The CRIT Study: Anemia and Blood Transfusion in the Critically Ill - Current Clinical Practice in the United States

Howard L. Corwin, MD; Andrew Gettinger, MD; Ronald G. Pearl, MD, PhD; Mitchell P. Fink, MD; Mitchell M. Levy, MD; Edward Abraham, MD; Neil R. MacIntyre, MD; M. Michael Shabot, MD; Mei-Sheng Duh, MPH, ScD; Marc J. Shapiro, MD


Crit Care Med. 2004;32(1) 

In This Article


The study was a prospective, multiple center, observational study of ICU patients in the United States. The enrollment period was August 2000 through April 2001. Patients were enrolled within 48 hrs of ICU admission. Inclusion criteria included: age of ≥18 yrs; admission to a medical, surgical, or combined medical-surgical ICU; anticipated ICU stay of >48 hrs; and informed consent. Exclusion criteria included: admission to a pediatric, cardiothoracic, cardiac, neurologic, or burn ICU; renal failure on dialysis; patients prohibited from receiving RBC transfusions; and patients involved in other transfusion research protocols. Patients were followed for either 30 days or until hospital discharge or death if these occurred before day 30. The study protocol was approved by the institutional review board of each participating institution.

Data collected included: hospital and ICU characteristics; patient demographics; admitting diagnostic categories; co-morbidities; ICU admission Acute Physiology and Chronic Health Evaluation (APACHE) II score; ICU admission and weekly Sequential Organ Failure Assessment (SOFA) scores; RBC transfusions; age of each RBC unit transfused; baseline (value closest to enrollment), weekly, and pretransfusion hemoglobin levels; mortality; ventilator days; ICU and hospital LOS; and clinical complications (see "APPENDIX 1" for definitions).

The primary end point of the study was to quantify the RBC transfusion practice in critically ill patients. The secondary end point was to describe the clinical outcomes and complications associated with anemia and RBC transfusions in these patients. SAS PROC MEANS procedure (SAS Institute, Cary, NC) was used to analyze the mean, standard deviation, and median of continuous variables, such as age and hemoglobin level. The significance of differences between two continuous measurements was determined by Student's t-test. Analysis of variance (ANOVA) was used when more than two measurements were compared. Results are presented as mean ± SD. SAS PROC FREQ procedure (SAS Institute) was used to tabulate the frequencies of categorical variables, such as number of transfusions and number of complications. Chi-square tests were used to test statistical significance. Pearson's correlation coefficients were used to assess the degree of linear correlation between two continuous variables. A two-sided alpha error of <.05 was considered to indicate statistical significance. Bonferroni adjustment was used when multiple comparisons were made.

Accelerated failure time models were used (PROC LIFEREG procedure, SAS Institute) to assess the factors associated with ICU LOS or hospital LOS. Adjustment was made for potential confounding factors, including patient demographic characteristics, RBC transfusion, nadir hemoglobin level or baseline hemoglobin level, the difference between the maximum and minimum hemoglobin values, mean age of blood transfused, mechanical ventilation status, baseline APACHE II and SOFA scores, origin of admission (e.g., emergency room, operating room), admitting diagnoses, and medical history. The median ICU and hospital LOS by transfusion status were generated conditional on the average values of other covariates in the model.

Mortality was analyzed and presented using two different models. First, logistic regression (PROC LOGISTIC procedure, SAS Institute) was used to examine transfusion and covariate effects after controlling for the duration on study. In a further confirmatory analysis of transfused patients, a Kaplan-Meier survival analysis and log-rank test (PROC LIFETEST procedure, SAS Institute) was performed, after 1:1 matching of transfused patients with nontransfused patients using propensity scores technique. Because the assignment of transfusion vs. no transfusion could not be randomized, potential selection bias was addressed by developing a propensity score for transfusion. Baseline attributes, including patient demographics, baseline APACHE II and SOFA scores, origin of admission, admitting diagnoses, medical history, and hospital LOS, that are potentially associated with receiving a transfusion were gathered into a single composite-predicted probability in logistic regression that summarized the likelihood for a patient with a given set of characteristics to receive a transfusion. A transfused patient was then matched to a nontransfused patient who had similar propensity (i.e., conditional probability) to receive a blood transfusion, using a greedy matching method. In this study, a subcohort of 44.8% of transfused patients had a match from the nontransfused patients. The remaining transfused patients with whom none of the nontransfused patients had a similar propensity to match were excluded from the propensity score analysis because their great baseline differences from the nontransfused patients hampered the ability to investigate the independent effect of transfusion on mortality.