Accepte

Reprint320 EastPA 1521

0039-60

� 2013

http://d

810 S

Risk of pulmonary embolism in traumapatients: Not all created equalElan Jeremitsky, MD,a Natasha St. Germain, MD,a Amy H. Kao, MD, MPH, MS,b Adrian W. Ong, MD,c

and R. Stephen Smith, MD,d Pittsburgh and Reading, PA, and Columbia, SC

Introduction. Patients with traumatic brain injury (TBI) are assumed to be at an increased risk forpulmonary embolism (PE). Delay in the initiation of chemoprophylaxis and prophylactic placement ofinferior vena cava filters have been advocated by some because of concerns for increased intracranialhemorrhage in the presence of prophylactic anticoagulation.We hypothesized that patients with isolated TBIwould not be at increased risk for the development of PE compared with the general trauma population.Methods. Patients from the National Trauma Data Bank from the year 2008 were analyzed. Patientdemographics, Injury Severity Score, and the prevalence of deep-vein thrombosis and PE were extracted.Studied injuries were assigned to six categories: thorax, abdominal solid organs, pelvic fracture, lowerextremity fracture, spine fracture, and TBI.Results. Of a total of 627,775 injured patients, 2,182 (0.35%) had a documented PE. The prevalenceof PE in patients with isolated TBI, lower extremity, pelvic fracture, liver and/or spleen, thorax, spine,multiple injuries, and none of the studied injuries were 0.25%, 0.36%, 0.35%, 0.37%, 0.52%,0.37%, 1.1%, and 0.12%, respectively. Using an age-, sex- and race-adjusted multivariable logisticregression model and controlling for interaction between inferior vena cava filters and injury types, wefound that isolated TBI was not associated with PE.Conclusion. Isolated TBI does not appear to be associated with an increased incidence of PE comparedwith other injuries. Patients with isolated TBI may not require early aggressive prophylaxis as is thestandard for other high-risk groups. (Surgery 2013;154:810-5.)

From the Departments of Surgerya and Medicine,b Allegheny General Hospital, Pittsburgh, PA; Department ofSurgery,c Reading Hospital, Reading, PA; and Department of Surgery,d University of South Carolina Schoolof Medicine, Columbia, SC

VENOUS THROMBOEMBOLISM (VTE) is a frequentoccurrence in patients with multiple injuries.1 Itappears that certain trauma populations and injurypatterns have an increased risk of VTE.2,3 In partic-ular, the literature describes an increased risk ofVTE in patients suffering a traumatic brain injury(TBI).4,5 Many articles in which the authors discussTBI also focus on thromboembolic prophylaxis,ranging from ‘‘safe and effective’’ in a stable TBI(by computed tomography [CT]) to potentiallyworsening hemorrhage, especially in the settingof progressing intracerebral hemorrhage.6,7

The diagnosis of pulmonary embolism (PE) inthe trauma population is approximately 0.4%.2

Although this prevalence is seemingly low, theresultant morbidity is considerable and has now

d for publication April 19, 2013.

requests: Elan Jeremitsky, MD, Department of Surgery,North Avenue, Allegheny General Hospital, Pittsburgh,2. E-mail: ejeremit@wpahs.org.

60/$ - see front matter

Mosby, Inc. All rights reserved.

x.doi.org/10.1016/j.surg.2013.04.045

URGERY

become part of a quality operative indicator. Theroutine use of spiral CT to diagnose PE has onlyincreased the prevalence at the expense of findinga corresponding deep-vein thrombosis (DVT), asonly 20% of patients with PE have an associatedDVT.2 Brain-injured patients are a special segmentof the injured population because conflicting ornot well-defined guidelines exist between traumasurgery and neurosurgery as to when it is safe toprovide chemical prophylaxis.8,9

In a recently published paper, Knudson et al2

showed a borderline-negative association betweenTBI and PE with an odds ratio of 0.87 (95% confi-dence interval 0.73�1.04). This unusual finding ina large National Trauma Databank (NTDB) evalu-ation is not well explained by the literature andseems counterintuitive. To further elucidate the in-fluence of TBI on VTE, we hypothesize that pa-tients with isolated TBI do not have the same riskof PE as patients with other injury patterns.

METHODS

The American College of Surgeons establishedthe NTDB to serve as a repository of trauma-related

SurgeryVolume 154, Number 4

Jeremitsky et al 811

data voluntarily reported on an annual basis byparticipating hospitals with trauma registries.Currently, the NTDB contains detailed data onmore than 2.7 million cases from more than 900U.S. trauma centers. The 2008 NTDB repository wasused to obtain our cohort of injured patients.10 Allinjured patients were included irrespective of injurymechanism. Injuries were classified into six cate-gories: thoracic injury (rib fractures), intra-abdominal solid organ (liver and/or spleen) injury,pelvic fracture, lower extremity fracture, spine frac-ture (cervical, thoracic, and lumber), and traumaticbrain injury (TBI). International Classification of Dis-ease, Version 9 diagnosis codes were used to createa broad base for each of the six injury categories:chest injury 807–807.4, solid-organ injury 864–865.19, pelvic fracture 808–808.9, lower-extremityfracture 820–826.1, spine fracture 805–806.9, andTBI 851–854.19. Complication codes for DVT andPE (no. 14 and no. 21, respectively) were examinedfor andmerged with the database. Patients who hadthe procedure code 38.7 for interruption of venacava or inferior vena cava filter (IVCF) also were ex-tracted and merged. Preinjury use of warfarin wasalso included from the list of NTDB comorbid con-ditions (no. 4) and incorporated into patient demo-graphics, along with injury severity. The isolatedTBIpatients were further classified into different levelsof injury severity via our use of the AbbreviatedInjury Scale. We then created groups of isolated in-juries from the six previously described injury cate-gories and created additional categories: acategory that had none of the studied injuries, anda category for themultiply injuredpatients (patientshaving two or more of the six injury categories). Pa-tients who had none of the studied injuries wouldhave been included in the database as the result ofinjuries not included in our list of studied injuries,including but not limited to upper-extremityinjuries, genitourinary injuries, or facial fractures.

Statistical analysis. Descriptive statistics werereported as means (with SD) and medians (withinterquartile range: 25th–75th percentile) based onthe distribution of the continuous variables and aspercentages for categorical variables. The means ormedians of continuous variables were comparedusing either t tests or Wilcoxon rank sum tests de-pending on the distribution of the data. Categoricalvariables were analyzed using v2 tests. Because ofconcern for the potential influence of IVCF on PEdiagnosis, we added an interaction term, which isthe product of two variables (‘‘IVCF’’ and ‘‘injurytypes’’). Because of the significant interaction be-tween IVCF and injury types on themultivariable lo-gistic regression model, the interaction term was

included in the final multivariable analysis. All testswere two-sidedwith a significance level of<0.05. An-alyses were performedusing STATASE 11.0 forWin-dows (Stata Corporation, College Station, TX).

RESULTS

A total of 2,182 patients with PE were identifiedfrom the initial patient cohort of 627,775 patients(0.35%). Patients who suffered a PE were morelikely to be severely injured or older and had alonger duration of hospital stay. In addition, pre-existing warfarin use was associated with PE. Only22% of patients with PE were demonstrated tohave a DVT and 30% of patients with PE had anIVCF placed (Table I). The prevalence of PE in pa-tients with isolated TBI, lower extremity, pelvicfracture, liver and/or spleen, thorax, spine, multi-ple injuries, and none of the studied injuries was0.25%, 0.36%, 0.35%, 0.37%, 0.52%, 0.37%,1.1%, and 0.12%, respectively, and was substan-tially different by v2 test (P < .001) as shown inTable II.

Table II also illustrates the prevalence of IVCFplacement in the study sample and also was signif-icantly different by v2 test (P < .001). An interac-tion term was created to investigate the effect ofIVCF on injury types. This interaction term wasdefined as the product of two variables (IVCF *the different injury types). The product was foundto be important and was included in the finalmultivariable logistic regression model. In themultivariable logistic regression model (TableIII), isolated TBI was not greatly associated withPE. Even after we separated the isolated TBI intodifferent severities by Abbreviated Injury Scale, iso-lated TBI regardless of injury severity remainednot substantially associated with PE in the multivar-iable logistic regression model. Other injuries inisolation, multiple injuries, use of warfarin, IVCFplacement, and the development of DVT wereassociated with PE. We re-ran the multivariable lo-gistic regression and included only patients whodid not have an IVCF placement. Isolated braininjury continued to be not associated with PE(P = .42), whereas other injury categoriescontinued to be associated with PE.

DISCUSSION

Our results from this large NTDB analysisshowed that isolated TBI was not associated witha diagnosis of PE, whereas the presence of multipleinjuries was associated with PE. Patients with mul-tiple injuries are clearly at greater risk for PEbecause other injury categories contribute to agreater risk of PE. The use of an IVCF clearly plays

Table I. Demographics and clinical features ofstudy cohort stratified by pulmonary embolism

Pulmonaryembolism

(n = 2,182)

Nopulmonaryembolism

(n = 625,593)P

value

Age, years 47.9 ± 28.9* 36.7 ± 32.1 <.001Sex, % female 30.1 35.1 <.001Race, %White 73.5 65.2 <.001

African-American 12.7 14.0

Other 13.8 20.8Median DOS, days

(IQR)16 (9–28) 3 (1–6) <.001

Median ICUDOS, days (IQR)

10 (4–20) 2 (1–6) <.001

Median ventilation,days

10 (4–20) 2 (1-8) <.001

Injury Severity Score 18.9 ± 12.5 9.6 ± 9.5 <.001DVT, present 22.0% 0.73% <.001IVCF placement 30.8% 1.3% <.001Warfarin use 5.1% 2.3% <.001

*Mean (±SD); median (IQR).DOS, Duration of stay; DVT, deep-vein thrombosis; ICU, intensive careunit; IQR, interquartile range (25th–75th percentile); IVCF, inferiorvena cava filter.

Table II. Injury types and prevalence of PE ininjured patients

Injury types

No.patients, %

(n = 627,775)PE, %

(n = 2,182)*IVCF, %

(n = 8,892)*

Isolated brain 64,267 (10.2) 0.25 1.4Lower extremity 109,123 (17.4) 0.36 0.54Pelvic fracture 12,678 (2.0) 0.35 1.63Liver and/or

spleen12,597 (2.0) 0.37 0.82

Thorax 31,019 (4.9) 0.52 1.75Spine 37,964 (6.1) 0.37 2.09Multiple injuries 85,867 (13.7) 1.1 6.25None of studied

injuriesy274,260 (43.7) 0.12 0.28

*v2 test, P < .001.yIncluded facial injuries, upper extremity injuries, etc.IVCF, Inferior vena cava filter; PE, pulmonary embolism.

Table III. Multivariable logistic regression forpulmonary embolism

Oddsratio

95% confidenceinterval P value

Age 1.00 1.00–1.01 <.001Sex, female 0.80 0.72–0.88 <.001Race 0.94 0.90–0.98 .005Injury Severity Score 1.02 1.02–1.03 <.001Deep vein thrombosis 6.72 5.90–7.66 <.001Inferior vena cava filter 33.06 26.77–40.83 <.001Warfarin use 1.38 1.12–1.70 .002Isolated injuries*Isolated brain 1.01 0.82–1.24 .92Lower extremity 2.90 2.49–3.39 <.001Pelvis 2.18 1.58–3.02 <.001Liver and/or spleen 2.73 1.99–3.75 <.001Thorax 3.73 3.01–4.55 <.001Spine 2.49 2.02–3.07 <.001Multiple injured 4.36 3.72–5.11 <.001

Interaction term betweeninferior vena cavafilter and injury types

0.77 0.75–0.80 <.001

*Referent: Patients with none of the six studied injuries.

SurgeryOctober 2013

812 Jeremitsky et al

an interactive role in the diagnosis of PE with braininjury, which may be attributable to the indicationof IVCF placement. We accounted for this by usingan interaction term in the final multivariablemodel. Stanwicki et al11 evaluated DVT and PE ina large database set and found a similar incidenceof PE (0.4–0.7%) in trauma patients. In theirmultivariable analysis, ‘‘ISS consistently showedpredictive value for DVT and PE.’’ They also hada similar grouping of high-risk trauma patients,with patients having had multiple injuries,

specifically those with a combined pelvic fracture,lower-extremity fracture, and closed head injuryhaving the greatest incidence of DVT.

Morbidity from brain hemorrhage remains aserious and sometimes catastrophic event. It isimperative that patients with TBI have theirmedicalcare optimized to facilitate healing without addedinsults that can further compromise their condi-tion, such as progression of brain trauma with theuse of prophylactic anticoagulation. TheChest 2012Practice guidelines for trauma with grade 2c recom-mendations for TBI state that neither use of an IVCFnor periodic screening with duplex ultrasound arerecommended.9 The neurosurgical literatureregarding the use of prophylactic anticoagulants af-ter brainoperationorTBI is not clear, butmanyneu-rosurgeons delay initiation of chemoprophylaxis for72 hours. 5-7,12-14 AmericanAssociation ofNeurolog-ical Surgeons guidelines for the management of se-vere TBI cite Level III evidence for prophylaxis:‘‘There is insufficient evidence to support recom-mendations regarding the preferred agent, dose,or timing of pharmacologic prophylaxis forDVT.’’8 In patients with TBI, there is a wide variationin practice among clinicians as to the optimal strat-egy for DVT prophylaxis. As such, the mode andtiming of DVT prophylaxis in this population isoften individualized, which makes it difficult toattempt to estimate the risk of PE retrospectivelyin a large database. This certainly would have impli-cations for patient safety and quality of trauma care.

SurgeryVolume 154, Number 4

Jeremitsky et al 813

Minshall et al15 evaluated patients with TBI bycomparing unfractionated heparin and low-molecular-weight heparin prophylaxis and foundthat the unfractionated heparin group had agreater severity of brain injury and a greater rateof PE and progression of injury. Scudday et al16

evaluated chemoprophylaxis in TBI and foundthat approximately one-half of their patientsreceived chemical prophylaxis, with a large portionof patients (60%) receiving some form of chemicalprophylaxis in the first 72 hours. Those whoreceived prophylaxis had a lower rate (1% vs 3%)of VTE.

Similar to the 2011 study by Knudson et al, wefound that chest injury was independently associ-ated with PE and that TBI was not.2 We alsoinitially performed a regression for PE with TBI(not isolated injuries) and similarly found a nega-tive association between TBI and PE in this 2008NTDB trauma population. This prompted us tore-evaluate the dataset with regard to isolated braininjuries by controlling for IVCF placement to havea ‘‘pure’’ sample. Upon doing so, no associationwith PE and isolated TBI was found regardless ofinjury severity.

Perhaps the primary reason why thoracic in-juries have such a strong association with PE iscurrent clinical practice standards. Patients withthoracic injury are closely monitored with pulseoximetry, and any signs of hypoxia along withtachycardia result in radiographic studies for PEevaluation. The clinical significance of PE in chesttrauma remains to be determined, but in ourpractice usually results in treatment for PE. Knud-son et al2 suggested that a prothrombotic milieuassociated with pulmonary injury results in detec-tion and development of PE.

Similarly, patients receiving warfarin beforeinjury were found to have an increased risk of PEin our study. This is probably the result of thepropensity of these patients to be in a hypercoag-ulable state or have a history of PE beforeadmission for injury. DVT was strongly associatedwith PE, but only 20% of patients with PE had apreviously identifiable DVT. This value is in linewith other studies.2,7 The explanation for this isperhaps that once the diagnosis for PE is made,there is little need for an additional workup forDVT.

There are limitations with this retrospectivestudy that may be addressed by future prospectivetrials. Although the NTDB research data set isexpansive, information about the type of DVTprophylaxis (both mechanical and chemical) islacking. The benefit of prophylactic IVCF use is

cryptic, and its influence on pulmonary embolismis difficult to discern from the current data. Theinjury types we selected to study were very broad innature and do encompass minor injuries as thereferent injury category. We chose to includeminor injuries as well because treatment practices,such as those for nonweight-bearing injuries, mayvary across institutions. Although our injury typeswere mutually exclusive by themselves, it is difficultto completely remove the influence of otherinjuries. To obtain a completely pure sample isdifficult because of the presence of multisysteminjuries. In this study, we attempted to isolate theseinjuries into six main categories of organ systeminjury involvement as best as we could, but theremight still be some overlap with injuries notaccounted for. Defining the role of prophylacticIVCF placement by using the argument that ‘‘therewas no PE identified in this patient, therefore IVCFmust have been placed prophylactically,’’ is prob-lematic. Our concern herein belies the fact thatone defines the IVCF role on the basis of knowingthe outcome. For this very reason, we chose not todefine ‘‘prophylactic IVCF placement’’ and insteadwe controlled for the ‘‘IVCF placement procedure’’and the interaction term in our multivariablelogistic regression model.

We have demonstrated that isolated TBI, unlikeother injury patterns, is not associated with PE.Whether this translates into liberal usage of pro-phylactic anticoagulation on the basis of tailoring aspecific regimen for individualized patient man-agement, or delaying chemoprophylaxis 2–3 daysin a patient with a completely stable head CTremains to be seen. It appears that this particularpatient group may not be at as increased a risk asother injury populations and although not associ-ated with PE compared with our designated low-risk population, DVT prophylaxis remains anessential component of postinjury care.

The authors thank Michael Anderson for his editorialassistance. The NTDB remains the full and exclusivecopyrighted property of the American College of Sur-geons (ACS). The ACS is not responsible for claimsarising from works based on the original data, text, ortables.

REFERENCES

1. Geerts WH, Code KI, Jay RM. A prospective study of venousthromboembolism after major trauma. N Engl J Med 1994;331:1601-6.

2. Knudson MM, Gomez D, Haas B, Cohen MJ, Nathens AB.Three thousand seven hundred thirty-eight posttraumaticpulmonary emboli: a new look at an old disease. AnnSurg 2011;254:625-32.

SurgeryOctober 2013

814 Jeremitsky et al

3. Maung AA, Schuster KM, Kaplan LJ, et al. Risk venousthromboembolism after spinal cord injury: not all levelsare the same. J Trauma 2011;71:1241-5.

4. Reiff DA, Haricharan RN, Bullington NM, Griffin RL,McGwin G Jr, Rue LW 3rd. Traumatic brain injury is associ-ated with the development of deep vein thrombosis inde-pendent of pharmacological prophylaxis. J Trauma 2009;66:1436-40.

5. Saadeh Y, Gohil K, Bill C, Smith C, Morrison C, Mosher B,et al. Chemical venous thromboembolic prophylaxis is safeand effective for patients with traumatic brain injury whenstarted 24 hours after the absence of hemorrhage progres-sion on head CT. J Trauma Acute Care Surg 2012;73:426-30.

6. Levy AS, Salottolo K, Bar-Or R, Offner P, Mains C, SullivanM, et al. Pharmacologic thromboprophylaxis is a risk factorfor hemorrhage progression in a subset of patients withtraumatic brain injury. J Trauma 2010;68:886-94.

7. Koehler DM, Shipman J, Davidson MA, Guillamondegui O.Is early venous thromboembolism prophylaxis safe intrauma patients with intracranial hemorrhage. J Trauma2011;70:324-9.

8. Guidelines for the Management of Severe Traumatic BrainInjury: Section V. Deep Vein Thrombosis Prophylaxis.J Neurotrauma 2007;24:S32-6.

9. Guyatt GH, Akl EA, Crowther M, Gutterman DD,Schu€unemann HJ, American College of Chest PhysiciansAntithrombotic Therapy and Prevention of ThrombosisPanel. Executive summary: Antithrombotic Therapy andPrevention of Thrombosis, 9th ed: American College ofChest Physicians Evidence-Based Clinical Practice Guide-lines[erratum in: Chest 2012;141:1129]. Chest 2012;141(2Suppl):7S-47S.

10. Committee on Trauma, American College of Surgeons.NTDB, Version 2008. Chicago, IL: ACS; 2008.

11. Stanwicki SP, Grossman MD, Cipolla J. Deep vein throm-bosis and pulmonary embolism in trauma patients: an over-statement of the Problem? Am Surg 2005;71:387-91.

12. Niemi T, Armstrong E. Thromboprophylactic managementin the neurosurgical patient with high risk for both throm-bosis and intracranial bleeding. Curr Opin Anaesthesiol2010;23:558-63.

13. Phelan HA, Wolf SE, Norwood SH, Aldy K, Brakenridge SC,Eastman AL, et al. A randomized, double-blinded, placebo-controlled pilot trial of anticoagulation in low-risk trau-matic brain injury: the Delayed Versus Early EnoxaparinProphylaxis I (DEEP I) study. J Trauma Acute Care Surg2012;73:1434-41.

14. Macdonald RL, Amidei C, Lin G, Munshi I, Baron J, WeirBK, et al. Safety of perioperative subcutaneous heparin forprophylaxis of venous thromboembolism in patients under-going craniotomy. Neurosurgery 1999;45:245-51.

15. Minshall CT, Eriksson EA, Leon SM, et al. Safety and effi-cacy of heparin or enoxaparin prophylaxis in blunt traumapatients with a head Abbreviated Injury Score >2. J Trauma2011;71:396-400.

16. Scudday T, Brasel K, Webb T, et al. Safety and efficacy ofprophylactic anticoagulation in patients with traumaticbrain injury. J Am Coll Surg 2011;213:148-54.

DISCUSSION

Dr Stephen Marshall (Peoria, IL): The treatment ofTBI is an integral part of the practice of most traumacenters. We have always felt that the presence of TBIhad increased the patient’s risk of a thromboembolic

event, a situation that has led to active discussion withour neurosurgical colleagues regarding the timing ofprophylactic anticoagulation.

The current paper reviews 627,775 injured patientsfrom the NTDB 2008, of which 2,182 had pulmonaryembolism. The authors divided the injuries into severalcategories and discovered that their hypothesis was true,that is, TBI was not associated with an increased risk ofpulmonary embolism, independent of the severity of theTBI.

On the basis of this work, I have a couple ofquestions.

In the results, you stated that 30% of the patients withPE had an IVCF placed. Do you have any way ofdetermining if this was pre- or post-PE?

You discussed an interaction term that was created tostudy the effect of the IVCF and, unless I misunderstoodit, it seems that the use of IVCF is associated with agreater risk of PE. Again, is this cause or effect? Can youclarify this?

Were you able to determine the rate of warfarin usein the TBI patients versus the other groups? This wouldbe interesting, as the use of anticoagulation may increasethe risk of intracranial bleed, and the cessation of themedicine tends to lead to a hypercoagulable state, whichthen increases the risk of PE.

And finally, it appears that since the patient withisolated TBI is not at an increased risk of PE, they stillhave some risk. How should this be incorporated intoclinical practice? Does this mean that we can be slightlymore comfortable when the neurosurgeons do not wantprophylactic anticoagulation?

Dr Natasha St. Germain: Your first question aboutIVCF, unfortunately, from the way the database is setup, we can’t actually extract the timing for these filters.And that would be something that we would love tolook at in the prospective study.

The second question, about warfarin in these patientswith head injuries, we didn’t actually look specifically atwhich patients in the database were given warfarin. Weincluded them as a whole separate group.

And your last question, about clinical practice, for myownclinicalpractice, I doplan to change things.As it standsnow, there is always an argument between the traumasurgeon and the newer surgeons as far as is it okay to startthesepatients, specifically traumaticbrain-injuredpatients,on chemoprophylaxis? Normally, posttrauma, day one,we’d like to start them on some type of chemoprophylaxis,and the neurosurgeons often tell us they need to haveanother five head CTscans before we allow that.

I think I’d feelmore comfortable with allowing them tohold off on starting anything in the first 24 to 72 hours. Idon’t think that this would change my practice as far asindefinitely starting these patients on some kind of pro-phylaxis, at least in the initial posttraumatic period.

Dr Frederick Luchette (Maywood, IL): I applaud youfor the daunting task of trying to take on analysis of theNTDB data. As you know, the input of data is veryinstitution-specific. And one of my frustrations with

SurgeryVolume 154, Number 4

Jeremitsky et al 815

looking at the NTDB for any type of data analysis is thelack of validity of compliance with complication entryinto the database. And you know everything about VTEin trauma patients at this moment, as we stand here inthis room, and you know the incidence in the literaturein prospective studies. It’s as high as 25%, despite me-chanical as well as pharmacologic prophylaxis. So my

question to you is, did you compare your findings fromthe NTDB with your own institution’s, Allegheny Gen-eral, incidence of VTE in the trauma population?

Dr Natasha St. Germain: Unfortunately, we did notactually compare the national trauma databank to ourown institution’s data. I’m not sure that it’s complete,so we were unable to do that.