Feature Articles

Clinical predictors of and mortality in acute respiratory distress

syndrome: Potential role of red cell transfusion*
Michelle Ng Gong, MD, MS; B. Taylor Thompson, MD; Paige Williams, PhD;
Lucille Pothier; Paul D. Boyce, MD; David C. Christiani, MD, MPH

Objective: Clinical predictors for acute respiratory distress breaths/min (ORadj 2.39, 95% CI 1.51–3.78), hematocrit >37.5%
syndrome (ARDS) have been studied in few prospective studies. (ORadj 1.77, 95% CI 1.14 –2.77), arterial pH <7.33 (ORadj 2.00, 95%
Although transfusions are common in the intensive care unit, the CI 1.31–3.05), and albumin <2.3 g/dL (ORadj 1.80, 95% CI 1.18 –
role of submassive transfusion in non-trauma-related ARDS has 2.73). Packed red blood cell transfusion was associated with
not been studied. We describe here the clinical predictors of ARDS ARDS (ORadj 1.52, 95% CI 1.00 –2.31, p ⴝ .05). Significant predic-
risk and mortality including the role of red cell transfusion. tors for mortality in ARDS included age (ORadj 1.96, 95% CI
Design: Observational prospective cohort. 1.50 –2.53), Acute Physiology and Chronic Health Evaluation III
Setting: Intensive care unit of Massachusetts General Hospital. score (ORadj 1.78, 95% CI 1.16 –2.73), trauma (ORadj 0.075, 95% CI
Patients: We studied 688 patients with sepsis, trauma, aspi- 0.006 – 0.96), corticosteroids before ARDS (ORadj 4.65, 95% CI
ration, and hypertransfusion. 1.47–14.7), and arterial pH <7.22 (ORadj 2.32, 95% CI 1.02–5.25).
Interventions: None. Packed red blood cell transfusions were associated with in-
Measurements and Main Results: Two hundred twenty-one creased mortality in ARDS (ORadj 1.10 per unit transfused; 95% CI
(32%) subjects developed ARDS with a 60-day mortality rate of 1.04 –1.17) with a significant dose-dependent response (p ⴝ .02).
46%. Significant predictors for ARDS on multivariate analyses Conclusions: Important predictors for the development of and
included trauma (adjusted odds ratio [ORadj] 0.22, 95% confidence mortality in ARDS were identified. Packed red blood cell transfu-
interval [CI] 0.09 – 0.53), diabetes (ORadj 0.58, 95% CI 0.36 – 0.92), sion was associated with an increased development of and in-
direct pulmonary injury (ORadj 3.78, 95% CI 2.45–5.81), hemato- creased mortality in ARDS. (Crit Care Med 2005; 33:1191–1198)
logic failure (ORadj 1.84, 95% CI 1.05–3.21), transfer from another KEY WORDS: acute respiratory distress syndrome; transfusion;
hospital (ORadj 2.08, 95% CI 1.33–3.25), respiratory rate >33 mortality; respiratory failure; acute lung injury

A cute respiratory distress syn- and outcomes of ARDS with each of these replacement has been associated with
drome (ARDS) is a common common disorders (1, 3–5) and none lung injury as a result of both transfu-
pulmonary disorder of criti- have used the American European Con- sion-related lung injury (TRALI) and
cally ill patients that usually sensus Conference definition of ARDS massive transfusions (1, 4, 7). Some have
occurs after an injury such as sepsis, that is used widely today. Such epidemi- advocated transfusion of critically ill pa-
trauma, or aspiration. The incidence, cy- ologic standardization is necessary before tients to improve oxygen delivery given
tokine profile, and mortality rate in ARDS we can assess the contribution of biomar- its properties in volume expansion and
differ by the type of injury that predis- kers to the development of ARDS. oxygen delivery (8, 9). However, transfu-
posed the individual to ARDS (1, 2), sug- There has been much recent debate sion has been linked to increased compli-
gesting that the pathogenesis and out- about transfusion practices in critically ill cations and infections among critically ill
come of ARDS may differ for different patients. The Transfusion Requirement
patients and in patients after cardiac sur-
predisposing clinical risks. Yet few pro- in Critical Care Study showed no benefit
gery (10 –13). ARDS patients often have
spective studies have examined the risk to a liberal transfusion strategy (6). Blood
evidence of oxygen debt, but conse-
quences of red cell transfusions in ARDS
patients have not been studied.
*See also p. 1420. MA); grant K23 HL67197 from the National Heart, We describe a large epidemiologic pro-
From the Pulmonary and Critical Care Unit, Depart- Lung, and Blood Institute (MNG, Massachusetts Gen-
ment of Medicine, Massachusetts General Hospital, eral Hospital and Mount Sinai School of Medicine, New spective study of critically ill patients at
Harvard Medical School, Boston, MA (BTT, PDB, DCC); York); and grant T32 HL07874 (PDB), Massachusetts risk for ARDS as a result of sepsis, pneu-
Environmental Health Department (Occupational Health General Hospital. monia, trauma, massive transfusions, and
Program) (LP, DCC) and Department of Biostatistics Address requests for reprints to: David C. Chris- aspiration. We report on the clinical fac-
(PW), Harvard School of Public Health, Boston, MA; and tiani, MD, Harvard School of Public Health, 665 Hun-
Division of Pulmonary, Sleep and Critical Care Medi- tington Avenue, Boston, MA 02115. E-mail: tors associated with the development of
cine, Department of Medicine, Mount Sinai School of dchristi@hsph.harvard.edu Phone: (617) 432–3323 and mortality in ARDS including an as-
Medicine, New York, NY (MNG). Copyright © 2005 by the Society of Critical Care sociation between transfusion of red cells
Supported, in part, by research grant RO1 Medicine and Lippincott Williams & Wilkins
HL60710 from the National Heart, Lung, and Blood
and the development of and mortality in
DOI: 10.1097/01.CCM.0000165566.82925.14 ARDS.
Institute (Massachusetts General Hospital, Boston,

Crit Care Med 2005 Vol. 33, No. 6 1191

MATERIALS AND METHODS sent was obtained from all subjects or their were blinded to the possibility that transfusion
appropriate surrogates. may be related to ARDS.
Study Population. As part of a molecular Baseline clinical data and demographics For each patient, research coordinators
epidemiology study of ARDS, admissions to such as age, past history of ARDS, diabetes or collected daily information on respiratory fail-
the neurologic, cardiac, medical and surgical liver disease, alcohol abuse, or tobacco use ure, arterial blood gas, presence of bilateral
intensive care units (ICUs) of the Massachu- were collected. Organ failure was defined ac- infiltrates on chest radiographs, and pulmo-
setts General Hospital (MGH, Boston, MA) cording to the Brussels Organ Dysfunction nary arterial occlusion pressure if right heart
were screened daily for risk factors for ARDS Score (14), whereby cardiovascular failure is catheter was present or notation of congestive
(Table 1). Exclusion criteria included age ⬍18 defined as systolic blood pressure ⬍90 mm Hg heart failure on progress notes. For each day
yrs, diffuse alveolar hemorrhage or chronic or need for vasopressor, renal failure is defined of the study, enrolled patients were screened
lung disease, which may mimic ARDS, and as creatinine ⬍2.0 mg/L, hepatic failure is for ARDS as defined by respiratory failure re-
directive to withhold intubation. Because in- defined by total bilirubin ⬎2.0 mg/dL, and quiring intubation and fulfillment of Ameri-
flammatory cytokines are examined in the par- can European Consensus Conference criteria
hematologic failure is defined as platelets
ent study, patients with neutropenia not sec- for ARDS as follows (16): a) presence of hy-
⬍80,000/m3. Vital signs and laboratory values
ondary to sepsis and immunosuppression poxemia as evidenced by PaO2/FIO2 ⱕ200 mm
in the first 24 hrs of ICU admission were
secondary to medication or diseases such as Hg; b) presence of bilateral infiltrates on chest
collected. Recollection and re-entry of the
HIV infection were excluded. Treatment with radiographs; and c) absence of left atrial hy-
clinical data from 89 (13%) subjects selected
granulocyte colony-stimulating factor or in- pertension as evidenced by pulmonary arterial
at random revealed a data-entry error rate of
hibitors of tumor necrosis factor was ex- occlusion pressure ⱕ18 mm Hg or lack of
1% and a data collection error rate of 2.8%. All
cluded. After November 2000, immunosup- notation for congestive heart failure as a prob-
data were collected onto clinical data forms
pression secondary to corticosteroid was lem in the progress note.
removed as an exclusion criterion due to the and entered into an ACCESS database at the
Harvard School of Public Health. Infiltrates on chest radiographs were de-
increasing use of steroids in sepsis. Patients fined as opacities that cannot be explained
enrolled after this change were more likely to As part of quality control for genotyping in
the parent study, the transfusion history in completely by pleural effusions, mass, body
be female (p ⫽ .01), Caucasian (p ⫽ .03), with habitus, or collapse. Upper zone redistribution
a history of steroid use (p ⬍ .001) and diabetes the 8 days before enrollment into the study
and pulmonary vascular congestion were not
(p ⫽ .02), and were more likely to have been was recorded (15). Ninety percent of the sub-
considered infiltrates. Daily chest radiographs
transfused with blood (p ⫽ .007). There was jects were enrolled within 7 days of ICU ad-
were interpreted by two pulmonary and criti-
no difference in the rate of or mortality in mission, with ARDS developing a median of 1
cal care physicians (MNG, PDB, BTT, or DCC).
ARDS, clinical risk factor, age, severity of ill- day after ICU admission (25–75% quartile 0 –3
Any disagreement went to a third intensivist
ness, and baseline comorbid condition. Adjust- days). In 66 (30%) of ARDS patients, at least
for arbitration. All physicians underwent a
ing for whether the patient was enrolled be- one transfusion during the period of observa-
consensus training session on the radiologic
fore or after the change in exclusion criteria tion occurred after development of ARDS.
criteria for ARDS. All were blinded to the clin-
changed the estimates in the final models for Thus, the analysis was repeated after exclud-
ical status of the patients and the presence of
ARDS and mortality in ARDS 0 –7% with no ing transfusion received after development of
other criteria for ARDS. The kappa score for
change in statistical significance. The change ARDS. As a result, the period of observation agreement between radiologic interpretation
in estimate for transfusion and the develop- for ARDS patients was shorter (median 7 days, of bilateral infiltrates was 0.75 (95% confi-
ment of ARDS or mortality in ARDS was ⬍1%. 25–75% quartile 6 –7 days) than for non-ARDS dence interval [CI] 0.62– 0.89), which is com-
Patients admitted to the ICU with at least patients. Data on type, date, and number of parable to other reports after consensus train-
one defined risk factor for ARDS (Table 1) and transfusions were collected from the blood ing (17). Patients with ARDS were followed for
no exclusion criteria were eligible for the bank’s computerized clinical database. Deci- all-cause 60-day mortality.
study. The Human Subjects Committees of sion to transfuse was left to the treating phy- Statistical Analysis. Serum albumin, bili-
the MGH and Harvard School of Public Health sicians, and no predetermined transfusion rubin, and arterial pH were missing for 90
approved the study, and informed written con- protocol was enforced. Research coordinators (13%), 87 (13%), and 50 (7%) subjects, respec-
tively. All other variables were complete in
ⱖ95% of the patients. Like the Acute Physiol-
Table 1. Study required risk factors for acute respiratory distress on admission to intensive care unit ogy and Chronic Health Evaluation (APACHE)
III study, missing values tended to occur in
Sepsis: As defined by SCCM (53) to be a known or suspected source of systemic infection and at patients with stable vital signs or laboratory
least two of the following: a) temperature ⬎38°C or ⬍36°C; b) heart rate ⬎90 beats/min; c) values (p ⱕ .03). Similar to the APACHE III
respiratory rate ⬎20 breaths/min or PaCO2 ⬍32 mm Hg; d) WBC ⬎12,000/mm3, ⬍4000/mm3, or study, normal values closest to the median
⬎10% bands. value for the cohort were assigned to missing
Septic shock: Fulfill requirements for sepsis and one of the following: a) SBP ⬍90 mm Hg or
values here (18). Nonphysiology data such as
reduction of ⱖ40 mm Hg from baseline for ⱖ30 mins unresponsive to 500 mL of fluid
history of diabetes or tobacco use were coded
resuscitation; b) need for vasopressors to maintain SBP 90 mm Hg or within 40 mm Hg of
as missing.
baseline.
Univariate analyses were performed using
Pneumonia: Fulfill two or more of the following: a) new infiltrate on CXR; b) temperature ⬎38.3°C
or ⬍36.0°C or WBC ⬎12,000 or ⬍4000 or ⬎10% bandemia; c) positive microbiologic culture.
Fisher’s exact test for dichotomous variables
Trauma: Defined as multiple fractures and/or pulmonary contusions. Multiple fractures are defined and Wilcoxon rank sum for continuous vari-
as a fracture of two long bones, an unstable pelvic fracture, or one long bone and a pelvic able as the data did not have a normal distri-
fracture. Pulmonary contusion is defined as infiltrates on CXR within 8 hrs of admission to the bution. Continuous variables were modeled
emergency room and evidence of blunt trauma to the chest such as fractured ribs or ecchymosis assuming a linear or step function after cate-
overlying the infiltrate. gorizing by quartiles. When no model ap-
Multiple transfusions: Defined as receiving ⱖ8 units of PRBCs within 24 hrs. peared to be the best, a linear relationship was
Aspiration: Defined as witnessed or documented aspiration event or the retrieval of gastric contents assumed. Clinical risks and predictors from
from the oropharynx, endotracheal tube, or bronchial tree the univariate analyses with p ⱕ .06 were
studied in a multiple logistic regression model
SCCM, Society of Critical Care Medicine; WBC, white blood cell count; SBP, systolic blood using a backward elimination algorithm and
pressure; CXR, chest radiograph; PRBC, packed red blood cells. eliminated if they did not meet p ⱕ 0.1. Final

1192 Crit Care Med 2005 Vol. 33, No. 6

models included predictors from backward ticipants were significantly more likely to 36.8, p ⫽ .01). Subjects with ARDS were
elimination as well as clinically relevant fac- be non-Caucasian and less likely to have enrolled a median of 2 days after ICU
tors such as age and APACHE III scores for the aspiration or sepsis without hypotension admission (25–75% quartile, 1– 4 days)
development of ARDS and alcohol abuse, on admission to the ICU. Otherwise, par- compared with a median of 1 day for
chronic liver disease, and septic shock for
ticipants and nonparticipants did not dif- non-ARDS patients (25–75% quartile 1–2
mortality in ARDS. The C-statistic was 0.80 for
the final model for development of ARDS and fer significantly in age, gender, severity of days, p ⬍ .001). Race, age, alcohol or
0.83 for the final model on mortality in ARDS. illness, comorbid diseases, and frequency tobacco abuse, corticosteroid treatment,
Interactions between transfusion and other of alcohol or tobacco abuse (p ⬎ .06). kidney or liver failure, mean arterial pres-
clinical risks factors for ARDS in the model Development of ARDS. A total of 221 sure, temperature, urine output, white
were tested with the addition of an interaction (32%) subjects developed ARDS a median blood cell count, serum sodium, potas-
term, but no significant interaction was found of 1 day after ICU admission (25–75% sium, blood urea nitrogen, glucose, and
in the final models for development of ARDS quartile 0 –3 days). Table 2 shows the bicarbonate were not significantly associ-
and mortality in ARDS (p ⬎ .2). In the final proportions of patients who developed ated with ARDS (p ⬎ 01).
model, p ⱕ .05 was considered statistically
ARDS with each precipitating condition. Table 4 shows the clinical predictors
significant. All analyses were conducted using
SAS version 8 (SAS Institute, Cary, NC). Sepsis and septic shock patients with for development of ARDS in the final
pneumonia were more likely to develop multivariate model after adjusting for
ARDS than those with extrapulmonary days between enrollment and admission.
RESULTS
sources of infections (odds ratio [OR] Trauma as a risk factor for ARDS and
Between September 9, 1999, and Au- 3.41, 95% CI 1.79 – 6.48 in sepsis and OR diabetes was associated with a decreased
gust 7, 2002, 1,072 admissions were eli- 2.84, 95% CI 1.77– 4.57 in septic shock). risk of ARDS, whereas direct pulmonary
gible for the study. In 58 subjects, the Septic shock increased the risk of ARDS injury, hematologic failure defined as
treating physician declined the study, in patients with pneumonia (OR 1.55, platelets ⱕ80,000/mm, transfer from out-
whereas 51 subjects died before consent. 95% CI 1.00 –2.42). side hospital, respiratory rate ⬎33
No appropriate surrogate was available The baseline characteristics between breaths/min, hematocrit ⬎37.5%, arte-
for 83 patients, and 151 patients or their ARDS patients and at-risk non-ARDS pa- rial pH ⬍7.33, serum albumin ⱕ2.3 g/dL,
surrogates declined the study. Three ad- tients are shown in Table 3. On univariate and transfusion of packed red blood cells
missions before November 2000 were ex- analysis of the physiology variables, pa- (PRBCs) were associated with increased
cluded after enrollment because cortico- tients who developed ARDS also had development of ARDS.
steroid was initiated. Thirty-eight higher heart rates (p ⬍ .001), respiratory ARDS patients were transfused more
individuals qualified for the study on rates (p ⬍ .001), and hematocrit (p ⫽ frequently (61%) than non-ARDS pa-
more than one ICU admission, but only .01), and lower albumin (p ⬍ .001) and tients (49%, p ⫽ .004), with ARDS pa-
the first admission was included in this arterial pH (p ⬍ .001) than did non-ARDS tients receiving a median of 1 unit of
study. Thus, 688 individuals were avail- subjects in the first 24 hrs of ICU admis- PRBCs (25–75% quartile 0 –3) compared
able for analyses. sion. The median hematocrit for patients with a median of 0 units (25–75% quar-
Clinical information was available in with ARDS was 34.4% (25–75% quartile tile 0 –3) among non-ARDS patients (p ⫽
299 (87%) eligible nonparticipants for 31.1–39.5) compared with 33.6% in those .04). On multivariate analysis (Table 4),
whom consent was not obtained. Nonpar- without ARDS (25–75% quartile 30.3– transfusion of any number of units of

Table 2. Etiology of acute respiratory distress syndrome (ARDS) in cohort

Development of ARDS 60-Day Mortality in ARDS

Non-ARDS ARDS Rate of Survivors Nonsurvivors

(n ⫽ 467), (n ⫽ 221), ARDS, p (n ⫽ 119), (n ⫽ 102), Mortality p
Risk for ARDS No. (%) No. (%) %a Value No. (%) No. (%) Rate, %b Value

Sepsis syndrome 174 (37) 71 (32) 29 NS 48 (40) 23 (23) 32 .006

Pneumonia source 91 (19) 56 (25) 38 ⬍.001 36 (30) 20 (20) 36 NS
Extrapulmonary source 83 (18) 15 (7) 15 12 (10) 3 (3) 20
Septic shock 205 (44) 121 (55) 37 .009 54 (45) 67 (66) 55 .003
Pneumonia source 91 (19e) 84 (38) 48 ⬍.001 37 (31) 47 (47) 56 NS
Extrapulmonary source 114 (24e) 37 (17) 25 17 (14) 20 (19) 54
Trauma 46 (10) 10 (5) 18 0.02 9 (8) 1 (1) 10 0.02
Multiple transfusions 57 (12) 23 (10) 29 NS 10 (8) 13 (13) 57 NS
Aspiration 38 (8) 23 (10) 38 NS 11 (9) 12 (12) 52 NS
⬎1 risk for ARDS 54 (12) 27 (12) 33 NS 13 (11) 14 (14) 52 NS
Direct pulmonary injuryc 226 (48) 155 (70) 41 ⬍.001 85 (71) 70 (69) 45 NS
Indirect pulmonary injuryd 241 (52) 66 (30) 22 34 (29) 32 (31) 48

NS, not statistically significant (p ⬎ .05). Numbers of patients with each risk add up to more than 688 patients because of multiple risks in 58 patients.
a
Number of subjects with risk with ARDS/total number of subjects with risk in cohort; bnumber of ARDS nonsurvivors with risk/TOTAL number of
ARDS subjects with risk; cpneumonia, aspiration, or pulmonary contusions were categorized as direct pulmonary injury; dsepsis from an extrapulmonary
source, trauma without pulmonary contusions, and multiple transfusions were categorized as indirect pulmonary injury: patients with both direct and
indirect pulmonary injuries were considered to have direct pulmonary injury; edoes not sum up to expected percentage because of rounding.

Crit Care Med 2005 Vol. 33, No. 6 1193

Table 3. Baseline characteristics in the cohort

Development of ARDS Mortality in ARDS

Non-ARDS ARDS Survivors Nonsurvivors

(n ⫽ 467) (n ⫽ 221) p Value (n ⫽ 119) (n ⫽ 102) p Value

Females, n (%) 183 (39) 102 (46) NS 50 (42) 52 (51) NS

White, n (%) 433 (93) 199 (90) NS 105 (88) 94 (92) NS
Age, median (range) 67 (18–94) 65 (18–97) NS 55 (18–89) 73 (22–97) ⬍.001
APACHE III, median 63 (10–137) 68 (8–136) .03 70 (8–128) 89 (40–150) ⬍.001
(range)a
Diabetes, n (%)b 123 (26) 41 (19) .03 25 (21) 16 (16) NS
History of alcohol 49 (10) 28 (13) NS 12 (10) 16 (16) NS
abuse, n (%)
Tobacco abuse, n (%)c 223 (60) 104 (63) NS 56 (67) 48 (64) NS
Chronic liver disease, 22 (5) 11 (5) NS 4 (3) 7 (7) NS
n (%)b
End-stage renal 25 (5) 6 (3) NS 2 (2) 4 (4) NS
disease, n (%)
History of steroid use, 39 (8) 24 (11) NS 9 (8) 15 (15) NS
n (%)
Transfer from another 82 (16) 78 (35) ⬍.001 40 (34) 37 (36) NS
hospital, n (%)
Transfusion of PRBCs, 228 (49) 134 (61) .001 65 (55) 69 (68) .05
n (%)
Number of PRBCs 0 (0–74) 1 (0–63) .04 1 (0–31) 2 (0–63) .02
transfused, median
(range)
Systolic BP ⬍90 mm 319 (68) 169 (76) .03 86 (72) 83 (81) NS
Hg, n (%)
Creatinine ⬎ 2.0 mg/ 159 (34) 68 (31) NS 29 (24) 39 (38) .03
L, n (%)
Bilirubin ⬎2.0 mg/dL, 55 (12) 34 (15) NS 14 (12) 20 (20) NS
n (%)
Hematologic failure 59 (13) 40 (18) .06 16 (13) 24 (24) .05
(platelets ⱕ80,000/
mm), n (%)

ARDS, acute respiratory distress syndrome; NS, not statistically important for final model (p ⱖ .1); APACHE, Acute Physiology and Chronic Health
Evaluation; PRBCs, packed red blood cells; BP, blood pressure.
a
For development of ARDS, APACHE III scores for patients and controls were calculated without the PaO2/FIO2 component: For mortality in ARDS, the
APACHE III score was calculated with all components; bchronic health information was missing on one patient and two controls; ctobacco history was
missing in 56 (26%) patients and 97 (22%) controls.

PRBCs was associated with increased of 0 units, 25–75% quartile 0 –2 units for 75% quartile 6 – 8) compared with the 8
odds of developing ARDS (ORadj 2.19, transfers compared with median of 1 days of observation for non-ARDS pa-
95% CI 1.42–3.36, p ⬍ .001). After cate- unit, 25–75% quartile 0 –3 units among tients (p ⬍ .001), the association between
gorizing by quartile, there was a trend nontransfers, p ⬎ .08). transfusion and the development of ARDS
toward increasing rate of ARDS with in- Because the period of observation in the restricted analysis remained signif-
creased transfusion (p ⫽ .05, Fig. 1). Al- overlapped with the development of icant (ORadj 1.52, 95% CI 1.00 –2.31, p ⫽
though the rate of ARDS appears to de- ARDS in some ARDS patients, the analy- .05). Transfusions were given a median of
cline in those patients transfused with sis was repeated after excluding transfu- 1 day before development of ARDS (25–
⬎3 units of PRBCs compared with those sions received after the development of 75% quartile, 4 days before to day of
transfused with ⱕ3 units, this is due to a ARDS. In one patient without ARDS and development of ARDS). Although prior
higher proportion of trauma patients in 16 patients with ARDS, the number of reports of transfusion and ARDS per-
(18% vs. 5%, p ⬍ .001), who had the units transfused or the timing of transfu- tained mostly to massive transfusions of
lowest rate of ARDS, and a lower percent- sions relative to the development of ⱖ10 units of blood (1, 2, 4), only 12% of
age of patients with septic shock (35% vs. ARDS could not be determined accurately patient in this study were massively
51%, p ⬍ .001) and direct pulmonary as these patients had received transfu- transfused (Table 2) and the majority of
injury (27% vs. 63%, p ⬍ .001), who had sions at another institution before trans- patients in this study received submassive
the highest rates of ARDS. There was no fer to MGH. These patients were also ex- amounts of PRBCs (median 3 units, 25–
difference between patients transferred to cluded, leaving 205 patients with ARDS 75% quartile 2– 8).
the study hospital and nontransfers in the and 466 patients without ARDS for the Mortality in ARDS. The 60-day mor-
frequency of transfusions (49% of trans- restricted analysis. Although the period tality rate for the 221 patients with ARDS
fers vs. 54% of nontransfers, p ⬎ .3) or in of observation for transfusion was shorter was 46%. The etiology for ARDS and
the number of PRBCs transfused (median for ARDS patients (median 7 days, 25– baseline characteristics between survi-

1194 Crit Care Med 2005 Vol. 33, No. 6

Table 4. Multivariate analysis for the development of acute respiratory distress syndrome (ARDS) PRBCs was associated with increased risk
of mortality in ARDS (ORadj 1.10 per unit
Odds Ratio transfused, 95% CI 1.04 –1.17). All trans-
(95% Confidence Interval) p Value
fusions in the 8 days before enrollment
Trauma 0.22 (0.09–0.53) ⬍.001 into the study both before and after de-
Diabetes 0.58 (0.36–0.92) .02 velopment of ARDS were used in this
Direct pulmonary injury 3.78 (2.45–5.81) ⬍.001 analysis. After we categorized the number
Transfer from another hospital 2.08 (1.33–3.25) .001 of units of blood transfused according to
Hematologic failure (platelets ⬍80,000/mm3) 1.84 (1.05–3.21) .03
HR ⬎99 beats/min 1.58 (0.96–2.61) NS
quartiles, there was a significant trend to
RR ⬎33 breaths/min 2.39 (1.51–3.78) ⬍.001 increasing mortality rate with increased
Hematocrit ⬎37.5% 1.77 (1.14–2.77) .01 transfusions (p ⫽ .02, Fig. 1).
Arterial pH ⬍7.33 2.00 (1.31–3.05) .001
Albumin ⱕ2.3 g/dL 1.80 (1.18–2.73) .006
APACHE III, per point increase 1.00 (0.99–1.01) NS DISCUSSION
Age, per year of age 1.00 (0.99–1.01) NS
Transfusion of PRBCs 2.19 (1.42–3.36) ⬍.001 We identified several clinically impor-
tant factors in the development and out-
HR, heart rate; NS, not statistically significant, p ⬎ .05; RR, respiratory rate; APACHE, Acute come of ARDS in a large prospective co-
Physiology and Chronic Health Evaluation; PRBCs, packed red blood cells. APACHE III score was hort of at-risk individuals including an
calculated without the arterial-alveolar oxygen gradient score as it is a part of the criteria for ARDS. association between transfusion and the
development of and mortality in ARDS.
This is consistent with findings from the
Transfusion Requirement in Critical Care
Study, in which patients randomized to
the liberal transfusion strategy had a
nonstatistically significant greater risk of
developing ARDS (11.4% vs. 7.7%, p ⫽
.06) and a greater mortality rate (23.3%
vs. 18.7%, p ⫽ .11) than the patients in
the restrictive group (6).
This study has the benefit of being a
prospective cohort at risk for ARDS,
which reduces several potential problems
Figure 1. Development of acute respiratory distress syndrome (ARDS) and 60-day mortality rate in ARDS such as recall or coding biases associated
by number of packed red cells transfused in the 8 days before enrollment. Number of patients analyzed is with retrospective studies. In addition,
specified within each bar. The p value for chi-square trend test was .05 for ARDS and .02 for mortality in the clear characterization of clinical risks
ARDS. Two patients were not included because of missing data on total number of units transfused. for ARDS in this study is advocated by
some in decreasing the clinical heteroge-
neity in studies of ARDS given that the
vors and nonsurvivors with ARDS are admission, lung injury score, and tidal incidence, natural history, cytokine pro-
shown in Tables 2 and 3, respectively. volume/kg of ideal body weight on first files, and outcomes of ARDS differ signif-
ARDS secondary to sepsis without hypo- day of ARDS (median 9.7 mL/kg 25–75% icantly by etiology of ARDS (19 –21). Fi-
tension and trauma had the lowest mor- quartile 8.2–11.5 for the cohort, p ⬎ .8). nally, the focus of this study on critically
tality rates (32% and 10%, respectively). The results for the multivariate anal- ill patients who have the opportunity to
All other risk factors were associated with ysis for mortality in ARDS are shown in develop the outcome is more clinically
similar mortality rate in ARDS. Direct Table 5. Trauma as a risk factor for ARDS relevant to the critical care physician.
pulmonary injury was associated with a was associated with decreased mortality The associations found between ARDS
greater risk of ARDS (p ⬍ .001) but no in ARDS. Significant predictors for in- and the clinical risk factors, low arterial
difference in mortality rate (p ⫽ .7). creased mortality in ARDS include age, pH, hematologic failure, hypoalbumine-
On univariate examination of physio- APACHE III score on admission, cortico- mia, and absence of diabetes, are consis-
logic variables in the first 24 hrs of ICU steroid treatment ⱖ15 mg/day pred- tent with previous reports (1, 5, 22–25).
admission, nonsurvivors were more likely nisone or ⱖ300 mg of prednisone in the Unlike previous studies (5, 26), we did not
to have lower temperatures (p ⫽ .02), 21 days before development of ARDS, ar- find an association between alcohol abuse
mean arterial blood pressures (p ⬍ .001), terial pH ⬍ 7.22, and transfusion of or chronic liver disease and ARDS, likely
urine output (p ⫽ .002), and arterial pH PRBCs. a result of insufficient power since only
(p ⫽ .008) and higher blood urea nitro- Among those with ARDS, 65 (55%) 5% patients had evidence of chronic liver
gen (p ⫽ .004). There were no statisti- survivors and 69 (68%) nonsurvivors disease and only 1% of patients had a
cally significant differences between the were transfused (OR 1.79, 95% CI 1.03– history of alcohol abuse in this cohort.
survivors and nonsurvivors for gender, 3.11, Table 3). Survivors received a me- The association between mortality in
race, time between ICU admission and dian of 1 unit of PRBCs (25–75% quartile ARDS and age and severity of illness is
study enrollment, history of diabetes, 0 –2) compared with 2 units (25–75% consistent with other reports (26 –30).
liver or kidney failure or disease, tobacco quartile 0 – 4) in nonsurvivors (p ⫽ .02). We report several novel observations.
or alcohol use, frequency of shock on On multivariate analysis, transfusion of In this study, direct pulmonary injury

Crit Care Med 2005 Vol. 33, No. 6 1195

Table 5. Multivariable analysis for predictors of mortality in acute respiratory distress syndrome sions (34, 35), and none of the ARDS
(ARDS) patients in this study was considered to
have TRALI. A possible association be-
Odds Ratio
(95% Confidence Interval) p Value
tween transfusion and ARDS in this study
raises the question of the true incidence
Trauma 0.075 (0.006–0.96) ⬍.05 of TRALI, especially in ICU patients,
Age, per decade of age 1.96 (1.50–2.53) ⬍.001 where other risk factors for lung injury
APACHE III, per 25-point increasea 1.78 (1.16–2.73) .008 are common. Although granulocyte and
History of corticosteroid treatment before ARDS 4.65 (1.47–14.7) .009
History of alcohol abuse 2.68 (0.93–7.74) NS human leukocyte antigen antibodies have
Chronic liver disease 3.07 (0.65–14.5) NS been implicated in TRALI (35, 36), the
Septic shock 1.24 (0.58–2.62) NS exact mechanism of lung injury in TRALI
Arterial pH ⬍7.22 2.32 (1.02–5.25) .04 is unclear. A “two-hit” model of TRALI
Pack red blood cells, per unit transfused 1.10 (1.04–1.17) .001
has been theorized whereby some predis-
APACHE, Acute Physiology and Chronic Health Evaluation. posing comorbid condition such as infec-
a
Calculated with all components of the APACHE III. tion results in sequestration of neutro-
phils in the lung (37, 38). Then, a second
hit consisting of exposure to lysophos-
progressed to ARDS significantly more patients, the transfusion of even one unit photidylcholine accumulated in stored
frequently than indirect pulmonary in- of PRBCs was associated with increased PRBCs results in lung injury. Such a
jury. However, when pneumonia is com- mortality in ARDS. model has important clinical implica-
plicated by a systemic insult such as sep- However, this study was not expressly tions with regard to transfusion of criti-
tic shock, the risk of ARDS increases designed to examine the role of transfu- cally ill patients.
significantly. This suggests an additive sion in ARDS. Thus, other potentially im- No study, to our knowledge, has ex-
role of direct alveolar injury from pneu- portant factors that may clarify the na- amined the relationship between transfu-
monia and endothelial injury from sys- ture of the association between sion and outcomes specifically in ARDS.
temic inflammatory response associated transfusion and ARDS, such as the age of Our finding that transfusion of PRBCs
with pneumonia-induced severe sepsis. stored blood products, transfusion of
was associated with increased mortality
However, although direct pulmonary in- other blood products, the percentage of
rate in ARDS in a dose-dependent man-
jury was significantly associated with the leukocyte-filtered blood given, and the
ner is consistent with prior reports of
development of ARDS, there was no effect transfusion history for the entire ICU
transfusion in other critically ill patients
on the outcome in ARDS. This finding is course of the patients, cannot be as-
(12, 39 – 42). This association has been
consistent with some, but not all, prior sessed. Although clinically suspected pul-
attributed to aged, stored blood (40, 45)
studies (27, 31). monary edema from volume overload
and the immunosuppressive effects of al-
An association between a high hemat- must be excluded for the diagnosis of
logenic blood (44), with a resultant in-
ocrit and an increased risk for developing ARDS, transfusions resulting in unsus-
ARDS has not been reported previously. pected pulmonary edema cannot be ex- creased risk of nosocomial infections and
This finding may represent hemoconcen- cluded as a pulmonary artery catheter ventilator-associated pneumonia (47). In-
tration from increased systemic and pul- was not required for ARDS. Last, al- deed, transfusions have been associated
monary capillary leak in those patients though age and severity of illness were with increased immunosuppression and
who progress to ARDS. The association adjusted for in the final analyses, it is down-regulation of inflammatory cyto-
between steroid treatment before admis- difficult to exclude confounding by indi- kine response in septic rats and after sur-
sion and mortality in ARDS has not been cation in an uncontrolled study. Indeed, gery (46 –51). The risk of infection is par-
described. Immunosuppression or rela- the odds ratio for the development of ticularly relevant here, as most patients
tive adrenal insufficiency may increase ARDS was reduced from 2.19 to 1.53 after with ARDS die of sepsis (52).
mortality in ARDS. Alternatively, steroid excluding transfusions given after devel- We acknowledge a few limitations to
treatment may be a marker of comorbid opment of ARDS. This could be secondary our study. MGH is a tertiary academic
disease. The importance of these two to the shorter period of observation for hospital and frequently accepts transfers
findings is unclear and will need to be transfusion in ARDS patients compared of patients with ARDS, so the results here
confirmed. with non-ARDS patients, which would may not be generalizable to community
Although massive transfusion of ⱖ10 bias the result toward the null. However, hospital setting. However, transfers are
units of blood over 12– 48 hrs has been although the association between trans- unlikely to explain the finding between
associated with ARDS, especially in fusion and ARDS remained significant af- transfusions and ARDS because they were
trauma patients (1, 3, 4, 11, 32), it is not ter exclusion of transfusion given after adjusted for in the analyses and there was
clear whether submassive transfusion of ARDS, we cannot exclude the possibility no association between transfer from an-
PRBCs may lead to increased develop- that transfusion is a marker of clinical other hospital and the frequency and vol-
ment of and increased mortality in ARDS. suspicion for ARDS or poor outcome not ume of transfusion received. In addition,
We found in this cohort an association captured by the data collected. Thus, because of the study design, the results
between transfusion of PRBCs and the these findings will need to be confirmed, may not be generalizable to minorities, to
development of and mortality in ARDS. preferably in randomized controlled immunocompromised hosts, or to pa-
The majority of these patients received study. tients without risk factors for ARDS or
modest transfusions (median 3 units, 25– TRALI is generally thought to be rare with different clinical risk factors for
75% quartile 2– 8 units). Among ARDS with an incidence of 1 in 5,000 transfu- ARDS. Nevertheless, the population stud-

1196 Crit Care Med 2005 Vol. 33, No. 6

 3. Pepe P, Potkin RT, Reus DH, et al: Clinical and Blood Institute working group. Am J

P
predictors of the adult respiratory distress Respir Crit Care Med 2003; 167:1027–1035
acked red blood syndrome. Am J Surgery 1982; 144:124 –130 20. Parson PE: Mediators and mechanisms of acute
4. Fowler, AA, Hamman RF. Adult respiratory dis- lung injury. Clin Chest Med 2000; 21:467– 476
cell transfusion tress syndrome: Risk with common predispo- 21. Moss M, Thompson, BT.: Definitions and
sitions. Ann Intern Med 1983; 98:593–597 clinical risk factors. In: Acute Respiratory
was associated 5. Moss M, Bucher B, Moore FA, et al: The role Distress Syndrome. Mathay MA, ed. New
of chronic alcohol abuse in the development York, Marcel Dekker, 2003, pp 7–36
with an increased develop- of acute respiratory distress syndrome in 22. Roupie E, Lepage E, Wysocki M, et al: Prev-
adults. JAMA 1996; 275:50 –54 alence, etiologies and outcome of the acute
ment of and increased mor- respiratory distress syndrome among hypox-
6. Herbert PC, Wells G, Blajchman MA, et al: A
multicenter randomized, controlled clinical emic ventilated patients. SRLF Collaborative
tality in acute respiratory Group on Mechanical Ventilation. Societe de
trial of transfusion requirements in critical
distress syndrome. care. N Engl J Med 1999; 340:409 – 417 Reanimation de Langue Francaise. Intensive
7. Kopko PM, Holland PV: Transfusion-related Care Med 1999; 25:920 –929
acute lung injury. Br J Haematol 1999; 105: 23. Moss M, Guidot DM, Steinberg KP, et al:
322–329 Diabetic patients have decreased incidence of
8. Haupt MT: Debate: Transfusing to normal acute respiratory distress syndrome. Crit
ied is relevant to a large proportion of hemoglobin levels improves outcome. Crit Care Med 2000; 18:2187–2192
critically ill patients. Care 2001; 5:64 – 66 24. Luhr OR, Antonsen K, Karlsson M, et al:
9. Rivers E, Nguyen B, Havstad S, et al: Early Incidence and mortality after acute respira-
Goal-Directed Therapy Collaborative Group. tory failure and acute respiratory distress
CONCLUSION syndrome in Sweden, Denmark, and Iceland.
Early goal-directed therapy in the treatment
This study describes differences in the of severe sepsis and septic shock. N Engl The ARF Study Group. Am J Respir Crit Care
J Med 2001; 345:1368 –1377 Med 1999; 159:1849 –1861
development and outcome of ARDS with
10. Leal-Noval SR, Marquez-Vacaro JA, Garcia- 25. Mangiarlardi RJ, Martin GS, Bernard GR, et
different clinical risk conditions and al: Hypoproteinemia predicts acute respira-
identifies several novel clinical predictors Curiel A, et al: Nosocomial pneumonia in
patients undergoing heart surgery. Crit Care tory distress syndrome development, weight
for the development of ARDS such as a gain and death in patients with sepsis. Crit
Med 2000; 28:935–940
high hematocrit and chronic corticoste- Care Med 2000; 28:3137–3145
11. Malone DL, Dunne J, Tracy JK, et al: Blood
roid treatment. Although massive trans- 26. Milberg JA, Davis DR, Steinberg KP, et al:
transfusion, independent of shock severity, is
fusions have been associated with ARDS, associated with worse outcome in trauma.
Improved survival of patients with acute re-
we report here a possible association be- spiratory distress syndrome (ARDS):
J Trauma 2003; 54:898 –905
tween submassive transfusion of PRBCs 1983–1993. JAMA 1995; 273:306 –309
12. Taylor RW, Manganaro L, O’Brien J, et al: Im-
and development of and mortality in 27. Bersten AD, Edibam C, Hunt T, et al: Austra-
pact of allogenic packed red cell transfusion on
lian and New Zealand Intensive Care Society
ARDS. A randomized control trial is nosocomial infection rates in the critically ill
Clinical Trials Group. Incidence and mortal-
needed to definitively determine whether patient. Crit Care Med 2002; 30:2249 –2254
ity of acute lung injury and the acute respi-
transfusions can increase the risk of lung 13. Corwin HL, Gettinger A, Pearl RG, et al: The
ratory distress syndrome in three Australian
injury and mortality in ARDS and CRIT Study: Anemia and blood transfusion in
States. Am J Respir Crit Care Med. 2002;
whether a simple change in transfusion the critically ill—Current clinical practice in
165:443– 448
practice can decrease the incidence of the United States. Crit Care Med 2004; 32:
28. Monchi M, Belllenfant F, Cariou A, et al:
39 –52
ARDS and improve mortality rates in at- Early predictive factors of survival in acute
14. Bernard GR, Doig G, Hudson L: Quantifica- respiratory distress syndrome: A multivariate
risk patients.
tion of organ failure for clinical trials and analysis. Am J Respir Crit Care Med 1998;
clinical practice. Abstr. Am J Respir Crit Care 159:1076 –1081
ACKNOWLEDGMENTS Med 1995; 151:A323 29. Zilberberg MD, Epstein SK: Acute lung in-
15. Gong MN, Sai Y, Zhou W, et al: Genotyping jury in the medical ICU. Am J Respir Crit
We thank Wei-Ling Zhang, Kelly Mc-
with recent blood transfusions. Epidemiol- Care Med 1998; 157:1157–1164
Coy, Thomas McCabe, and Barbara Bean ogy 2003; 14:744 –747 30. Ely EW, Wheeler AP, Thompson BT, et al:
for patient recruitment; Andrea Solomon, 16. Bernard GR, Artigas A, Brigham KL, et al: Recovery rate and prognosis in older persons
Andrea Shafer, and Lia Shimada for re- The American-European Consensus Confer- who develop acute lung injury and the acute
search support; Janna Frelich, Marcia ence on ARDS. Definitions, mechanisms, rel- respiratory distress syndrome. Ann Intern
Chertok, Sal Mucci, and Richard Rivera evant outcomes, and clinical trial coordina- Med 2002; 136:25–36
for data management; Paige Williams, tion. Am J Respir Crit Care Med 1994; 149: 31. Vieillard-Baron A, Girou E, Valente E, et al:
PhD, for statistical assistance; and David 818 – 824 Predictors of mortality in acute respiratory
Miller, ScD, for his epidemiologic advice. 17. Meade MO, Cook RJ, Guyatt GH, et al: Inter- distress syndrome: Focus on the role of right
observer variation in interpreting chest ra- heart catheterization. Am J Respir Crit Care
diographs for the diagnosis of acute respira- Med 2000; 161:1597–1601
REFERENCES
tory distress syndrome. Am J Respir Crit 32. Miller PR, Croce MA, Kilgo PD, et al: Acute
1. Hudson LD, Milberg JA, Anardi D, et al: Clin- Care Med 2000; 161:85–90 respiratory distress syndrome in blunt trau-
ical risks for development of the acute respi- 18. Knaus WA, Wagner DP, Draper EA, et al: The ma: Identification of independent risk fac-
ratory distress syndrome. Am J Respir Crit APACHE III prognostic system. Risk prediction tors. Am Surg 2002; 68:845– 850
Care Med 1995; 151:293–301 of hospital mortality for critically ill hospital- 33. Malone DL, Dunne J, Tracy JK, et al: Blood
2. Moss M, Gillespie MK, Ackerson L, et al: ized adults. Chest 1991; 100:1619 –1636 transfusion, independent of shock severity, is
Endothelial cell activity varies in patients at 19. Matthay MA, Zimmerman GA, Esmon C, et associated with worse outcome in trauma.
risk for the adult respiratory distress syn- al: Future research directions in acute lung J Trauma 2003; 54:898 –905
drome. Crit Care Med 1996; 24:1782–1786 injury: Summary of a National Heart, Lung, 34. Popovsky MA, Chaplin HC Jr, Moore SB:

Crit Care Med 2005 Vol. 33, No. 6 1197

 Transfusion-related acute lung injury: A ne- fused in septic ICU patients. Can J Anaesth Study Group. Red blood cell transfusion and
glected, serious complication of hemo- 1997; 44:1256 –1261 ventilator-associated pneumonia: A potential
therapy. Transfusion 1992; 32:589 –592 41. Flores JM, Jimenez PI, Rincon MD: Early risk link? Crit Care Med 2004;32:666 – 674
35. Kopko PM, Marshall CS, MacKenzie MR: factors for sepsis in patients with severe blunt 48. Iwagaki H, Yagi T, Urushihara N, et al: Blood
Transfusion-related acute lung injury: Re- trauma. Injury Int J Care Injured 2001; 32: transfusion and postoperative plasma cytokine
port of a clinical look-back investigation. 5–12 antagonist levels in colorectal cancer patients.
JAMA 2002; 287:1968 –1971 42. Mezrow CK, Bergstein I, Tartter PI: Postop- Hepatogastroenterology 2001; 48:1351–1354
36. Palfi M, Berg S, Ernerudh J: A randomized erative infections following autologous and 49. Salinas JC, Cabezalli R, Torcal J, et al: Im-
controlled trial of transfusion-related acute homologous blood transfusions. Transfusion mune response and cytokines in septic rats
lung injury: Is plasma from multiparous 1992; 32:27–30 undergoing blood transfusion. J Surg Res
blood donors dangerous? Transfusion 2001; 43. Heiss MM, Mempel W, Jauch KW, et al: Ben-
1998; 80:295–299
41:317–322 eficial effect of autologous blood transfusion
50. Shao W, Edelman LS, Sullivan DJ, et al:
37. Silliman CC, Voelkel NF, Allard JD, et al: on infectious complications after colorectal
Long-term cytokine alterations following al-
Plasma and lipids from stored packed red blood cancer surgery. Lancet 1993; 342:1328 –1333
logenic blood transfusion. J Investig Med
cells causes acute lung injury in an animal 44. Jensen LS, Andersen AJ, Christiansen PM, et
1998; 46:161–167
model. J Clin Invest 1998; 101:1458 –1467 al: Postoperative infection and natural killer
38. Wyman TH, Bjornsen AJ, Elzi DJ, et al: A two- cell function following blood transfusion in 51. Sousa R, Salinas JC, Navarro M, et al: Autol-
insult in vitro model of PMN-mediated pulmo- patients undergoing elective colorectal sur- ogous blood transfusion as an immuno-
nary endothelial damage: Requirements for ad- gery. Br J Surg 1992; 79:513–516 modulator in experimental sepsis. Int J Surg
herence and chemokine release. Am J Physiol 45. Marik PE, Sibbald WJ: Effect of stored-blood Investigs 2000; 1:365–371
Cell Physiol 2002; 283:C1592–C1603 transfusion on oxygen delivery in patients 52. Montgomery A, Stager M, Carrico C: Causes
39. Taylor RW, Manganaro L, O’Brien J, et al: Im- with sepsis. JAMA 1993; 269:3024 –3029 of mortality in patients with the adult respi-
pact of allogenic packed red cell transfusion on 46. Kaplan J, Sarnaik S, Gitlin J: Diminished ratory distress syndrome. Am Rev Respir Dis
nosocomial infection rates in the critically ill helper/suppressor lymphocyte ratios and nat- 1985; 132:485– 489
patient. Crit Care Med 2002; 30:2249 –2254 ural killer activity in recipients of repeated 53. Bone R, Sibbald W, Sprung C: The ACCP-
40. Purdy FR, Tweedale MG, Merrick PM: Asso- blood transfusions. Blood 1984; 64:308 –310 SCCM consensus conference on sepsis and
ciation of mortality with age of blood trans- 47. Shorr AF, Duh MS, Kelly KM, et al: CRIT organ failure. Chest 1992; 101:1481–1483

1198 Crit Care Med 2005 Vol. 33, No. 6