Prolonged Blood Storage and Risk of Posttransfusion Acute Kidney Injury

We performed a planned substudy of the INFORM trial.^15,33  Because the original study did not assess kidney function, this substudy required the collection of additional laboratory data—creatinine and hemoglobin—from blood sample measurements before and after transfusion. INFORM recruited patients from six hospitals in four countries, all of which participated in our substudy: Hamilton General Hospital (Canada), Juravinski Hospital (Canada), St. Joseph’s Healthcare Center (Canada), Cleveland Clinic (United States), Flinders Medical Center (Australia), and Meir Medical Center (Israel). This substudy was approved by the institutional review board at each participating site.

Between April 2012 and October 2015, the INFORM trial enrolled 31,497 patients. They were randomized 1:2 to receive either the freshest available blood (experimental group) or the oldest available blood (control group), up to the maximum accepted shelf life of 42 days. No other aspect of patient management was controlled.

Adults were eligible for INFORM if they required an erythrocyte transfusion during their hospital admission (surgery was not required). The trial excluded patients expected to receive massive transfusions, patients who required uncrossmatched erythrocyte units or autologous erythrocyte units, patients with complex antibody profiles, and patients deemed to have a clinical indication for the freshest available blood. Among the randomized patients, 24,736 met all enrollment criteria and were included in the original INFORM analysis. No crossovers occurred during the study.

For this substudy, we additionally excluded patients who did not have both pretransfusion and posttransfusion serum creatinine measurements, had baseline creatinine greater than 4.0 mg/dl, had an estimated glomerular filtration rate of less than 30 ml/min based on the Modification of Diet in Renal Disease equation, were dependent on renal dialysis, had procedures on the urinary system, or received 10 erythrocyte units or more during their hospital admission. Patients with inadequate serum creatinine measurements or severely impaired baseline renal function were excluded because we would have been unable to calculate our primary outcome of postTransfusion AKI. Patients who received 10 erythrocyte units or more during their hospital admission were also excluded because the transfused erythrocyte units would likely span a large range of storage durations, which would blur the separation we were trying to achieve between patients who received fresh versus old blood.

Our primary exposure was transfusion with either the freshest available blood (short-term storage group) or the oldest available blood (long-term storage group). Our primary outcome was the development of posttransfusion AKI, defined using the creatinine component of the Kidney Disease: Improving Global Outcomes (KDIGO) criteria.³⁴  The baseline creatinine value was defined as the mean serum creatinine measurement in the week before transfusion. AKI was defined as an increase of at least 1.5 times the baseline creatinine or an increase of 0.3 mg/dl in any rolling 48-h window in the 7 days after transfusion.³⁴  Therefore, at minimum, patients needed to have a baseline serum creatinine measurement and another serum creatinine measurement within 7 days of transfusion to be included in our study. We chose to limit our timing to 7 posttransfusion days—rather than including serum creatinine measurements from the entire hospital stay—to remain consistent with the KDIGO criteria. Furthermore, we excluded patients with baseline serum creatinine values greater than 4.0 mg/dl or who were already on renal dialysis because their baseline renal function was so poor that they were already classified as having the most severe AKI under KDIGO guidelines, so we would be unable to calculate further injury using the KDIGO definitions. Since we did not have data regarding urine output, we did not use this parameter for an alternative definition of AKI. Nonetheless, serum creatinine is both sufficient and more common for AKI definition.

In the primary analysis, we tested the association between treatment group (short-term vs. long-term storage) and AKI using a log-binomial regression to report relative risk in this randomized trial. We assessed whether the treatment effect varied across levels of various baseline factors, including age, sex, patient blood type, and study site, by assessing interaction between treatment group and the baseline factor. Likelihood ratio tests were used to test for heterogeneity of the treatment effect.

For the secondary analysis, we analyzed blood storage duration as a continuous variable. The mean duration of erythrocyte storage was used for patients who received more than one unit. A log-binomial model was fitted with AKI as the outcome and the actual age of the erythrocytes (in days) as the exposure. This model was adjusted for hospital site and patient blood type to account for imbalance. The time interval for the erythrocyte transfusions to be included in the analysis was the time from the first erythrocyte transfusion to the earliest time at which patients were determined to meet the criteria for AKI or 7 days from the first transfusion (for subjects who did not meet the criteria for AKI).

Additionally, a sensitivity analysis was conducted using the age of the oldest erythrocyte unit (in days) given to each patient as the exposure. The model was adjusted for hospital site, patient blood type, and number of erythrocyte transfusions to account for imbalance. A Poisson regression model (with AKI as the outcome) with robust error variance was fitted for this analysis in lieu of the log-binomial model, which failed to converge. Poisson regression models are usually used for estimating relative rates for count data. This can be easily extended to binomial data, in which case the relative rates can be thought of as relative risks. However, Poisson regression tends to be conservative when applied to binomial data. This can be rectified by using robust error variances obtained using sandwich estimation. The time interval for erythrocyte transfusions to be considered for inclusion in the sensitivity analysis was the same as that for the secondary analysis.

For the primary analysis, balance of demographic and clinical characteristics between the randomized groups was assessed using absolute standardized difference. The absolute standardized difference is roughly the absolute difference between means or proportions divided by the pooled SD.^35,36  For the secondary and sensitivity analyses, balance was assessed using absolute Pearson correlation (for continuous and ordinal covariates) or absolute standardized difference (for nominal covariates), because the exposures (e.g., blood storage duration) were continuous.³⁷  The absolute standardized difference in this case was calculated by treating the nominal covariates as the stratifying groups and the treatment variables as the outcome. For both primary and secondary analyses, an absolute standardized difference or absolute Pearson correlation greater than 0.1 indicated imbalance between the two treatment groups, and covariate adjustment would be used to account for the imbalance.

Relative risk and 95% CI were obtained from all models. The confidence level for all tests was set at 95% (P < 0.05 was considered significant). This study had 4,777 patients in the short-term storage group (treatment) and 9,684 patients in the long-term storage group (lower). Using the observed control group incidence of 15.4%, we had 90% power to detect a relative risk of 0.86 or lower at the 0.05 significance level. The analyses were conducted using R version 4.0.2 (https://www.R-project.org/, accessed December 22, 2020).

A total of 24,736 patients from the INFORM study were considered for this substudy. After excluding patients who did not meet the inclusion criteria, 14,461 patients were left for analysis. Among them, 4,777 were randomized to receive the freshest available blood (short-term storage group) and 9,684 to receive the oldest available blood (long-term storage group; fig. 1). The average age of the patients was 67 yr, and most had blood types O and A. Before transfusion, the average serum creatinine was approximately 1.0 mg/dl, hemoglobin was approximately 8.0 g/dl, and approximately 7% of the patients in each group had a history of chronic kidney disease. Overall, the patients in the two treatment groups were well balanced at baseline for the primary analysis (all with an absolute standardized difference of less than 0.10; table 1).

For the secondary and sensitivity analyses, balance on potential confounding variables was assessed separately, because the exposures (mean age of transfused erythrocytes and age of the oldest erythrocyte unit) were continuous (table 2). Imbalance was detected for hospital site and patient blood type for both of these exposures. Imbalance of the number of erythrocytes transfused was also detected for the age of the oldest erythrocyte unit. These variables were thus adjusted for in the corresponding analyses.

The 14,461 patients included in our analysis received a total of 40,077 erythrocyte units: 13,202 in the short-term storage group and 26,875 in the long-term storage group (fig. 2). Patients in each group received a median of 2 erythrocyte units [interquartile range, 1 to 3]. For patients who received more than one erythrocyte unit, we determined both the mean age of the transfused units and the age of the oldest unit. The median of the mean age of erythrocyte units transfused per patient was 11 days [interquartile range, 8 to 15] in the short-term storage group and 23 days [interquartile range, 17 to 30] in the long-term storage group. The median of the oldest erythrocyte unit transfused per patient was 12 days [interquartile range, 9 to 18] in the short-term storage group and 26 days [interquartile range, 18 to 35] in the long-term storage group. The proportion of patients who received other transfusions—platelets, plasma, and cryoprecipitate—was similar in each group (table 3).

The criteria for AKI were met by 688 of 4,777 patients (14.4%) in the short-term storage group and 1,487 of 9,684 patients (15.4%) in the long-term storage group. We did not find an association between the treatment group and AKI in the primary analysis, with an estimated relative risk (95% CI) of 0.94 (0.86 to 1.02; P = 0.132). The estimated CI suggests that the risk of developing AKI in the short-term storage group ranged from 14% lower to 2% higher than in the long-term storage group. No significant interactions were found between treatment assignment and baseline factors such as age, sex, patient blood type, and study site (fig. 3).

For the secondary analysis, after adjusting for hospital site and patient blood type, the mean age of erythrocytes was not associated with AKI (P = 0.978), with an estimated relative risk (95% CI) of 1.00 (0.96 to 1.04) for a 10-day increase in the mean age of erythrocytes. Similarly, for the sensitivity analysis, after adjusting for hospital site, patient blood type, and number of erythrocyte transfusions, the age of the oldest erythrocyte unit was not associated with AKI (P = 0.885), with an estimated relative risk (95% CI) of 1.00 (0.96 to 1.04) for a 10-day increase in the age of the oldest erythrocyte unit.

Among hospitalized adults without severely impaired baseline renal function who received fewer than 10 erythrocyte transfusions, the duration of blood storage had no appreciable effect on the incidence of AKI. This finding was consistent when comparing patients who were randomized to receive either the freshest or oldest available blood and when comparing the actual age of the transfused blood cells. It is thus unlikely that erythrocyte storage influences renal function over the typical range of storage durations.

Previous studies investigated the potential renal consequences of transfusing older blood (appendix).^17–32  Most were observational and focused on particular subsets of patients, such as those having cardiac surgery^{19,22,25–28,32}  or needing intensive care.^24,30,31  Some of the older studies reported conflicting results, perhaps because of small sample sizes and the heterogeneous definitions of blood age and kidney injury. Notably, most of the studies with greater than 500 patients (and all three randomized trials) have shown that there is no significant association between blood storage duration and adverse renal outcomes.

Our trial was by far the largest assessing erythrocyte storage duration and kidney injury. The enrollment of a broad patient population, regardless of reason for transfusion or surgical status, allows our results to be generalized to most hospitalized patients. Additionally, with our pragmatic design, we achieved a 12-day separation between the two randomized treatment groups, which was better than the 10 days targeted by the original INFORM trial. With this large, broad patient sample and adequate storage-duration separation achieved between the treatment groups, we found no effect of blood storage duration on AKI in our primary analysis. Furthermore, there was no evidence of harm in our continuous analysis even at extreme durations, up to the 42-day storage limit allowed in most developed countries. We therefore conclude that blood storage duration has no clinically meaningful effect on kidney injury in patients who do not have severely impaired renal function at baseline, which is consistent with the underlying INFORM trial that clearly showed no difference in mortality. Given that blood is a limited and valuable resource, it is reassuring to find that the patients in our study could safely receive erythrocyte units of any age, according to the current storage practices. Based on these findings, the way we maintain our blood supply appears to be safe, and we see no reason to routinely discard or limit the use of older blood if it has not yet reached the storage limit of 42 days.

The most common indication for erythrocyte transfusion is anemia, which is inherently associated with AKI. In addition, patients with preexisting renal disease may be more susceptible to further kidney injury. Thus, it was important to assess for balance of anemia and baseline kidney function because these are potential confounders. It was reassuring that both absolute standardized difference analyses and absolute Pearson correlations showed that pretransfusion hemoglobin and creatinine were similar for patients in the primary, secondary, and sensitivity analyses. It is important to note that we excluded patients with baseline serum creatinine values greater than 4.0 mg/dl or who were already on renal dialysis because we were unable to calculate further kidney injury in these patients. Our study population, therefore, was made of patients who were considered “sick” enough to warrant measuring their serum creatinine, but not so sick that they had severely impaired renal function at baseline.

We included patients with all ABO blood types in our analysis, although most had type O or A, and we expected patients with type B and AB to be more likely to receive non–type-specific units. Patient blood types were comparably distributed between the treatment groups of the primary analysis, but we observed slight imbalances in the secondary and sensitivity analyses. Imbalances were unsurprising because while the primary analysis was based on randomization of patient allocation, the secondary and sensitivity analyses used the actual blood storage duration. We therefore adjusted for patient blood type in our secondary and sensitivity models. Notably, the treatment effect for our primary outcome was similar across all blood types.

Because this is a substudy of the INFORM trial, we were limited in the variables we could include in our analysis. The original INFORM trial did not systematically record detailed information on coexisting illnesses, reasons for transfusion, and concomitant interventions. We also did not have information on which patients were surgical versus nonsurgical, although the underlying pathophysiology is often different between those patient populations. We were therefore unable to adjust for those possible confounders or perform those subgroup analyses with our models. However, in a large randomized trial, it is highly unlikely that baseline characteristics and ancillary treatments differed systematically between the groups.

Because we were using a different outcome than the original INFORM study, we observed a significant percentage of missing outcome data. Specifically, 14,461 of 24,736 (58.5%) of the original INFORM patients remained available for analysis in this substudy, which invites the possibility that our patient population might have different characteristics than the INFORM study population. Most patients were excluded for missing serum creatinine measurements, so it seems very unlikely that there was a systematic bias in the missing data related to the patient’s allocation. Additionally, we maintained our 1:2 allocation ratio, which further highlights the unlikeliness of bias; however, this cannot be ruled out conclusively and should be noted as a limitation of the study.

Another limitation was the lack of blinding because regulatory agencies require all erythrocyte products to be labeled with the date of collection or expiration. However, lack of blinding seems unlikely to influence transfusion practice, because clinicians would not know blood age until the requested units arrived at the transfusion venue, and at that point, the blood age would be unlikely to influence a previous decision to transfuse.

In summary, there was no difference in the incidence of posttransfusion AKI in our study population of patients who received the freshest available blood compared to patients who received the oldest available blood. This result aligns with the standard blood bank inventory management practice of providing the oldest available blood for routine transfusions.

Dr. Adegboye thanks all of the members of her thesis committee, without whose clinical knowledge and research resources, this study would not have been completed. Dr. Adegboye also thanks the research staff, fellows, and assistants at the Department of Outcomes Research, Cleveland Clinic, Cleveland, Ohio. Their help was invaluable in acquiring, analyzing, and interpreting the data. Additionally, Dr. Adegboye thanks the participating researchers from the original Informing Fresh versus Old Red Cell Management (INFORM) trial in Canada, Israel, and Australia who collaborated and shared their data.

Supported by internal funding of the Department of Outcomes Research, Cleveland Clinic, Cleveland, Ohio.

Dr. Eikelboom has financial relationships with Astra-Zeneca (Cambridge, United Kingdom), Bayer (Leverkusen, Germany), Boehringer-Ingelheim (Ingelheim am Rhein, Germany), Bristol-Myer-Squibb (New York, New York), Daiichi-Sankyo (Chuo City, Tokyo, Japan), Eli-Lilly (Indianapolis, Indiana), Glaxo-Smith-Kline (Brentford, United Kingdom), Pfizer (New York, New York), Janssen (Beerse, Belgium), Sanofi-Aventis (Paris, France), and Servie (Porsgrunn, Norway). Dr. Heddle has financial relationships with Canadian Institute of Health Research (Ottawa, Ontario, Canada), Haemonetics (Braintree, Massachusetts), and CSL Behring (Ottawa, Ontario, Canada). The other authors declare no competing interests.