Anesth Analg 2004;99:1239-1244
© 2004 International Anesthesia Research Society
doi: 10.1213/01.ANE.0000132928.45858.92
GENERAL ARTICLES
Predicting Allogeneic Blood Transfusion Use in Total Joint Arthroplasty
Saifudin Rashiq, MB MSc, FRCPC,
Meera Shah,
Ava K. Chow, MSc,
Paul J. O’Connor, MB FFARCSI, and
Barry A. Finegan, MB FRCPC
Department of Anesthesiology and Pain Medicine, University of Alberta, Edmonton, Alberta, Canada
Address correspondence to Saifudin Rashiq, MB, MSc, 3B2.32, 8440–112 Street, Edmonton AB Canada T6G 2B7. Address email to srashiq{at}ualberta.ca
 |
Abstract
|
|---|
Total joint arthroplasty (TJA) patients often receive allogeneic blood transfusion. In this study we sought to create and validate a clinical prediction rule for transfusion in TJA using data that are easily available when scheduling the procedure. Logistic regression modeling was applied to retrospective data from all TJA procedures performed in Edmonton, Alberta in 2000 (n = 1875). The area under the receiver operating curve for the resulting model in the training and validation data sets was 0.80 and 0.76 respectively. By assigning a simple score based on six independent predictors (age, gender, weight, hemoglobin, ASA operative risk classification and whether revision surgery was planned), it was possible to classify a given subject’s risk of receiving allogeneic transfusion. We conclude that accurate prediction of transfusion risk in TJA is possible using a rule based on simple preoperative clinical and laboratory data. Such prediction could allow transfusion prevention strategies to be applied selectively to those at greatest risk.
IMPLICATIONS: The use of allogeneic blood in patients undergoing joint replacement surgery was modeled statistically. A prediction rule was created from this model. It estimates a given patient’s risk of transfusion during total joint arthroplasty and can be used to target transfusion risk reduction measures more effectively.
|
Introduction
|
|---|
Allogeneic blood transfusion is frequently necessary during the perioperative period of total joint arthroplasty (TJA). The potential hazards of allogeneic blood transfusion, although rare, are serious and have been the focus of a highly public debate (1).
A number of interventions can be electively performed before TJA that have been shown to be effective in reducing exposure of patients to allogeneic blood. These include administration of erythropoietin and supplemental iron and folic acid, and autologous blood donation. However, the collection and transfusion of autologous blood is expensive (2), and approximately 30% of autologous units that are collected are not used (3). Perioperative tranexamic acid infusion (4), acute normovolemic hemodilution (5), and perioperative red cell salvage (6) are examples of other strategies, involving significant added cost, that have been successfully used to reduce allogeneic transfusion in some centers. To make the best use of these interventions, a method for identifying those patients who are at the highest risk of transfusion is required. The purpose of this study was to use the transfusion experience of TJA patients to develop and validate a clinical prediction rule for allogeneic red cell transfusion in TJA using data that are readily available to the clinician at the time of scheduling the procedure.
|
Methods
|
|---|
The Health Research Ethics Board of the University of Alberta approved the study. All TJAs performed in our health region in the year 2000 were identified and a retrospective chart review of these cases was then performed. Research personnel collected the required data from patient charts and entered them directly onto computerized spreadsheets. These were amalgamated into a single database, and the spreadsheets used for data collection were then destroyed. The final database was stored under password protection on CD-ROM in a secure location. Biometric and clinical variables collected included age, weight, height, preoperative hemoglobin, ASA risk class, anesthetic type, surgeon, surgery time, postoperative hemoglobin, and length of hospital stay. Whether subjects received autologous or allogeneic red cells during or after the procedure was also recorded. In a 10% random sample of charts, the data were collected independently by two separate individuals and compared by a third person for abstraction and coding errors; less than 1% errors were found.
The data set was imported into SAS for Windows Version 8 (SAS, Cary, NC). Subjects who had donated autologous blood were removed from the analysis. Univariate analysis was performed to look for associations between receiving allogeneic red cells (in any amount) and each of the predictor variables in the database. Predictor variables were compared in the transfused and nontransfused series by Student’s t-tests or 
2 test, as appropriate. A statistical correlation between number of cases per surgeon and transfusion rate was sought using Pearson’s coefficient.
Predictor variables that showed a statistically significant relationship (P < 0.05) with the risk of transfusion on univariate testing and that could have been known at the time of scheduling the procedure were considered for inclusion in the regression modeling process.
The data set was randomly divided into a training set and a validation set with a selection probability for either set of 0.5. Data from subjects in the validation set were quarantined until the completion of the modeling process. A logistic regression model was constructed using the maximum likelihood function. The probability modeled was that the subject received allogeneic red cell products, in any amount, as a binary variable. Nominal variables were used in binary format. One ordinal variable was used (ASA physical status classification), and this was dichotomized into high (ASA III–V) and low (ASA I–II). Continuous variables (age, starting hemoglobin, weight, height) were categorized into a priori numerically convenient groups and indicator variables created from each category. Indicator variables were also constructed depending on whether the surgeon concerned had performed fewer than 50, between 50 and 100, or more than 100 TJAs in the year of study.
Forward and backward stepwise logistic regression modeling was performed. The two models were compared and further minor adjustments made by hand. The goal of this iterative process was to achieve a reasonable model fit in the training set with as parsimonious a variable set as possible.
The odds ratio for transfusion for each predictor variable was multiplied to yield an arithmetically convenient integer, and these new covariates were used to create a scoring system. The scores were categorized into 4 groups, corresponding to approximate predicted allogeneic transfusion risk respectively of 10% or less, 10%–30%, 30%–50%, and >50%. Then, the resulting prediction rule was applied only once to the validation set after completion of all phases of the modeling and the scoring system. Receiver operating characteristic curves were constructed for the model in both data sets.
Finally, we compared the discriminative ability of our clinical prediction rule with a previously published rule (7) on the same data set (our validation set).
|
Results
|
|---|
Fifty-seven autologous blood donors were excluded, leaving 1818 subjects to be included in the analysis (Table 1). Five-hundred-five subjects (28%) were transfused with banked blood and the mean amount used was 2.36 U per transfused case, resulting in a mean donor unit exposure for the whole series of 0.67 U per case. Three percent of transfused units were given intraoperatively, and the majority of the remainder was given between 18 and 48 h postoperatively.
Positive associations were found between transfusion risk and increasing age, female gender, low body weight, short stature, preexisting anemia, surgical time, the use of only general anesthesia, ASA class more than II, and revision surgery (Table 2). There was no difference in transfusion rate between hip or knee arthroplasty. Individual surgeons performed a median of 58 arthroplasties during the year (range, 1–185). Surgeon-specific transfusion rates varied from 0% to 50%, but there was no correlation between surgical volume and transfusion rate (r = 0.093). A hemoglobin measurement was recorded no more than 12 h before the commencement of blood transfusion in 471 of the 505 transfused cases. The mean hemoglobin before transfusion (trigger hemoglobin) was 81 g/dL (SD 8 g/dL), and 80% of transfused cases had a trigger hemoglobin between 71 and 89 g/dL. Surgeon-specific transfusion triggers did not vary greatly. There was no difference in the risk of transfusion between hospital sites (P > 0.05).
The division of the data set resulted in creation of two sets (training set and validation set) that, although slightly unequal in size, were closely comparable in all other respects (Table 1). The final model is shown in Table 3. The area under the receiver operating characteristic curve for the model in the training set was 0.80. The corresponding statistic when the model was applied to the validation set was 0.76 (Fig. 1).
Table 4 shows the clinical prediction rule for transfusion, developed from the odds ratios generated by the model. Table 5 shows the ability of this scoring system to identify patients who are at increased risk of perioperative transfusion. The rule affords the ability to discriminate between those at high and low risk of transfusion. It is able to identify a large subgroup (46%) for whom transfusion risk is approximately 13% and for whom special measures to prevent transfusion are therefore likely to have a low yield. At the other extreme, 9% of TJA patients have a predictable transfusion risk in excess of 60% and would therefore be good candidates for such measures.
|
Discussion
|
|---|
We have developed a clinical prediction rule that allows a clinician to make an accurate assessment of his patient’s risk of receiving allogeneic transfusion during TJA. The prediction rule uses only simple, intuitive clinical and laboratory data and effectively separates those at the lowest and highest risks. The rule could be adapted to specific situations by creating narrower risk bands from the model to obtain more precise estimates of individual risk or to separate patients across other a priori risk thresholds.
In TJA, prediction of transfusion risk is important for both the individual and the health care system. These procedures can be scheduled in advance, allowing time for preoperative interventions to increase circulating red cell mass. Such prediction and preventative treatment would assist those who manage our blood resources in two tangible ways: by reducing or eliminating the need for blood in a large proportion of individual TJA patients and by allowing better estimates of the blood product needs of TJA population before they enter hospital.
Several previous studies have identified variables that may contribute to an increased transfusion risk in TJA such as low preoperative hemoglobin (8,9), low body weight (10), increasing age (11), higher ASA classification (12), female gender (13), and revision surgery (second and subsequent operations performed after TJA to replace malpositioned or nonfunctioning hardware components) (14). We concur with the importance of each of these factors individually. Our model gives the largest predictive weight of all to low preoperative hemoglobin, a predictive factor identified as uniquely important in previous work (9).
The use of regional anesthesia reduces the risk of transfusion in hip arthroplasty (15) and we were surprised to see that it exerted such a weak protective effect against transfusion in our TJA group as a whole. Secondary analysis indicates that regional anesthesia protects against transfusion in hip but not in knee arthroplasty (data not shown). In any event, whether or not regional anesthesia will successfully be used in TJA cannot be known at the time of scheduling the case, and this variable could not therefore be used in the model.
We excluded the 57 autologous blood donors from the analysis because we wanted our prediction rule to be applicable to the many centers that perform TJA but that do not have access to an autologous blood transfusion program.
One example of published models in this area similar to ours is that of Larocque et al. (7). Their model was derived using from a smaller series of 599 selected TJAs performed in their center over a 5-year period. It was then validated prospectively on a consecutive sample of 460 TJAs. Inexplicably, their patient population had very different transfusion rates depending on which of the two hospitals performed the treatment, which was not the case in our study. The number of actual transfusions analyzed was corrected for appropriateness (based on extant clinical guidelines) and it was concluded that 65% and 90% of the transfusions given at each of the two sites, respectively, were clinically inappropriate. The analysis also included a small number of individuals who underwent simultaneous bilateral arthroplasties (which are not performed in our center). Their model does not incorporate three factors that we found to be independent predictive factors of the likelihood of receiving a transfusion (age, gender, and ASA class) but includes arthroplasty type (hip versus knee), which we found to be irrelevant. We believe our model to be more robust based on these considerations. To confirm this hypothesis, we applied both models to the same data set and found that our rule provides superior discrimination, particularly at the higher end of transfusion risk (data not shown).
Our data were collected retrospectively from sources not specifically intended for research. The data on blood product use, however, were drawn from the computerized blood product inventory management system and confirmed by manual review of the patient record and are therefore likely to be accurate. We did not specifically record the use of intraoperative cell salvage and reinfusion, as this technique is used very rarely in our region. The decision to transfuse in any given subject was made by the clinician and not in accordance with a rigorous transfusion protocol specifically established for the duration of this study. There was, however, a very high degree of uniformity in transfusion practice. Each of the 28 surgeons in our study initiated allogeneic transfusion with a mean trigger hemoglobin in the range 76–86 g/dL (the mean trigger hemoglobin for the entire series was 81 g/dL). These data suggest that transfusion practice closely followed current Canadian guidelines for allogeneic transfusion (16).
The analysis is limited by the number of variables collected; it cannot be extrapolated to situations in which the risk of transfusion is likely to be affected by factors such as bleeding diathesis, anticoagulant therapy, or nonclinical reasons for avoiding blood transfusion. In addition, although we have performed internal validation of our model, we have not tested its generalizability to other populations in which clinical practices (e.g., the use of intraoperative red cell salvage, autologous transfusion, and acute therapeutic hemodilution) might differ.
What might be the practical consequence of our findings? One hypothesis might be that the use of our prediction rule could be used to target resources more effectively. For example, 505 of our 1818 subjects (29%) received at least one unit of allogeneic blood during or after TJA. The only modifiable preoperative risk factor for transfusion is low preoperative hemoglobin. If everyone in our series had been treated to increase his or her hemoglobin to 140 g/dL or more, the number of subjects requiring transfusion would have decreased by 25% to 377, but 981 subjects would have received treatment. Instead, if the prediction rule had been applied to selectively increase hemoglobin to 140g/dL only in those with a point score of 200 or more, slightly more subjects (385 versus 377) would have been transfused, but far fewer (318 versus 981) would have received treatment. This selection process might therefore have saved 663 subjects from the cost and inconvenience of treatment that would have made little difference in their risks of transfusion. An alternative way to use the prediction rule might be to try to achieve the greatest possible sparing of allogeneic blood. Such a plan might, for example, call for everyone to be treated to achieve a hemoglobin level of at least 140 g/dL, but additionally for those at the highest risks of transfusion (those with a points score of 150 or more) to be treated to achieve 150 g/dL. Hypothetically, this process would have reduced the number transfused subjects to 307, although 1040 would have required treatment to achieve this. The projected results of this very ag- gressive strategy illustrate the limitations of this process because 17% of patients would still have been transfused. Reducing this number further would require approaches other than optimization of preoperative hemoglobin levels.
We conclude that prediction of allogeneic blood transfusion risk during TJA is possible using a clinical prediction rule. We offer this strategy as a potential way to reduce transfusion by identifying those who would most likely benefit from preoperative intervention. Such a strategy would be worthy of evaluation as part of an overall orthopedic perioperative blood conservation program.
|
Acknowledgments
|
|---|
Supported, in part, by the Department of Anesthesiology and Pain Medicine.
|
References
|
|---|
- Commission of Inquiry on the Blood System in Canada. Final report (Krever Commission). Catalog number CP32–62–3–1997E. Ottawa: Canadian Government Publishing, 1997.
- Marchetti M, Barosi G. Cost-effectiveness of epoetin and autologous blood donation in reducing allogeneic blood transfusions in coronary artery bypass graft surgery. Transfusion 2000; 40: 673–81.[Medline]
- Savoia HF, Metz J, Maxwell EL, et al. Utilization of preoperative autologous blood donation in elective surgery. Aust N Z J Surg 2002; 72: 557–60.
- Husted H, Blond L, Sonne-Holm S, et al. Tranexamic acid reduces blood loss and blood transfusions in primary total hip arthroplasty: a prospective randomized double-blind study in 40 patients. Acta Orthop Scand 2003; 74: 665–9.[ISI][Medline]
- Shulman G, Grecula MJ, Hadjipavlou AG. Intraoperative autotransfusion in hip arthroplasty. Clin Orthop Rel Res 2002; 396: 119–30.
- Zarin J, Grosvenor D, Schurman D, Goodman S. Efficacy of intraoperative blood collection and reinfusion in revision total hip arthroplasty. J Bone Joint Surg Am 2003; 85: 2147–51.[Abstract/Free Full Text]
- Larocque BJ, Gilbert K, Brien WF. Prospective validation of a point score system for predicting blood transfusion following hip or knee replacement. Transfusion 1998; 38: 932–7.[ISI][Medline]
- de Andrade JR, Jove M, Landon G, et al. Baseline hemoglobin as a predictor of risk of transfusion and response to Epoetin alfa in orthopedic surgery patients. Am J Orthop 1996; 25: 533–42.[Medline]
- Faris PM, Spence RK, Larholt KM, et al. The predictive power of baseline hemoglobin for transfusion risk in surgery patients. Orthopedics 1999; 22: s135–40.[Medline]
- Salido JA, Marin LA, Gomez LA, et al. Preoperative hemoglobin levels and the need for transfusion after prosthetic hip and knee surgery: analysis of predictive factors. J Bone Joint Surg Am 2002; 84: 216–20.[Abstract/Free Full Text]
- Marx RG, Wotherspoon S, Stephens D, Davey JR. Patient factors affecting autologous and allogeneic blood transfusion rates in total hip arthroplasty. Am J Orthop 2001; 30: 867–71.[Medline]
- Grosflam JM, Wright EA, Cleary PD, Katz JN. Predictors of blood loss during total hip replacement surgery. Arthritis Care Res 1995; 8: 167–73.[Medline]
- Keating EM, Meding JB, Faris PM, Ritter MA. Predictors of transfusion risk in elective knee surgery. Clin Orthop 1998; 357: 50–9.
- Cushner FD, Foley I, Kessler D, et al. Blood management in revision total knee arthroplasty. Clin Orthop 2002; 404: 247–55.
- D’Ambrosio A, Borghi B, Damato A, et al. Reducing perioperative blood loss in patients undergoing total hip arthroplasty. Int J Artif Organs 1999; 22: 47–51.[ISI][Medline]
- Expert Working Group. Guidelines for red blood cell and plasma transfusion for adults and children. Can Med Assoc J 1997; 156 (11 suppl): S1.
Accepted for publication May 4, 2004.
Top
Abstract
Introduction
