Transfusion-related acute lung injury: incidence and risk factors

Since 2003, the FDA has documented that the leading cause of transfusion-related fatality has been transfusion-related acute lung injury (TRALI),¹  defined as acute lung injury (ALI)²  that develops during or within 6 hours after transfusion of one or more units of blood or blood components.^3,4  Included in this definition are cases of ALI after multiple transfusions, a well-known ALI risk factor.⁵  The condition has been under-reported since the first description in 1985 by Popovsky and Moore,⁶  and the overall incidence has been reported only by passive surveillance studies. Although risk factors have been identified in subgroups, such as critically ill patients⁷  and cardiac surgery patients,⁸  risk factors in recipients and in transfused blood products (eg, antibody, bioreactive substances, older RBC storage age⁹ ) have not been identified in the general population of transfused patients because of the lack of a large prospective, case-controlled study. The goal of this study was to prospectively determine incidence by an active surveillance system¹⁰  implemented at 2 large academic centers. During the course of this study, the American Association of Blood Banks recommended the reduction of transfusion of plasma and platelets from donors potentially harboring alloantibodies,^11,12  thus making it possible to measure any change in TRALI incidence that was concurrent with implementation of this recommendation. The other goal of this study was to determine transfusion and recipient risk factors by enrolling concurrent transfused controls without TRALI.

See supplemental Methods (available on the Blood Web site; see the Supplemental Materials link at the top of the online article).

Active surveillance of TRALI was conducted at 2 tertiary care medical centers: the University of California–San Francisco (UCSF), San Francisco, CA and the Mayo Clinic, Rochester, MN. Enrollment began on March 1, 2006; the case-control study ended on August 31, 2009 and surveillance ended on December 31, 2009. All RBC and platelet units transfused during the study period were leuko-reduced. All patients older than 6 months were prospectively evaluated in real time for hypoxemia (PaO₂/FiO₂ < 300 mmHg) within 12 hours after issue of any blood component from the blood bank, by continuous electronic surveillance of arterial blood gas (ABG) results, and blood bank records in the hospital laboratory information system (Oztech Systems).¹⁰  Given the 6-hour window for the acute onset of TRALI by definition, most cases would have an ABG result within 12 hours. Cases without FiO₂ in ABG reports would be missed. After receiving an electronic alert in real time, coordinators gathered and entered patient data into a Web-based database (QuesGen Systems) and sent an electronic summary, including history, laboratory data, chest radiographs, and radiologist reports of chest radiographs, to the expert panel for review, usually within 72 hours of the alert.

Reviewers were blinded to all information regarding transfused units, including component type. TRALI was diagnosed by concurrence of 2 expert physicians by predetermined criteria (Table 1). At least monthly, conference calls were conducted with site investigators and coordinators, and disputable cases were reviewed by the full panel of 4 experts by conference call.

Implementation of plasma from male donors (Mayo), and plasma from male and never pregnant female donors (UCSF) occurred in 2007 to 2008, whereas reduction of platelets from previously pregnant females (UCSF) and predominantly male donors (Mayo) was partially implemented in 2008. Recipient tracing of components from donors associated with TRALI cases was not performed.

Because the study was observational and testing was performed on residual clinical laboratory specimens, cases and controls were enrolled without written informed consent. The protocol was approved by the institutional review board at each institution.

TRALI was defined as new ALI that developed during or within 6 hours of transfusion of one or more units, not attributable to another ALI risk factor.⁴  The study was designed to detect cases of TRALI that, by definition (PaO_2/FiO₂ < 300 mmHg), had severe enough hypoxemia that the patient's physician on any service would probably obtain ABG measurement(s) for patient management. The study was not designed to detect mild cases¹³  that do not meet the PaO_2/FiO₂ less than 300-mmHg criterion for ALI. To include patients with the highest likelihood of TRALI, we excluded patients who had another major ALI risk factor (pneumonia, sepsis, aspiration, multiple fractures, and pancreatitis), except if that patient was stable, the risk factor did not appear to cause ALI, and the ALI was precipitated by transfusion. “Possible TRALI”³  cases with another major ALI risk factor were designated “transfused ALI” and not included in the case-control study.

Control patients were randomly selected transfusion recipients who had no pulmonary signs or symptoms within 12 hours after transfusion of the last unit. We stratified sampling by number of units transfused, because this was probably a strong risk factor⁵  and having too few controls with large numbers of units transfused would make assessment of other risk factors more difficult. Controls were stratified according to the number of units transfused to the recipient (regardless of component type): low (1-2 units), medium (3-9 units), and high volume (≥ 10 units) transfusions.

Case and control samples were performed in the same runs to reduce reagent lot variability effects.

Low resolution class I HLA-A, B, Cw and class II HLA-DRB1, DRB3/4/5, DQB1 typing was performed using the LABType SSO (One Lambda).

Screening for antibody to HLA class I (anti–HLA-class I) and antibody to HLA class II (anti–HLA-class II) was performed using LABScreen Mixed LSM12 (One Lambda). A normalized background (NBG) ratio more than 2.2 was considered a positive result and the positive samples were further tested for antibody specificity using the LABScreen Single-antigen Bead assay (One Lambda) to identify antibody specificity. A specific antibody was considered positive if the MFI was more than 300 mmHg and strong if the MFI was more than 2500 mmHg for class I, or more than 1500 mmHg for class II.¹⁴  HLA antibody strength was also assessed by the NBG ratio result of the screening test.¹⁵ 

Screening for anti–human neutrophil antigen (HNA) was performed using the granulocyte agglutination test (GAT) and granulocyte immunofluorescence test (GIFT), analyzed by a flow cytometer (EPICS XL/MCL Flow Cytometer Beckman Coulter). GIFT was interpreted as positive if cells from at least 1 panel donor were interpreted as positive. Quantitative determination of antibody strength by the GIFT was assessed (see supplemental Methods).

Anti-HNA specificity was determined by 2 methods: (1) by examining the reactivity pattern obtained from samples interpreted as positive in the GAT and/or GIFT (eg, for anti–HNA-3a); and (2) by the monoclonal antibody immobilization of neutrophil antigens assay, where monoclonal antibodies were used to capture glycoproteins expressing HNA-1a, HNA-1b, HNA-1c, HNA-2a, HNA-4a, and HNA-5a alloantigens.

Lysophosphatidylcholine concentrations were determined by liquid chromatography-tandem mass spectrometry. Cytokine levels, including vascular endothelial growth factor in plasma of patients, and donor units were determined using Luminex multiplex bead assays as described by the manufacturer (Bio-Rad). The levels of soluble CD40 ligand and platelet factor 4 in plasma samples were determined using standard plate-bound ELISA format assays. Methods were used as described by the manufacturer (soluble CD40 ligand kit from Bender MedSystems; platelet factor 4 kit from R&D Systems).

To determine neutrophil priming activity, samples were examined on a dual-laser flow cytometer (FACscan Flow Cytometer, BD Biosciences) and analyzed using FlowJo Version 9.2 software (TreeStar). To detect plasma-induced CD11b/CD66 up-regulation, donor neutrophils were incubated with test plasma and then stained for CD11b/CD66b: (1) plasma from the neutrophil donor incubated with his/her own neutrophils, negative control; (2) donor neutrophils incubated with 1nM fMLF/BSA, positive control; (3) donor neutrophils incubated with plasma test sample, showing lack of reactivity; and (4) donor neutrophils incubated with plasma test sample, showing significant up-regulation of CD11b/CD66b. Numbers in each quadrant designate percentage of positive cells based on negative control.

Steps to reduce the incidence of TRALI (TRALI mitigation) were implemented from 2007 to 2008. In 2006, neither center had started mitigation. In 2009, both sites had completed mitigation. Incidences in 2006 and 2009 were compared. To assess trend, monthly counts of TRALI cases were modeled by Poisson regression, controlling for the number of transfusions in each month.

TRALI occurs in patients rather than in units; thus, our analyses treated each patient as one observation. In keeping with the case-control design, the primary outcome was whether the patient was a TRALI case or a control. Regarding potential recipient risk factors, we analyzed measurements made in the recipient before, and not during, the transfusion period of 6 hours before pulmonary edema, to avoid measuring the effects of any TRALI prodrome.

We defined and evaluated a large number of potential risk factors. Recipient factors evaluated were potential or known risk factors for ALI. Transfusion factors evaluated were donor gender, donor pregnancy, component type, storage age of RBC, platelet and plasma units, patient and donor ABO compatibility, anti-HNA, anti-HLA, MICA antibody, lysoPC, leukocyte and platelet cytokines, and neutrophil priming activity. The quantity of a potential transfusion predictor (antibody or bioreactive substance) in a transfused unit was calculated by the predictor strength/concentration multiplied by the plasma volume estimated for each positive component. All candidate risk factors were evaluated by logistic regression. To permit estimation of the effect of number of involved units, the SAS SurveyLogistic procedure¹⁶  was used for initial analyses. Because initial results indicated that virtually all the risk of increasing number of transfused units could be explained by other variables in multivariate models, simpler stratified conditional logistic regression methods were subsequently used for the primary results. Multiple imputation methods¹⁷  were used to prevent deletion of patients when any one of their units had a missing measurement, to preserve more information and introduce less bias than alternative methods.

We used as predictors the numbers of RBC units received by each patient that were above certain unit age cutoffs. These can never be reduced by addition of fresher units, and they were well defined (zero) for those who did not receive any units above the specified unit age cutoff.

We hypothesized that the number of units transfused might be an important risk factor for TRALI. To prevent confounding with the number of units from producing spurious results for other risk factors, we controlled for the number of units received during or within 6 hours of TRALI (or the corresponding artificial TRALI time created for controls), which we called “involved units,” in all models used to initially evaluate potential risk factors. Thus, a patient characteristic associated with needing many units would not appear to be risky unless it also increased risk by some other mechanism. We also controlled for study site (UCSF or Mayo), as is usual practice for multisite studies. Multivariate models were built stepwise, selecting risk factors with small P values for addition and/or factors that no longer had small P values for deletion at each step, taking into account biologic plausibility.

We do not provide statistical comparisons of descriptive summaries (Tables 2 to 6) because (1) they do not take account of the stratified sampling, and (2) they would reverse the correct roles of dependent and independent variables. TRALI versus control is the dependent variable in risk factor analyses.

During the active surveillance period of 45 months, 463 207 units of blood and blood components were transfused at the 2 centers, and 89 TRALI cases were identified (Figure 1). Seventy cases received one or more high plasma volume blood products. Of the 89 TRALI cases, only one had a major ALI risk factor (sepsis); the septic patient was stable until transfusion precipitated pulmonary edema. Nine had minor ALI risk factors (4 receiving amiodarone, 3 disseminated intravascular coagulation, 1 after lung resection, 1 acute central nervous system injury/stroke). Of the 89 TRALI cases, only 40 (45%) were reported to the blood banks as a transfusion reaction. Five cases were identified after transfusion reaction reports only, as an ABG was ordered more than 12 hours after blood issue or a FiO₂ was not entered.

The annual TRALI incidence (March 1, 2006 to December 31, 2009) decreased from 2.57 (95% confidence interval [CI], 1.72-3.86) per 10 000 units transfused (23 cases/89 321 units) in 2006 to 0.81 (95% CI, 0.44-1.49) per 10 000 units transfused (10 cases/123 731 units) in 2009 (P = .002; Figure 2). Reductions from pre- to post-mitigation periods were larger at Mayo (P = .014), where TRALI incidence also decreased in patients who received plasma from only male donors, associated with concurrent patient risk factor mitigation.¹⁸  By trend analysis, there was an overall 35% reduction per year (95% CI, 21%-47%, P < .0001). In contrast, the decrease in incidence of transfused ALI cases (“possible TRALI”) was not statistically significant (P = .38, p for trend = 0.78).

Incidences of TRALI by transfused patient-days (number of unique patients transfused per 24 hours) in 2006 versus 2009 were 7.46 (95% CI, 4.99-11.19) versus 2.29 (95% CI, 1.26-4.22 per 10 000 patient-days, P = .002).

The demographic characteristics (March 1, 2006 to August 31, 2009) of the 89 cases and 164 controls are in Table 2. Descriptive data show more units were transfused to TRALI patients (Table 3). Contents in units transfused to TRALI and control patients are described in Tables 4 to 6. Components received by TRALI patients are in Table 7. Statistically significant univariate patient factors are in Table 8, statistically significant univariate transfusion factors are in Table 9, and statistically insignificant factors including RBC storage age are in Table 10. Univariate analyses of anti-HLA are in supplemental Table 8 and anti-HNA in supplemental Table 9. Cognate anti-HLA specificities are in supplemental Table 10. The transfusion and patient risk factors with independent predictive value by multivariate analysis are in Tables 11 and 12.

Initial estimates showed that increased number of transfusions were associated with increased risk, with OR of 4.5 (95% CI, 2.4-8.4, P < .0001) for 3 to 9 versus 1 to 2 units, and an additional OR of 1.32 (95% CI, 1.23-1.42, P < .0001) per unit beyond 9. To determine whether the risk of greater number of units could be explained by transfusion and patient risk factors we identified, we controlled for these factors in the primary multivariate model (Table 12) and found that the risk of even 10 or more units became small and was no longer statistically significant at OR of 1.05 (95% CI, 0.91-1.20) per unit more than 9 (P = .53).

We found evidence against longer storage of leuko-reduced RBC units being an important risk (Table 10). In multivariate models, estimated ORs for older versus fresher RBC units were in the protective directions for all age cut-offs (ORs, 0.80-0.92).

Plasma (including plasma in whole blood) from female donors has been a focus of TRALI risk, and we found this to be a strong risk factor in multivariate analysis controlling for patient risk factors only (OR = 4.5, 95% CI, 1.85-11.2, P = .001; Table 11). Replacing plasma from female donors with antibody risk factors in the multivariate model, we found that 2 antibody measures were strong risk factors: quantity of cognate anti–HLA-class II (MFI > 1500) and volume of units anti-HNA positive by GIFT (Table 12). When female donor plasma and the 2 antibodies were in the same model, female donor plasma risk was smaller and no longer statistically significant (OR = 2.4, 95% CI, 0.87-6.9, P = .09).

Concerning the importance of cognate HLA antibody, total quantity of cognate anti–HLA-class II (MFI > 1500) was associated with risk in the multivariate analysis (Table 12). However, for anti–HLA-class II (MFI > 1500), total quantity, strongest MFI in any unit, and largest quantity in any unit were all collinear and highly predictive (OR = 3.2-3.5, P = .0035-.0052), and which is most important cannot be reliably determined from our data. In contrast, when controlled for cognate anti–HLA-class II (MFI > 1500), no substantial risk was associated with the quantity of noncognate anti–HLA-class II (MFI > 1500, OR = 0.41, 95% CI, 0.10-1.59, P = .19).

Regarding the importance of class I cognate HLA antibody, no substantial risk was associated with the presence of any cognate anti–HLA-class I (MFI > 2500, OR = 1.12, 95% CI, 0.20-6.20, P = .90), when controlled for cognate anti–HLA-class II (MFI > 1500).

With regard to the relative importance of strong (MFI > 1500) versus weak (MFI 300-1500) antibody, no substantial risk was associated with the presence of any weak cognate anti–HLA-class II (MFI 300-1500, OR = 0.32, 95% CI, 0.09-1.20, P = .09), when controlled for cognate anti–HLA-class II (MFI > 1500).

Regarding cognate HLA antibody specificities (supplemental Table 10), no single strong cognate HLA class I or II specificity was statistically significant when added to the primary multivariate model (Table 12) as an any versus no dichotomous predictor, although many were too rare to be evaluated individually.

Among study units, 84% (1712 of 2030) were tested for HLA antibody by the screening test. To facilitate applying the findings to clinical practice, we investigated what level of antibody (NBG ratio) found by the anti-HLA donor screening test would pose a significantly increased risk for transfusion recipients. After testing multiple assay thresholds (in models that did not include the cognate HLA class II variable), we found that a larger volume of strong anti–HLA-class II (NBG ratio > 27.5) increased risk (OR = 1.92 per 100 mL, 95% CI, 1.08-3.4, P = .03), a threshold equivalent to greater than 5 SDs in a cohort of nontransfused males.¹⁵  However, similar to the specific anti-HLA antibodies, the volume of weak anti–HLA-class II positive on the screening test (NBG ratio, 2.2-27.5) was not associated with risk (OR = 0.81 per 100 mL, 95% CI, 0.34-1.93, P = .63), and the volume of anti–HLA-class I did not appear to increase risk, even for strong antibody (NBG ratio > 59.3, OR = 0.98 per 100 mL, 95% CI, 0.46-2.1, P = .97).

Among study units, 78% (1574 of 2030) were tested for anti-HNA by GAT and GIFT. Patients were not tested for HNA type, except for the 2 recipients of anti–HNA-3a. By multivariate analysis, plasma volume of all GIFT-positive units received was a predictor of risk (OR = 1.71 per 100 mL, 95% CI, 1.18-2.5, P = .004; Table 12). GAT was not a statistically significant predictor when controlled for GIFT. Anti-HNA with defined specificity was rare (Table 4), including anti–HNA-3a. Of the 1574 tested units, only 2 (0.1%) contained anti–HNA-3a: 1 in an RBC and 1 in a plasma unit. A TRALI patient received this RBC unit and other units that contained strong cognate anti–HLA-class II; a control patient received the plasma unit. Both recipient genotypes were homozygous for HNA-3a (courtesy of Dr Brian Curtis, Blood Center Wisconsin). GIFT strength could not be quantified (see supplemental Discussion).

Among the 89 TRALI patients, 52% received one or both antibody predictors (10% received cognate anti–HLA-class II with MFI > 1500, 12% received anti-HNA positive by GIFT, and 30% received both).

Bioreactive substances were significant by univariate analysis (Table 9), but when added to the primary multivariate model (Table 12), the total amounts of MICA antibody and candidate bioreactive molecules were not statistically significant. For bioreactive substances, half of the units were not returned, and missing data required multiple imputation (“Statistical methods”).

On univariate analysis, we identified patient risk factors that increased the risk of TRALI or were protective (Table 8). However, on multivariate analysis. the protective factors dropped out of the model, leaving independent patient risk factors of shock, liver surgery (mainly transplantation), chronic alcohol abuse, positive fluid balance, peak airway pressure greater than 30 cm H₂O if mechanically ventilated before transfusion, and current smoking (Tables 11 and 12). The patient's plasma IL-8 level measured before transfusion was also a predictor of risk by multivariate analysis.

The major findings of this first study of TRALI by active surveillance of the general transfusion population can be summarized as follows. First, implementation of a predominantly male plasma supply was associated with reduced incidence of TRALI determined by active surveillance, thus supporting the effectiveness of this approach. Second, 2 predictive transfusion risk factors identified were the quantities of strong cognate anti–HLA-class II and volumes of anti-HNA in the blood units. No substantial risk was found for noncognate or weak cognate anti–HLA-class II, or anti–HLA-class I. Third, the data support using a cut-off of NBG ratio more than 27.5 for screening of anti–HLA-class II in female apheresis platelet donors. Fourth, the combination of transfusion and patient risk factors appeared to explain most of the risk of developing TRALI known to be associated with multiple transfusions.⁵  Fifth, we found evidence against longer RBC storage age being an important risk factor.⁹ 

We found that receipt of plasma (including whole blood) from female donors is a strong risk factor, and reduction of this risk factor was concurrent with a decrease in TRALI incidence from approximately 1:4000 units to approximately 1:12 000 units. The premitigation incidence of approximately1:4000 units we found in 2006 was close to the approximately1:5000 units found by a careful study in 1985 at the Mayo Clinic where a transfusion team performed and monitored transfusions,⁶  but 10-fold higher than the 2005 premitigation incidence of approximately1:40 000 units distributed found by passive surveillance (26.3 cases in 10⁶ units).¹⁹  In our study, other factors may have contributed to the decrease in incidence (eg, institutional improvements in critical care delivery that reduced patient risk factors we found that appear to render patients susceptible to TRALI, such as improving treatment of septic shock and decreasing high peak airway pressure > 30 cm H₂0 while being mechanically ventilated).¹⁸  Decreases in TRALI after conversion to plasma from predominantly male donors have been reported by passive surveillance studies from the United Kingdom,²⁰  the FDA,¹  and the American Red Cross.²¹  No similar decrease in “possible TRALI” incidence after implementation of the TRALI mitigation strategy suggests that plasma from alloimmunized donors is not a strong risk factor in “possible TRALI.”

Although units transfused to TRALI and control patients were similar in terms of percent positive for antibody (Table 4), TRALI patients received more units of every component (Table 3) and thus probably received larger quantities/volumes of antibody. Multivariate analysis confirmed that antibody strength and quantity/volume are important (Table 12).

Regarding the relative importance of class I versus class II HLA antibody, we found class II to be more important. The association of anti–HLA-class II with TRALI was first reported by Kopko et al.²²  Case series have reported that cognate anti–HLA-class II was the most frequent antibody implicated in TRALI.^23,24  This predominance occurs despite the fact that frequencies of class I (10%) and class II (12%) antibodies are comparable in female donors.²⁵  The greater importance of anti–HLA-class II appears to also be true in renal transplantation.²⁶ 

Regarding anti–HLA-class I, we found evidence against anti–HLA-class I being an important risk, even for strong cognate antibody with MFI more than 2500 (OR = 1.12, 95% CI, 0.20-6.20, P = .90). Others have reported similar results⁷  and similar conclusions.²⁷  Although the estimated risk for class I was very small and not statistically significant, we cannot be certain that there is absolutely no risk, as the upper confidence bound of 6.20 does not preclude risk. There are reports that cognate anti–HLA-class I can be associated with TRALI.^23,24,28  However, retrospective studies have found that this is rare.^29,30 

Regarding cognate versus noncognate HLA antibody, we found evidence against substantial risk from noncognate antibody. This finding argues against the hypothesis that noncognate HLA antibody is a marker for blood donors who make other antibodies that increase TRALI risk.

Volume of anti-HNA positive by GIFT was a strong predictor (Table 12). The GIFT detects antibodies to HNA, antibodies to unknown HNA, and anti–HLA-class I. Because we found that anti–HLA-class I did not appear to produce substantial risk, the risk detected by the GIFT is probably the result of anti-HNA. Because specific anti-HNA was rare, we could not examine the contribution of anti-HNA specificities. There may be unidentified HNA antigens that may be important.

The 2 antibody predictors in Table 12 are applicable to all component types. For example, for the same cognate anti–HLA-class II with MFI more than 1500, the estimated risk is 3.2 times greater in 200 mL of plasma (plasma, apheresis platelets) than in 20 mL of plasma (RBCs, cryoprecipitate) because the OR is 3.2 per 10-fold increase in amount transfused. For anti-HNA positive by GIFT, the estimated risk is 2.6 times greater for 200 mL than for 20 mL of plasma because the OR is 1.71 per 100-mL increase.

Whether longer RBC storage is associated with increased risk for lung injury and mortality is considered the most critical issue currently facing transfusion medicine.⁹  In this prospective case-control study, we found evidence against longer storage of leuko-reduced RBC units being an important risk for TRALI.

The risks associated with bioreactive substances were not statistically significant in the multivariate model (Table 12), similar to the results in a study of cardiac surgery patients.⁸  This result was surprising, given many pioneering basic studies by Silliman et al indicating that bioreactive substances are important in the development of TRALI.^31-34  However, the CIs of the odds ratios in our results were wide, and the upper bounds would correspond to substantial risk. Thus, we cannot conclude that these substances do not carry risk. Future clinical studies should test for these important potential mediators in the setting of ongoing TRALI mitigation.

Why do some patients develop TRALI and others who receive blood from the same donor do not? We found that cognate antibody matters, and patients who developed TRALI may have received cognate antibody and others did not. However, cognate versus noncognate antibody is not the only reason, as several look-back studies have found patients who received cognate antibody but did not develop TRALI.^29,30,35  We found 4 additional factors that influence why some patients develop TRALI and others do not: first, the volume of antibody transfused, second, the HLA class of the cognate antibody, third, the strength of anti–HLA-class II, and fourth, the presence or absence of patient factors that increase the risk and lowers the threshold³⁶ for TRALI.

Approximately half of the TRALI patients received cognate anti–HLA-class II (MFI > 1500) or anti-HNA. Some remaining cases may have received transfusion factors for which we found evidence against substantial risk, but the upper bounds of the CIs still included the possibility of some risk. Some cases may have received other transfusion factors we did not study (supplemental Discussion). In some cases with patient risk factors, it may be possible that transfusion was coincidental.

The diverse patient-associated risk factors we found for TRALI in this study are consistent with the known underlying comorbidities that predispose and lower the threshold for ALI, thus supporting the validity of this study. Shock results in tissue injury,³⁷  perhaps predisposing to TRALI through priming of the recipient's neutrophils. Chronic alcohol abuse increases risk, probably because of reduced levels of the antioxidant glutathione in the lung,³⁸  reduced phagocytosis of apoptotic cells, and the resulting enhanced pulmonary inflammatory response.³⁹  Patients with intravascular volume overload are more likely to manifest clinical pulmonary edema when there is ALI.⁴⁰  Previous studies have documented the risk for developing ALI with peak airway pressure greater than 30 cm H₂0⁴¹  and current smoking.⁴² 

We found higher IL-8 in patients before transfusion of involved units increased risk. Higher IL-8, a marker of inflammation and increased mortality risk,⁴³  may prime neutrophils and the lung endothelium. Acute contemporaneous events that increase inflammation and tissue injury could be “first hits” as first suggested by Silliman et al.³¹  Experimental models of TRALI have shown that host inflammation may be necessary to produce ALI before challenge with cognate antibody.^32,44  Inflammation (first hit) may up-regulate expression of HLA class II antigens on activated classic antigen-presenting cells (monocytes,⁴⁵ macrophages, dendritic cells), activated neutrophils,⁴⁶  and activated lung endothelial cells,⁴⁷  and exposure to large quantities of strong cognate anti–HLA-class II may then lead to ALI (second hit).⁴⁸ 

Liver surgery (transplantation) was a patient risk identified, even when controlled for severe liver disease, volume of transfusions, and alcohol abuse.^49,50  Cardiac surgery⁸  and spine surgery were not significant factors in multivariate analysis.

The primary strength of this study is that it is the largest prospective, case-controlled study of TRALI identified by active surveillance in a general population of transfused patients at academic centers, and a large number of biologically plausible risk factors were studied. Limitations include possible missed TRALI patients who had no Fi0₂ data available or an ABG was obtained after 12 hours of blood issue, and no report of the reaction was made to the blood bank. The wide CIs of the odds ratios, because of missing data, the decrease in TRALI incidence, and possibly differences in storage times of units transfused to cases and controls, limited the strength of evidence that this study could provide, particularly for negative findings (eg, bioreactive substances). In addition, we could not assess the effect of measures to reduce transfusion of platelets from alloimmunized donors because it was partially implemented late in this study.

In conclusion, this prospective study indicates that TRALI incidence, as determined by active surveillance, decreased after reduction of transfusion of plasma from female donors. The decrease in TRALI incidence was probably the result in part of reduced transfusion of strong cognate HLA class II antibodies and HNA antibodies to patients who are susceptible to ALI, although decreases in patient risk factors probably also played a role. To further reduce TRALI risk, our clinical evidence supports consideration of screening donors for strong HLA class II antibodies¹⁵  and the development of high-throughput GIFT methods to screen for antibodies to known and unknown human neutrophil antigens. Importantly, reduction of modifiable patient risk factors should also reduce the risk for developing TRALI.¹⁸ 

The authors thank Charlene Anderson for preparation of the manuscript and Dr Brian Curtis for performing HNA-3a genotyping of the 2 recipients of anti–HNA-3a.

This work was supported by the National Heart, Lung, and Blood Institute (Transfusion Medicine SCCOR P50HL081027; P.T.) and Clinical and Translational Science Award (grant UL1 RR024131).

The authors honor the assistance and inspiration of the late Dr S. Breanndan Moore, who passed away in 2009 during the study period.

Contribution: P.T., O.G., E.L.M., M.A.M., P.B., M.R.L., R.B.W., R.H., C.A.L., and M.A.G. conceived and designed the study; G.W., M.K., R.S.R., R.S., D.L., and P.W. acquired data; P.B., B.G., P.T., M.A.M., M.R.L., and R.B.W. analyzed and interpreted data; P.B. and B.G. performed statistical analysis; P.T., P.B., O.G., E.L.M., M.A.M., M.R.L., R.B.W., R.H., and C.A.L. obtained funding; G.W., M.K., R.S.R., R.S., D.L., and P.W. provided technical or material support; P.T. supervised the study; O.G. supervised the Mayo Clinic site; J.L.W. supervised collection of blood donor samples and data at Mayo Clinic; N.H. supervised collection of blood donor samples and data at Blood Centers of the Pacific; M.J.G. supervised the laboratory testing for HLA typing and antibody; P.J.N. supervised data download of and assisted in the analyses of HLA antibody screening and single-antigen bead test results; D.M. supervised the testing for antibody to human neutrophil antigens; C.A.L. supervised laboratory testing for cytokines and neutrophil priming activity; and all authors drafted the manuscript and critically revised the manuscript for important intellectual content.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Pearl Toy, Department of Laboratory Medicine, Box 0451, University of California–San Francisco, San Francisco, CA 94143-0451; e-mail: pearl.toy@ucsf.edu.