Shah MD, Goldstein DP, McCluskey SA, Miles BA, Hofer SO, Brown DH, Irish JC, Gullane PJ, Gilbert RW. Blood Transfusion Prediction in Patients Undergoing Major Head and Neck Surgery With Free-Flap Reconstruction. Arch Otolaryngol Head Neck Surg. 2010;136(12):1199-1204. doi:10.1001/archoto.2010.202
Author Affiliations: Departments of Otolaryngology[ndash]Head and Neck Surgery and Surgical Oncology, Princess Margaret Hospital (Drs Shah, Goldstein, Miles, Brown, Irish, Gullane, and Gilbert), Perioperative Blood Conservation Program, Department of Anesthesia and Pain Management (Dr McCluskey), and Division of Plastic Surgery, Department of Surgery (Dr Hofer), University Health Network, Toronto, Ontario, Canada.
Objective To develop a clinically useful perioperative blood transfusion prediction model for patients undergoing a major head and neck surgical procedure requiring free-flap reconstruction.
Design Retrospective observational study.
Setting Tertiary care university-affiliated teaching hospital (University Health Network, Toronto, Ontario, Canada).
Patients All patients with a head and neck malignant neoplasm undergoing major head and neck surgery requiring free-flap reconstruction.
Main Outcome Measure Perioperative single-unit red blood cell transfusion.
Results All the preoperative variables were tested for an association with perioperative blood transfusion using univariable and multivariable analyses. After multivariable regression analysis, the following preoperative variables were found to be significantly associated with perioperative transfusion: sex, body mass index, T stage, preoperative hemoglobin level, and type of free-flap reconstruction used (ie, osseous vs nonosseous). The regression model was used to develop a transfusion risk score. Receiver operating characteristic curve analysis confirmed adequate discrimination of risk using the transfusion risk score.
Conclusions We have developed a reliable model for predicting perioperative blood transfusion requirements in patients undergoing major head and neck surgery requiring free-flap reconstruction. This model can be used for accurate preoperative risk stratification.
Perioperative blood transfusion is commonly required in patients undergoing major head and neck surgical procedures, with the reported incidence in the literature between 12% and 84%.1 Despite the fact that the safety of allogenic blood transfusion has improved dramatically over time, significant risk remains for transmission of blood-borne viral pathogens and adverse transfusion reactions.2 Furthermore, results of several studies2- 6 have suggested that perioperative blood transfusion may increase the risk of cancer recurrence owing to transfusion-related immunomodulation.
For these reasons, there has been significant interest in reducing the exposure of patients to allogenic blood transfusions. Preoperative autologous blood donation (PAD), treatment with recombinant erythropoietin, and acute normovolemic hemodilution have been explored.2
Knowledge about a patient's probability of requiring a perioperative blood transfusion is useful for patient counseling. Alternatives to allogenic blood transfusion, such as autologous donation and recombinant erythropoietin treatment, have inherent limitations and potential risks. Comprehensive blood conservation algorithms have been shown to be valuable in effectively using these resources to reduce allogenic blood transfusion.7 Perioperative risk stratification is essential to identify patients who are at high risk of requiring a transfusion and who would benefit from such programs. Therefore, the objective of our study was to develop a clinically useful perioperative blood transfusion prediction model for patients undergoing a major head and neck surgical procedure requiring free-flap reconstruction.
Data were obtained from a database in which prospective perioperative data about all patients undergoing major head and neck surgery requiring free-flap reconstruction at the University Health Network (Toronto, Ontario, Canada) are collected. Dates of study inclusion were 1999 to 2009. This database has been checked for accuracy by separate observers and has been used successfully in previous studies.8,9 The study was approved by the institutional ethics review board at the University Health Network. To maintain the homogeneity of the study population, patients were excluded from the study if they had a nonmalignant diagnosis (eg, facial reanimation surgery or osteoradionecrosis) or if adequate clinical tumor staging or perioperative transfusion data were unavailable. Five hundred eighty-five patients were eligible for inclusion in the study.
The primary outcome measure was perioperative single-unit red blood cell transfusion. This was defined as any allogenic blood transfusion given during surgery or during the first 24 hours after surgery.
Preoperative clinical characteristics were tested as potential predictors of perioperative blood transfusion. The following variables were tested: sex, age, body mass index (BMI), T stage, N stage, treatment with preoperative radiation therapy or chemotherapy, the presence of significant medical comorbidity, preoperative hemoglobin level, type of free-flap reconstruction used (osseous vs nonosseous), and the site of the primary tumor. To simplify development of the model and to allow for creation of an easy-to-use model in clinical settings, variables were reclassified as binary outcomes wherever possible. Dichotomization of variables was based on clinically significant divisions or median values for that variable. Body mass index (calculated as weight in kilograms divided by height in meters squared) was reclassified as underweight (<18.5) vs normal or overweight ([ge]18.5). Medical comorbidity was assessed using the Kaplan-Feinstein Comorbidity Index,10 which has been shown to be valid in the stratification of comorbidity among patients with head and neck cancer.11 The index was classified as none or mild vs moderate or severe. Preoperative hemoglobin level was classified as low or normal based on accepted reference values for men (>13.0 g/dL) and for women (>12.0 g/dL) (to convert hemoglobin to grams per liter, multiply by 1.0). The type of free-flap reconstruction used was reclassified as osseous vs nonosseous.
All the statistical analyses were performed using commercially available software (STATA, version 8.2; StataCorp LP, College Station, Texas). Univariable analysis was performed using the [khg];2 test, with a significance level of P[nbsp]<[nbsp].05. All potential preoperative predictor variables were tested for an association with the primary outcome measure, perioperative single-unit red blood cell transfusion. Multivariable analysis was conducted using logistic regression techniques, with significance determined by the likelihood ratio test. The goal was to create the simplest predictive regression model that best fit the study data. To create the predictive model, forward and backward stepwise logistic regression techniques were used. Potential predictor variables were included in the modeling if their associated P value on univariable analysis was less than 0.1 or if the variable was thought to be clinically relevant. Similar results were obtained using the forward and backward techniques. The bootstrap method was used to account for generalizability error, and bias-corrected 95% confidence intervals were calculated based on 1000 resamples of the data.12
A transfusion risk score (TRS) was then developed by assigning each variable in the final model an integer score. This score was calculated by dividing the regression coefficient of each variable by the smallest coefficient of all the variables in the final model and then rounding to the nearest integer. The TRS for each patient was the sum of the scores for each variable in the final model. Receiver operating characteristic (ROC) curves were generated to examine the predictive value of the regression model and the TRS. The area under the ROC curve can quantify the predictive value of a model; greater than 0.7 is deemed reasonable or fair, while greater than 0.8 is deemed good.13
Of 585 patients in the study, 144 (24.6%) required a blood transfusion. The mean age of the patient sample was 60 years (age range, 12-93 years), and most patients were male (66.7%). The preoperative and transfusion characteristics of the study population are summarized in Table 1. The results of the univariable analysis are given in Table 2. Sex, BMI, T stage, N stage, preoperative hemoglobin level, type of free-flap reconstruction used, and the Kaplan-Feinstein Comorbidity Index were significantly associated with perioperative transfusion.
The multivariable stepwise logistic regression analysis determined that sex (female vs male), T-stage group (T3 or T4 vs T1 or T2), BMI (underweight vs normal or overweight), preoperative hemoglobin level (low vs normal), and type of free-flap reconstruction (osseous vs nonosseous) were independently associated with an increased risk of perioperative transfusion. The regression coefficients, odds ratios, P values, and calculated integer scores are summarized in Table 3. The bootstrap method yielded bias-corrected 95% confidence intervals for the regression coefficients. None of these intervals crossed 0, supporting inclusion of each variable in the final regression model. Using the derived final regression model, the predicted probability of requiring a perioperative blood transfusion was calculated for women and for men based on strata of the variables in the final model (Table 4).
The TRS was calculated for each patient as the sum of the integer scores for that patient. For example, a female patient (score, 1) with a normal BMI (score, 0), low preoperative hemoglobin level (score, 2), T1 or T2 tumor (score, 0), and receipt of an osseous free-flap reconstruction (score, 1) would be assigned a score of 4. The proportion of patients in the study population who underwent transfusion for each TRS is given in Table 5. Also given in this table is the range of probabilities predicted by the regression model that corresponds to each TRS (based on the calculations used to generate the data given in Table 4).
Our goal was to create a TRS that was more easily used in a clinical setting but was still meaningful for risk stratification. Transfusion risk scores of 0 and 1 were considered very low and low risk, respectively. The proportions of patients who underwent blood transfusion with a TRS of 2 or 3 were similar; therefore, we combined these 2 groups to create an intermediate-risk group. Because few patients had a TRS of 5 or 6, these 2 groups were combined with patients having a TRS of 4 to create a high-risk group (Table 6).
The area under the ROC curve for the regression model was 0.754. When the 7-category TRS was applied to the data set, the area under the ROC curve was 0.747 (95% confidence interval, 0.710-0.782). The value changed only minimally with the 4-category TRS (area under the ROC curve, 0.744; 95% confidence interval, 0.706-0.779). Any further simplification of the TRS led to significant decreases in the area under the ROC curve (<0.700).
The incidence of perioperative blood transfusion among patients undergoing head and neck surgery varies widely in the literature, likely owing to variations in study populations by tumor site and stage, as well as nonuniform transfusion criteria. The incidence in our study was similar to that reported in the literature for patients requiring free-flap reconstruction.14,15
Weber14 developed a transfusion prediction model in patients undergoing major head and neck surgical procedures. Tumor stage, preoperative hemoglobin level, and requirement of free-flap reconstruction were found to be strong predictors of blood transfusion requirement. This model was later validated using data from a separate institution.15 Most patients in these studies did not require a microvascular free flap for reconstruction. As such, the studies contained too few patients to conduct an analysis of predictors among only those patients requiring free-flap reconstruction. Therefore, the objective of our study was to develop a clinically useful perioperative blood transfusion prediction model for patients undergoing a major head and neck surgical procedure requiring free-flap reconstruction.
Eleven variables were examined as potential preoperative predictors of perioperative blood transfusion risk. After multivariable analysis, the following 5 preoperative variables remained significant independent predictors: sex, BMI, T stage, preoperative hemoglobin level, and type of free-flap reconstruction used (Table 3). It is unclear why women were more likely to receive a blood transfusion compared with men. This may be related to an inherent higher requirement for blood transfusion among women or to application of a similar hemoglobin level transfusion trigger in men and women, despite differences in accepted normal hemoglobin levels. Many patients with head and neck cancer experience significant weight loss due to poor oral intake before surgery and are malnourished; therefore, it was not surprising that a low BMI was predictive of a perioperative blood transfusion requirement. Advanced T-stage tumors typically require extensive surgical procedures that may involve substantial intraoperative blood loss. Advanced T stage has been found to be predictive of blood transfusion requirement by other authors.14,15 Use of an osseous free flap for reconstruction vs a nonosseous flap resulted in a higher risk of blood transfusion. This may have been secondary to additional blood loss from the free-flap harvest, or this variable may be a surrogate for tumor ablative procedures that involved significant resection of bone. Preoperative hemoglobin level is often low in patients with head and neck cancer,1,14 and this was seen commonly (among 27.4% of patients) in our study. Low preoperative hemoglobin level was the strongest independent predictor of blood transfusion risk, similar to findings in other studies.14,15
Using our multivariable model, a TRS was developed for easier clinical use. Using the TRS, our study showed that we can reliably predict the risk of perioperative blood transfusion among patients scheduled to undergo a major head and neck surgical procedure requiring free-flap reconstruction. We believe that the 4-category TRS is preferable to the 7-category one, as by reducing the number of risk categories, it is more straightforward for use in clinical settings; however, it does not sacrifice the ability to stratify patients by risk. Using the 4-category TRS, individual patients can be classified as being at very low risk, low risk, intermediate risk, or high risk of requiring a perioperative blood transfusion.
Because of the risks associated with allogenic blood transfusions, alternatives have been investigated to minimize exposure to allogenic blood. PAD avoids several risks associated with allogenic blood such as transmission of blood-borne viruses and incompatibility reactions. Furthermore, autologous blood may avoid transfusion-related immunomodulation and any potential effect on cancer recurrence16; however, this remains controversial.2 PAD is not without limitations. The time from diagnosis to surgery for oncologic patients may allow insufficient time for donation and compensatory erythropoiesis. Even so, compensation does not completely replace donated units, and patients with cancer are more likely to experience anemia before surgery.2 Furthermore, PAD is significantly more expensive than the use of allogenic blood,17 and it is difficult to predict the number of units that will be required, resulting in donated units being discarded.18,19 Finally, many patients with head and neck cancer have anemia before surgery and are not candidates for PAD.
Several studies1,2,20,21 have examined the feasibility of preoperative treatment with recombinant erythropoietin. Scott et al1 conducted a double-blind, placebo- controlled, randomized trial among patients with head and neck cancer and found a trend toward decreased transfusion requirements in patients treated before surgery with recombinant erythropoietin. Furthermore, studies among other patient populations have found that recombinant erythropoietin can facilitate PAD in patients with anemia22,23 and can reduce allogenic blood transfusion.24 The time from diagnosis to surgery for oncologic patients may limit the use of recombinant erythropoietin, as there may be insufficient time to see a meaningful effect.
PAD and recombinant erythropoietin are expensive resources, with limitations and potential adverse effects. Comprehensive blood conservation programs allow for safe and effective use of blood transfusion[ndash]related resources and have been shown to be effective in reducing the use of allogenic blood transfusions.7 Formal blood conservation programs and resources such as recombinant erythropoietin and PAD are not routinely used among patients with head and neck cancer in our institution at this time. As programs are developed, risk stratification will be essential for identification of patients at high risk and for appropriate and safe allocation of these resources.
Our study is not without limitations. Our model was derived from patients treated within a single institution and, as such, may reflect the institutional philosophy with regard to administration of allogenic blood products. This can vary significantly among institutions, as there are no widely accepted criteria for the administration of blood transfusions. This may limit the generalizability of our results. Ideally, our model should be validated using a data set from another institution.
In conclusion, we have developed an easy-to-use model that reliably predicts the risk of requiring a perioperative blood transfusion among patients who are to undergo a major head and neck surgical procedure requiring free-flap reconstruction. This model can be used for preoperative patient counseling and for accurate risk stratification.
Correspondence: Ralph W. Gilbert, MD, FRCSC, Departments of Otolaryngology[ndash]Head and Neck Surgery and Surgical Oncology, Princess Margaret Hospital, University Health Network, 3-955 Wharton Head and Neck Centre, 610 University Ave, Toronto, ON M5G 2M9, Canada (email@example.com).
Submitted for Publication: March 30, 2010; final revision received July 13, 2010; accepted August 18, 2010.
Author Contributions: Drs Shah, Goldstein, McCluskey, Gullane, and Gilbert had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Shah, Goldstein, Hofer, Irish, Gullane, and Gilbert. Acquisition of data: McCluskey, Miles, and Brown. Analysis and interpretation of data: Shah, Goldstein, and Gilbert. Drafting of the manuscript: Shah, Goldstein, and Miles. Critical revision of the manuscript for important intellectual content: Goldstein, McCluskey, Hofer, Brown, Irish, Gullane, and Gilbert. Statistical analysis: Shah, Goldstein, McCluskey, and Miles. Administrative, technical, and material support: Goldstein, McCluskey, Hofer, and Gilbert. Study supervision: Goldstein, Brown, Irish, Gullane, and Gilbert.
Financial Disclosure: None reported.
Previous Presentation: This study was presented at the annual meeting of the American Head and Neck Society; April 28, 2010; Las Vegas, Nevada.
This article was corrected for errors on July 10, 2013.