We estimated the hospital costs associated with bleeding-related complications in 8 surgical cohorts, including cardiac, vascular, non-cardiac thoracic, solid organ, general surgery, knee/hip replacement, reproductive organ, and spinal surgery, using patient-level data from Premier's PCD. Patients undergoing the surgical procedures of interest during calendar years 2006 and 2007 were identified using ICD-9-CM procedure codes. For each surgical cohort, patients with bleeding-related complications and/or blood product transfusions and those without these events were further delineated using a combination of diagnosis and procedure codes. Mean hospital costs were calculated for each study group after adjusting for differences in baseline demographic and clinical characteristics deemed to have an impact on costs. The incremental difference in adjusted mean costs between patients with a bleeding event and/or blood product transfusion and those without these events was assumed to be an estimate of the hospital costs attributable to having a bleeding event and/or blood product transfusion.
Results of our analyses indicate that bleeding events and/or blood product transfusions were relatively common during the surgical hospitalization and varied according to surgical cohort (general: 27.5%; cardiac: 47.4%; solid organ: 28.5%; non-cardiac thoracic: 34.3%; vascular: 31.5%; knee/hip replacement: 29.8%; reproductive organ: 7.5%; and spinal surgery: 15.0%). Among the types of events identified in this study, blood product transfusions were the most common, occurring in approximately 21.2% of the patient cohort. Interventions for bleeding control, in contrast, were less common and occurred in only 2.5% of patients.
In analyses of costs adjusted for disparities in baseline and clinical characteristics, we observed large variations in the incremental difference in mean total costs between patients with a bleeding-related complication and/or blood product transfusion and those without these events for each surgical cohort. Patients with bleeding-related complications and/or blood product transfusions experienced longer hospital lengths of stay and spent more time in the ICU compared to patients without these events.
Studies investigating costs related to bleeding-related complications reveal that they vary widely depending on the surgical cohort. For example, hospital costs attributable to bleeding in trauma patients are reported to be much higher ($38,628) compared to patients undergoing more routine procedures such as PCI ($5,883) and knee and hip surgery ($7,593) [21–23]. Our inpatient cost estimates attributable to bleeding-related complications and/or blood product transfusions ranged from $2,805-$17,279 for reproductive and spinal surgery, respectively. Considering that the surgical procedures examined in this study were less complex compared to trauma surgery, it makes sense that our estimates are lower than what is previously reported for trauma patients.
This study is subject to several limitations. First, the identification of patients who had a bleeding event and/or blood product transfusion was based on an algorithm utilizing a combination of ICD-9-CM procedure and diagnosis codes as well as billing charges appearing on the hospital record. These coding systems are primarily used for administrative purposes in obtaining reimbursement for the services provided by the hospital. Therefore, we chose to incorporate both bleeding events as well as blood product transfusions in our definition of the study groups. This distinction is important because, with certain surgical procedures, blood product transfusions are routinely administered as part of the surgery and not necessarily because the patient had unexpected bleeding requiring transfusion. Thus, we felt we could not have reliably distinguished one transfusion type from the other. Additionally, it was not possible to differentiate between transfusions occurring as a consequence of the surgery from transfusions that were the result of an underlying or presenting condition such as vascular trauma or ruptured aortic aneurysms if the transfusion occurred on the same day of the surgery as only data pertaining to the day of the transfusion were available. Therefore, it is possible that some patients within the bleeding-event and/or blood product transfusion group had a transfusion that did not occur as a consequence of the surgery and were misclassified. We did not exclude patients with conditions such as vascular trauma or ruptured aortic aneurysms from this analysis in order to facilitate bleeding occurring as a consequence of the surgery. Patients presenting with aortic aneurysms probably would have undergone either open aortic resection with replacement (38.44) or endovascular repair (39.71). The proportion of vascular surgery patients undergoing open aortic resection with replacement was only 1.8% and patients undergoing endovascular repair were not selected for this analysis. Therefore, we do not expect the inclusion of patients with ruptured aortic aneurysm and the possible misclassification of the bleeding outcome to change results significantly. With respect to vascular trauma, the inclusion of trauma cases that likely had a transfusion as a result of their presenting condition and not the surgical procedure probably introduced some misclassification bias into this study. Since trauma patients have much higher costs relative to patients without trauma, the inclusion of these cases would have had the effect of inflating our bleeding-related complication cost estimates . We suspect that this upward bias is probably minimized by the fact that patients presenting with trauma probably represented only a small percentage of the total patients included in this study. A study from the University of Michigan comparing outcomes between trauma and general surgery patients reported enrolling only 525 patients admitted to the Trauma service in comparison to 54, 478 general surgery patients during 2004 .
Second, patients were excluded from analyses if they were transferred from another hospital or an unknown source to ensure that complete hospitalization data were available for every patient. We did not feel that we could reliably identify and link multiple hospitalizations for patients transferred to another hospital using the administration codes in the database. As these transferred patients probably had higher costs relative to the current sample of patients, the exclusion of these patients likely had the effect of underestimating the total inpatient episode of care. Furthermore, if more patients with a bleeding-related complication were transferred to other hospitals versus those without bleeding-related complications, the incremental difference in costs between these groups may in fact be larger than what is currently reported.
Third, pediatric patients < 18 years of age were included in analyses. Although age was included as a covariate in multivariate cost models, one could argue that, because pediatric patients likely underwent different procedures compared to adult patients, separate analyses of these distinct patient groups is warranted. Thus, multivariate cost models examining adult and pediatric patients separately were constructed to provide insight regarding the differences in costs between these two distinct subpopulations. Among adults, patients with bleeding-related complications had higher costs relative to patients without complications (cardiac: $38,686 vs. $28,914, vascular: $30,640 vs. $16,027, non-cardiac thoracic: $36,150 vs. $23,494, solid organ: $31,807 vs. $18,878, general surgery: $18,880 vs. $14,427, knee/hip replacement: $18,248 vs. $15,247, reproductive organ: $9,269 vs. $6,493, and spinal: $37,978 vs. $20,847); results were similar to main analyses and statistically significant. Among the subset of pediatric patients, those with bleeding-related complications had higher costs relative to patients without complications (cardiac: $58,239 vs. $29,514, vascular: $113,822 vs. $39,506, non-cardiac thoracic: $79,898 vs. $35,680, solid organ: $61,122 vs. $28,742, general surgery: $104,505 vs. $37,316, reproductive organ: $34,703 vs. $16,999, and spinal: $54,369 vs. $31,984). A multivariate model was not created for pediatric knee/hip surgery patients because of the small sample size relative to the number of parameters in the model. Overall, costs in patients < 18 years of age were higher compared to the adult population. The incremental difference in costs between patients with bleeding-related consequences and those without bleeding-related consequences was also higher in pediatric patients compared to adults.
Fourth, our analyses of costs were adjusted for differences in baseline clinical and demographic characteristics using multivariate ordinary least squares regression with Duan's smearing back retransformation. The approach used for retransformation should be dependent upon the nature of the error term on the transformed scale . Since the distribution of the error term is usually unknown, reliance on the assumption of normality or homoskedasticity can lead to inconsistent estimates . Therefore, we also examined the sensitivity of results to retransformation using subgroup-specific smearing factors as proposed by Manning . Using this alternate retransformation method, cost estimates for patients with bleeding-related complications remained statistically significantly higher versus those without complications. In fact, the incremental differences in mean total costs (bleeding-related complication - no complication) for each of the surgical subgroups were higher compared to the main study results (cardiac: $17,055, vascular: $25,795, non-cardiac thoracic: $30,963, solid organ: $25,153, general: $16,567, knee/hip replacement: $4,009, reproductive organ: $7,707, and spinal surgery: $21,326).
It should be noted that a small percentage of patients (approximately 5%) were included in more than one study subgroup because they were operated on in surgical sites spanning multiple surgical categories during the index hospitalization. A sensitivity analysis excluding these patients was conducted to examine the effect they might have had on cost results. Results show that their effect was minimal as the incremental difference in mean total costs (bleeding-related complication - no complication) were similar to the main analyses for each of the surgical subgroups (cardiac: $10,391, vascular: $14,639, non-cardiac thoracic: $12,366, solid organ: $12,434, general: $3,974, knee/hip replacement: $3,148, reproductive organ: $3,266, and spinal surgery: $17,448).
All baseline and clinical characteristics deemed to have an impact on total hospital costs were included in our models. Nevertheless, we could not control for certain covariates known to affect costs because this information was not available in the PCD. These covariates included sociodemographic parameters such as smoking status and whether the patient was living alone. Although we identified patients who were obese using ICD-9-CM diagnosis codes (278.00-278.02), to the best of our knowledge it is unknown to what extent these codes can be used to accurately identify individuals who are obese. Additionally, data on certain chronic comorbid conditions were likely underestimated. Comorbidity data were only available if a patient was treated for the condition during the index hospitalization or if a patient was treated in a prior admission to a hospital within Premier's data capture network. Because data on outpatient physician visits are not included in the PCD, for some comorbidities including thrombocytopenia, MI, hypertension, non-MI coronary disease, renal disease, congestive heart failure, and deep vein thrombosis it was impossible to determine whether the condition was chronic and unrelated to the index surgical hospitalization or occurred as an adverse event related to the surgery or bleeding complication. Many patients also did not have prior hospitalization data, therefore many comorbidities were identified within the index surgical hospitalization. As a result, the assumption of independence between covariates in our multivariate analyses may have been violated if, for example, certain conditions such as MI or renal failure developed as a result of the bleeding complication and were not actually pre-existing conditions. As a result regression models could have produced inaccurate estimates of regression coefficients, variability, and P-values. Therefore, multivariate cost analyses were also run using only the conditions that were deemed to be obvious chronic conditions identifiable in the database. These conditions included diabetes, obesity, COPD, cancer, and cirrhosis. The incremental cost per hospitalization associated with bleeding-related complications was similar to the main study results ($15,931 for vascular, $15,410 for solid organ, $14,653 for non-cardiac thoracic, $11,715 for cardiac, $5,114 for general, $3,119 for knee/hip replacement, $2,961 for reproductive organ, and $17,707 for spinal surgery) and the difference in costs among patients with bleeding-related complications and those without complications was statistically significant for every surgical cohort (P <0.001). We also did not control for conditions such as coagulopathy, anemia, or the use of bone marrow suppressants. If a higher percentage of patients in the bleeding complication group had these prior conditions or procedures, our cost estimates related to bleeding complications may have been overstated especially if a prior hospitalization for these conditions or procedures was associated with higher costs during the study index surgical hospitalization. Finally, information relating to the severity of the comorbid condition was not available. As severity could have an impact on hospital costs, the inability to control for severity is another limitation of our analysis.