This meta-analysis was conducted in accordance with the PRISMA guidelines. Two authors (Zhou X and Xu Z) independently searched the PubMed, OVID, Web of Science and Cochrane databases of clinical trials for RCTs comparing transfusion of fresher versus older RBCs in critically ill adult patients admitted to the ICU from database inception to December 2017. A comprehensive search was conducted to avoid missing relevant studies, and references from other publications were reviewed. The medical subject headings (MeSH) term included: “erythrocyte transfusion”, “erythrocytes”, “blood component transfusion”, “blood transfusion”, “intensive care units”, “intensive care”, “critical illness”, “critical care”, “time factors”. Other non-MeSH terms included: “red cell”, “red blood cell”, “RBC”, “blood”, “retransfus*”, “transfuse*”, “infuse”, “blood preservation”, “age*”, “aging”, “fresh*”, “old*”, “new*”, “young*”, “store*”, “storage”, “storing”, “preserv*”, “critical care”, “critically ill”, “critical*”, “intensive*”. The detailed search strategy is shown in appendix 1. There was no language restriction in this meta-analysis, and all articles in other languages that provided English abstracts were also considered.

All records searched from the above-mentioned four databases were initially filtered through an automated function to exclude duplicates. Three authors (Wang Y, Sun L and Zhou W) independently screened the titles and abstracts of the remaining records by using a questionnaire, and any disagreements were resolved by consensus. Then, eligible articles were included in a full-text review, and we contacted the authors by email for the full text when the full text was unavailable online. The inclusion criteria included the following: prospective RCTs comparing transfusion of fresher RBCs versus older RBCs and critically ill adult patients (aged≥18 years) admitted to the ICU. The primary endpoint was short-term mortality, which included 28-day mortality, ICU mortality and in-hospital mortality. The secondary endpoints were the duration of ICU stay and hospital stay. Articles assessing at least one of the above-mentioned indicators were included in this study. Pediatric studies, animal studies and retrospective, observational, non-randomized or cohort studies were excluded. Studies lacking data on the endpoints and duplicate published studies were also excluded, and the most recent updated data were included in the final analysis.

Two authors (Zhou X and Liu X) independently extracted data from the included studies using a custom-made form. Data regarding the name of the first author, publication year, study population, study design, number of subjects, number of RBC units transfused per patient, acute physiology and chronic health evaluation (APACHE) II or III score, hemoglobin level before transfusion, duration of RBC storage, short-term mortality and duration of ICU stay and hospital stay were recorded. In studies in which various short-term mortality were reported, the longer follow-up short-term mortality was included for analysis.

The two above-mentioned independent authors (Zhou X and Liu X) evaluated the internal validity and risk of bias of the included studies for the primary outcome (short-term mortality) according to the Cochrane Collaboration methods13 using standardized criteria, which assigned a value of high, unclear, or low to the following six domains: adequate sequence generation, allocation concealment, blinding, incomplete outcome data, free of selective reporting and free of other bias. We defined in advance that studies with a high or unclear risk of bias in less than two domains would be considered to be high quality.14

This meta-analysis was performed using Stata/SE 11.0 (StataCorp, College Station, TX, USA). A pooled odds ratio (OR) with corresponding 95% confidence interval (CI) was calculated to assess the effect of storage age of transfused RBCs on mortality using the Mantel–Haenszel method. Data on duration of ICU stay and hospital stay were pooled by using the mean difference (MD) with the inverse variance method. Heterogeneity among included trials was evaluated by visual inspection of forest plots and quantified with inconsistency factor (I2) statistics using chi-squared test. Both fixed-effects and random-effects meta-analyses were conducted for all endpoints (supplementary Table 1). Given the potential clinical and statistical heterogeneity between the trials (in sample size, primary diagnosis and definition of fresher RBCs), we reported the random-effects pooled result as the main result. However, if one or two trials accounts for approximately 80% or more of the total weight in a fixed-effect meta-analysis, we reported the fixed-effects result as the main result.15 For continuous data reported as the median with the interquartile range (IQR), the mean was estimated using the method of Hozo et al.,16 and the standard deviation (SD) was calculated using the formula proposed by the Cochrane handbook as: SD=IQR/1.35.13 The duration of ICU stay and hospital stay were log transformed before being included in meta-analysis due to the skewed distribution. We used approximations to calculate the mean and standard deviation based on logarithmic transformation using method 1 in Higgins et al.,17 and then the difference in geometric means between the fresher RBCs and older RBCs group was exponentiated to obtain the ratio of geometric means on the un-logged scale to offer interpretable results.18 Subgroup analysis was also conducted for the primary endpoint based on the study design (multi-center or single-center) and the risk of bias of the studies included.

We performed a trial sequential analysis (TSA) to assess the risk of type I error and prevent the increased risk of random error from repeated significance testing.19 TSA was conducted using both fixed-effects and random-effects with a 5% risk of type I error and 80% power for mortality in all studies and high-quality studies with low risk of bias, and we reported the random-effects result as the main result. The required information size was also calculated based on a relative risk reduction of 15% in mortality.

Sensitivity analyses were performed by sequentially excluding one trial at a time to assess the robustness of pooled results for the primary endpoint. A two-tailed P-value <0.05 was considered statistically significant.

A total of 3858 records were searched from the above-mentioned four databases through a combination of a MeSH terms search strategy and a liberal terms search strategy; an additional 16 records were also searched from references from the other publications. Among the 3874 total records, 285 duplicate records were filtered; then, 3589 records were included for review of the title and abstract, and 3560 records were ineligible according to our inclusion criteria and were excluded from our study. Subsequently, the remaining 29 records were included for screening the full text. After careful review of the full text, a total of six articles met the eligibility criteria and were included in the final analysis.11,12,20–23 The reasons for excluding the 23 ineligible studies are listed in appendix 2. The PRISMA flow chart of this study is shown in Fig. 1.

Figure 1.

The PRISMA flow diagram of the study selection process.

(0.26MB).

Among these six studies,11,12,20–23 four studies11,12,20,21 were multi-center RCTs and the remaining two studies22,23 were single-center RCTs; they were all published after 2012, and the number of subjects ranged from 51 to 10578. All six studies reported data on the primary endpoint; however, only four studies reported data on the duration of ICU stay and hospital stay.11,12,21,23 Three studies clearly reported that the subjects in their study were critically ill patients admitted to the ICU.11,12,23 Although the subjects in the other two studies were trauma patients21 and mechanically ventilated patients,22 respectively, all patients in both studies were admitted to the ICU, and therefore, these patients were firmly considered critically ill patients and included in our analysis. However, the subjects in the study by Heddle et al.20 were general hospitalized patients; fortunately, the authors conducted a subgroup analysis for patients in the ICU, and we enrolled these patients in our final analysis. The average number of units of RBCs transfused per patient was 2–4 units and was similar in both groups of patients who received fresher RBCs or older RBCs. The mean or median storage age of transfused RBCs ranged from 4 to 13 days in patients who were allocated to receive fresher RBCs and 22–36 days in patients allocated to receive older RBCs. The detailed study characteristics for the individual trials are summarized in Table 1.

The six domains for assessing the quality and risk of bias for the primary outcome for individual studies are described in detail in Table 2. Regarding random sequence generation, all six trials (100%) had a low risk of bias. In addition, both allocation concealment and selective outcome reporting were assessed as low risk of bias in all trials (100%). Regarding blinding method, five trials11,12,21–23 had a low risk of bias, and one trial by Heddle et al.20 had a high risk of bias. Incomplete outcome data were assessed as unclear risk of bias in all trials (100%). Overall, five trials were judged as having a low risk of bias and were considered to be of high quality.11,12,21–23

There were six studies, containing 18240 patients in total, that included data on short-term mortality; the pooled results indicated that the transfusion of fresher RBCs was not associated with a reduction in short-term mortality compared with transfusion of older RBCs [random-effects OR=1.04, 95% CI: 0.96–1.13, P=0.312; I2=0.0%] (Fig. 2). The random-effects TSA-adjusted 95% CI was 0.90–1.21, the required information size was 7534 randomized patients and the boundaries for futility were crossed, which indicated that future trials are unlikely to show a 15% relative risk reduction in short-term mortality (Fig. 3).

Figure 2.

Forest plot of subgroup analysis for short-term mortality based on the study design. OR, odds ratio; RBC, red blood cell.

(0.26MB).

Figure 3.

Trial sequential analysis using random-effects for short-term mortality in all studies included. Short-term mortality of 16.4% in the control arm, diversity of 0.0%, a relative risk reduction of 15%, a risk of type I error of 5% and power of 80%; the required information size of 7534 patients is reached and the boundaries for futility are crossed. The trial sequential analysis adjusted 95% confidence interval for a relative risk of 1.04 is 0.90–1.21.

(0.2MB).

We conducted a stratified analysis based on the study design, and the results revealed a neutral benefit of transfusion of fresher RBCs compared with that of older RBCs on short-term mortality, regardless of whether the studies were multi-center (random-effects OR=1.04, 95% CI: 0.96–1.13, P=0.292; I2=0.0%; four trials; 18090 patients) or single-center (random-effects OR=1.16, 95% CI: 0.28–4.71, P=0.839; I2=56.7%; two trials; 150 patients) (Fig. 2). Furthermore, a similar result was found in the subgroup analysis for high-quality studies with a low risk of bias (random-effects OR=1.04, 95% CI: 0.94–1.16, P=0.445; I2=0.0%; five trials; 7622 patients) (Fig. 4). The TSA was also conducted in the subgroup analysis for high-quality studies with low risk of bias and the random-effects TSA-adjusted 95% CI was 0.89–1.22. Similarly, future trials are also unlikely to detect a 15% relative risk reduction in short-term mortality because the required information size of 4924 patients was reached and the boundaries for futility were crossed (Fig. 5).

Figure 4.

Forest plot of subgroup analysis for short-term mortality based on risk of bias of studies included. OR, odds ratio; RBC, red blood cell.

(0.26MB).

Figure 5.

Trial sequential analysis using random-effects for short-term mortality in five studies with low risk of bias. Short-term mortality of 22.9% in the control arm, diversity of 0.0%, a relative risk reduction of 15%, a risk of type I error of 5% and power of 80%. The required information size of 4924 patients is reached and the boundaries for futility are crossed. The trial sequential analysis adjusted 95% confidence interval for a relative risk of 1.04 is 0.89–1.22.

(0.21MB).

There were four studies (7550 patients) that reported data on the duration of ICU stay and the duration of hospital stay. As two trials11,12 account for more than 90% of the total weight in the fixed-effects meta-analysis, we reported the fixed-effects result as the main result. The pooled results showed that transfusion of fresher RBCs was not associated with a reduction in the mean log duration of ICU stay (mean log 0.01, 95% CI: −0.03 to 0.05, P=0.638; I2=81.5%), corresponding to 1.0% (95% CI: −3.0 to 5.1%) increase in the geometric mean, or the mean log duration of hospital stay (mean log 0.00, 95% CI: −0.04 to 0.04, P=0.957; I2=7.4%), corresponding to 0.0% (95% CI: −3.9 to 4.1%) increase in the geometric mean, compared with transfusion of older RBCs (supplementary Figure 1 and supplementary Figure 2).

As shown in supplementary Figure 3, sensitivity analysis suggested that the overall effect of storage age of transfused RBCs on short-term mortality was not changed after sequentially removing one study at a time.