Subjects were all fee-for-service Medicare beneficiaries (n = 24,789) who received CABG surgery (International Classification of Diseases, Clinical Modification, ninth edition (ICD-9) procedure codes 36.1×) from 2003 to the end of 2006. Only patients 65 years of age or older, who were Michigan residents or received their surgery in a Michigan hospital, were included. In this retrospective cohort study, patients were followed from hospital admission to 30 days after hospital discharge. Inpatient standard analytical files and denominator files were obtained from the Centers for Medicare and Medicaid Services (CMS), and contained information regarding hospitalisations and Medicare enrolment.
There were two main areas of investigation. The first was the evaluation of differences in transfusion use and infection rates in hospitals and to calculate the intraclass (that is, intrahospital) correlation coefficient. The second was the assessment of the relationship between transfusion and patient outcomes. The primary outcome was infection during hospitalisation. Secondary outcomes were death (in-hospital and 30-days post discharge) and readmission to a hospital (for any reason and for reason of infection). For post-discharge outcomes, only those individuals who survived to hospital discharge were included in the analyses. Since infection was the primary outcome, we excluded those patients who were initially admitted for reason of infection (prior to the CABG procedure) and those with evidence of pre-existing infection (for example, acquired immunodeficiency syndrome) during the hospital stay when the CABG procedure occurred. This constituted 0.4% of the sample (n = 115 patients).
Data regarding blood transfusions were extracted from procedure codes (99.0×), as well as revenue codes for blood products and services (38× for purchased blood and 39× for donated blood). For purposes of these analyses, the receipt of an allogeneic transfusion could have included any of the following components at any time during hospitalisation: red blood cells, whole blood, platelets, plasma or cryoprecipitates. The use of autologous blood (where donor and recipient were the same individual) was also obtained from two procedure codes (99.00: perioperative autologous transfusion of whole blood or blood components; 99.02: transfusion of previously collected autologous blood).
We determined infection by using ICD-9 codes that explicitly stated infection (for example, 0xx.xx) or provided evidence of infection (purulent, suppurative, septic, pyogenic or abscess). Data were also extracted regarding age, gender, race, secondary diagnoses, type of admission (elective, urgent, emergency), and length of stay. Less than 1% of values for race and type of admission were missing and were imputed using best subset regression. We examined race at both the patient and hospital levels; specifically, for purposes of this investigation, hospitals were classified as African-American if ≥ 50% of the patients who received CABG surgery annually were African-American.
Surgeon volume was determined by summing the number of Medicare CABG procedures per operating physician, calculating the annual mean, and categorising into 2 equal groups based on the median number of cases per year (60 CABG procedures/year). Hospital volume was determined by summing the number of Medicare CABG procedures and calculating the annual mean. We then categorised hospitals into 2 equal groups based on the median number of cases per year (240 CABG procedures/year). For the analyses of hospital measures and intraclass correlation coefficients, the analyses were restricted to those hospitals that performed at least 50 CABG procedures (n = 40 hospitals).
Patient characteristics were evaluated first by receipt of allogeneic blood transfusion. Bivariate associations were assessed using Pearson χ2
tests for categorical data and the Wilcoxon rank sum test for differences in median length of hospital stay. Multilevel mixed-effects logistic regression was used to evaluate the associations between transfusion and study outcomes (in-hospital infection, 30-day readmission, 30-day mortality). A two-level hierarchical model was used in which patients were nested within hospitals. The hospital was modelled as a random intercept with transfusion included as a fixed effect. The structure of the covariance matrix for the random effect was specified using the identity structure (uncorrelated random effects with common variance). In postestimation, predicted probabilities were calculated based on the linear predictor of both fixed and random effects. The intraclass correlation coefficient for the multilevel logistic model was calculated as described by Snijders and Bosker [9
In order to address the possible confounding effect of comorbid conditions, propensity scores were calculated. Specifically, we estimated the propensity for each person to receive a transfusion in order to address the possibility that recipients of blood transfusion had more underlying illnesses than those not receiving transfusions. The probability of receiving an allogeneic blood transfusion was based on the predicted values generated from logistic regression using the following covariates: age, gender, race, type of admission (elective, urgent, emergency), congestive heart failure, diabetes mellitus, renal failure, hypertension, chronic pulmonary disease, malignancy, peripheral vascular disease, cerebrovascular disease, and myocardial infarction (area under the receiver operating characteristic curve = 0.7368). The scores were categorised into deciles. Mean propensity scores were not different among patients transfused and not transfused within each block. In addition to adjustment for propensity decile, all results controlled for surgeon volume and hospital volume. The α was set at 0.05, and all tests were two-tailed. Stata/SE 10.0 software was used for all analyses (Stata, College Station, TX, USA).
This study was approved by the Institutional Review Board on Human Subjects at the University of Michigan at Ann Arbor and by the Privacy Review Board at CMS.