SPINE surgery is considered at risk of significant intraoperative bleeding in adult patients.1–3
Important variability in intraoperative bleeding and erythrocyte requirements has been reported in adult patients undergoing major spine surgery.4
Therefore, predicting the need for allogeneic erythrocyte transfusion based on patient preoperative characteristics would be helpful (1) to identify the patient subpopulations undergoing spine surgery at risk of massive perioperative transfusion and encourage erythrocyte-saving strategies in these patients, (2) to improve patient information on their perioperative erythrocyte requirements, and (3) to properly allocate blood and mobilize donors. In adult patients, several lines of evidence support that blood loss is particularly frequent and important during surgery of spine tumors and arthrodesis with posterior incision.5–8
Age, anemia, osteotomy, and fusion have been identified as risk factors for bleeding in the context of spine surgery.5–7
In most of these studies, however, measurement of blood loss was restricted to the intraoperative period, which is likely to have underestimated total perioperative blood loss. The goal of the current study was to derive a model based on transfusion up to 5 days after surgery. For this purpose, the Predictive Model of Transfusion in Spine Surgery (PMTSS), based on preoperative characteristics, was generated to determine the individual probability of erythrocyte transfusion in adult patients undergoing spine surgery.
Materials and Methods
The protocol was approved by the institutional review board of the Groupe Hospitalo-Universitaire de Paris Nord, Bichat University Hospital, Paris, France, and informed consent was obtained from patients.
Patients scheduled to undergo elective spine surgery were included. More than 50% of the patients had an American Society of Anesthesiologists physical status of II or III. Exclusion criteria were exclusive cervical spine surgery, one-level laminectomy, and polytrauma. Clinical management was at the discretion of the attending anesthesiologists and surgeons. In the operating room, patients were continuously monitored with electrocardioscopy, blood pressure monitoring, pulse oximetry, capnography, and esophageal temperature monitoring. A Bair Hugger device (Arizant, Eden Prairie, MN) was used, and fluids were warmed. Anesthesia was induced with propofol (1.5–2.5 mg/kg), sufentanil (15 μg), and atracurium (0.7 mg/kg) and was maintained by a continuous infusion of sufentanil and atracurium, with desflurane in a 50%–50% vol/vol O2
O gas mixture. The rate of the sufentanil infusion and the inspired concentration of desflurane were adjusted to maintain mean blood pressure between 50 and 70 mm Hg without decreasing lower than 20% of the preanesthetic value measured immediately before induction of anesthesia. Patients were carefully placed in the decubitus, procubitus, or knee–chest position, and great attention was paid to protect the eyes from mechanical injury and to preserve thorax movements and inferior vena cava venous return. Tranexamic acid (1-g bolus before incision followed by a 10-mg · kg−1
continuous infusion until skin closure) was used at the discretion of the attending anesthesiologist. Hemoglobin blood levels were repeatedly checked by a Hemocue device (Hemocue France, Meaux, France) (at least before surgical incision and before erythrocyte administration). A cell saver for intraoperative blood salvage was available in the operating room. Transfusion criteria were those established on the basis of the recommendations made by the Agence Française de Sécurité Sanitaire des Produits de Santé and the American Society of Anesthesiologists Task Force on perioperative blood transfusion and adjuvant therapies.9,10
Briefly, erythrocyte transfusion was initiated intraoperatively or postoperatively in all cases when the hemoglobin level was lower than 7 g/dl. It was considered for hemoglobin levels between 7 and 10 g/dl depending on the cardiopulmonary reserve of the patients. After completion of surgery, patients were discharged to the postanesthesia care unit. Postoperative thromboprophylaxis with low-molecular-weight heparin molecules were started as soon as the hemorrhagic risk seemed minimal.
The following parameters were recorded from the patients' intraoperative and postoperative charts: demographic data (age, sex, weight, body mass index, American Society of Anesthesiologists physical status), duration of the surgery, number of levels, type of operation (laminectomy, fusion, transpedicular osteotomy, osteosynthesis), underlying disease (posttraumatic, tumoral, septic, degenerative), intraoperative heart rate and blood pressure recorded every 10 min, necessity of reoperation, hemoglobin levels measured the day before surgery and on postoperative day 5, use of antifibrinolytics and epoetin, number of patients undergoing allogeneic (autologous, respectively) blood transfusion and number of allogeneic (autologous, respectively) erythrocyte units transfused, hemoglobin value immediately before transfusion, number of patients undergoing blood salvage, and volume of blood transfused. In the operating room, blood loss was estimated by hourly measurement of the volume of fluids aspirated by the surgeon and the hemoglobin level. Total erythrocyte loss (including both compensated and uncompensated loss) from the day of operation until postoperative day 5 was calculated by the appropriate formulas11–13
Total erythrocyte loss = compensated + uncompensated erythrocyte loss
Uncompensated blood loss = total blood volume × (hematocrit D0 − hematocrit D5),
where total blood volume = 70 ml/kg (65 ml/kg) in males (females),18
hematocrit at day 0 (D0) corresponds to the preanesthetic hematocrit, and hematocrit D5 is the hematocrit on postoperative day 5.
Compensated blood loss = sum of all erythrocyte received from all sources of transfusion (allogeneic, autologous, cell saver, etc.).
A 250-ml erythrocyte unit with a hematocrit = 60% corresponds to 150 ml of pure erythrocytes (100% hematocrit). The mean hemoglobin level of blood obtained via the cell saver was 20 g/dl.
Compensated erythrocyte loss = (number of erythrocyte units transfused × 150) + (volume salvaged × 0.3).
Between January 2006 and March 2007, all consecutive adult patients having undergone major elective thoracolumbar spine surgery were retrospectively included. We hypothesized that the population of inference would be future comparable patients in our center or possibly in other spine surgical centers. Because the transfusion rate in our spine surgical population was approximately 25%, and taking into consideration that we planned to derive a four- or five-variable logistic regression model, at least 50 patients receiving erythrocyte transfusion would need to be present in the sample, 50 × 4 = 200 patients at least to be enrolled.14
Univariate descriptive statistics were performed to describe the patient demographics, characteristics of surgery, intraoperative and postoperative bleeding and erythrocyte transfusion (incidence and number of erythrocyte units). Bivariate statistics were used to examine the relation between the outcome variables (transfusion of erythrocytes) and other variables related to patients and/or surgery. Multivariable analysis was performed by backward stepwise selection of a restricted number of independent variables selected from the bivariate analysis.15
Only independent preoperative variables related either to the patients (age, sex, body mass index, weight, size, preoperative hemoglobin level) or to surgery (reoperation, number of levels for laminectomy, number of levels for fusion, osteotomy) were entered into the multivariable analysis after intervening variables (intraoper-ative blood loss, duration of surgery) and underlying disease had been excluded. Interaction between these factors (data not shown) were tested and considered nonsignificant. Only predictors significantly associated with homologous erythrocyte transfusion (P
< 0.05) in the multivariable analysis were factors to be included in the PMTSS. The detailed process of PMTSS generation is reported in the appendix
. The discriminating capacity of the PMTSS to predict the probability of erythrocyte transfusion was estimated by use of a receiver operating characteristic (table and curve) analysis.16,17
Correlation of the PMTSS with the number of homologous erythrocyte units transfused was analyzed by the Spearman ρ correlation coefficient.
The PMTSS was then tested in a prospective validation cohort. There was a theoretical risk (called contamination bias
) that practice may have been influenced by the score value collected during the validation period, because both predictors of erythrocyte transfusion and outcome (erythrocyte transfusion) were not assessed in a blinded fashion. Therefore, the duration of the validation study had to be minimized.18
Because the transfusion rate in the derivation sample was 32%, dealing with a four-variable model (10 events per variable, 1 event = 1 transfused patient), 125 patients were required in the validation sample.14
The same descriptive analysis (univariate and bivariate statistics, multivariable model checking) as in the derivation cohort was performed to describe the validation sample. The calibration of the score in predicting the individual probability of erythrocyte transfusion was examined by fitting the data (i.e.
, transfusion rates at each score level in the validation set) to expected transfusion frequencies (probabilities) obtained from the derivation set. Then, the observed frequencies of transfusion at each score cutoff value in the validation set were plotted against the corresponding predicted positive values of erythrocyte transfusion. The predictive positive values were calculated from the observed prevalence of transfusion and intrinsic diagnostic characteristics (sensitivity, specificity, likelihood ratios) derived from the derivation study.
Statistical analysis was performed using Excel® version 11.3.5 (Microsoft Corporation, Redmond, WA; 2004) and JMP version 7 (SAS Institute Inc., Cary, NC; 1989–2007). Data are expressed as mean and SD for variables following normal distribution and as median [25–75 interquartile range] for others. Normality of distributions and equality of variances were checked by using the goodness-of-fit (normal distribution) Shapiro–Wilk W test before parametric statistics were undertaken. Bivariate analysis was performed by analysis of variance and the Student t test for continuous dependent variables (corrections for unequal variances were used when needed). Ordinal and nonnormally distributed dependent variables were analyzed by the nonparametric rank sum Wilcoxon/Kruskal–Wallis tests. The Fisher exact test (two-tailed) was used for both binary dependent and independent variables. Simple logistic regressions were performed when the outcome variable was binary (transfusion) and independent variables were continuous. Pearson–Yates chi-square tests were used for nominal variables. The Spearman rank correlation coefficient was calculated for studied correlations. Multivariable analyses were performed using linear regression and nominal logistic models according to the dependent variables selected. Estimate parameters were tested by Wald tests. Recursive partitioning was used after identifying predictors to help determine optimal cutoff points taking natural significance and size conditions into consideration. Confidence intervals (CIs) for areas under receiver operating characteristic curves were calculated according to the Hanley and Hill methods. Those corresponding to proportions were calculated by the Wilson method. P < 0.05 was considered the threshold for significance.
Two hundred thirty consecutive patients were included in the derivation population. The median [range] total erythrocyte loss was 1,437 [106–9,070] ml. Intraoperative blood salvage was used in 32 of 230 patients (13.9%). The allogeneic transfusion rate was 74 in 230 (32%), and the number of erythrocyte units administered in transfused patients was 3.5.1,13
Of the 286 transfused erythrocyte units, 212 (74%) were administered intraoperatively. The remaining erythrocytes were administered postoperatively (14.3% after postoperative day 1). The preoperative hemoglobin level was 13.2 [7.4–16.6] g/dl. This value was less than 12 g/dl in 45 of 230 patients (19.5%) and less than 10 g/dl in 23 of 230 patients (10%). The incidence of the use of antifibrinolytics in this cohort was 17.4% (66.7% in the validation cohort).
Independent variables associated with allogeneic erythrocyte transfusion identified by bivariate analysis and the results from multivariable analysis with selected factors for allogeneic transfusion are shown in table 1
. Four independent preoperative predictors of homologous erythrocyte transfusion were identified: age older than 50 yr (adjusted odds ratio [aOR] = 4.9; 95% CI, 2–13.5), preoperative hemoglobin level less than 12 g/dl (aOR = 6.9; CI, 3.1–17.2), fusion of more than two levels (aOR = 6.7; CI, 3.1–15.2), and transpedicular osteotomy (aOR = 19.9; CI, 5.6–98.2). The discriminant capacity of the PMTSS obtained from transformation of the four factors identified as independent predictors for allogeneic erythrocyte transfusion in the multivariable analysis was good (area under the curve = 0.86; 95% CI, 0.81–0.92; fig. 1
). Of note, when the score was greater than 2, it had a sensitivity of 0.74, a specificity of 0.87, a positive likelihood ratio of 5.8, a negative likelihood ratio of 0.29, a positive predictive value of homologous erythrocyte transfusion of 0.73, and a predictive negative error of 0.12. The corresponding individual probabilities of perioperative erythrocyte transfusion were 0, 7, 19, 54, and 90% for PMTSS values of 0, 1, 2, 3, and 4, respectively.
Patient characteristics were similar compared with the derivation sample (narrow validation; table 2
). No comparison was justified. The transfusion rate was slightly increased in the validation set (37% vs.
32% in the derivation set), and osteotomy was less frequently performed (6.4% vs.
12.6%). Bivariate and multivariate statistics revealed similar trends to identify preoperative independent predictors of erythrocyte transfusion. There was a good fit between the observed rates of transfusion and the probabilities of transfusion obtained from the model (regression line: y = 0.98x + 0.04, P
< 0.0001; goodness of fit χ2
= 0.125). We found a good fit between the observed transfusion rates at different score thresholds and the corresponding positive predictive values generated by the score (y = 0.90x + 0.04, P
< 0.0001; figs. 2 and 3
). The score also correlated with the number of erythrocyte units transfused (Spearman = 0.61, P
< 0.0001). The receiver operating characteristic curve analysis in the validation set showed a very good discriminating capacity of the PMTSS (area under the curve = 0.83 [0.75–0.91]; fig. 4
). The corresponding individual probabilities of perioperative erythrocyte transfusion were 0, 4, 35, 64, and 88% for PMTSS values of 0, 1, 2, 3, and 4, respectively.
Data Presented in the Appendix
This section contains material for readers specifically interested in epidemiologic or methodologic aspects of this study. Details of the generation of the score from the four selected independent predictors of erythrocyte transfusion are given in the appendix.
The main original findings of the current study can be summarized as follows: A 0–4 score based on preoperative patient characteristics, the PMTSS, was derived from a large cohort of adult patients undergoing major elective thoracolumbar spine surgery and prospectively validated. The strategy used here was pragmatic, and differed from a classic explanative attitude when using a multivariable analysis insofar as a selection process of the variables was decided on specific criteria, and the study was powered to address this goal. The PMTSS was found a reliable predictive model of allogeneic transfusion requirements in adult spine surgery. It may be useful in clinical practice to identify patients undergoing spine surgery at risk of particularly important transfusion, properly allocate blood, and encourage erythrocyte-saving strategies in these patients.
Four independent predictors for erythrocyte homologous transfusion were identified in patients undergoing major spine surgery and entered into the process of generating the PMTSS. Our study confirms and extends the prevalence of preoperative anemia (defined as a hemoglobin level less than 13 g/dl for males and 11.5 g/dl for females), which is very close to those reported by others for patients presenting for major orthopedic surgery, including spine surgery.5,19
Measurements of blood loss were obtained from the day of operation until postoperative day 5. This represents an original finding because most of the studies had focused only on intraoperative blood loss.5–8
The intraoperative median blood loss reported in our study was also consistent with the values previously reported for spine surgery.5–7
Interestingly, the range of blood loss was broad, indicating that some patients have undergone massive perioperative hemorrhage. Of note, Murrey et al. 4
reported an average intraoperative blood loss of 2,342 ml for transpedicular osteotomy, with maximal bleeding of 9,000 ml in some cases. Intraoperative blood salvage was used in a minority of patients (13.6%). We used validated criteria recommended to initiate erythrocyte transfusion in both the derivation and the validation cohorts.9
These criteria were fairly consistent with those published later by the American Society of Anesthesiologists Task Force.10
The adequacy of this strategy is supported by the median hemoglobin levels found at postoperative day 5 (10.6 g/dl in the derivation cohort and 10.8 g/dl in the validation cohort). These moderate values support that erythrocyte transfusion was not overused in our study.
We found that fusion surgery involving more than two levels was an independent risk factor predictive of allogeneic transfusion. The number of levels fused has been previously shown to be a risk factor for intraoperative bleeding in spine surgery.4–6,20
Our results confirm and extend these findings by showing that this risk factor also applies to postoperative bleeding. We also found that transpedicular osteotomy was a major independent risk factor of blood loss in this context. Transpedicular osteotomy represents major reconstructive surgery and is associated with a high rate of complications, including severe hemorrhage.21,22
Indeed, because of the nature of the resection, bleeding originates from the vertebra itself. This explains that control of intraoperative bleeding is particularly difficult for the surgeon in this situation.
Age older than 50 yr was the third independent predictive factor of allogeneic erythrocyte transfusion, as was suggested previously in one study.6
This factor was still present after adjustment of blood loss to the type of surgical procedure. Several hypotheses can be proposed to account for this finding. This could be explained by an increased prevalence of comorbidities such as hypertension or situations for which medications interfering with coagulation or platelet function were required. Aspirin intake has been reported to increase the risk for bleeding and transfusion in the case of hip arthroplasty.23
On the other hand, data on the effect of clopidogrel in patients scheduled to undergo noncardiac surgery are scarce, and their levels of proof are low.24
Alternatively, patients older than 50 yr had more posterior fusions (46% vs.
22.4%), fewer anterior fusions (4.3% vs.
14.9%), and less posttraumatic disease (3.1% vs.
20.9%) than those aged 50 yr or younger, which may contribute to explain this result.
The generation of a simplified transfusion risk score, the PMTSS, represents a major original result of this study. A classification bias was unlikely to be present, because the variables collected were objective ones (age, hemoglobin level, fusion of more than two levels, transpedicular osteotomy). This supports reproducibility of our results. Similarly, it could be argued that an observation bias may have contributed to decrease the incidence of homologous erythrocyte transfusion in the validation population and subsequently to decrease the discriminant capacity of the score. This seems unlikely to be the case, because the discriminant capacity of the PMTSS was very good (area under the curve = 0.83) in the validation population. The cutoff points were defined a posteriori to comply with clinical coherence and sample size. This choice emphasizes the importance of the prospective validation of the PMTSS. The discriminant capacity of the PMTSS was remarkable (area under the curve > 0.80) in the derivation population. Small sample size and insufficient overlapping could explain the nonsignificant statistical value of osteotomy when adjusted to the other cofactors of the model in the validation population. Indeed, the PMTSS is a four-variable model with three of them giving two possibilities for a patient (age > or ≤ 50 yr, osteotomy or no osteotomy, fusion > 2 levels or ≤ 2 levels) and one giving three possibilities for the last variable (preoperative hemoglobin > 14 or between 12 and 14 or ≤ 12). Therefore, 3 × 2 × 2 × 2 = 24 possible patterns were identified. The rate of osteotomy in the validation sample was decreased in comparison with the derivation sample (15% vs. 35%) and was too low to cover all possible patterns (insufficient overlapping). The discriminant capacity of the PMTSS remained also excellent in the validation population despite that slightly different characteristics were present in this population compared with the derivation set. Cell salvage was more frequently used in the validation sample (22% vs. 14%). Also, the use of tranexamic acid was threefold in comparison with the derivation population. Therefore, it is unlikely that the use of either blood salvage or antifibrinolytics impaired the performance of the PMTSS. The calibration of the score for predicting the individual probability of transfusion, as well as the number of erythrocyte units to be transfused, was robust.
Our study has limitations. One of them is the retrospective nature of a major part of this work. Also, it can be argued that the external validity of this single-center study could be limited.25
Nevertheless, our findings may be relevant to other surgical centers routinely practicing major spine surgery in adults, provided that their transfusion rate remains within a 30–40% range. Finally, the clinical impact of using the PMTSS in routine practice has not been examined as a primary endpoint and deserves further investigation in a prospective, multicenter trial. It can be suggested that preoperative calculation of the PMTSS could be helpful and cost-saving by identifying the patient subpopulations at risk of massive bleeding, therefore encouraging erythrocyte-saving strategies in these patients. Determination of the individual probability of erythrocyte transfusion may improve patient information on their perioperative erythrocyte requirements and give them the opportunity to start with a preoperative transfusion-sparing strategy (autotransfusion and/or epoetin). It can be speculated that patients with a PMTSS value greater than 2 could benefit from postponing surgery if sufficient blood amounts were not available (e.g.
, for rare phenotypes). Finally, preoperative calculation of the PMTSS may help to identify those patients for whom antifibrinolytics and/or intraoperative blood salvage could be beneficial.
In conclusion, we have found four independent predictive factors of intraoperative and postoperative allogeneic erythrocyte transfusion in thoracolumbar spine surgery. A score of individual probability of allogeneic transfusion for spine surgery, the PMTSS, has been generated and prospectively validated. This strategy may contribute to improve proper allocation of blood in the perioperative context and mobilize donors.
Appendix: Generation of 0–4 the Score from the Four Selected Independent Predictors of Erythrocyte Transfusion
Guidelines for reporting observational studies were followed throughout.26
Hosmer and Lemeshow goodness-of-fit of the model, parameter estimates, and adjusted odds ratios were first determined. Cutoff points for interval variables (age, hemoglobin levels, and number of fusion levels) were defined a posteriori
after the variables were selected from a first multivariable analysis using continuous independent factors, because influential observations were found. Cutoff points were chosen by using both discriminating (recursive partitioning) and “natural” cutoff points (age, hemoglobin levels) or equal-size groups (fusion levels). The goal was then to generate a clinically relevant and easy-to-use four-variable model for predicting erythrocyte transfusion. Therefore, we started from ordering multivariable estimates (additive in a multiplicative scale) and assessment of the probability of distribution of allogeneic transfusion according to the logistic model which seemed multimodal. We also noticed a 90% transfusion rate in case of transpedicular osteotomy along with a relative risk of transfusion of 3.8 (adjusted odds ratio = 24, which is similar to the unadjusted odds ratio = 27). A 4-point score (0–4) was then proposed and defined as follows: 4 points (maximum value) were allocated to transpedicular osteotomy. Other estimates were close to each other (1.6, 1.6, and 1.8, which corresponds to an adjusted odds ratio for transfusion of 5). Because preoperative hemoglobin level has a major impact on perioperative transfusion rate and treatment decision, new cutoff points corresponding to maximal statistical gain were investigated for this variable. Therefore, three intervals defined by preoperative hemoglobin levels (hemoglobin > 14 g/dl, 12 g/dl < hemoglobin < 14 g/dl, hemoglobin < 12 g/dl) were considered. One point (zero, respectively) was attributed to age > 50 yr (≤ 50 yr, respectively), spine fusion levels > 2 (≤ 2, respectively), and preoperative hemoglobin ≤ 14 g/dl (> 14 g/dl, respectively); two points were allocated when preoperative hemoglobin < 12 g/dl. The score was calculated as the arithmetic sum of points corresponding to each item. The maximal total score was 4, which defined five classes of probability levels. Transpedicular osteotomy was allocated 4 points whatever the other items. The probability of transfusion (P
) was estimated by considering the score as an internal variable as follows:
P(transfusion) = 1/1 + E-(b*score + A),
where B = 1.61, A = −4.56, or (delta + 1 point of the score = 5).
Correlations of probability distributions between the original model and the PMTSS demonstrated a good fit (Spearman correlation coefficient = 0.97). Patient characteristics, bleeding, and transfusion rates in the derivation and validation sets are reported in table 2
. Details of the model and of the receiver operating characteristic analysis are given in tables 3–8
. Plots of the model performance are displayed in figures 5–7
. Cited Here...
1. Bess RS, Lenke LG: Blood loss minimization and blood salvage techniques for complex spinal surgery. Neurosurg Clin North Am 2006; 17:227–34
2. Wass CT, Long TR, Faust RJ, Yaszemski MJ, Joyner MJ: Changes in red blood cell transfusion practice during the past two decades: A retrospective analysis, with the Mayo database, of adult patients undergoing major spine surgery. Transfusion 2007; 47:1022–7
3. American Society of Anesthesiologists Task Force on Perioperative Blindness: Practice advisory for perioperative visual loss associated with spine surgery. Anesthesiology 2006; 104:1319–28
4. Murrey DB, Brigham CD, Kiebzak GM, Finger F, Chewning SJ: Transpedicular decompression and pedicle substraction osteotomy (eggshell procedure): A retrospective review of 59 patients. Spine 2002; 27:2338–45
5. Nuttall GA, Horlocker TT, Santrach PJ, Oliver WC Jr, Dekutoski MB, Bryant S: Predictors of blood transfusions in spinal instrumentation and fusion surgery. Spine 2000; 25:596–601
6. Zheng F, Cammisa FP Jr, Sandhu HS, Girardi FP, Khan SN: Factors predicting hospital stay, operative time, blood loss, and transfusion in patients undergoing revision posterior lumbar spine decompression, fusion, and segmental instrumentation. Spine 2002; 27:818–24
7. Willems KF, Slot GH, Anderson PG, Pavlov PW, de Kleuver M: Spinal osteotomy in patients with ankylosing spondylitis: Complications during first postoperative year. Spine 2005; 30:101–7
8. Suk SI, Kim JH, Kim WJ, Lee SM, Chung ER, Nah KH: Posterior vertebral column resection for severe spinal deformities. Spine 2002; 27:2374–82
9. Agence Française de Sécurité Sanitaire des Produits de Santé: Transfusion of erythrocyte homologues: Products, indications, alternatives. Ann Fr Anesth Reanim 2003; 22:67–81
10. American Society of Anesthesiologists Task Force on Perioperative Blood Transfusion and Adjuvant Therapies: Practice guidelines for perioperative blood transfusion and adjuvant therapies. Anesthesiology 2006; 105:198–208
11. Mercuriali F, Inghilleri G: Proposal of an algorithm to help the choice of the best transfusion strategy. Curr Med Res Opin 1996; 13:465–78
12. Brecher ME, Monk T, Goodnough LT: A standardized method for calculating blood loss. Transfusion 1997; 37:1070–4
13. Segal JB, Guallar E, Powe NR: Autologous blood transfusion in the United States: Clinical and non-clinical determinants of use. Transfusion 2001; 41:1539–47
14. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR: A simulation of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996; 49:1373–9
15. Katz MH: Multivariable analysis: A primer for readers of medical research. Ann Intern Med 2003; 138:644–50
16. Hanley JA, McNeil JA: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143:29–36
17. Obuchwski NA, Lieber ML: Confidence intervals for the receiver operating characteristic area in studies with multiple small samples. Acad Radiol 1998; 5:561–71
18. McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Stiell IG, Richardson WS: Users' guides to the medical literature, XXII: How to use articles about clinical decision rules. Evidence-based Medicine Working Group. JAMA 2000; 284:79–84
19. Saleh E, McClelland DB, Hay A, Semple D, Walsh TS: Prevalence of anaemia before major joint arthroplasty and the potential impact of preoperative investigation and correction on perioperative blood transfusions. Br J Anaesth 2007; 99:801–8
20. Lentschener C, Cottin P, Bouaziz H, Mercier FJ, Wolf M, Aljabi Y, Boyer-Neumann C, Benhamou D: Reduction of blood loss and transfusion requirement by aprotinin in posterior lumbar spine fusion. Anesth Analg 1999; 89:590–7
21. Boachie-Adjei O, Ferguson JAI, Pigeon RG, Peskin MR: Transpedicular lumbar wedge resection osteotomy for fixed sagittal imbalance: Surgical technique and early results. Spine 2006; 34:485–92
22. Wang MY, Berven SH: Lumbar pedicle substraction osteotomy. Neurosurgery 2007; 60 (suppl 1):ONS-140–6
23. Nuttall GA, Santrach PJ, Oliver WC, Horlocker TT, Shaughnessy WJ, Cabanela MF, Bryant S: The predictors of red cell transfusion in total hip arthroplasties. Transfusion 1996; 36:144–9
24. Chassot PG, Delabays A, Spahn DR: Perioperative antiplatelet therapy: The case for continuing therapy in patients at risk of myocardial infarction. Br J Anaesth 2007; 99:316–28
25. Justice AC, Covinsly KE, Berlin JA: Assessing the generalizability of prognostic information. Ann Intern Med 1999; 130:515–24
26. Von Elm E, Altman DG, Pocok SJ, Goetzsche PC, Vandenbroucke JP: Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: Guidelines for reporting observational studies. BMJ 2007; 335:806–9
© 2009 American Society of Anesthesiologists, Inc.