Racial disparities in the use of blood transfusion in major surgery

The source of data was the University HealthSystem Consortium (UHC) database (January 2009 - September 2011). The time frame was chosen to represent the most contemporary data available and to minimize the effect of transfusion practice change over time. The UHC is a national alliance of academic medical centers and their affiliated hospitals. The UHC database consists of administrative records for all hospitalizations in its member institutions. This represents about 90% of the nation’s hospitalizations at academic medical centers. The UHC database includes patient demographic characteristics, admission status (emergent, urgent, elective, and trauma), ICD-9-CM diagnostic codes, ICD-9-CM procedure codes, present-on-admission (POA) codes, 3 M APR-DRG severity of illness, AHRQ comorbidities [39], in-hospital death, cost, and detailed blood use information. The race information in the UHC database is comprehensive and complete. A unique strength of the UHC database is that it contains full inpatient transfusion information [40]. The UHC dataset also includes encrypted hospital identifiers. No specific patient, physician, or hospital was examined in the analysis, and the study proposal was granted exemption by the University of Rochester Research Subjects Review Board.

Black or white patients who received one of the following three surgical procedures were included in the analysis: isolated coronary artery bypass surgery (CABG), isolated total hip replacement (THR), or isolated colectomy. We used two steps to identify the patients who underwent isolated surgery. First, we used ICD-9-CM procedure codes to identify CABG (36.10 - 36.19), THR (81.51), and colectomy (45.73-45.76). Second, we applied the Healthcare Cost and Utilization Project (HCUP) clinical classification software (CCS) for ICD-9-CM (available at: http://​www.​hcup-us.​ahrq.​gov/​toolssoftware/​ccs/​ccs.​jsp) to identify the patients who underwent primary isolated CABG, THR, or colectomy surgery during the hospitalization. These procedures were selected because they are commonly performed major surgical procedures representing cardiac surgery, orthopedics, and general surgery; and patients undergoing these procedures have relatively high transfusion rates. We excluded patients whose age was less than 18 years old and patients whose important demographic characteristics (e.g., gender) or important healthcare information (e.g., procedure date, transfer out status, transfer in status, admission POA indicator, and admission status) were missing. The final study sample comprised 25,849 isolated CABG patients, 42,933 isolated THR patients, and 8,255 isolated colectomy patients in 87 hospitals (see Figures 1, 2, 3).

Baseline characteristics were summarized as frequency, percentage, median (inter- quartile range) as appropriate for each racial group for each surgical procedure. Comparison between racial groups (white vs. black) were made using Chi square test or Fisher exact test for categorical variables, and t test for continuous variables (see Table 1).

The outcome variable was whether a patient received any perioperative RBC transfusion. A patient-level binary variable was coded as 1 if the patient received RBC transfusion perioperatively and as 0 otherwise. The primary independent variable of interest was race (white vs. black). The unit of analysis in the study was the patient.

To examine racial disparities in blood transfusion, a series of multivariable logistic regression models for each surgical procedure were fit to the patient-level data in a sequential manner. The baseline model (model 1) examined the association between race and transfusion, controlling for patient demographics (age and gender), admission status, APR-DRG severity of illness, POA status, and AHRQ comorbidities (including anemia). To determine whether black-white differences in transfusion persisted within payer groups, we interacted race and payer status (model 2).

In order to examine whether the source of black-white differences in blood transfusion was due to between-hospital or within-hospital effects, we estimated two additional models. In the first model, we added dummy variables representing the hospital quartile of black patient concentration to the baseline model (model 3). In the second model, we added hospital fixed effects to the baseline model (model 4). Robust variance estimators were used because patient outcomes within the same hospital may be correlated [41]. Data management, analyses, and regression diagnostics were performed using Stata MP version 11.0 (StataCorp, College Station, TX). All statistical tests were 2-sided and P < 0.05 was considered significant.

Confronting Racial and Ethnic Disparities in Healthcare summarized the striking finding that blacks tended to receive lower quality healthcare across a wide range of clinical services. For example, blacks were more likely to receive lower-extremity amputations [45]. This influential report also concluded that the racial differences may stem from patient characteristics (e.g., minority patients may be more likely to delay seeking healthcare), clinical encounters (e.g., medical uncertainty can “open the door” for physicians’ stereotypes and biases to affect their clinical judgment of minority patients), and health systems (e.g., fragmented healthcare delivery systems present cultural and linguistic barriers for minority patients). In response, numerous high-profile initiatives at both national and local levels were designed and implemented to mitigate or eliminate racial and ethnic disparities in healthcare. However, the 2010 National Healthcare Disparities Report demonstrated that racial disparities in healthcare have persisted despite these efforts [15].

The stubborn persistence of racial disparities over time may, in part, reflect a lack of understanding of the underlying causes of racial disparities in healthcare. More recently, a new explanation for racial disparities in health care has drawn great attention because it leads to different policy implications to overcome disparities. Supported by empirical studies, this explanation proposes that racial and ethnic disparities are determined by both discrimination (i.e., “within provider variation”, “who you are”) and segregation (i.e., “across provider variation”, “where you live”) [2, 7, 12, 42, 46]. The former highlights the importance of race in the clinical encounter, whereas the latter relates more to geographic variations in the quality of care patterns for all patients. Using this framework, new studies have found that some types of racial disparities in healthcare were largely caused by discrimination [47, 48] while others were mainly attributable to segregation [21, 49]. To reduce the former type of racial disparities, efforts to improve patient-physician communication and to enhance “patient-centered” care during the clinical encounter are recommended. These efforts include physician cultural competency training, expansions in the numbers of minority physicians in the hospital, hospital’s adoption of patient-centered information technology, and hospital’s efforts to improve effective communication and to promote “communities of care” [45, 50‐53]. To reduce latter type of racial disparities, interventions to improve performance of particular hospitals which served disproportionately high concentrations of minority patients but provided suboptimal quality of care are needed. These interventions could involve system-level quality initiatives such as pay for performance to incentivize quality improvement in low-quality black-serving hospitals [49, 54].

Our findings of racial disparities in transfusion in isolated CABG and THR surgeries not only fill the knowledge gap of contemporary transfusion practices but also suggest the main cause of such disparities - discrimination (“within hospital variation”) during the clinical encounter. This may reflect, in part, complexity of the medical decision making process underlying transfusion decisions. The decision to transfuse a patient can be complex (whether, what, and how much to transfuse) and can involve more than one physician (surgeon, anesthesiologist, critical care medicine physician). Some of the variation in transfusion practices may be due to variation in surgical blood loss, and may account for some of the racial variation in the use of blood transfusion. Surgical procedure time, surgeon technical skill, case complexity, and anesthetic management can contribute to the variation in operative blood loss in complex ways. For example, Silber et al reported that black race was associated with increased anesthesia time and significantly longer procedure time in noncardiac surgery [55]. Our results also indicate that patient insurance coverage and racial segregation (“across hospital variation”) were not major contributors to the detected racial disparities in blood transfusion. Therefore, improving insurance coverage and focusing on black-serving hospitals only might have very limited power in reducing racial disparities in blood transfusion.

Our findings have important policy implications. First, as blood transfusion is gaining recognition as an important domain of quality of care, our study suggests that efforts to monitor and reduce racial disparities in transfusion practices are necessary and important. Since blood transfusion is widely employed in surgery, improving quality of care in transfusion therapy with a particular focus on minorities may lead to improved perioperative outcomes. Second, given that these racial gaps are largely due to “within hospital variation”, interventions with the goal of improving patient-physician communication and enhancing patient-centered health care are more likely to be effective and successful in narrowing these disparities. Fueled by the Patient Protection and Affordable Care Act’s emphasis on patient preferences and medical decision making, there is growing enthusiasm in promoting patient-centered care and addressing racial and ethnic disparities. Third, our findings provide further empirical evidence of black-white disparities in health care. Historically, minorities tend to receive fewer interventions associated with improved health outcomes (e.g. PCI for acute myocardial infarction) [4, 5]. However, in some cases, black patients are more likely to undergo interventions suggestive of less than optimal management of chronic diseases (e.g. it was also reported that black patients had significantly higher rates of lower-extremity amputations) [14, 15]. Since blood transfusion is frequently associated with higher risk of mortality and complications in surgical patients, our findings that black patients undergoing surgery are more likely to receive blood transfusion suggests there is a need to examine the clinical appropriateness of transfusion therapy in minority patients [56, 57].

Our study has significant limitations. Most importantly, preoperative hematocrit values were not available in the UHC database. We were able to control only for the presence but not for the severity of anemia. Thus it is likely that some of the black-white differences in blood transfusions may be due to unmeasured variation in patient preoperative hematocrits. Second, it is possible that a small portion of blacks could be misclassified as whites and vice versa [58]. Thus, our results could be biased towards the null and may underestimate the actual racial disparities in blood transfusion. Third, the present study is based on an administrative database. Therefore, it shares drawbacks of using administrative datasets in evaluating quality such as inability to specify clinical definitions for risk factors (e.g., forced reliance on ICD-9 codes), and limited ability to distinguish between complications of care and pre-existing conditions. Fourth, we cannot rule out the possibility of residual confounding because of other unmeasured risk factors which could account for some of the black-white differences in blood transfusion. Fifth, the blood usage information in the UHC database has not been audited. Finally, because only academic hospitals are included in the UHC database, our findings are not necessarily generalizable to non-academic centers.