Abstract
Background:
Venous thromboembolic disease, including deep venous thrombosis and pulmonary embolism, is a serious and potentially life-threatening complication following orthopaedic surgical procedures. We sought to investigate the prevalence of thromboembolism as well as the efficacy and complications of various prophylactic measures in a population of patients who had undergone elective spine surgery.
Methods:
A meta-analysis and univariate logistic regression were performed on selected studies to determine the prevalence of and risk factors for deep venous thrombosis and pulmonary embolism following elective spine surgery. Studies were included on the basis of the selection criteria (specifically, the inclusion of only patients undergoing spine surgery, or the treatment of patients undergoing spine surgery as an independent cohort; the use of an objective diagnostic modality for the diagnosis of deep venous thrombosis, including Doppler ultrasonography or venography; the use of an objective diagnostic modality for the diagnosis of pulmonary embolism, including computed tomography of the chest or a ventilation-perfusion scan; and a study population of more than thirty patients). Patients with a known spinal cord injury were excluded.
Results:
Fourteen studies (including a total of 4383 patients) met our selection criteria. On the basis of the meta-analysis, the prevalence of deep venous thrombosis was 1.09% (95% confidence interval, 0.54% to 1.64%) and the prevalence of pulmonary embolism was 0.06% (95% confidence interval, 0.01% to 0.12%) following elective spine surgery. The use of pharmacologic prophylaxis significantly reduced the prevalence of deep venous thrombosis relative to either mechanical prophylaxis (p = 0.047) or no prophylaxis (p < 0.01). One fatal pulmonary embolism was reported. An epidural hematoma requiring surgical evacuation was reported in eight of 2071 patients receiving pharmacologic prophylaxis; three of these patients had a permanent neurologic deficit.
Conclusions:
The risk of deep venous thrombosis and pulmonary embolism is relatively low following elective spine surgery, particularly for patients who receive pharmacologic prophylaxis. Unfortunately, pharmacologic prophylaxis exposes patients to a greater risk of epidural hematoma. More evidence is needed prior to establishing a protocol for prophylaxis against venous thromboembolic disease in patients undergoing elective spine surgery. Future prospective studies should seek to define the safety of various prophylactic modalities and to identify specific subpopulations of patients who are at greater risk for venous thromboembolism.
Level of Evidence:
Therapeutic Level II. See Instructions to Authors for a complete description of levels of evidence.
Venous thromboembolic disease, including deep venous thrombosis and pulmonary embolism, is a serious and potentially life-threatening complication in patients undergoing orthopaedic surgical procedures. Venous thromboembolic disease has been well studied following joint arthroplasty and orthopaedic trauma, but less is known about the prevalence of venous thromboembolic disease following elective spine surgery, and the optimum method of prophylaxis remains controversial.
Mechanical and pharmacologic prophylactic measures decrease the prevalence of deep venous thrombosis and pulmonary embolism and are widely advocated for most hospitalized orthopaedic patients, particularly those undergoing joint arthroplasty and those with trauma1,2. Pharmacologic agents include aspirin, low-molecular-weight heparin, and warfarin. These agents expose patients to an increased risk of bleeding complications, which may be acceptable risks following total joint replacement or trauma. Following elective spine surgery, however, bleeding complications include epidural hematoma and spinal cord or cauda equina compression, the consequences of which can be disastrous.
Recent patient safety standards from The Joint Commission (formerly the Joint Commission on the Accreditation of Healthcare Organizations [JCAHO]) and pay-for-performance criteria from the Centers for Medicare and Medicaid Services (CMS) have highlighted venous thromboembolic disease as an area in which systematic improvements may reduce morbidity and mortality. To comply, the American Academy of Orthopaedic Surgeons (AAOS), the American College of Chest Physicians (ACCP), and many other medical organizations have established guidelines for the routine use of prophylaxis in various patient populations. The recommendations often include the administration of pharmacologic prophylaxis, particularly in populations at higher risk, including patients managed with joint arthroplasty and those with spinal cord injury.
Patients undergoing elective spine surgery are generally thought to be at lower risk for venous thromboembolic disease than other orthopaedic patients are. In addition, these patients are at risk for major adverse neurologic events resulting from postoperative epidural hematomas related to pharmacologic prophylaxis. Therefore, the aforementioned recommendations may not apply to this select group of patients. Because of the overall increased emphasis on pharmacologic prophylaxis in hospitalized patients, we performed the present analysis to determine the prevalence of venous thromboembolic disease and attempted to measure the risk of pharmacologic prophylaxis as it relates to the population of patients undergoing elective spine surgery. These data could guide future studies aimed toward establishing a safe protocol for prophylaxis against venous thromboembolic disease in patients undergoing elective spine surgery.
The present review was performed with adherence to a prospective protocol, which established the research questions, search strategy, selection criteria, data extraction, and statistical analysis3.
Search Strategy
We searched the electronic databases MEDLINE, Cumulative Index to Nursing and Allied Health Literature (CINAHL Plus), and the Cochrane Database of Systematic Reviews for all manuscripts related to spine surgery and thromboembolic disease that had been published from January 1966 to January 2007. The search protocols and keywords are given in a table in the Appendix. Searches were limited to publications in the English language and those investigating human subjects.
In addition to the studies that were retrieved with the above search criteria, the reference lists of all studies chosen for inclusion in this meta-analysis were hand-searched for relevant publications. Reference lists of review articles relating to the subject matter were also searched.
Selection Criteria
The selection criteria for inclusion in the meta-analysis were (1) the inclusion of only patients undergoing spine surgery, or the treatment of patients undergoing spine surgery as an independent cohort, (2) the use of an objective diagnostic modality for the diagnosis of deep venous thrombosis, including Doppler ultrasonography or venography, (3) the use of an objective diagnostic modality for the diagnosis of pulmonary embolism, including computed tomography of the chest or a ventilation-perfusion scan, and (4) a study population of more than thirty patients. Patients with a spinal cord injury were excluded.
Data Extraction
Data were extracted independently by two reviewers (J.M.S. and A.M.d.R.), with discrepancies resolved by a third reviewer (P.A.A.). Information that was extracted from studies (when available) included the number of patients enrolled, age, weight, sex, surgical indications/diagnosis, operative level and type of operation, duration of surgery, duration of hospital stay, time to walking postoperatively, method of prophylaxis against deep venous thrombosis, complications related to prophylaxis, deep venous thrombosis and pulmonary embolism screening protocols, deep venous thrombosis and pulmonary embolism diagnostic methods, prevalence of deep venous thrombosis, prevalence of pulmonary embolism, location of deep venous thrombosis, pulmonary embolism-related mortality, individual characteristics of patients with deep venous thrombosis or pulmonary embolism, postoperative timing of deep venous thrombosis or pulmonary embolism, and duration of postoperative follow-up.
Statistical Methods
Meta-Analysis
Fixed and random-effects meta-analyses of the prevalence of deep venous thrombosis and pulmonary embolism were calculated. The level of significance was set at p < 0.05 (two-sided). Heterogeneity between studies was tested with the Q statistic. Prevalence was reported with use of a random-effects model if studies were heterogeneous according to the Q statistic (p < 0.05) and was reported with use of a fixed-effects model if studies were homogeneous according to the Q statistic (p > 0.05). Forest, influence analysis, funnel, and radial (Galbraith) plots were obtained. All computations and figures were done with Metagraphiti (StataCorp, College Station, Texas) in R for Windows4.
Notably, the term meta-analysis is typically applied to statistical analyses of randomized controlled clinical trials. For the purposes of the present study, no such trials exist, and thus the term meta-analysis is used hereafter to describe a method of analyzing observational data.
Univariate Logistic Regression Model (Cochran-Mantel-Haenszel Test)
Deep venous thrombosis and pulmonary embolism were each tested for association against each of three variables: (1) prophylaxis (none, mechanical, aspirin, low-molecular-weight heparin, or warfarin), (2) operative level (cervical, thoracolumbar), and (3) diagnosis (deformity, degenerative disease, other). Because each study contributed a table of each risk factor plotted against outcome and it was possible that the distribution varied across studies (strata), association was calculated with use of a Cochran-Mantel-Haenszel test. For computational convenience, the test is the Wald test for the risk factor in a logistic regression model. Because not all of the studies evaluated the risk factors uniformly, the logistic regression models are univariate in the sense that they model the outcome of interest (deep venous thrombosis or pulmonary embolism) as a function of a single risk factor, stratified by study. For each risk factor, the first category in alphabetical order was used as the baseline. The level of significance was set at p < 0.05 (two-sided).
Source of Funding
There were no external funding sources for this study.
Of the 1592 articles that were found to match our initial search criteria, fourteen qualified for inclusion in this analysis (Fig. 1). The fourteen studies included 4383 patients and are summarized in a table in the Appendix.
Demographic Data
The average age of patients across all fourteen studies was forty-nine years. Age was not recorded in two reports5,6. The male:female ratio was 48.2%:51.8%, with sex not being recorded in four studies5-8. The average patient weight was 69.2 kg, but weight was only reported in four studies9-12.
Meta-Analysis: Deep Venous Thrombosis
The Q statistic demonstrated the studies to be heterogeneous (p < 0.0001), so a random-effects model was applied. With use of this method, the prevalence of deep venous thrombosis was 1.09% (95% confidence interval, 0.54% to 1.64%). Forest plots assessed the homogeneousness of the effect sizes and the variability in the studies (Fig. 2). No publication bias was seen on the funnel plot (Fig. 3). Influence analysis revealed that the study by Gerlach et al.6 had a significant effect on the overall calculated prevalence. When that study was omitted, the prevalence of deep venous thrombosis as determined with our meta-analysis increased from 1.09% to 1.91%.
Meta-Analysis: Pulmonary Embolism
The Q statistic demonstrated the studies to be homogeneous (p = 0.427), so a fixed-effects model was applied. The prevalence of pulmonary embolism was 0.06% (95% confidence interval, 0.01% to 0.12%). Forest plots assessed the homogeneousness of the effect sizes and the variability in the studies (Fig. 4). No publication bias was seen on the funnel plot (Fig. 5). Influence analysis revealed that the study by Gerlach et al.6 had a significant effect on the overall calculated prevalence. When that study was omitted, the prevalence of pulmonary embolism as determined with our meta-analysis increased from 0.06% to 0.27%.
Univariate Logistic Regression Model: Deep Venous Thrombosis (Table I)
The prevalence of deep venous thrombosis was 5.8% (twenty-one of 361) when no prophylaxis was used, 1.8% (thirty-six of 1951) when only mechanical prophylaxis was used, 0% (n = 117) when mechanical prophylaxis and aspirin were used, <0.01% (one of 1919) when mechanical prophylaxis and low-molecular-weight heparin were used, and 0% (n = 35) when mechanical prophylaxis and warfarin were used. The univariate analysis demonstrated that use of aspirin, warfarin, or low-molecular-weight heparin significantly reduced the prevalence of deep venous thrombosis relative to the use of either mechanical prophylaxis (p = 0.047) or no prophylaxis (p < 0.01).
The prevalence of deep venous thrombosis was 1.3% (eleven of 840) for patients undergoing cervical procedures and 1.3% (thirty-six of 2838) for patients undergoing thoracolumbar procedures. Univariate analysis revealed that operative level did not significantly affect the prevalence of deep venous thrombosis.
The prevalence of deep venous thrombosis was 0.9% (one of 116) for patients undergoing surgery for the treatment of spinal deformity, 1.4% (forty-three of 2991) for those undergoing surgery for the treatment of degenerative conditions, and 0.3% (one of 321) for those with other diagnoses, including tumor, infection, pseudarthrosis, intraspinal hematoma, syrinx, and arteriovenous fistula. Univariate analysis revealed that diagnosis did not significantly affect the prevalence of deep venous thrombosis.
Univariate Logistic Regression Model: Pulmonary Embolism (Table I)
The prevalence of pulmonary embolism was 0% (n = 361) when no prophylaxis was used, 0.4% (eight of 1951) when mechanical prophylaxis was used, 0% (n = 117) when mechanical prophylaxis and aspirin were used, 0% (n = 1919) when mechanical prophylaxis and low-molecular-weight heparin were used, and 0% (n = 35) when mechanical prophylaxis and warfarin were used. Univariate analysis could not be performed for the method of prophylaxis because of the low reported prevalence of pulmonary embolism.
The prevalence of pulmonary embolism was 0.4% (three of 840) for patients undergoing cervical spine procedures and 0.2% (five of 2838) for patients undergoing thoracolumbar procedures. Univariate analysis demonstrated that the operative level had no significant influence on the prevalence of pulmonary embolism (p = 0.5).
The prevalence of pulmonary embolism was 0% (n = 116) for patients undergoing surgery for the treatment of spinal deformity, 0.1% (four of 2991) for those with degenerative conditions, and 0% (n = 321) for those with other diagnoses, including tumor, infection, pseudarthrosis, intraspinal hematoma, syrinx, or arteriovenous fistula. Univariate analysis could not be performed for patient diagnosis because of the low reported prevalence of pulmonary embolism.
Venous Thromboembolic Disease-Specific Data
The mean time to the diagnosis of venous thromboembolic disease was eight days (range, one to sixteen days)6,12-15. Deep venous thrombosis was symptomatic in only two of twenty-four patients for whom documentation of symptomatic as opposed to asymptomatic deep venous thrombosis was recorded6,9. Only eight of forty-five deep venous thromboses were found to be proximal to the popliteal fossa, whereas the majority were located distally6,7,9,13-17.
Complications of Prophylaxis
In three studies6,10,11, a total of 2071 patients received pharmacologic postoperative prophylaxis in the form of aspirin, low-molecular-weight heparin, or warfarin. Eight cases of epidural hematoma developed; all occurred in a group of patients receiving postoperative low-molecular-weight heparin6. All patients underwent surgical evacuation, and three patients had a persistent neurologic deficit.
Mortality
One fatal pulmonary embolism occurred nine days after a staged anterior and posterior thoracolumbar arthrodesis.
Comorbidities
The meta-analysis to assess the effect of medical comorbidities on the prevalence of venous thromboembolic disease was not possible as quantifiable information was lacking. Various authors reported identifiable risk factors in patients who experienced deep venous thrombosis, including factor V Leiden mutation (hypercoagulability syndrome), obesity, a history of deep venous thrombosis, a history of malignancy, advanced age, and circumferential fusion11-18. None of the authors performed a multivariate statistical analysis to determine the significance of these various patient-related factors as it related to venous thromboembolic disease risk.
Recently, an emphasis has been placed on the prevention of venous thromboembolic disease in hospitalized patients, and many orthopaedic subspecialties have already defined standards for prophylaxis against venous thromboembolic disease2,19. However, no such standards have been defined in the field of spine surgery.
The most recently published guidelines of the ACCP fail to strictly define recommendations for patients undergoing elective spine surgery as they suggest the use of anti-embolic stockings or low-dose unfractionated heparin, and possibly intermittent pneumatic compression stockings, with the additional suggestion of one of six different regimens for patients with one or more risk factors1. Heck et al.20, in a recent review article, suggested the use of low-molecular-weight heparin for all patients undergoing elective spine surgery, except those with no risk factors, which included an age of more than sixty years and a body mass index of >30. Some authors have even suggested preoperative placement of inferior vena cava filters in high-risk patients21. Unfortunately, we are aware of no prospective data on the preoperative placement of inferior vena cava filters in patients undergoing elective spine surgery, precluding their analysis in this meta-analysis.
Despite the recommendations of the ACCP, there is a scarcity of existing data regarding specific risk factors and the safety of pharmacologic prophylaxis. Furthermore, the highly variable range of spine surgery and the diversity of the patient population have contributed to an inability to arrive at a consensus regarding recommendations for prophylaxis against venous thromboembolic disease22.
We found the overall prevalence of deep venous thrombosis to be relatively low (1.09%; 95% confidence interval, 0.54% to 1.64%) and the prevalence of pulmonary embolism to be a fraction of this (0.06%; 95% confidence interval, 0.01% to 0.12%). Most deep venous thromboses were distal to the popliteal fossa, and most were asymptomatic. Only one case of fatal pulmonary embolism was documented. Many patients with an extraordinarily low risk of venous thromboembolic disease were included among those managed with pharmacologic prophylaxis.
Most authors have agreed that minor spine surgery (discectomy and single-level laminectomy) presents minimal risk of venous thromboembolic disease, but major spinal reconstruction (multiple-level operations, instrumented fusions) places the patient at greater risk22,23. Our data suggest a relatively low risk of venous thromboembolic disease in the population of patients managed with elective spine surgery, but we had difficulty trying to identify specific subpopulations in which clinical risk is disproportionately higher. In our analysis, there was a trend toward greater risk with combined anterior and posterior arthrodesis; however, the small number of subjects in that category did not allow statistical analysis. In addition, we found that patients who received mechanical prophylaxis alone were at a significantly higher risk (albeit still minimal) of venous thromboembolic disease as compared with their counterparts receiving pharmacologic prophylaxis.
It is difficult to truly define the exact prevalence of disease with use of meta-analysis. The results are greatly affected by interstudy variability and the differential influence of specific studies on the cumulative data. Although our inclusion criteria were uniform, the results at each center were highly variable because of differences in practice patterns, patient selection, and the availability of medical technology. Also, venous thromboembolic disease in this population appears to be exceedingly rare, rendering accurate prevalence and risk factor quantification difficult, even with large sample sizes.
Gerlach et al.6 reported an extraordinarily low rate of venous thromboembolic disease in a disproportionately large population of patients, thus decreasing our calculated prevalence of venous thromboembolic disease. This is exemplary of the interstudy variability that makes the results of meta-analyses less reliable than those of prospective, randomized clinical trials. To that end, care must be taken not to alter clinical practice on the basis of the trends observed by means of meta-analysis.
We were unable to resolve various questions that we sought to answer. Namely, our data could not define the risk of venous thromboembolic disease with respect to patient diagnosis, and the risk of pulmonary embolism could not be predicted on the basis of the method of prophylaxis. In addition, specific risk profiles taking into account patient, disease, and treatment-related variables could not be established.
We reported morbidity in association with the use of pharmacologic prophylaxis. In one study, the use of aspirin did not result in any reported untoward effects, although the study population was small (n = 117), thereby limiting our ability to draw strong conclusions. Conversely, there were eight epidural hematomas in a population of patients receiving low-molecular-weight heparin and there were two complications (excessive postoperative blood loss and disseminated intravascular coagulopathy) in a group receiving warfarin6,10,11. Notably, in the population receiving low-molecular-weight heparin, the risk of epidural hematoma was nearly eightfold greater than the risk of pulmonary embolism6.
We hope that these data serve as a foundation for additional research that seeks to define a safe protocol for prophylaxis against venous thromboembolic disease in patients undergoing elective spine surgery and that these data identify specific populations at risk for this complication. The risk of venous thromboembolic disease in this group of patients appears to be low, perhaps suggesting that the risk of pharmacologic prophylaxis outweighs its benefit. We caution readers, however, that instituting change in existing clinical protocols on the basis of these limited data would be premature.
In conclusion, the population of patients managed with elective spine surgery appears to be at low risk for the development of postoperative venous thromboembolic disease. However, prospective research is needed to definitively stratify risk according to patient, diagnosis, and treatment-related variables and to establish a safe, standardized, and universally applicable protocol for prophylaxis against venous thromboembolic disease.
A table listing the search strategy and keywords and a table listing the fourteen studies that were evaluated are available with the electronic version of this article on our web site at jbjs.org (go to the article citation and click on "Supporting Data").
Geerts
WH;
Pineo
GF;
Heit
JA;
Berqvist
D;
Lassen
MR;
Colwell
CW;
Ray
JG. Prevention of venous thromboembolism: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest.
2004;126(
3 Suppl):338S-400S.[PubMed][CrossRef]
American Academy of Orthopaedic Surgeons. American Academy of Orthopaedic Surgeons clinical guideline on prevention of pulmonary embolism in patients undergoing total hip or knee arthroplasty. .
Moher
D;
Cook
DJ;
Eastwood
S;
Olkin
I;
Rennie
D;
Stroup
DF. Improving the quality of reports of meta-analyses of randomised controlled trials: the QUOROM statement. Quality of Reporting of Meta-analyses. Lancet.
1999;354:1896-900.[PubMed]
The R Project for Statistical Computing. .
Flinn
WR;
Sandager
GP;
Silva
MB;
Benjamin
ME;
Cerullo
LJ;
Taylor
M. Prospective surveillance for perioperative venous thrombosis. Experience in 2643 patients. Arch Surg.
1996;131:472-80.[PubMed]
Gerlach
R;
Raabe
A;
Beck
J;
Woszczyk
A;
Seifert
V. Postoperative nadroparin administration for prophylaxis of thromboembolic events is not associated with an increased risk of hemorrhage after spinal surgery. Eur Spine J.
2004;13:9-13.[PubMed]
Ferree
BA;
Wright
AM. Deep venous thrombosis following posterior lumbar spinal surgery. Spine (Phila Pa 1976).
1993;18:1079-82.[PubMed]
Smith
MD;
Bressler
EL;
Lonstein
JE;
Winter
R;
Pinto
MR;
Denis
F. Deep venous thrombosis and pulmonary embolism after major reconstructive operations on the spine. A prospective analysis of three hundred and seventeen patients. J Bone Joint Surg Am.
1994;76:980-5.[PubMed]
Lee
HM;
Suk
KS;
Moon
SH;
Kim
DJ;
Wang
JM;
Kim
NH. Deep vein thrombosis after major spinal surgery: incidence in an East Asian population. Spine (Phila Pa 1976).
2000;25:1827-30.[PubMed]
Nelson
LD
Jr;
Montgomery
SP;
Dameron
TB
Jr;
Nelson
RB. Deep vein thrombosis in lumbar spinal fusion: a prospective study of antiembolic and pneumatic compression stockings. J South Orthop Assoc.
1996;5:181-4.[PubMed]
Rokito
SE;
Schwartz
MC;
Neuwirth
MG. Deep vein thrombosis after major reconstructive spinal surgery. Spine (Phila Pa 1976).
1996;21:853-9.[PubMed]
Wood
KB;
Kos
PB;
Abnet
JK;
Ista
C. Prevention of deep-vein thrombosis after major spinal surgery: a comparison study of external devices. J Spin Disord.
1997;10:209-14.
Epstein
NE. Intermittent pneumatic compression stocking prophylaxis against deep venous thrombosis in anterior cervical spinal surgery: a prospective efficacy study in 200 patients and literature review. Spine (Phila Pa 1976).
2005;30:2538-43.[PubMed]
Epstein
NE. Efficacy of pneumatic compression stocking prophylaxis in the prevention of deep venous thrombosis and pulmonary embolism following 139 lumbar laminectomies with instrumented fusions. J Spinal Disord Tech.
2006;19:28-31.[PubMed]
Ferree
BA. Deep venous thrombosis following lumbar laminotomy and laminectomy. Orthopedics.
1994;17:35-8.[PubMed]
Ferree
BA;
Stern
PJ;
Jolson
RS;
Roberts
JM
5th;
Kahn
A
3rd. Deep venous thrombosis after spinal surgery. Spine (Phila Pa 1976).
1993;18:315-9.[PubMed]
Oda
T;
Fuji
T;
Kato
Y;
Fujita
S;
Kanemitsu
N. Deep venous thrombosis after posterior spinal surgery. Spine (Phila Pa 1976).
2000;25:2962-7.[PubMed]
Dearborn
JT;
Hu
SS;
Tribus
CB;
Bradford
DS. Thromboembolic complications after major thoracolumbar spine surgery. Spine (Phila Pa 1976).
1999;24:1471-6.[PubMed]
Janku
GV;
Paiement
GD;
Green
HD. Prevention of venous thromboembolism in orthopaedics in the United States. Clin Orthop Relat Res.
1996;325:313-21.[PubMed]
Heck
CA;
Brown
CR;
Richardson
WJ. Venous thromboembolism in spine surgery. J Am Acad Orthop Surg.
2008;16:656-64.[PubMed]
Leon
L;
Rodriguez
H;
Tawk
RG;
Ondra
SL;
Labropoulos
N;
Morasch
MD. The prophylactic use of inferior vena cava filters in patients undergoing high-risk spinal surgery. Ann Vasc Surg.
2005;19:442-7.[PubMed]
Brambilla
S;
Ruosi
C;
La Maida
GA;
Caserta
S. Prevention of venous thromboembolism in spinal surgery. Eur Spine J.
2004;13:1-8.[PubMed]
Samama
CM;
Albaladejo
P;
Benhamou
D;
Bertin-Maghit
M;
Bruder
N;
Doublet
JD;
Laversin
S;
Leclerc
S;
Marret
E;
Mismetti
P;
Samain
E;
Steib
A; Committee for Good Practice Standards of the French Society for Anaesthesiology and Intensive Care (SFAR). Venous thromboembolism prevention in surgery and obstetrics: clinical practice guidelines. Eur J Anaesthesiol.
2006;23:95-116.[PubMed]