Adjustment for confounding factorsFrom a univariate analysis, a list of biologically plausible and statistically significant this explanation confounders were identified, including severity of illness (APACHE III score), leukodepletion status, pre-ICU transfusions, cardiac surgery, other transfused blood components, and pretransfusion hemoglobin concentration preceding the first transfusion. We further adjusted for clustering of study sites. The APACHE III scores were first obtained by linkage of the study database with the ANZICS Adult Patient Database and were available for 432 study patients. Second, multiple requests for the missing APACHE III scores were sent retrospectively to the study sites, ending up with 713 surviving patients (94.2%) and 141 out of 146 patients who died (96.

6%) with an APACHE III score (compared with <1% of missing values in other study data). Hospital discharge status was re-checked at the same time.Finally, given a possible relationship between exposure to older blood and increased mortality, we sought to further explore this relationship. A series of binomial variables were created for each possible maximum age of blood (<2days, <3days, and so forth), and a cumulative graph was plotted indicating the mortality rate for each binomial cut-off point. To visually show the relationship between mortality and the maximum age of red blood, we also provided a plot of the predicted risk of death (as derived from the multivariate logistic regression model) against the maximum age of RBCs, and a locally weighted nonparametric smoother (LOWESS) was fitted to the data.

LOWESS fits simple models to localized subsets of the data to build up a function that describes the deterministic part of the variation in the data, point by point.Statistical analysisStatistical analysis was performed using SAS version 9.1 (SAS Institute Inc., Cary, NC, USA). Descriptive statistics were computed separately for all study variables for all study patients. Univariate analysis was performed using chi-square tests for equal proportions, Student t-tests for normally distributed outcomes and Wilcoxon rank-sum tests otherwise, with results reported as percentages (n), means (standard errors), or medians (interquartile ranges). The results from logistic regression analysis were reported as odds ratios (ORs) (95% confidence interval (CI)). Two-sided P = 0.

05 was considered statistically significant.Multivariate logistic regression models were constructed using both stepwise selection and backward elimination procedures Dacomitinib with statistically significant covariates (P < 0.05) remaining in the model. Models included the identified list of covariates firstly using the maximum age of blood as a continuous variable and then secondly as a predetermined categorical variable in quartiles.