Telephone: 416-340-5164; FAX: 416-340-3698
Derivation of a risk index for the prediction of massive blood transfusion in liver transplantation
Article first published online: 1 SEP 2006
Copyright © 2006 American Association for the Study of Liver Diseases
Volume 12, Issue 11, pages 1584–1593, November 2006
How to Cite
McCluskey, S. A., Karkouti, K., Wijeysundera, D. N., Kakizawa, K., Ghannam, M., Hamdy, A., Grant, D. and Levy, G. (2006), Derivation of a risk index for the prediction of massive blood transfusion in liver transplantation. Liver Transpl, 12: 1584–1593. doi: 10.1002/lt.20868
- Issue published online: 20 OCT 2006
- Article first published online: 1 SEP 2006
- Manuscript Accepted: 16 MAY 2006
- Manuscript Received: 29 NOV 2005
Massive blood transfusion (MBT) remains a serious and common occurrence in liver transplantation surgery. This retrospective cohort study was undertaken to identify preoperative predictors of MBT and to develop a risk index for MBT in liver transplantation. Data were retrospectively collected on all liver transplantations carried out at a single institution between January 1998 and March 2004. Multivariable logistic regression analysis was used to identify independent predictor variables of MBT, defined as ≥6 units of red blood cell concentrate (RBC) in the first 24 hours of surgery. The model was internally validated by bootstrapping. Of the 460 liver transplant recipients, 193 (42%) received ≥6 units of RBC within 24 hours of surgery. Unadjusted analyses identified 12 preoperative predictors of MBT: age, height, gender, repeat transplantation, etiology of liver failure, and preoperative laboratory values (hemoglobin concentration, platelet count, international normalized ratio for prothrombin activity [INR], albumin, total bilirubin, and creatinine). In multivariable logistic regression, 7 independent predictors of MBT were identified: age (>40 years), hemoglobin concentration (≤10.0 g/dL), INR (1.2-1.99, and >2.0), platelet count (≤70 × 109/L), creatinine (≥110 μmol/L for female subjects and ≥120 μmol/L for male subjects), albumin (< 28 g/L), and repeat transplantation. The area under the receiver-operating characteristic curve (ROC) for the model was 0.82. By using the regression β coefficients to derive weights for each of these predictors, a risk index was developed that assigned each patient a score between 0 and 8. The ROC for this risk index was 0.79. MBT in liver transplantation surgery can be accurately predicted by 7 readily available preoperative predictors. Liver Transpl, 2006. © 2006 AASLD.
Prevention of excessive blood loss is an important objective in the perioperative management of liver transplantation surgery because it is associated with prolonged recovery, increased morbidity, and reduced graft survival.1, 2 Over the last 20 years, changes in surgical and anesthetic management have reduced intraoperative blood transfusion in liver transplantation surgery.3–5 Although massive transfusion remains a common event, 10 to 50% of surgeries are conducted without requiring a blood transfusion.2, 3, 5–8 To date, an accurate prediction rule for blood transfusion in liver transplantation surgery has not been developed. The lack of such an index precludes the tailoring of inventions aimed at reducing blood loss.
One reason for the inability to predict blood loss in liver transplantation is that until recently, intraoperative variables leading to blood loss have been overwhelming, thus making transfusion requirements impossible to predict. As a result of improvements in the intraoperative management, however, the relative importance of preoperative determinants of blood loss should have increased.
The primary objectives of this study were to identify preoperative predictors of massive blood transfusion (MBT) (>6 units red blood cell concentrate [RBC]) within the first 24 hours of liver transplantation, and to develop a risk index to predict MBT in liver transplantation.
MATERIALS AND METHODS
Population and Databases
After we received institutional ethics approval, data were retrospectively collected on consecutive patients who underwent liver transplantation (cadaveric or living-donor) between January 1, 1998, and March 31, 2004, at the Toronto General Hospital, University Health Network (Toronto, Ontario, Canada). Database accuracy was measured by reabstracting the medical records of a randomly selected 10% (n = 50) of the study sample. In addition, outlying values were compared to patients' records to identify and correct errors in the database. Whenever possible, missing values were completed from the medical records; otherwise, patients with missing categorical variable values were excluded. For continuous variables, missing values were imputed on the basis of the mean for the entire sample.
Study Setting and Clinical Practice
The Toronto General Hospital is a tertiary-care teaching hospital affiliated with the University of Toronto. The liver transplant program was started in 1985 and currently conducts between 80 and 120 transplantations per year. The live-donor program was initiated in May 2000 and accounted for approximately 20% of liver transplantations in 2004. During the study period, patients were managed according to standardized clinical protocols, as described below.
Anesthetic Procedure and Monitoring
The anesthetic induction and maintenance was performed with a combination of propofol, midazolam, fentanyl, and pancuronium. Patients were ventilated with an oxygen-air mixture. Hemodynamic monitoring consisted of an arterial line and a pulmonary artery catheter. Body temperature was maintained with warming blankets and intravenous (IV) fluid warmers with a target temperature of 36-37°C.
The most common surgical technique involved the use of the donor inferior vena cava and is referred to as the “classic technique.” In live-donor transplantations and in cadaveric donor cases that were not able to tolerate a complete inferior vena cava cross-clamp, the piggyback technique was used. Venovenous bypass was used in one case, which was excluded from the final analysis.
Blood Conservation Strategies
Cell salvage was used in all cases not involving hepatocellular cancer or sepsis. Antifibrinolytic use was not standardized and was used at the discretion of the surgical team. Those patients that did receive antifibrinolytics received 1 of 3 regimens: (1) tranexamic acid 1 g as a single bolus, (2) tranexamic acid 10 mg/kg/h until 2 hours after liver reperfusion, and (3) aprotinin 1 × 106 KIU bolus followed by 0.5-1.0 × 106 KIU per hour until 2 hours after liver reperfusion.
Blood Transfusion Guidelines
Transfusion of RBC was based on clinical assessment, hemodynamic monitoring, and laboratory measurement of hemoglobin and hematocrit. The target hemoglobin concentration was 8.0-10.0 g/dL (hematocrit 25-30%). Cell saver units were used instead of RBC when available, and all cell saver units were returned to the patient.
Fresh frozen plasma (FFP) (2 units) was indicated for an international normalized ratio for prothrombin activity (INR) between 1.5 and 2.0 during the preanhepatic phase or if associated with blood loss. If the INR exceeded 2.0, we administered 4 units of FFP before repeating a coagulation profile. Platelets (5 units) were indicated to keep the platelet count above 80 × 109/L. Cryoprecipitate was indicated for fibrinogen level below 1.0 g/L in the bleeding patient. Crystalloids, normal saline, and colloids (10% Pentaspan, albumin 5%, and albumin 25%) were used for volume replacement and to maintain adequate urine output (> 0.5 mL/kg/h) at the discretion of the attending anesthesiologist.
Blood transfusion data (RBC, FFP, platelets, cryoprecipitate) were collected from the anesthesia, intensive care unit, and transfusion records over the first 48 hours after skin incision. These data were validated by comparison with the hospital blood bank database (Hemocare). For the blood bank database, any blood product issued and not returned was considered transfused.
Potential Predictor Variables and Dependent Variables
Potential preoperative predictors were selected on the basis of a review of prior studies that attempted to define predictors of blood loss in liver transplantation (Table 1). The variables used in this study are defined in Table 2.
|Study||Liver transplant population||Data collection and analysis||Outcome||Variables in final model||Accuracy of final model|
|Mor et al. (1993)1||205 consecutive (Jan 1988-Dec 1998)||Retrospective univariate and multivariable logistic regression||≤10 units RBC||Creatinine, platelet count, PT||Sensitivity = 60% Specificity = 69%|
|Cacciarelli et al. (1996)3||306 primary (Jan 1992-Dec 1994)||Retrospective univariate and multivariable logistic regression||Transplantation without RBC||Preoperative Hct, UNOS status, piggyback technique, operative time and year of transplantation|
|Hendriks et al. (2000)17||164 primary, cholestatic and noncholestatic (Jan 1989-Dec 1996)||Retrospective univariate and multivariable logistic regression||RBC transfusion requirement||Gender, Child-Pugh classification, serum urea levels, year of transplant, length of cold ischemic time. use of autologous (call saver) blood|
|Steib et al. (2001)18||410 consecutive transplants (Sep 1988-Dec 1998)||Retrospective univariate and multivariable stepwise analysis||High blood loss (>12 units RBC)||Preoperative Hb, preoperative FDP, previous abdominal surgery||Sensitivity = 18% Specificity = 98% PPV = 69% NPV = 82%|
|Findlay and Rettke (2000)16||583 sequential (Jun 1986-Nov 1995)||Univariate and multivariable linear regression||Units of blood transfused||Age, pseudocholinesterase, bilirubin, creatinine||R2 = 0.22|
|Ramos et al. (2003)2||122 consecutive (Sep 1998-Nov 2000)||Prospective univariate and multiple linear regression||1 or more units RBC >6 units RBC||Preoperative Hb UNOS Classification, intraoperative portocaval shunt|
|Massicotte et al. (2004)19||206 liver transplant (Jan 1998-Apr 2002)||Retrospective univariate and multivariable logistic regression||>4 units of RBC||INR, platelet count, and duration of surgery|
|Variable||Definition and unit of measurement|
|Variables related to RBC volume|
|Preoperative hemoglobin||Continuous variable; g/dL|
|Age||Continuous variable; yr|
|Sex||Binomial variable; Male/Female|
|Height||Continuous variable; cm|
|Weight||Continuous variable; kg|
|Previous abdominal surgery||Binomial variable; Yes/No|
|Coronary artery disease||Binomial variable; Yes/No|
|Hypertension||Binomial variable; Yes/No. Defined as a history of hypertension and on treatment at the time of surgery|
|Diabetes||Binomial variable; Yes/No. Pharmacologically treated diabetes|
|Renal dysfunction||Binomial variable; Yes/No. From history and unrelated to liver failure|
|Chronic obstructive pulmonary disease||Binomial variable; Yes/No. Defined as a requirement for pharmacologic therapy for the treatment of chronic pulmonary disease, or FEV1 <75% of predicted value|
|Smoking||Binomial variable; Yes/No. Ex-smokers or current smokers|
|Variables related to liver disease etiology|
|Viral hepatitis||Binomial variable; Yes/No|
|Primary biliary cirrhosis||Binomial variable; Yes/No|
|Fulminant hepatic failure/acute liver failure||Binomial variable; Yes/No|
|Primary sclerosing cholangitis alcoholic liver disease||Binomial variable; Yes/No|
|Cryptogenic cirrhosis||Binomial variable; Yes/No|
|Hepatorenal syndrome||Binomial variable; Yes/No|
|Ascites||Binomial variable; Yes/No. None vs. Moderate/Severe|
|Encephalopathy||Binomial variable; Yes/No. Mild vs. Moderate Severe|
|Esophageal varices||Binomial variable; Yes/No. None vs. Present|
|Preoperative blood work||All blood work performed on blood drawn immediately before surgery|
|Hematology-hemoglobin concentration, platelet count, INR, PTT||Continuous variables|
|Liver function studies—aspartate aminotranferase, alanine aminotranferase, alkaline phosphatase, albumin, total bilirubin, amylase||Continuous variables|
|Renal function-creatinine||Continuous variable|
The Model for End-Stage Liver Disease (MELD) score has not been adopted in Canada. The MELD score was calculated from the immediate preoperative values for INR, serum creatinine, total bilirubin, and primary etiology of liver failure.9 The formula for the MELD score is 3.8 ln(bilirubin [mg/dL]) + 11.2 ln(INR) + ln(creatinine [mg/dL]) + 6.4(etiology: 0 if cholestatic or alcoholic, 1 otherwise). The highest value for creatinine was 4 mg/dL and the lowest value for the other variables in the MELD score was 1. The maximum MELD score was 40.
MBT, the dependent variable, was defined as transfusion of ≥6 units of RBC in the first 24 hours of liver transplantation (i.e., 24 hours from skin incision).
Statistical analyses were performed by SAS version 8.02 (SAS Institute, Cary, NC). Categorical variables were summarized as frequencies and percentages; continuous variables as means with standard deviation.
The unadjusted association of potential predictors of massive transfusion was evaluated for both continuous (t test, Wilcoxon rank sum test) and categorical (χ2 test, Fisher's exact test, Mantel-Haenszel χ2 test) variables.
Independent predictors of MBT were identified by multivariable logistic regression analysis. Initially, the mathematical relationships between the continuous potential predictor variables and the probability of massive transfusion (logit transformation) were assessed by cubic spline functions.10, 11 These preliminary analyses were used to guide categorization of continuous variables in the multiple logistic regression analysis.12 All variables associated with MBT in this univariate (P ≤ 0.05) analyses were included in the multivariate analysis. Backward stepwise variable selection was employed to construct the final multivariable regression model (criterion for selection: P ≤ 0.05). The calibration of the logistic regression model was assessed by the Hosmer-Lemeshow goodness-of-fit statistic (larger P values imply better fit).13 The model's discriminative performance was assessed by the c-statistic, which is equivalent to the area under the receiver-operating characteristic curve (ROC). An area of 0.5 indicates no discrimination, whereas an area of 1.0 indicates perfect separation of patients with different outcomes.14
Risk Index Development
The regression β coefficients of independent predictors were used to develop a risk index for MBT. To derive a simple practical scoring scheme, the regression β coefficients were rounded to the nearest integer. The discriminative performance of this risk index for predicting massive transfusion was measured by ROC, sensitivity, specificity, positive predictive value, and negative predictive value.
The validity of the final risk index was further described by bootstrap techniques. Initially, 1000 computer-generated samples, each including 459 individuals (i.e., study sample less one patient), were derived from the study sample by random selection with replacement. The bootstrap samples were used to estimate the 95% confidence interval for the ROC for the risk index. The reliability of the independent predictors included in the index was also described by bootstrap bagging. In summary, 1000 bootstrap samples were generated as described above. Within each bootstrap sample, forward stepwise variable selection (criterion for inclusion: P ≤ 0.05) was employed using all potential independent variables. The reliability of predictor variables in the final regression model was estimated by how often they were retained as independent predictors in the bootstrap samples. Reliable predictors were expected to be retained in a higher proportion of bootstrap samples.
During the study period, 471 liver transplantations were performed. Eleven operations were repeat liver transplantations, and only data from the first transplantation was kept in the database. Preoperative patient characteristics for the entire 460 patients are listed in Table 3.
|Variable||Total sample (n = 460)|
|Age (yr)||50.4 ± 13.2|
|Height (cm)||169.7 ± 12.5|
|Weight (kg)||77.2 ± 17.5|
|Sex (female)||146 (32%)|
|MELD score, median (interquartile range)||15.9 (11.7-22.7)|
|Coronary artery disease||18 (4%)|
|Renal dysfunction||17 (4%)|
|Thyroid disease||13 (3%)|
|Viral hepatitis||259 (56%)|
|Alcoholic liver disease||119 (26%)|
|Primary sclerosing cholangitis||45 (10%)|
|Primary biliary cirrhosis||31 (7.0%)|
|Fulminant hepatic failure||38 (8.0%)|
|Thrombotic liver disease||2 (0.4%)|
|Cryptogenic cirrhosis||40 (9%)|
|Comorbidities related to liver disease|
|Hepatorenal syndrome||43 (9%)|
|Esophageal varices||205 (45%)|
|Previous abdominal surgery||168 (37%)|
|Living donor||75 (16%)|
|Repeat transplant||22 (5%)|
|Hemoglobin concentration (g/dL)||10.9 ± 2.3|
|Platelet count (× 109/L)||100 ± 73|
|INR||1.79 ± 0.94|
|PTT (s)||49 ± 17|
|AST (U/L)||200 ± 622|
|ALT (U/L)||171 ± 573|
|ALP (U/L)||152 ± 151|
|Albumin (g/L)||31 ± 7.5|
|Total bilirubin (μmol/L)||108 ± 160|
|Amylase (U/L)||76 ± 63|
|Creatinine (μmol/ml)||108 ± 91|
If the first operation occurred before the data collection period, then the second operation was kept in the database (n = 21). Height was imputed for 4 patients and weight for 3 patients. The accuracy of the database exceeded 95% in comparison to reabstracted hospital records, and no patients were excluded for missing data.
Donors were largely deceased in origin (83%) but 17% were living related; 55% were male and 45% female; mean age was 44 ± 16 years (range 9-88 years), and cold ischemia time was 8 ± 3 hours (range 1-28 hours). The major cause of death was cerebrovascular accident 49%. Routine donor liver biopsies were not performed. If there was any suspicion of fatty liver, a biopsy was performed, and if the fat content was greater than 30%, the graft was not used.
No antifibrinolytics were used in 60% (n = 274) of the study sample. Forty patients received a tranexamic acid infusion (10 mg/kg/mL) until 2 hours after the liver was transplanted; 143 patients received a single bolus of tranexamic acid 1 g, and 7 patients received aprotinin (2 × 106 kIU bolus and 0.5-1.0 × 106 kIU/h until 2 hours after reperfusion of the graft).
Blood and Blood Product Transfusion
In the first 24 hours, 85% of patients were transfused RBC with a mean (± standard deviation) of 7.2 ± 7.6 units and a median of 5 units (interquartile range 3-9 units) per patient transfused (Fig. 1A). Intraoperative RBC transfusion accounted for 91% of the RBC transfusions within the first 24 hours; transfusions within the first 24 hours accounted for 96% of the total RBC transfusions occurring within the first 48 hours. The incidence of MBT was 42% (n = 193).
More than 90% of patients received FFP (median 9, interquartile range 6-14 units) (Fig. 1B). Intraoperative FFP transfusion accounted for 90% of all FFP transfusions, whereas 97% of all FFP transfusions occurred in the first 24 hours.
Sixty-four percent of patients received a platelet transfusion (median 5, interquartile range 0-10) (Fig. 1C). Intraoperative platelet transfusion accounted for 87% of all platelet transfusions, with 96% of platelet transfusions occurring within the first 24 hours.
The unadjusted association between potential predictors and MBT are presented in Tables 4 and 5. Of these predictor variables, 12 were considered for inclusion in the logistic regression model: age, height, gender, repeat transplantation surgery, etiology of liver failure, and preoperative laboratory values (hemoglobin concentration, platelet count, INR, albumin, total bilirubin, and creatinine). Although height, weight, and gender were not associated with massive transfusion in the unadjusted analyses, they were included in the logistic regression analyses because of their clinical relevance in transfusion medicine. Continuous variables that were retained in the final model were dichotomized as described in Table 6. Other continuous variables that did not remain in the final model were dichotomized as follows: total bilirubin >22 μmol/L, and partial thromboplastin time >42 seconds. The etiology of liver failure was categorized as primarily obstructive (primary biliary cirrhosis, and primary sclerosing cholangitis) or hepatocellular (alcoholic liver disease, viral hepatitis, and cryptogenic cirrhosis).
|Variable||Transfused <6 RBC 24 h||Transfused ≤6 RBC 24 h||P value|
|Age (yr)||47.4 ± 14.1||52.2 ± 11.8||0.0149|
|Height (cm)||170.1 ± 13.9||169.2 ± 10.1||0.3881|
|Weight (kg)||76.3 ± 17.3||78.5 ± 17.8||0.1772|
|Hemoglobin concentration (g/dL)||11.7 ± 2.2||9.9 ± 2.0||<0.0001|
|Platelet count (× 109/L)||115 ± 79||79 ± 55||<0.0001|
|INR||1.61 ± 0.93||2.02 ± .91||<0.0001|
|PTT (sec)||46 ± 15||52 ± 18||0.0004|
|AST (U/L)||196 ± 596||206 ± 658||0.8626|
|ALT (U/L)||193 ± 649||140 ± 438||0.3236|
|ALP (U/L)||161 ± 174||139 ± 111||0.1143|
|Albumin (g/L)||33.2 ± 7.2||28.7 ± 7.1||<0.0001|
|Total bilirubin (μmol/L)||84 ± 139||141 ± 179||0.0001|
|Amylase (U/L)||71.3 ± 46.4||80.3 ± 80.5||0.1462|
|Creatinine (μmol/ml)||92 ± 73||130 ± 108||<0.0001|
|Variable||Transfused <6 RBC 24 h||Transfused ≤6 RBC 24 h||P value|
|Gender (female)||85 (32%)||61 (32%)||0.9540|
|MELD score, median, (interquartile range)||13.3 (10.5, 17.9)||22 (15.6, 28.8)||<0.0001|
|Coronary artery disease||8 (3%)||10 (5%)||0.2330|
|Diabetes||74 (28%)||62 (32%)||0.3065|
|Renal dysfunction||7 (3%)||10 (5%)||0.1510|
|COPD||25 (9%)||31 (16%)||0.0301|
|Previous abdominal surgery||91 (34%)||77 (40%)||0.2012|
|Living donor||39 (15%)||36 (19%)||0.2420|
|Repeat transplant||7 (3%)||15 (8%)||0.0106|
|Viral hepatitis||146 (55%)||113 (59%)||0.4295|
|Alcoholic liver disease||67 (25%)||52 (27%)||0.6552|
|Primary sclerosing cholangitis||36 (13%)||9 (5%)||0.0015|
|Primary biliary cirrhosis||19 (7%)||12 (6%)||0.7043|
|Fulminant hepatic failure||21 (8%)||17 (8%)||0.8977|
|Cryptogenic cirrhosis||18 (7%)||22 (11%)||0.802|
|Hepatorenal syndrome||13 (5%)||31 (16%)||0.0001|
|Ascites||139 (52%)||136 (70%)||0.0004|
|Encephalopathy||88 (33%)||99 (51%)||<0.0001|
|Esophageal varices||115 (43%)||902 (47%)||0.4483|
|Variable||Description||Estimate||Standard error||P value||Reliability||Risk index†|
|Age (yr)||> 40||0.832||0.393||0.0219||65%||1|
|Hemoglobin (g/dL)||<10.0 g/L||0.756||0.274||0.0068||96%||1|
|Platelet (× 109/L)||≤70||0.682||0.265||<0.0099||98%||1|
|Creatinine (μmol/mL)||Women > 110||1.171||0.292||<0.0001||98%||1|
Independent predictors of MBT that were identified by multivariable logistic regression are listed in Table 6. The regression model had good discrimination (ROC 0.82) and calibration (Hosmer-Lemeshow P value 0.44). Bootstrap validation indicated that all independent predictors retained in the final model had reliabilities in excess of 60% (Table 6). Independent predictors were assigned points in the risk index on the basis of their individual regression β coefficients (Table 6). This risk index had good discriminative performance for predicting MBT with an ROC of 0.79 (bootstrap 95% confidence interval 0.78-0.85).
The MELD score was also found to be associated with risk of MBT (Table 5). When the MELD score was substituted for the independent variables creatinine, INR, total bilirubin, and etiology of liver failure, the model was somewhat less discriminative, with an ROC of 0.791 (Hosmer-Lemshow 0.3532).
We separated patients into 3 categories on the basis of their risk index score of MBT (Fig. 2). Transplant recipients with a risk index ≥3, who comprised 34% of the study sample, had a probability of MBT that exceeded 80%; in contrast, the overall MBT rate in the study sample was 42%. With respect to predicting MBT, a risk index ≥3 had a sensitivity of 58%, specificity of 87%, positive predictive value of 76%, and negative predictive value of 73%.
By using readily available preoperative clinical variables, we have developed and internally validated a clinical prediction risk index that can be used to accurately identify liver transplant recipients at risk of MBT. The discriminative strength of the risk index is consistent with other preoperative predictive risk indices such as those used to predict perioperative cardiac risk in noncardiac surgery.15
All of the 7 independent predictors of MBT have been found to be related individually to blood loss in other studies (Table 1). These predictors are related to disease severity or coagulation status, which may explain their association with MBT. Age, as it is so often in risk analysis studies, is likely a surrogate marker for a predictor variable not measured. In our model, age is one of the least reliable predictors (Table 6), and therefore it is a weak predictor of MBT.
The MELD Score is used to predict short-term survival in patients with end-stage liver failure, and it may have simplified our model. However, we elected not to use the MELD score because it did not improve the strength of the clinical prediction risk index for MBT. Importantly, all of the components of the MELD score, INR, serum creatinine, total bilirubin, and etiology of liver failure, were considered in our model development.
Previous studies that attempted to identify preoperative predictors of MBT in liver transplantation had several important limitations. First, most lacked sufficient statistical power to allow firm conclusions to be drawn. Second, many studies were undertaken in an era of liver transplant surgery where intraoperative factors, such as surgeon experience or the use of venovenous bypass, had an overwhelming influence on perioperative blood loss.1, 3, 16–18 Finally, some investigators included duration of surgery in their statistical modeling; such intraoperative information would not be useful for clinicians attempting to preoperatively assess patients' risk for MBT.19 In the current study, we used a large sample size, carried out appropriate model-building strategies, used preoperative predictors only, and used a clinically relevant outcome.
The transfusion of ≥6 units of RBC has been associated with reduced survival after liver transplantation.2 Furthermore, this definition is consistent with ones employed in cardiac surgery.20 Nonetheless, it is important to note that although there is an association between massive transfusion and poor perioperative outcomes in liver transplant surgery, it remains unclear whether RBC transfusion is an independent predictor of poor outcomes, or simply a surrogate marker for other comorbidities and complications.1, 2, 8, 17, 21
Despite the strengths of our present study, it has limitations inherent to all observational studies. Potentially important preoperative predictors were not included in our analyses because the data were not available or were difficult to interpret retrospectively. For example, portal hypertension, which may significantly affect bleeding in the early stages of liver transplant surgery, is difficult to reliably ascertain by retrospective chart review. In addition, information on graft quality (e.g., prolonged cold storage time, fatty liver, or elderly donor) that may affect postreperfusion and posttransplantation hemostasis was not included in the model.
The surgical technique was not included in the development of the risk index for massive transfusion. At our institution, the standard of care is the classical technique where the recipient's transhepatic vena cava is replaced with the donor vessel. The piggyback technique, where the donor hepatic veins are grafted to the recipient's vena cava, was reserved for all living-donor recipients and those cadaveric donor recipients who were not able to tolerate a complete cross-clamp on the vena cava. This confounding by indication selection bias therefore precluded the inclusion of surgical technique in the prediction rule.
Similarly, we did not include antifibrinolytic use in our risk index, although there is evidence that antifibrinolytics may reduce RBC transfusion in liver transplantation surgery by 30% and 60%.6, 22 However, given that antifibrinolytic use was at the discretion of the surgical team and therefore targeted higher-risk patients, the inclusion of antifibrinolytics in the analysis would have been affected by confounding by indication. Furthermore, only a small fraction of the study sample actually received antifibrinolytics at doses shown to be effective in reducing blood transfusion requirements: tranexamic acid infusions (10 mg/kg/mL) or aprotinin.5, 6, 22
Although the risk index was internally validated by bootstrapping, external validation is needed. Given the lack of standardized protocol for antifibrinolytic management, the generalizability of the model should not be assumed until external validation is performed. If the risk index is externally valid, it may serve several important functions. First, the categorization of patients preoperatively based on risk of MBT may guide modifications to the intraoperative management based on risk-benefit considerations.
One example is the use of a procoagulant, recombinant factor VIIa, that has been found to reduce blood loss in a number of situations.23, 24 In liver transplantation surgery, however, single-dose factor VIIa does not reduce blood transfusions,25 and a multicenter international trial of multiple doses of factor VIIa found only modest reductions in perioperative transfusion rates.26 Nonetheless, by using a preoperative risk index, it may be possible to identify high-risk patients in whom this therapy may be more beneficial and for whom the risk-benefit balance of this therapy may favor benefit.26
Second, the early identification of patients at risk of MBT may provide the opportunity to develop unique clinical pathways and test novel approaches to prevent blood loss during liver transplant surgery. Previous studies investigating the efficacy of interventions (antifibrinolytics and factor VIIa) to reduce blood loss were underpowered, in part because of the heterogeneity of transfusion requirements of the enrolled subjects.5, 26 Preoperative identification of a high-risk group may improve the design of future clinical trials assessing antifibrinolytics or other interventions (e.g., 1,desamino-8-D-arginine vasopressin, factor VIIa).
In conclusion, we developed a prediction rule for MBT in liver transplantation surgery based solely on preoperative predictor variables. If externally generalizable, this index has the potential to guide clinical management of liver transplant surgery and improve the design of clinical trials evaluating interventions to reduce transfusion requirements.
- 10Spline functions for logistic regression modeling. Annual SAS Users Group International Conference. Cary, NC: SAS Institute; 1986: 646–651., .
- 11http://biostat.mc.vanderbilt.edu/twiki/bin/view/main/SASmacros. Accessed July 1, 2006.. SAS macros and data step programs useful in survival analysis and logictic regression.
- 12Multivariable Analysis: A Practical Guide for Clinicians. Cambridge, Mass: Cambridge University Press; 1999..
- 13FeinsteinAR, FeinsteinAR, ed. Multiple Logistic Regression, Multivariable Analysis: An Introduction. New Haven, Conn: Yale University Press; 1996.
- 14LezzoniLI, ed. Evaluating the Performance of Risk Adjustment Methods: Dichotomous Outocmes. Risk Adjustment for Measuring Healthcare Outcomes. 2nd ed. Chicago, Ill: Health Aministration Press; 1997.