 days post randomisation (MD ‐0.50, 95% CI ‐1.38 to 0.38; 1069 participants; Analysis 1.15).

Foss 2009 (107 participants) assessed QoL by recording ambulation scores on days one to three post operation and symptoms of anaemia (fatigue and dizziness) during physiotherapy (on days one to three post operation) and reported the data as medians and interquartile ranges. Table 5 presents full details. These indicated that the only measure with a statistically significant difference (i.e. P value < 0.05) between the two transfusion threshold groups was the fatigue score at day two post operation whereby the liberal transfusion threshold group reported lower levels of fatigue than the restrictive transfusion threshold group (reported P value = 0.04).

Table 5. Quality of life outcome data (Foss 2009)

Quality of life dimension¹	Liberal transfusion threshold (n = 54)	Restrictive transfusion threshold (n = 53)	P value (as reported by study)
Ambulation score (0‐6)²
Day 1	3 (3 to 4)	3 (2.25 to 3.75)	0.75
Day 2	3 (3 to 5)	3 (3 to 4.75)	0.35
Day 3	3 (3 to 6)	3 (3 to 5)	0.67
Fatigue (0‐4)³
Day 1	1 (1 to 2)	1 (1 to 2)	0.37
Day 2	1 (1 to 2)	2 (1 to 3)	0.04
Day 3	1 (1 to 2)	2 (1 to 2)	0.11
Cumulated fatigue score (0‐12)⁴	4 (3 to 6)	5 (3 to 7)	0.46
Dizziness score (0‐4)³
Day 1	0 (0 to 1)	0 (0 to 1)	0.94
Day 2	0 (0 to 1)	0 (0 to 1)	0.48
Day 3	0 (0 to 1)	0 (0 to 1)	0.82
Cumulated dizziness score (0‐12)⁴	0 (0 to 1)	0 (0 to 1)	0.64

¹ Values reported as medians, with 25 to 75 percentiles.
² Ambulation score was a composite score evaluating independence in walking or getting up from a chair; higher scores = better mobility.
³ Via a verbal rating scale with scores (0‐4); higher scores = worse outcome (4 = very severe).
⁴ Cumulated scores were the cumulated values from days 1‐3.

Gregersen 2015 used the Modified Barthel Index to assess the range of independence or dependence (substantial or complete) a person had in their activities of daily living 10 days after hip fracture surgery. There was no evidence of a difference between the liberal and restrictive transfusion threshold groups in people who were dependent (RR 1.05, 95% CI 0.92 to 1.20; 284 participants; event rate 108/140 in the liberal transfusion threshold group vs. 106/144 in the restrictive transfusion threshold group) or completely dependent (RR 0.94, 95% CI 0.69 to 1.27; 284 participants; event rate 50/140 in the liberal transfusion threshold group vs. 55/144 in the restrictive transfusion threshold group) (Analysis 1.16).

Length of stay in hospital

Although four studies provided length of hospital stay data, we decided not to pool the four trials because length of stay in hospital is usually protocol‐based and can vary between countries.

Two studies (1304 participants) reported length of stay data from participants hospitalised in the USA (Carson 1998; Carson 2011). There was little evidence of a difference in length of hospital stay between the liberal and restrictive transfusion threshold groups (MD ‐0.27 days, 95% CI ‐0.66 to 0.12; I² = 0%; Analysis 1.17). Carson 2011 also reported length of stay data from 791 participants hospitalised in Canada and found little evidence of a difference in the mean length of stay between the liberal and restrictive transfusion threshold groups (MD ‐0.67 days, 95% CI ‐1.98 to 0.64). Likewise, there was little evidence of a difference in the mean length of hospital stay between the liberal and restrictive transfusion threshold groups in Foss 2009 based in Denmark (MD 1.80 days, 95% CI ‐3.23 to 6.83; 107 participants) or Parker 2013 based in the UK (MD ‐1.50 days, 95% CI ‐7.81 to 4.81; 200 participants).

Length of hospital stay was reported to be similar in the two transfusion groups of Gregersen 2015, but dependent more on whether the participants were from nursing homes (median two days length of hospital stay in both groups) or sheltered housing median 10 vs. 11 days; reported P value = 0.35).

Adverse effects of transfusion

Only Gregersen 2015 (284 participants) and Parker 2013 (200 participants) commented on adverse effects of transfusion. Gregersen 2015 reported that no complications were observed during or after transfusion and Parker 2013 reported that no participant had an adverse reaction to transfusion.

Discussion

Summary of main results

The objective of this systematic review was to assess the effects (benefits and harms) of red blood cell transfusion in people undergoing surgery for hip fracture. All six included trials compared 'liberal' versus 'restricted' red blood cell transfusion thresholds. Hence, we found no trials comparing red blood cell transfusion with no transfusion, red blood cell transfusion versus alternative methods or different protocols for red blood cell transfusion administration.

summary of findings Table for the main comparison shows the main findings of the review. With the exception of myocardial infarction, we judged the quality of the evidence for each outcome to be low, which means that we consider that further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate for all outcomes presented in this table. We judged the evidence for myocardial infarction to be very low indicating greater uncertainty regarding these results.

The low quality evidence from the data pooled from five trials (2683 participants) provided no evidence of a difference between the two transfusion thresholds in mortality at 30 days' follow‐up. Assuming an illustrative baseline risk at 30 days of 50 per 1000 people in the restricted threshold group, the result was compatible with 17 fewer deaths and 13 more deaths in the liberal threshold group. A similar conclusion applied for the low quality evidence on mortality at 60 days from three trials (2283 participants). The low quality evidence from the data pooled from the two trials (2083 participants) reporting inability to walk 10 feet (3 m; or across a room) without human assistance at 60 days showed little difference between liberal versus restricted threshold transfusion at 60 days' follow‐up. Assuming an illustrative baseline risk at 60 days of 283 per 1000 people unable to walk 10 feet unsupported in the restricted threshold group, this result was compatible with 37 fewer and 43 more people with this outcome at 60 days in the liberal threshold group.

summary of findings Table for the main comparison presents postoperative morbidity for the following complications: thromboembolism (four trials; 2416 participants); stroke (four trials; 2416 participants); wound infection (three trials; 2332 participants); cardiovascular events, measured as myocardial infarction (three trials; 2217 participants) and new‐onset congestive heart failure (three trials; 2332 participants); respiratory infection, measured as pneumonia (four trials; 2416 participants) and risk of developing congestive heart failure (three trials; 2332 participants). In all, we pooled low quality evidence and we found no evidence of a difference between the two transfusion thresholds in all of these postoperative morbidities except myocardial infarction. We found some low quality evidence for a reduction in the incidence of myocardial infarction in the liberal transfusion threshold group in comparison with the restrictive transfusion threshold group. Assuming an illustrative baseline risk of myocardial infarction of 24 per 1000 people in the restricted transfusion threshold group, this result was compatible with between 1 and 15 fewer myocardial infarctions in the liberal transfusion threshold group.

Overall completeness and applicability of evidence

This review only identified trials that compared red blood cell transfusion thresholds (liberal vs. restrictive). One of these trials, which randomised 18 participants, did not report any outcome data; therefore, the quantitative findings of the review are based on data from five trials. In the five trials,1349 participants were randomised to the liberal transfusion threshold group and 1355 participants were randomised to the restrictive transfusion threshold group. The mean age of participants was similar across all the trials, ranging from 81 to 87 years and far fewer men than women were included in the trials: the percentage of men in each trial ranged from 16% in Parker 2013 to 25% in Gregersen 2015. In both these characteristics, these trials are comparable with the typical epidemiology of people with a hip fracture.

The timing of randomisation varied, with one trial randomising people preoperatively at admission to hospital (Foss 2009), one trial randomising participants after surgery (Gregersen 2015), and four trials randomising participants when their haemoglobin concentration dropped postoperatively (between one and three days post surgery) (Carson 1998; Carson 2011; Palmer 1998; Parker 2013). Thus, the majority of evidence applied to postoperative transfusion.

A key factor that should be acknowledged when interpreting the findings of this review is the dominance of the results by those of one relatively large study (Carson 2011), which randomised 2016 participants. Our post‐hoc sensitivity analyses exploring the impact this trial on the pooled effect estimates for each outcome found the only outcome affected was postoperative morbidity cardiovascular events; without Carson 2011, there was no longer a difference in the risk of a myocardial infarction between the two threshold groups (only two events occurred in the two other trials with this outcome).

There was heterogeneity in how the trials defined restrictive and liberal thresholds, while, there was some consistency in how the trials defined a liberal threshold for transfusion: receipt of 1 unit of packed red blood cells at the time of random assignment and as much blood as necessary to maintain the haemoglobin concentration greater than 10 g/dL (Carson 1998; Carson 2011; Palmer 1998; Parker 2013), when the haemoglobin concentration was less than 10 g/dL (Foss 2009), or receipt of 1 or 2 units of red blood cells when the haemoglobin threshold was at or below 11.3 g/dL within the first three weeks following surgery (Gregersen 2015). There was less consistency across the restrictive transfusion threshold groups. The restrictive transfusion threshold was receipt of a red blood cell transfusion if participants showed symptoms of anaemia or if their haemoglobin dropped to less than 8 g/dL in Carson 1998 and Carson 2011; when the haemoglobin concentration was 8 g/dL or less (with transfusion not based on symptoms or presence of clinical anaemia) in Foss 2009; when the haemoglobin threshold was at 9.7 g/dL or less within the first three weeks following surgery in Gregersen 2015; when perceived necessary by the physicians in Palmer 1998; and when participants were symptomatic of anaemia in Parker 2013. While such heterogeneity reflects variability in clinical practice, it introduced a limitation when we pooled outcome data across these studies. We did not explore the impact that this heterogeneity brought to the pooled effect estimates and overall findings in this review, but it may be an important consideration in future updates. With the exception of Gregersen 2015, these restrictive thresholds are broadly consistent with recommendations in many guidelines for red blood cell transfusion; although practice is increasingly advocating lower thresholds (Goodnough 2014; Rohde 2014).

There were differences in participants' cardiac history between the trials, with the largest trial only including participants with cardiovascular or cerebrovascular history (or both) or cardiovascular risk factors (Carson 2011). Three trials did not list pre‐existing cardiac disease as an eligibility criterion (Carson 1998; Gregersen 2015; Parker 2013), and one trial excluded people with acute cardiac conditions (Foss 2009). This is important to consider when interpreting the findings of postoperative morbidity in terms of cardiovascular events.

The timing of the intervention differed between the trials. In Foss 2009, the study protocol required blood tests daily from admission to the fifth postoperative day with any indication for transfusion acted on immediately. In Gregersen 2015, participants were randomised after surgery. In the restrictive transfusion threshold group in Carson 1998 and in the other two trials (Carson 2011; Parker 2013), both the monitoring of haemoglobin concentration and transfusion of red blood cells started postoperatively. In the liberal transfusion threshold group in Carson 1998, participants received a red blood cell transfusion at the time of randomisation and as much additional blood as needed thereafter to keep the haemoglobin concentration above 10 g/dL. These differences in the timing of the intervention may be relevant if there are significant acute factors, particularly perioperative haemodynamic instability, which may relate to blood loss or to other common factors such as inadequate volume replacement, effects of medication and anaesthesia. A transfusion given in the immediate perioperative period (e.g. in theatre recovery) may be associated with more intensive surveillance and attention to other variables such as electrocardiograph monitoring, electrolytes or blood pressure monitoring compared with the clinical setting of a transfusion given up to four to five days later.

Quality of the evidence

The key methodological limitations of these studies that places these at risk of bias lie in the lack of blinding of study personnel, the baseline imbalance observed in two trials (Foss 2009 due to either type of surgery received or ASA classification and in Parker 2013 the proportion of participants presenting with cardiac disease) and the protocol violations of Carson 2011. Minimising these limitations in any further studies in this area would certainly improve the quality of the available trial evidence.

We assessed the quality of the evidence for each of our primary outcomes using the GRADE approach (GRADE 2011), see summary of findings Table for the main comparison. With the exception of myocardial infarction, all outcomes were graded as of low quality of evidence, meaning that we have limited confidence in the reliability of the outcome findings. For most outcomes, the grading of low quality was as a result of imprecision (typically small number of outcome events resulting in wide CI values) and concerns over risk of bias, in particular resulting from a statistically significant difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups in the largest trial (Carson 2011). There was no consistency between the trials in how functional recovery was measured by the trials. Although this outcome was measured by all of the five principal studies (2672 participants) in the review, the varied way in which it was measured prevented a pooling of outcome data for an outcome of critical importance to this participant population. While we acknowledge that the blinding of all involved in the study to treatment allocation is a difficulty with these interventions, the noted existence of performance bias for what is an often subjective outcome is a further limitation to the quality of the evidence for this outcome; which was represented by the inability to walk 10 feet (3 m; or across a room) unsupported in the summary of findings Table for the main comparison. We downgraded the evidence for myocardial infarction a further level to very low quality because of the additional risk of attrition bias from incomplete electrocardiographic results in 13% of the trial population and lack of blood samples for troponin tests in 18% of the trial population of Carson 2011.

Potential biases in the review process

The strengths of this review lie in the robust and comprehensive methodology employed to find and assess all relevant trials. We have followed standard Cochrane Collaboration methods for data extraction and analysis with reference to the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011c), and have referred to a statistician where necessary. We have had a clinician and methodologist working on all stages, independently of each other to control any bias that ensues due to clinical or systematic review methodology knowledge. When we were limited by a lack of reporting data to allow inclusion and assessment of trials, we contacted authors directly to obtain the necessary data.

In the protocol, we had planned the transfusion protocol and threshold comparisons to be restrictive versus liberal. We changed this order to bring it in line with the ordering of the comparisons used in the included studies and to minimise the risk of transcription errors. Given that both thresholds have been considered the control intervention in the literature, with variation in current practice, other authors may have chosen the reverse order as presented in our protocol. These changes were made during the preparation of data for analysis and have been documented in the Differences between protocol and review. We do not consider this change to have biased the findings of our review.

Agreements and disagreements with other studies or reviews

Evidence for the safety and efficacy of red blood cell transfusion thresholds has been published across a variety of surgical settings (Carson 2012; Carson 2013; Hajjar 2010; Shokoohi 2012a; So‐Osman 2013; Villanueva 2013; Walsh 2013).

One Cochrane meta‐analysis of 19 randomised controlled trials compared clinical outcomes of higher versus lower red blood cell transfusion triggers in over 6000 people and found no evidence for benefit of liberal red blood cell transfusion policies (Carson 2012). Two trials reported evidence of harm with liberal red blood cell policies, with postoperative infection incidence being higher in the liberal transfusion group in comparison with the restrictive group in elective orthopaedic knee and hip replacement surgery (So‐Osman 2013); and rate of further bleeding, mortality at 45 days and overall rate of complications being lower in the restrictive group in a trial in people with acute upper gastrointestinal bleeding (Villanueva 2013). Two pilot feasibility trials have enrolled people with symptomatic coronary artery disease (Carson 2013), and older mechanically ventilated critically ill people (Walsh 2013), but neither were sufficiently powered to demonstrate a difference in clinical outcomes. One retrospective audit of practice in an elderly population (919 people followed, of which 313 received a red blood cell transfusion) undergoing surgery for hip fracture found that receiving a red blood cell transfusion was not associated with changes in mortality, but, unlike this review, reported an association with an increased rate of postoperative infection in the transfused group (Shokoohi 2012a).

Figure 1

Study flow diagram

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Analysis 1.1

Comparison 1 Liberal versus restrictive threshold, Outcome 1 30‐day mortality.

Analysis 1.2

Comparison 1 Liberal versus restrictive threshold, Outcome 2 60‐day mortality.

Analysis 1.3

Comparison 1 Liberal versus restrictive threshold, Outcome 3 Longer‐term mortality: at 90, 120 and 365 days.

Analysis 1.4

Comparison 1 Liberal versus restrictive threshold, Outcome 4 Inability to walk 10 feet (3 m; or across a room) without human assistance.

Analysis 1.5

Comparison 1 Liberal versus restrictive threshold, Outcome 5 Regaining functional independence during hospitalisation.

Analysis 1.6

Comparison 1 Liberal versus restrictive threshold, Outcome 6 Poor mobility/physical ability (at 10 days).

Analysis 1.7

Comparison 1 Liberal versus restrictive threshold, Outcome 7 Mean change in mobility scores (score 0 to 9: best mobility).

Analysis 1.8

Comparison 1 Liberal versus restrictive threshold, Outcome 8 Thromboembolism (in hospital).

Analysis 1.9

Comparison 1 Liberal versus restrictive threshold, Outcome 9 Stroke (inpatient).

Analysis 1.10

Comparison 1 Liberal versus restrictive threshold, Outcome 10 Wound infection (in hospital).

Analysis 1.11

Comparison 1 Liberal versus restrictive threshold, Outcome 11 Cardiovascular events.

Analysis 1.12

Comparison 1 Liberal versus restrictive threshold, Outcome 12 Respiratory infections (namely pneumonia).

Analysis 1.13

Comparison 1 Liberal versus restrictive threshold, Outcome 13 Lower Extremity Physical Activities of Daily Living (ADL) (0 to 11: total dependency).

Analysis 1.14

Comparison 1 Liberal versus restrictive threshold, Outcome 14 Instrumental Activities of Daily Living (ADL) (higher scores = higher dependency).

Analysis 1.15

Comparison 1 Liberal versus restrictive threshold, Outcome 15 FACIT‐Fatigue score (higher scores = more energy).

Analysis 1.16

Comparison 1 Liberal versus restrictive threshold, Outcome 16 Dependency (based on Modified Barthel Index) at 10 days.

Analysis 1.17

Comparison 1 Liberal versus restrictive threshold, Outcome 17 Length of stay in hospital (days).

Summary of findings for the main comparison. Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery

Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery
Patient or population: people undergoing hip fracture surgery¹ Settings: hospital Intervention: liberal threshold red blood cell transfusion² Comparison: restrictive threshold red blood cell transfusion³
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Restrictive threshold	Liberal threshold
30‐day mortality Follow‐up: mean 30 days	50 per 1000⁴	46 per 1000 (33 to 63)	RR 0.92 (0.67 to 1.26)	2683 (5 studies)	⊕⊕⊝⊝ low^5,6	‐
Inability to walk 10 feet (3 m; or across a room) without human assistance Follow‐up: mean 60 days	283 per 1000⁴	283 per 1000 (246 to 326)	RR 1.00 (0.87 to 1.15)	2083 (2 studies)	⊕⊕⊝⊝ low^7,8	‐
Thromboembolism (in hospital)	20 per 1000⁴	23 per 1000 (11 to 47)	RR 1.15 (0.56 to 2.37)	2416 (4 studies)	⊕⊕⊝⊝ low^6,9	‐
Stroke (in hospital)	2 per 1000⁴	5 per 1000 (2 to 14)	RR 2.4 (0.85 to 6.79)	2416 (4 studies)	⊕⊕⊝⊝ low^6,9	‐
Wound infection (in hospital)	8 per 1000⁴	13 per 1000 (6 to 27)	RR 1.61 (0.77 to 3.35)	2332 (3 studies)	⊕⊕⊝⊝ low^6,10	‐
Cardiovascular events ‐ myocardial infarction	24 per 1000⁴	14 per 1000 (9 to 23)	RR 0.59 (0.36 to 0.96)	2217 (3 studies)	⊕⊝⊝⊝ very low^6,11	‐
Respiratory infections (namely pneumonia)	18 per 1000⁴	24 per 1000 (17 to 35)	RR 1.35 (0.95 to 1.92)	2416 (4 studies)	⊕⊕⊝⊝ low^6,12	‐
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
1. Although we included evidence for pre‐operative, peri‐operative and postoperative transfusion, the majority of the evidence applied to postoperative transfusion. 2. The liberal transfusion threshold was a haemoglobin concentration of about 10 g/dL in four trials, and 11.3 g/dL in one trial. 3. The restrictive transfusion threshold in four trials was a haemoglobin concentration of about 8 g/dL or symptoms of anaemia, and 9.7 g/dL in one trial. 4. The assumed risk was the median control risk across studies. 5. We downgraded the evidence one level because of risk of bias: 57% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 6. We downgraded the evidence one level because of imprecision: generally because of the small number of events in the studies reporting data for this outcome has resulted in wide confidence intervals for these studies. 7. We downgraded the evidence one level because of risk of bias: 96% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 8. We further downgraded the evidence one level because of risk of bias: both participants and study personnel "were aware of study group assignment after randomisation" in the two studies reporting data for this subjective outcome. Given that the participants themselves are involved in assessing this outcome, knowledge of treatment allocation may influence outcome measurement. 9. We downgraded the evidence one level because of risk of bias: 60% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 10. We downgraded the evidence one level because of risk of bias: 70% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups. 11. We downgraded the evidence two levels because of risk of bias: 94% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups; and, although numbers were comparable between the two transfusion threshold groups in this trial, overall 265 (13%) participants had incomplete electrocardiographic results and in 355 (18%) participants there was no blood sample for troponin testing. Thus, attrition bias was a potential problem. 12. We downgraded the evidence one level because of bias: 93% of the weighting for this outcome came from one trial in which there was a statistical difference (P value = 0.003) in the number of major protocol violations post randomisation between the two transfusion threshold groups.

Summary of findings for the main comparison. Liberal versus restrictive threshold transfusion for people undergoing hip fracture surgery

Table 1. Type of surgical procedure and type of fracture

Study ID	Type of surgical procedure	Type of hip fracture
Carson 1998	Multiple screws/plates (n = 59) Hemiarthroplasty (n = 12) Bipolar hemiarthroplasty (n = 13)	Femoral neck (n = 30)* Intertrochanteric (n = 55) Subtrochanteroc (n = 6)
Carson 2011	Not stated	Femoral neck (n = 854)^# Intertrochanteric (n = 1034) Subtrochanteroc (n = 183) Reverse oblique (n = 21)
Foss 2009	Screws/pins (n = 10) Arthroplasty (n = 46) Sliding hip screw (n = 46) Intermedullary hip screw (n = 18)	Not stated
Gregersen 2015	Internal fixation (n =221) Arthroplasty (n = 57) Other (n = 6)	Not stated
Palmer 1998	Not stated	Not stated
Parker 2013	Not stated Percutaneous screw fixation was an exclusion criterion	Not described in full Intracapsular (n = 68)
* Seven participants (five in the liberal and two in the restrictive transfusion threshold groups) in the trial had more than one type of hip fracture. ^# Several participants had more than one type of hip fracture: there was an excess of 41 fractures in the liberal group and 39 in the restricted group (data not available for four participants).

Table 1. Type of surgical procedure and type of fracture

Table 2. Baseline characteristics of randomised participants

Study ID	Liberal transfusion threshold	Restrictive transfusion threshold
Carson 1998	Number of participants reporting these data: 42 Age: mean (SD): 81.3 (8.1) years; range: 50‐94 years Gender: number (%) men: 9 (21.4) Number (%) people withcardiac conditions: Any: 19 (45.2) Coronary artery disease: 12 (28.6) Congestive heart failure: 6 (14.3) Transfusion history: Number of red blood cell transfusions received before randomisation, mean (SD): 0.5 (1.0) Last preoperative haemoglobin concentration, mean (SD): 11.7 (1.6) g/dL Randomisation haemoglobin concentration, mean (SD): 9.1 (0.6) g/dL	Number of participants reporting these data: 42 Age: mean (SD): 83.3 (10.8) years; range: 32‐95 years Gender: number (%) men: 11 (26.2) Number (%) people withcardiac conditions: Any: 19 (45.2) Coronary artery disease: 12 (28.6) Congestive heart failure: 6 (14.3) Transfusion history: Number of red blood cell transfusions received before randomisation, mean (SD): 0.3 (0.6) Last preoperative haemoglobin concentration, mean (SD): 11.6 (1.0) g/dL Randomisation haemoglobin concentration, mean (SD): 9.1 (0.6) g/dL
Carson 2011	Number of participants reporting these data: 1007 Age: mean (SD): 81.8 (8.8) years Gender: number (%) men: 250 (24.8) Number (%) people withcardiac conditions: Any: 637 (63.3) Coronary artery disease: 402 (39.9) Congestive heart failure: 184 (18.3) Hypertension: 824/1003 (82.2) Transfusion history: transfusions before randomisation: number/total number (%) of participants receiving 0 red blood cell transfusions: 754/1006 (75.0) number/total number (%) of participants receiving ≥ 1 red blood cell units: 252/1006 (25.0) Haemoglobin concentration before surgery, mean (SD): 11.3 (1.5) g/dL Haemoglobin concentration during eligibility screening, mean (SD): 9.0 (0.8) g/dL	Number of participants reporting these data: 1009 Age: mean (SD): 81.5 (9.0) years Gender: number (%) men: 239 (23.7) Number (%) people withcardiac conditions: Any: 631 (62.5) Coronary artery disease: 403 (39.9) Congestive heart failure: 167 (16.6) Hypertension: 821/1005 (81.7) Transfusion history: transfusions before randomisation: number/total number (%) of participants receiving 0 red blood cell transfusions: 720/1008 (71.4) number/total number (%) of participants receiving ≥ 1 red blood cell units: 288/1008 (28.6) Haemoglobin concentration before surgery, mean (SD): 11.3 (1.5) g/dL Haemoglobin concentration during eligibility screening, mean (SD): 9.0 (0.8) g/dL
Foss 2009	Number of participants reporting these data: 60 Age: mean (SD): 81 (6.8) years Gender: number (%) men: 14 (23) Number (%) people withcardiac conditions: Hypertension: 14 (23) Atrial fibrillation: 3 (5) Congestive heart failure: 5 (8) Ischemic heart disease: 4 (7) Cardiovascular disease: 21 (35) Transfusion history: not stated Haemoglobin concentration on admission: reported graphically and not all the data were readily interpretable, but mean was clearly 13.0 g/dL	Number of participants reporting these data: 60 Age: mean (SD): 81 (7.3) years Gender: number (%) men: 14 (23) Number (%) people withcardiac conditions: Hypertension: 20 (33) Atrial fibrillation: 3 (5) Congestive heart failure: 3 (5) Ischemic heart disease: 10 (17) Cardiovascular disease: 28 (47) Transfusion history: not stated Haemoglobin concentration on admission: reported graphically and not all the data were readily interpretable, but mean was between 13.0 and 14.0 g/dL
Gregersen 2015	Number of participants reporting these data: 140 Age: mean (SD): 88 (6.9) years Gender: number (%) men: 34 (24) Number (%) people withcardiac conditions: Cardiovascular disease: 25 (18) Transfusion history: not stated Number (%) of people with anaemia at baseline: 68 (49) Postoperative haemoglobin levels between 9.7 g/dL and 11.3 g/dL during the first 6 postoperative days Repeated measurements of haemoglobin concentration levels showed maintained mean haemoglobin concentration levels of 12.2 g/dL (95% CI 12.2 to 12.3)	Number of participants reporting these data: 144 Age: mean (SD): 86 (6.8) years Gender: number (%) men: 36 (25) Number (%) people withcardiac conditions: Cardiovascular disease: 34 (24) Transfusion history: not stated Number (%) of people with anaemia at baseline: 70 (49) Postoperative haemoglobin levels between 9.7 g/dL and 11.3 g/dL during the first 6 postoperative days Repeated measurements of haemoglobin concentration levels showed maintained mean haemoglobin concentration levels of 11.3 g/dL (95% CI 11.3 to 11.4)
Palmer 1998	No data reported	No data reported
Parker 2013	Number of participants reporting these data: 100 Age: mean (range): 84.4 years (60‐104) Gender: number (%) men: 17 (17) Number (%) of people withcardiac conditions: Any cardiac disease: 37 (37) Hypertension: 42 (42) Angina: 7 (7) Previous myocardial infarction: 4 (4) Previous congestive cardiac failure: 5 (5) Other cardiac disease: 21 (21) Transfusion history: not stated Haemoglobin concentration (mean) on admission: 11.5 g/dL Haemoglobin concentration (mean) after surgery (time point post surgery not defined): 8.7 g/dL	Number of participants reporting these data: 100 Age: mean (range): 84.2 years (60‐97) Gender: number (%) men: 15 (15) Number (%) of people with cardiac conditions: Any cardiac disease: 50 (50) Hypertension: 46 (46) Angina: 9 (9) Previous myocardial infarction: 9 (9) Previous congestive cardiac failure: 3 (3) Other cardiac disease: 29 (29) Transfusion history: not stated Haemoglobin concentration (mean) on admission: 11.8 g/dL Haemoglobin concentration (mean) after surgery (time point post surgery not defined): 8.9 g/dL
SD: standard deviation.

Table 2. Baseline characteristics of randomised participants

Table 3. Quantity of red blood cell units received (post randomisation)

Study ID	Liberal transfusion threshold		Restrictive transfusion threshold
Study ID	Number (%) of participants transfused	Quantity of red blood cell units transfused	Number (%) of participants transfused	Quantity of red blood cell units transfused
Carson 1998	41 (98)	Median number of red blood cell units transfused: 2, maximum of 4 (interquartile range 1 to 2)	19 (45)	Median number of red blood cell units transfused: 0, maximum of 6 (interquartile range 0 to 2) 45% (n = 19) participants received a red blood cell transfusion
Carson 2011	970 (97)	Median number of red blood cell units transfused: 2 (interquartile range 1 to 2) 96.6% (n = 970) participants received a red blood cell transfusion 3.3% (n = 33) participants received 0 units of red blood cells 41.9% (n = 420) participants received 1 unit of red blood cells 34.5% (n = 346) participants received 2 units of red blood cells 13.2% (n = 132) participants received 3 units of red blood cells 7.2% (n = 72) participants received ≥ 4 units of red blood cells	413 (41)	Median number of red blood cell units transfused: 0 (interquartile range 0 to 1) 41% (n = 413) participants received a red blood cell transfusion 59.0% (n = 594) participants received 0 units of red blood cells 24.4% (n = 246) participants received 1 unit of red blood cells 12.6% (n = 127) participants received 2 units of red blood cells 2.4% (n = 24) participants received 3 units of red blood cells 1.6% (n = 16) participants received ≥ 4 units of red blood cells
Foss 2009	44 (74)	Median number of red blood cell units transfused: 2 (interquartile range 1 to 2) 74% (n = 44) participants received a red blood cell transfusion	22 (37)	Median number of red blood cell units transfused: 1 (interquartile range 1 to 2) 37% (n = 22) participants received a red blood cell transfusion
Gregersen 2015	Not stated	Median number of red blood cell units per patient was 3.0 (interquartile range 2 to 5)	Not stated	Median number of red blood cell units per patient was 1.0 (interquartile range 1 to 2)
Palmer 1998	9 (100)**	Not reported	1 (11)**	Not reported
Parker 2013	100 (100)	A mean of 1.9 units of red blood cells transfused 16 participants received 1 unit of red blood cells 92 participants received 2 units of red blood cells 1 participant received 3 units of red blood cells 4 participants received 4 units of red blood cells	11 (11)	Participants received either 1 or 2 units of red blood cells: no further details reported
* 4 of these 42 participants received a transfusion in violation of the protocol (i.e. they did not have symptoms of anaemia or a haemoglobin of < 8 g/dL). ** Data taken from Table 25.2 (McClelland 2009).

Table 3. Quantity of red blood cell units received (post randomisation)

Table 4. Main cause of death (Gregersen 2014)

Cause of death	Liberal threshold	Restrictive threshold
	30 deaths	40 deaths
Stroke	8 (27%)	2 (5%)
Heart failure	3 (10%)	11 (28%)
Pneumonia	8 (27%)	18 (45%)
Sepsis	3 (10%)	5 (12%)
Dementia	3 (10%)	4 (10%)
Liver failure	5 (17%)	0 (0%)
Percentages do not add up to 100 in the liberal threshold column because of rounding errors

Table 4. Main cause of death (Gregersen 2014)

Table 5. Quality of life outcome data (Foss 2009)

Quality of life dimension¹	Liberal transfusion threshold (n = 54)	Restrictive transfusion threshold (n = 53)	P value (as reported by study)
Ambulation score (0‐6)²
Day 1	3 (3 to 4)	3 (2.25 to 3.75)	0.75
Day 2	3 (3 to 5)	3 (3 to 4.75)	0.35
Day 3	3 (3 to 6)	3 (3 to 5)	0.67
Fatigue (0‐4)³
Day 1	1 (1 to 2)	1 (1 to 2)	0.37
Day 2	1 (1 to 2)	2 (1 to 3)	0.04
Day 3	1 (1 to 2)	2 (1 to 2)	0.11
Cumulated fatigue score (0‐12)⁴	4 (3 to 6)	5 (3 to 7)	0.46
Dizziness score (0‐4)³
Day 1	0 (0 to 1)	0 (0 to 1)	0.94
Day 2	0 (0 to 1)	0 (0 to 1)	0.48
Day 3	0 (0 to 1)	0 (0 to 1)	0.82
Cumulated dizziness score (0‐12)⁴	0 (0 to 1)	0 (0 to 1)	0.64
¹ Values reported as medians, with 25 to 75 percentiles. ² Ambulation score was a composite score evaluating independence in walking or getting up from a chair; higher scores = better mobility. ³ Via a verbal rating scale with scores (0‐4); higher scores = worse outcome (4 = very severe). ⁴ Cumulated scores were the cumulated values from days 1‐3.

Table 5. Quality of life outcome data (Foss 2009)

Comparison 1. Liberal versus restrictive threshold

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 30‐day mortality Show forest plot	5	2683	Risk Ratio (M‐H, Fixed, 95% CI)	0.92 [0.67, 1.26]

2 60‐day mortality Show forest plot	3	2283	Risk Ratio (M‐H, Fixed, 95% CI)	1.08 [0.80, 1.44]

3 Longer‐term mortality: at 90, 120 and 365 days Show forest plot	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.1 90‐day mortality	2	484	Risk Ratio (M‐H, Fixed, 95% CI)	0.80 [0.55, 1.16]
3.2 120‐day mortality	1	200	Risk Ratio (M‐H, Fixed, 95% CI)	1.18 [0.56, 2.51]
3.3 365‐day mortality	1	200	Risk Ratio (M‐H, Fixed, 95% CI)	1.04 [0.65, 1.65]
4 Inability to walk 10 feet (3 m; or across a room) without human assistance Show forest plot	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

4.1 At 30 days	1	1995	Risk Ratio (M‐H, Fixed, 95% CI)	0.93 [0.84, 1.03]
4.2 At 60 days	2	2083	Risk Ratio (M‐H, Fixed, 95% CI)	1.00 [0.87, 1.15]
5 Regaining functional independence during hospitalisation Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6 Poor mobility/physical ability (at 10 days) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

6.1 Unable to walk (CAS)	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
6.2 Bedridden (CAS)	1		Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7 Mean change in mobility scores (score 0 to 9: best mobility) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

8 Thromboembolism (in hospital) Show forest plot	4	2416	Risk Ratio (M‐H, Fixed, 95% CI)	1.15 [0.56, 2.37]

9 Stroke (inpatient) Show forest plot	4	2416	Risk Ratio (M‐H, Fixed, 95% CI)	2.40 [0.85, 6.79]

10 Wound infection (in hospital) Show forest plot	3	2332	Risk Ratio (M‐H, Fixed, 95% CI)	1.61 [0.77, 3.35]

11 Cardiovascular events Show forest plot	4		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

11.1 Myocardial infarction	3	2217	Risk Ratio (M‐H, Fixed, 95% CI)	0.59 [0.36, 0.96]
11.2 Congestive heart failure (new diagnosis)	3	2332	Risk Ratio (M‐H, Fixed, 95% CI)	0.77 [0.48, 1.23]
12 Respiratory infections (namely pneumonia) Show forest plot	4	2416	Risk Ratio (M‐H, Fixed, 95% CI)	1.35 [0.95, 1.92]

13 Lower Extremity Physical Activities of Daily Living (ADL) (0 to 11: total dependency) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

13.1 30 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
13.2 60 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
14 Instrumental Activities of Daily Living (ADL) (higher scores = higher dependency) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

14.1 30 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
14.2 60 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
15 FACIT‐Fatigue score (higher scores = more energy) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

15.1 30 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
15.2 60 days	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
16 Dependency (based on Modified Barthel Index) at 10 days Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

16.1 Substantially or completely dependent	1	284	Risk Ratio (M‐H, Fixed, 95% CI)	1.05 [0.92, 1.20]
16.2 Completely dependent	1	284	Risk Ratio (M‐H, Fixed, 95% CI)	0.94 [0.69, 1.27]
17 Length of stay in hospital (days) Show forest plot	4		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

17.1 USA	2	1304	Mean Difference (IV, Fixed, 95% CI)	‐0.27 [‐0.66, 0.12]
17.2 Canada	1	791	Mean Difference (IV, Fixed, 95% CI)	‐0.67 [‐1.98, 0.64]
17.3 Denmark	1	107	Mean Difference (IV, Fixed, 95% CI)	1.80 [‐3.23, 6.83]
17.4 UK	1	200	Mean Difference (IV, Fixed, 95% CI)	‐1.50 [‐7.81, 4.81]

Comparison 1. Liberal versus restrictive threshold