Blood Transfusion in Major Abdominal Surgery for Malignant Tumors: A Trend Analysis Using the National Surgical Quality Improvement Program

Importance  Blood transfusion can be a lifesaving treatment for the surgical patient, yet transfusion-related immunomodulation may underlie the association of allogeneic transfusion with increased perioperative morbidity and possibly poorer long-term oncologic outcomes.

Objective  To evaluate trends in transfusion rates for major abdominal oncologic resections to assess changes in recent clinical practice (given the accumulating evidence of the deleterious effects of blood transfusion).

Design, Setting, and Participants  Retrospective review of a population-based registry of all hospitals participating in the American College of Surgeons National Surgical Quality Improvement Project (2005-2013 Participant Use Data Files), which was queried for patients who underwent major resection of a pancreatic, hepatic, or gastric malignant tumor. Data analysis was performed from July to August 2015.

Main Outcome and Measures  The primary outcome was the transfusion of any quantity of packed red blood cells. Transfusion rates were calculated for the perioperative period, which was defined as the time from the start of surgery to 72 hours after surgery. Secondary outcomes included wound infection, myocardial infarction, and renal insufficiency, and the rates of these complications were calculated as well. Trend analysis was performed for each year of data to evaluate for changes over the study period.

Results  A total of 19 680 patients (median age, 65.0 years [interquartile range, 57.0-73.0 years]) were identified, of whom 5900 (30.0%) received a blood transfusion (of 13 657 patients who underwent a pancreatic resection, 4074 required transfusion [29.8%]; of 1605 patients who underwent a gastric resection, 378 required transfusion [23.6%]; and of 4418 patients who underwent a hepatic resection, 1448 required transfusion [32.8%]). There was a significant trend toward decreasing rates of transfusion during the study period (z = −7.89, P < .001), which corresponded to an absolute 6.1% decrease in the rate of transfusion of packed red blood cells from 2005 to 2013 (ie, from 32.8% to 26.7%). There was no significant change in the rates of postoperative wound infection or renal insufficiency during this time period, but there was an increased rate of perioperative myocardial infarction during the study period (0.33% absolute increase; z = 3.15, P = .002).

Conclusions and Relevance  Over 9 years of contemporary practice, a trend of less perioperative blood transfusions for oncologic abdominal surgery was observed. Further studies are needed to assess whether these trends reflect changes in operative techniques, hospital cohorts, or transfusion thresholds.

Over the course of the last decade, data suggesting deleterious effects of allogeneic blood have prompted efforts to limit erythrocyte transfusion.^1,2 The Transfusion Requirements in Critical Care trial³ was the first large randomized, controlled clinical trial to establish that lower transfusion thresholds improve clinical outcomes in critical care patients. In the specific case of the surgical patient, several nonrandomized studies^4-6 have confirmed the association between blood transfusion and perioperative morbidity, although no significant differences in major morbidity or mortality have been observed in randomized trials evaluating restrictive transfusion strategies for surgical patients.^7-10

The relative risks and benefits of allogenic blood transfusion are particularly complex for the oncological surgery patient.^11,12 Transfusion-related immunomodulation has been recognized for more than 30 years, when it was observed that renal allograft survival was improved with preoperative blood transfusion.¹³ Whether the immunosuppressive effects of allogenic transfusion are associated with an increased risk of cancer recurrence has not been definitively established, although associations in a variety of abdominal malignant tumors have been observed.^14-17

Given the momentum in the literature for more judicious use of blood transfusion, we evaluated national trends in transfusion rates for major abdominal oncologic resections. We also evaluated concurrent trends in rates of perioperative wound infection, renal failure, and myocardial infarction (MI) to determine whether changes in utilization of transfusion were associated with an increased or decreased rate of perioperative complications.

The American College of Surgeons National Surgical Quality Improvement Project (ACS NSQIP) is a nationally validated, outcomes-based, risk-adjusted, peer-controlled program for the measurement and enhancement of the quality of surgical care.¹⁸ Using selected diagnosis and procedure codes from the International Classification of Diseases, Ninth Revision (ICD-9) and Current Procedural Terminology (CPT), we identified patients 18 years of age or older who underwent 1 of 3 cancer operations from 2005 through 2013 in the NSQIP data files: pancreatectomy (CPT codes 48140, 48145, 48146, 48148, 48150, 48152, 48153, 48154, 48155, and 48160), total gastrectomy (CPT codes 43620, 43621, and 43622), or formal hepatic resection (CPT codes 47122, 47125, and 47130). Because procedure indication is not available in the ACS NSQIP, specific abdominal oncologic resections were narrowly defined. Each case included an organ-specific resection performed during the same admission as the diagnosis of a malignant tumor at the site (pancreas: ICD-9 code 157.x; stomach: ICD-9 code 151.x; or liver: ICD-9 codes 155.0, 155.1, 153.x, 154.x, and 197.7). Liver wedge resections and any liver resection performed with concurrent bowel resection were excluded.

The primary outcome was the transfusion of any quantity of packed red blood cells (PRBCs). Transfusion rates were calculated for the perioperative period, which was defined as the time from the start of surgery to 72 hours after surgery. Secondary outcomes included rates of wound infection, renal insufficiency, and MI. Wound, or surgical site, infection was defined as the occurrence of a superficial, deep incisional, or organ space surgical site infection within 30 days from the time of the procedure. Patients with an existing infection at the surgical site at the time of surgery were excluded from this analysis. Myocardial infarction was diagnosed if there were changes on the electrocardiogram indicative of infarction and/or if there was a new elevation in serum troponin of more than 3 times the upper level of the reference range in the setting of suspected myocardial ischemia, occurring within 30 days of surgery. Renal insufficiency was diagnosed if the kidney showed a reduced capacity to perform its function, as evidenced by an increase in creatinine of more than 2 mg/dL (to convert to micromoles per liter, multiply by 88.4) from the preoperative level and/or the initiation of renal replacement therapy within 30 days of surgery. Patients with preoperative renal failure or who received dialysis were excluded from this analysis. Our study was deemed exempt from continuing review by the institutional review board of the University of Pennsylvania in Philadelphia. Informed consent was not obtained because the data were deidentified.

Descriptive statistics were performed. Categorical variables were described as totals and frequencies; continuous variables were described as median values with interquartile ranges (IQRs). The Cuzick test for trend was performed to evaluate for annual changes over the study period from 2005 through 2013 in the rates of PRBC transfusion, wound infection, renal insufficiency, and MI. The results of trend analysis are presented as z scores, in which positive values indicate increasing trends, negative values indicate decreasing trends, and the absolute value denotes the magnitude of change. Univariate comparisons were assessed for variables clinically relevant to the studied outcomes (ie, PRBC transfusion, wound infection, MI, and renal insufficiency) using the χ² test or the Mann-Whitney U test, as appropriate. Blood transfusion was evaluated for associations with the clinical and demographic characteristics of the patients (ie, age, race, Hispanic ethnicity, American Society of Anesthesiologists [ASA] classification, Charlson Comorbidity Index score, history of bleeding disorder, and preoperative chemotherapy), preoperative laboratory values (ie, hematocrit level, platelet count, and coagulation test results), and type of procedure.

The occurrence of a surgical site infection was evaluated for associations with the clinical and demographic characteristics of the patient (ie, age, race, Hispanic ethnicity, ASA classification, Charlson Comorbidity Index score, obesity, diabetes, cigarette smoking, steroid use, and preoperative weight loss), preoperative laboratory values (ie, serum albumin level), type of procedure, surgical wound classification (ie, clean, clean-contaminated, contaminated, or infected), and the occurrence of perioperative blood transfusion. Postoperative MI was evaluated for associations with the clinical and demographic characteristics of the patient (ie, age, race, Hispanic ethnicity, ASA classification, Charlson Comorbidity Index score, history of cigarette smoking, angina, prior MI, previous cardiac surgery, previous percutaneous coronary intervention, and congestive heart failure), type of procedure, and the occurrence of perioperative blood transfusion.

Renal insufficiency was evaluated by univariate analysis for significant associations with the clinical and demographic characteristics of the patient (ie, age, race, Hispanic ethnicity, ASA classification, Charlson Comorbidity Index score), preoperative laboratory values (ie, serum creatinine level), type of procedure, and the occurrence of perioperative blood transfusion. Variables for which data were missing for more than 15% of the study cohort were excluded. Variables that demonstrated a trend toward significance by univariate testing (P ≤ .10) were entered into a forward, stepwise logistic regression model (P < .05 for entry; P > .10 for exit). All variables identified as significantly associated with the outcomes of interest by multivariable analysis were analyzed using the Cuzick test for trend to assess for changes over the study period. Analyses were performed using STATA software, version 13 (StataCorp).

A total of 19 680 patients were identified who underwent a major abdominal oncologic resection during the study period. Median patient age was 65.0 years (IQR, 57.0-73.0 years), and the majority were male (10 427 [53.0%]). The greatest proportion of patients underwent a pancreatectomy (13 657 [69.4%]), followed by a formal hepatic resection (4418 [22.4%]) and a total gastrectomy (1605 [8.2%]). The distribution of the 13 657 pancreatic resections was as follows: pancreatoduodectomy (10 128 [74.2%]), distal pancreatectomy (3217 [23.6%]), and total pancreatectomy (312 [2.3%]). There were no significant trends in the proportions of specific types of pancreatic resections over the 9 years of the study (P = .49). The distribution of 4418 hepatic resections was as follows: left-sided lobectomy (1087 [24.6%]), right-sided lobectomy (2239 [50.7%]), and trisegmentectomy (1092 [24.7%]). There were no significant trends in the proportions of specific types of liver resections over the 9 years of the study (P = .22).

A total of 5897 patients (30.0%) received a perioperative PRBC transfusion (of 13 657 patients who underwent a pancreatic resection, 4074 required transfusion [29.8%]; of 1605 patients who underwent a gastric resection, 378 required transfusion [23.6%]; and of 4418 patients who underwent a hepatic resection, 1448 required transfusion [32.8%]). The median time to transfusion was 0 days (IQR, 0-0 days) for the 3431 patients for whom the data were available. By univariate analysis, the following variables were associated with an increased rate of transfusion: increasing age (P < .001), female sex (P = .01), nonwhite race (P < .001), ASA classification of higher than 2 (P < .001), increasing Charlson Comorbidity Index score (P < .001), history of bleeding disorder (P < .001), a preoperative platelet count of less than 100 × 10³/µL (to convert to ×10⁹ per liter, multiply by 1.0) (P < .001), a preoperative international normalized ratio of 1.5 or higher (P < .001), type of procedure (P < .001), and decreasing preoperative hematocrit level (P < .001).

The following factors remained significantly associated with blood transfusion by multivariable regression modeling: increasing age (odds ratio [OR], 1.01 [95% CI, 1.01-1.01]), female sex (OR, 1.12 [95% CI, 1.05-1.21]), nonwhite race (white: OR, 0.82 [95% CI, 0.76-0.88]), ASA classification of higher than 2 (OR, 1.37 [95% CI, 1.26-1.50]), Charlson Comorbidity Index score of 2 or higher (OR, 1.21 [95% CI, 1.11-1.34]), a preoperative platelet count of less than 100 × 10³/µL (OR, 1.72 [95% CI, 1.33-2.21]), a preoperative international normalized ratio of 1.5 or higher (OR, 1.34 [95% CI, 1.04-1.72]), type of procedure (total gastrectomy: OR, 0.56 [95% CI, 0.48-0.65]; hepatic resection: OR, 1.36 [95% CI, 1.25-1.48]), and preoperative hematocrit level (increasing hematocrit level: OR, 0.88 [95% CI, 0.87-0.88]) (Table 1).

Despite trends of lower preoperative hematocrit levels (z = −2.70, P = .01) and increasing patient comorbidity (increasing patient age: z = 2.41, P = .02; ASA classification of >2: z = 11.3, P < .001; platelet count of <100 × 10³/µL: z = 2.11, P = .04), there was a significant trend toward decreasing rates of PRBC transfusion during the study period (z = −7.89, P < .001). This trend was driven primarily by the decreasing use of blood transfusion for pancreatectomy (z = −7.63, P < .001), with significant decreases for total gastrectomy (z = −2.50, P = .01) and major hepatic resection (z = −3.60, P < .001). There was a 6.1% decrease (from 32.8% to 26.7%) in the annual rate of PRBC transfusion overall for these major oncologic surgical procedures from 2005 to 2013 (Figure 1). The rate of blood transfusion decreased 8.9% (from 35.2% in 2005 to 26.3% in 2013) after pancreatic resections and 6.0% (from 25.0% in 2005 to 19.0% in 2013) after gastric resections from the first year of the study period to the last year. There was an 11.7% decrease (from 41.3% to 29.6%) in the annual rate of PRBC transfusion for major hepatic resections from 2006 through 2012.

The rate of surgical site infection in this cohort was 16.3% (of 13 450 patients who underwent a pancreatic resection, 2387 required transfusion [17.8%]; of 1590 patients who underwent a gastric resection, 241 required transfusion [15.2%]; and of 4367 patients who underwent a hepatic resection, 525 required transfusion [12.0%]). There was a significant association between PRBC transfusion and the incidence of wound infection (P < .001), as well as several other variables (Table 2). There was no significant trend in the annual rate of wound infection during the study period (z = 1.59, P = .11), despite trends of higher ASA classification (z = 11.7, P < .001), obesity (z = 4.87, P < .001), and higher wound classification (class III: z = 6.32, P < .001), all of which were associated with increased odds of wound infection by multivariable analysis (Figure 2).

The rate of perioperative MI in the study cohort was 0.85% (of 13 657 patients who underwent a pancreatic resection, 127 required transfusion [0.9%]; of 1605 patients who underwent a gastric resection, 15 required transfusion [0.9%]; and of 4418 patients who underwent a hepatic resection, 26 required transfusion [0.6%]). The median time to MI was 3 days (IQR, 1-7 days) and was not significantly different for patients who received a transfusion compared with those who did not receive a transfusion (P = .06). There was a significant association between the incidence of PRBC transfusion and postoperative MI (P < .001). There was a significant increase in the annual rate of perioperative MI during the study period (z = 3.15, P = .002), which corresponded to a 0.33% absolute change. Of the variables associated with MI by multivariable analysis, there was a significant trend toward older patient age (z = 2.41, P = .02), white race (z = 56.4, P < .001), and higher Charlson Comorbidity Index score (1: z = 3.28, P = .001) (Figure 2).

The rate of renal insufficiency was 2.0% (of 13 604 patients who underwent a pancreatic resection, 230 required transfusion [1.7%]; of 1598 patients who underwent a gastric resection, 26 required transfusion [1.6%]; and of 4405 patients who underwent a hepatic resection, 150 required transfusion [3.4%]). There was a significant association between the incidence of PRBC transfusion and renal insufficiency (P < .001). There was no significant trend in the annual rate of renal insufficiency over the study period (z = 1.74, P = .08). Of the variables associated with renal insufficiency by multivariable analysis, the study cohort evidenced a temporal increase in the number of patients with a higher ASA classification (>2: z = 11.0, P < .001) and decrease in preoperative serum creatinine (z = −10.2, P < .001) (Figure 2).

Multiple national groups have issued evidenced-based recommendations in support of the restrictive use of blood transfusion, yet little is known regarding current national practice. In this large observational study of patients undergoing pancreatic, gastric, or hepatic surgery for malignant tumors, we demonstrated that (1) there was a significant trend of decreasing rates of perioperative PRBC transfusion over time, (2) the rates of PRBC transfusion decreased despite decreasing preoperative hematocrit levels and increasing patient comorbidity, and (3) annual changes in transfusion rates did not coincide with any significant changes in perioperative rates of surgical site infection or renal insufficiency, whereas annual rates of perioperative MI increased.

The precise etiology of these trends is likely multifactorial. One possibility is that changes in operative and anesthetic techniques are resulting in improved perioperative outcomes. Given that the vast majority of transfusions in this cohort occurred on the day of the procedure, we suspect that safer surgery is a strong driver of this trend.¹⁹ Alternatively, this trend may reflect the clustering effect, in which major oncologic resections are increasingly performed by surgeons at high-volume hospital centers. In the case of surgical metastasectomy, high-volume institutions are driving national trends toward increased rates of metastasectomy in the last decade, with decreasing in-hospital mortality despite increasing patient comorbidity.²⁰ Lastly, changes in transfusion trends may represent increasing concern regarding the purported deleterious effects of blood transfusion and a commitment to more restrictive transfusion thresholds.

We had hypothesized that potentially impaired tissue oxygenation with decreasing utilization of PRBC transfusion could result in poorer wound healing, MI, and renal insufficiency. While each of these secondary outcomes was, in fact, associated with the use of PRBC transfusion, the retrospective nature of our study cannot exclude the likely possibility of confounding by indication, in which transfusion serves as a surrogate for other variables (eg, hemorrhage) more likely to be directly related to morbidity. Of the 3 large randomized controlled trials evaluating a liberal transfusion strategy compared with a restrictive strategy for surgical patients, rates of ischemic and infectious complications were reported for both the Transfusion Requirements After Cardiac Surgery study⁷ and the Transfusion Indication Threshold Reduction trial,⁸ with no significant differences in the occurrence of these outcomes.⁹ More recently, a single-institution randomized controlled trial²¹ evaluating transfusion strategies for major oncologic surgery requiring postoperative intensive care unit admission found that a restrictive transfusion practice was associated with an increased risk of death or major ischemic complications. The generalizability of this study²¹ for elective oncologic procedures is unclear, given its inclusion of emergency procedures, and given that all study participants required intensive care unit admission postoperatively—an uncommon postoperative destination following colectomy and many of the other included procedures. Nevertheless, in our study, there was a significant trend toward increasing rates of perioperative MI as rates of transfusion decreased. Although this trend is likely explained by changes in the patient cohort over the study period, we cannot exclude a causal relationship between these outcomes.

Several limitations of the present study warrant discussion. The analyses rely on the quality of data abstraction and entry, and there remains the possibility of error related to miscoding and omission. In this regard, ACS NSQIP may be preferable to other large administrative data sets given the quality control provided by trained surgical clinical reviewers. One particular limitation of the database is the inclusion of cell saver blood as intraoperative transfusion (where every 500 mL of cell saver fluid returned to the patient was coded as a unit of blood). Second, changes in transfusion rates over the study period may reflect changes in the specific constituency of hospitals participating in the ACS NSQIP. That is, there may have been a preferential exit of hospitals that performed a high level of transfusions and/or the addition of hospitals that performed a low level of transfusions over the study period. It is also possible that trends in the coding process itself have changed, which could introduce bias to the reported trends. Because risk-adjusted outcomes are increasingly analyzed, there may be a trend toward improved capture of patient data—although increasing coding sensitivity would more likely lead to transfusion trends in the opposite direction to that observed. Lastly, while determining preoperative hematocrit levels allowed for multivariable analysis controlling for the degree of anemia, data on the last hematocrit level prior to the initiation of transfusion are not available. While most transfusions occurred on the day of surgery, the exact hematocrit “trigger” for transfusion could not be assessed in this cohort.

Transfusion practices have been observed to vary significantly among hospitals and among health care professionals within the same hospital.^22,23 To our knowledge, the present study represents the first analysis of national trends of transfusion practices associated with major abdominal oncologic surgery to rely on all participating institutions of the ACS NSQIP. Further studies are needed to assess whether these trends reflect changes in operative techniques, hospital cohorts performing the procedures, and transfusion thresholds.

Accepted for Publication: September 30, 2015.

Corresponding Author: Giorgos C. Karakousis, MD, Department of Surgery, Hospital of the University of Pennsylvania, 3400 Spruce St, 4 Silverstein, Philadelphia, PA 19104 (giorgos.karakousis@uphs.upenn.edu).

Published Online: January 13, 2016. doi:10.1001/jamasurg.2015.5094.

Author Contributions: Dr Ecker had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Ecker, Drebin, Fraker, Kelz, Roses, Karakousis.

Acquisition, analysis, or interpretation of data: Ecker, Simmons, Zaheer, Poe, Bartlett, Kelz, Roses, Karakousis.

Drafting of the manuscript: Ecker, Zaheer.

Critical revision of the manuscript for important intellectual content: Ecker, Simmons, Poe, Bartlett, Drebin, Fraker, Kelz, Roses, Karakousis.

Statistical analysis: Ecker, Simmons, Zaheer, Poe.

Administrative, technical, or material support: Fraker, Karakousis.

Study supervision: Drebin, Fraker, Kelz, Karakousis.

Conflict of Interest Disclosures: None reported.

Disclaimer: The ACS NSQIP and the hospitals participating in the ACS NSQIP are the source of the data used herein; they have not verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors.