The accuracy of the enzyme-linked immunosorbent assay D-dimer test in the diagnosis of pulmonary embolism: A meta-analysis

P U L M O N A R Y / O R I G I N A L C O N T R I B U T I O N

Michael D. Brown, MDBrian H. Rowe, MD, MScMathew J. Reeves, PhDJ. Michelle Bermingham, MDSamuel Z. Goldhaber, MD

From the Grand Rapids MERC/Michigan State University Pro-gram in Emergency Medicine,Grand Rapids, MI (Brown,Bermingham); the Division ofEmergency Medicine, Univer-sity of Alberta, Edmonton,Alberta, Canada (Rowe); theDepartment of Epidemiology,Michigan State University, EastLansing, MI (Reeves); and theCardiovascular Division, Depart-ment of Medicine, Brigham andWomen’s Hospital, and HarvardMedical School, Boston, MA(Goldhaber).

Study objective: We sought to determine the sensitivity and specificity of theenzyme-linked immunosorbent assay (ELISA) D-dimer test in the diagnosis of pul-monary embolism (PE) in the adult emergency department population.

Methods: A search of MEDLINE, EMBASE, and bibliographies of previous system-atic reviews was conducted, with no language restriction. Experts in the field of PEresearch were contacted to identify unpublished studies. Prospective investigationsinvolving a predominately outpatient population suspected of PE that used ELISAD-dimer tests were included. Two authors extracted data independently andassessed study quality on the basis of the patient spectrum and reference standard.Consensus was reached by means of conference. The analysis was based on a sum-mary receiver operating characteristic curve and pooled estimates for sensitivity andspecificity by using a random-effects model.

Results: The search yielded 52 publications. No unpublished studies were found.Eleven studies met the inclusion criteria and provided a sample of 2,126 patients. Thesummary receiver operating characteristic curve analysis found significant hetero-geneity among the 11 studies. Subgroup analysis of the 9 studies that used traditionalELISA D-dimer methods yielded the most valid pooled estimates, with a sensitivity of0.94 (95% confidence interval [CI] 0.88 to 0.97) and a specificity of 0.45 (95% CI 0.36 to0.55). Advanced age resulted in a lower specificity. A prolonged duration of symp-toms decreased the sensitivity and specificity.

Conclusion: The ELISA D-dimer test is highly sensitive but nonspecific for thedetection of PE in the clinical setting. This test might help clinicians safely rule outPE, especially in the face of low and low-to-moderate pretest probabilities.[Brown MD, Rowe BH, Reeves MJ, Bermingham JM, Goldhaber SZ. The accuracy of the enzyme-linkedimmunosorbent assay D-dimer test in the diagnosis of pulmonary embolism: a meta-analysis. Ann EmergMed. August 2002;40:133-144.]

The Accuracy of the Enzyme-Linked

Immunosorbent Assay D-Dimer Test in the

Diagnosis of Pulmonary Embolism: A Meta-Analysis

A U G U S T 2 0 0 2 4 0 : 2 A N N A L S O F E M E R G E N C Y M E D I C I N E 1 3 3

Copyright © 2002 by the AmericanCollege of Emergency Physicians.

0196-0644/2002/$35.00 + 047/1/124755doi:10.1067/mem.2002.124755

E L I S A D - D I M E R M E T A - A N A L Y S I SBrown et al

1 3 4 A N N A L S O F E M E R G E N C Y M E D I C I N E 4 0 : 2 A U G U S T 2 0 0 2

I N T R O D U C T I O N

The goal of any diagnostic test is to allow the clinicianto revise the patient’s probability of having disease to alevel greater than a treatment threshold or less than atest threshold.1 For conditions such as pulmonaryembolism (PE), the diagnostic threshold is low becausethe cost of missing the diagnosis is high. The 3-monthmortality rate for untreated PE has been reported to beas high as 17.5%.2 Given the inconsistent continuity ofcare and unreliable follow-up, the emergency depart-ment evaluation of patients who present with symp-toms and signs suspicious for PE is often expensive andcomplex.

PE can present with a variety of symptoms and signs,which have recently been grouped into models that pre-dict low, moderate, and high pretest probabilities.3,4

When faced with patients who might have PE, manyemergency physicians elect to obtain a ventilation-per-fusion (V/Q) lung scan to identify V/Q mismatching.Unfortunately, such testing is rarely definitive and usu-ally requires further complementary testing. Alterna-tive diagnostic options, which are available to varyingdegrees at EDs across North America, include helicalcomputed tomography (CT) scans, pulmonary angiog-raphy, and/or imaging of the extremities (eg, Dopplerultrasonography, impedance plethysmography, venog-raphy). The complex algorithms published for the diag-nostic workup of PE reflect physicians’ fear of missingthe diagnosis and the lack of universally accepted diag-nostic approaches for this condition.3,5

Fortunately, several new diagnostic tests have beenintroduced to assist in the ED workup of suspectedcases of PE. D-Dimer has been advocated as a diagnostictool that might preclude additional diagnostic tests,such as V/Q scanning, helical CT scanning, and angiog-raphy.6 D-Dimer is a fibrin degradation product that isusually increased in the presence of thromboembolicdisease. Unfortunately, it is also increased in such com-mon diseases as inflammatory arthritidies, cancer, andinfection. It might also be increased after surgery ortrauma. A number of different methods are currentlyavailable to measure D-dimer, including latex aggluti-

nation, whole-blood agglutination, and enzyme-linkedimmunosorbent assay (ELISA).6,7 Until very recently,the rapid latex tests and bedside assays have had inade-quate sensitivity to rule out a life-threatening condi-tion, such as PE.8,9

Because the published studies on the use of D-dimerto rule out PE to date are of various size and quality, theaccuracy and utility of the test is still debated. Previoussystematic reviews evaluating the utility of D-dimer inthe diagnosis of thromboembolism were very broad inscope7,10,11 and included deep vein thrombosis and PE,multiple testing methods, and both inpatient and out-patient populations. Thus, these findings might havelimited applicability to ED settings. The topic also war-rants an updated meta-analysis to include more recentinvestigations with a rapid ELISA D-dimer testing methodthat is more practical in the ED setting. To limit theproblems of clinical heterogeneity12 found in previoussystematic overviews and to eliminate the problems ofinterobserver reliability associated with qualitativetests, we chose to study only quantitative ELISA D-dimertests in this review. The primary objective of this sys-tematic review was to determine the accuracy of theELISA D-dimer test in the diagnosis of PE in the ED. Asecondary objective was to determine whether the testcharacteristics change with respect to covariates, suchas age, comorbidity, or duration of symptoms.

M A T E R I A L S A N D M E T H O D S

The following research question was posed: What isthe accuracy (eg, sensitivity, specificity, likelihoodratios) of the ELISA D-dimer test in the diagnosis of PEin the adult patient presenting to the ED with a sus-pected PE?

Computerized searching was performed with MED-LINE (January 1980 to January 1, 2001) to identify clin-ical studies assessing the utility of an ELISA D-dimertest in the diagnosis of PE. The search used the follow-ing MeSH terms: (pulmonary-embolism OR PE ORVTE) AND (D-dimer OR fibrin OR fibrinogen-degrada-tion OR FDP OR fibrinogen-degradation-products)AND (ELISA OR enzyme-linked-immunosorbent-assay)


V/Q scan, (2) CT scan result positive for PE, or (3) posi-tive lower extremity imaging study result (eg, ultra-sonography, impedance plethysmography, venogram).A negative angiography result was considered the crite-rion standard for ruling out PE. Acceptable surrogatereference standards for a negative diagnosis were nor-mal or very low probability V/Q scan or clinical follow-up documenting the absence of a thromboembolicevent over a minimum of 3 months.5 If a reference stan-dard is not used in all patients, the study is susceptibleto verification bias (workup bias).18-20 To minimize theeffect of verification bias and to provide the most con-servative estimate for test sensitivity, any study that didnot apply a reference standard to all patients had theresults analyzed on a worse-case assumption (ie, eachpatient lost to follow-up was assumed to have the worstoutcome).

The rigorous inclusion-exclusion criteria functionedas the primary quality filter in this meta-analysis. Themeta-analysis focused the appraisal of study quality onthe potential for differential reference standard bias20

and spectrum bias.19 Differential reference standardbias is a more subtle form of verification bias and mightoccur when a negative test result is verified by using aless rigorous standard than that used to verify a positivetest result.20,21 For example, a patient with a negativeELISA D-dimer result has his or her result verified with asingle lower extremity ultrasonogram and outpatientfollow-up, whereas a patient with a positive ELISAD-dimer result has his or her result verified by means ofserial lower extremity ultrasonographic examinationsand hospitalization. The reference standard and patientspectrum for each study was graded in regard to qualityparameters (A, excellent; B, susceptible to some bias; C,indeterminate or poor), as outlined below:

• The potential for differential reference standardbias was assessed as follows20: grade A, those studiesusing the same reference standard, regardless of theELISA D-dimer result; grade B, studies using differentreference standards depending on the results of theELISA D-dimer test; grade C, indeterminate or notmeeting the study protocol definition of an appropriatereference standard.

AND Sensitivity and Specificity. A search of EMBASEwas also performed with a similar approach.

Two reviewers (MDB, BHR) independently exam-ined the titles and abstracts of the references identifiedin the initial MEDLINE and EMBASE searches to deter-mine whether the study was relevant to the clinicalquestion (relevance search).13 Reviews and editorialswere excluded immediately. The reference list of thearticles chosen for inclusion in the meta-analysis andthe reference list of prior systematic reviews10,11 werealso screened to identify further studies for inclusion.In an attempt to identify other so-called grey literature,14

experts in the area of PE and the companies that marketlaboratory equipment that use ELISA D-dimer methodswere contacted. Non–English-language articles passingthe initial screening were translated before full review.

To be included in the meta-analysis, the study musthave been a prospective investigation involving a pre-dominately outpatient population presenting withsymptoms and signs suspicious for PE. If a study in-cluded any inpatients, the study population must havebeen composed of at least 80% outpatients, or data musthave been available to calculate sensitivity and speci-ficity for the outpatient component of the study popula-tion.

After the relevance search, the 2 primary authors metto compare their exclusion logs to determine whetherthere was any discordance. Where there was disagree-ment, consensus was reached by means of conference. Adata collection form was used to abstract data from eachstudy meeting the inclusion criteria. If a study met theinclusion criteria, reviewers attempted to contact theauthor to identify additional articles, to confirm dataextraction-estimation for correctness and complete-ness, and to obtain missing data. Two reviewers (MDB,BHR) independently confirmed numeric calculationsand graphic extrapolations. The data were evaluated forthe presence of publication bias by using statisticalmethods.15-17

Although a positive angiogram or autopsy result isconsidered the criterion standard for the diagnosis ofPE, we considered any one of the following as accept-able surrogate reference standards: (1) high-probability



• The external validity (generalizability) of the meta-analysis depends on the spectrum of disease included ineach study and how the patient population was assem-bled21: grade A, patient spectrum would be expected toinclude a consecutive or random sampling of a typicaloutpatient population presenting with symptoms andsigns suspicious for PE; grade B, studies that selectedonly a small subgroup of patients suspected of PE; gradeC, indeterminate or not meeting the study protocol def-inition of an appropriate patient spectrum.

The blind interpretation of the test under investiga-tion and the reference standard are typically consideredimportant components in the critical appraisal of diag-nostic tests.22 However, because the interpretation of aquantitative ELISA D-dimer test result is an objectivemeasurement, it is less critical that the technician per-forming the D-dimer quantitative analysis be blinded tothe clinical history or the reference standard.18 How-ever, there is potential for interpretation bias if the radi-ologist performing the reference standard was not blindto the ELISA D-dimer result.19 This information wasobtained from the article or by author query.

To provide the most conservative estimate of testcharacteristics, after each study was scored for quality,grade C studies were excluded from the analysis.

The statistical analysis was based on a summaryreceiver operating characteristic (SROC) curve.23,24

When studies use different thresholds for positive andnegative results, the reported sensitivity and specificitywill differ among studies. A graphic display of the vari-ability of the test characteristics between studies can beassessed with the SROC curve.21 Only single testthresholds and dichotomous results were used in theanalysis to simplify calculations. If a study reportedresults for more than 1 test threshold, a test threshold of500 ng/mL was used because this is the most commoncut-off point used in clinical practice. The sensitivityand specificity for the single test threshold identifiedfor each study was used to plot an unweighted SROCcurve.24,25 A correction factor of one half was added toeach cell to avoid calculation problems by having avalue of zero in the 2 × 2 table.24 This correction has notbeen found to significantly alter the results of the SROC


curve.24 The SROC curve analysis is based on a logittransformation of the data, which plots the difference(D=logit TPR–logit FPR) on the y axis and the sum(S=logit TPR+logit FPR) on the x axis. The y axis (D) isequivalent to the log diagnostic odds ratio, and the xaxis (S) is a measure of how the test characteristics varywith the test threshold. A regression equation (D=α+β*S)derived from the SROC curve analysis can be used toassess the heterogeneity among study results. If the βcoefficient is near zero and not statistically significant,then evidence of significant heterogeneity is not pre-sent.23 When there is little variability of test results (ie,homogeneity), the SROC curve does not provide infor-mation in addition to average sensitivity or specificityvalues.25 A random-effects model was used to calculatethe average sensitivity and specificity across stud-ies.21,25,26 The random-effects model accounts forbetween-study variability and provides a more conser-vative estimation compared with that of the fixed-effects model.23 Statistical tests related to the SROCcurve were performed by using Meta-Test (version 0.6,Joseph Lau, MD, New England Medical Center, Boston,MA). All other statistical tests were performed by usingthe SAS statistical application program (version 8.0,SAS, Inc., Cary, NC).

Sensitivity analysis was used to assess the effect ofstudy quality on the overall results. The SROC curveswere compared with and without the specified method-ologic flaw. A priori subgroup analyses were performedon studies using traditional ELISA methods, rapid ELISAmethods, the 500 ng/mL cut-off value, age, comorbid-ity, and duration of symptoms. Comorbidity was de-fined as having surgery, trauma, myocardial infarction,stroke, acute infection, disseminated intravascularcoagulation, pregnancy, postpartum, or active cancerwithin the 10 days preceding the ED evaluation.27

To assess for the presence of publication bias, where-by smaller studies show effects different from those oflarger studies, we used methods previously proposed byGalbraith15 and Egger et al.16 The standard normal devi-ate (SND) of the odds ratio (calculated by dividing theodds ratio by its SE) was regressed against its precision(as measured by the inverse of the SE; ie, SND=α+β*pre-


another that focused specifically on an elderly popula-tion47 with a mean age of 81 years. Excluding these 2outliers,47,49 the mean age of the patients in the studieswere similar, ranging from 54 to 63 years. The regres-sion analysis showed no statistical evidence of publica-tion bias on the basis of a P value of .19, although theabsolute value of the intercept was large (ie, 98).16,17

Only 2 studies were given a grade A with respect toboth of the key quality parameters, patient spectrumand reference standard (Table 3).28,48 One of thesestudies included 20% inpatients, which was the maxi-mum percentage of inpatients a study could have and

cision). The intercept α value provides a measure of thedegree of asymmetry resulting from publication bias.Data from a homogeneous or symmetric set of trials willscatter around a line that runs through the 0 origin,whereas in the presence of publication bias, the inter-cept will deviate from 0. The absolute magnitude of α, aswell as its 95% confidence interval (CI), can therefore beused as one measure of the presence of publication bias.

R E S U L T S

The MEDLINE search yielded 52 references. Eighteenwere immediately deemed ineligible for full review (1letter to the editor, 14 reviews, and 3 studies evaluatingD-dimer in patients suspected of having only deep veinthrombosis). The relevance search had excellent agree-ment between the 2 reviewers, with a simple agreementof 92% and a κ value of 0.83 (95% CI 0.67 to 0.99). AnEMBASE search yielded 71 references; 7 additional ref-erences were identified as eligible for full review (Figure1). The search for grey literature yielded 2 additionalpublished articles.28,29

A complete article review was performed on the re-maining 43 articles. Dutch (n=1), French (n=4), andGerman (n=1) articles required translation. After fullreview, 3 studies were identified as duplicates, and 17others were excluded for various reasons (Table 1). Anattempt was made to contact the authors of the 23remaining studies to clarify important missing informa-tion and confirm data extraction. Seventy-eight percent(18/23) of authors responded in some form to thesequeries. After obtaining additional information, a fur-ther 12 studies were excluded (Table 2).30-41 Elevenstudies therefore met the inclusion criteria and pro-vided a total study population of 2,126 patients. A sum-mary of the major characteristics of each study is pro-vided in Table 3.27-29,42-49

The prevalence of disease among outpatients sus-pected of PE ranged from 17% to 58%, with most fallingin the 20% to 40% range. In almost all studies, femalepatients were represented slightly more than malepatients. Most studies included a broad range of ages,except for one study with a mean age of 72 years49 and


Figure 1. Search, inclusion, and exclusion flow diagram. The grey lit-erature was defined as those studies that were unpublished orthat had limited distribution.14

Grey literaturesearch

2

MEDLINE search52

34 Ineligible18

EMBASEsearch

7

Full review43

Excluded17

(Table 1)Potential inclusion

23

Duplicates3

(Table 1)

Final inclusion11

(Table 3)

Excluded12

(Table 2)

Author contact18/23


still meet the inclusion criteria, as defined in the re-search protocol.48 Three of the studies included weregiven an excellent rating with regard to the referencestandard but were rated grade B with respect to thepatient spectrum. In contrast, 6 studies were given agrade A with respect to the patient spectrum but grade Bon the reference standard. In the majority (7/11) ofstudies, the radiologist was blind to the D-dimer resultsand any other clinical information.

The sensitivity and specificity of each included studywere calculated, with the 95% CI displayed (Figure 2and Table 3). The pooled summary estimate with a ran-dom-effects model resulted in a sensitivity of 0.95 (95%CI 0.90 to 0.98) and a specificity of 0.45 (95% CI 0.38 to0.52). However, this pooled estimate demonstrated sta-tistically significant heterogeneity, with a β value of .43(95% CI .004 to .87). The visual display provided by theSROC curve shows moderate variability in the results(Figure 3).

A sensitivity analysis on the basis of the key method-ologic quality parameters was performed to explore theheterogeneity found in the overall results (Table 4).When the SROC curve was derived by using only thosestudies with an excellent or grade A quality rating inregard to the reference standard, the test for hetero-geneity was not significant, with a β value of .18 (95%CI –.52 to .88). The pooled summary estimate for thissubgroup using a random-effects model resulted in asensitivity of 0.90 (95% CI 0.83 to 0.94) and a speci-


ficity of 0.40 (95% CI 0.26 to 0.55). In contrast, the sub-group of studies with a grade A patient spectrum wasfound to have statistically significant heterogeneity.The proper blinding of the radiologist to the D-dimerresults had minimal effect on the results (Table 4).There were only 2 studies that used a rapid form of theELISA D-dimer test, and therefore, an SROC curve anal-ysis could not be performed. The average sensitivity forthese 2 studies was higher (100%) when compared withthe pooled results by using traditional methods (sensi-tivity 0.94 [95% CI 0.88 to 0.97], specificity 0.45 [95%CI 0.36 to 0.55)]). Using a worse-case assumption didnot alter the pooled estimates because, in the majorityof studies, very few patients were lost to follow-up. Theonly exception was the subgroup with the rapid ELISAmethod.29,45 When the worse-case assumption wasapplied to this subgroup, the average sensitivity de-creased to 95%.

Although there were only a few studies with sufficientdata for subgroup analysis on the basis of age, comor-bidity, and symptom duration, all 3 of these covariateshad a considerable effect on the test characteristics ofthe ELISA D-dimer test (Table 4). Older age (≥70 years)

Table 1. Reasons for exclusion after full article review.

Primary Reason No. of Reports

Review article 4Duplicate publication 3Latex D-dimer method 3Grade C patient spectrum* 3Grade C reference standard† 7Total 20*Patient spectrum: grade C, indeterminate or not meeting protocol definition for ade-quate patient spectrum.†Reference standard: grade C, indeterminate or not meeting protocol definition foradequate reference standard.

Table 2. Articles excluded (n=12) after quality rating.

Author Year Reason

Sijens30* 2000 Grade C spectrum†‡

Quinn et al31 1999 Grade C spectrum‡

Brimble and Ginsberg32 1997 Duplicate‡

Duet et al33 1998 Grace C spectrumReber et al34 1995 Duplicate‡

Bonnin et al35 1997 Grade C spectrum‡

van Beek et al36 1996 Data missingRochemaure et al37 1995 Grade C spectrumFlores et al38 1995 Grade C reference standard§

Goldhaber et al39 1993 Grade C spectrum‡

van Beek et al40 1993 Data missingBounameaux et al41 1990 Grade C spectrum‡

*Search result exclusively from EMBASE.†Patient spectrum: grade C, indeterminate or not meeting protocol definition for ade-quate patient spectrum.‡After correspondence with author.§Reference standard: grade C, indeterminate or not meeting protocol definition foradequate reference standard.


the duration of symptoms was greater than 3 days inthis study population, both the sensitivity and speci-ficity decreased (Table 4).

D I S C U S S I O N

This systematic review attempted to identify the cur-rent published and unpublished literature regardingthe use of the ELISA D-dimer test in the diagnosis of PE

was the only specific age-related subgroup that could beanalyzed. The specificity of the ELISA D-dimer test inthis elderly study population47 was much lower (14%)than for all of the other individual studies included inthe meta-analysis (Figure 2). The subgroup analysis ofthe 3 studies with data on the population withoutcomorbidity had a higher specificity (55%) but a lowersensitivity (89%). Only one study evaluated the effect ofsymptom duration on the ELISA D-dimer results.27 If


Table 3. Eleven studies of ELISA D-dimer in the diagnosis of PE: Study characteristics and diagnostic test performance.

Test Study Quality StudyPrevalence N (% Mean Male Sex,* Threshold, Sensitivity, Specificity, (Reference Quality Radiologist

Study, Year of PE, % Outpatient) Age,* y % ng/mL % (95% CI) % (95% CI) Standard)† (Spectrum)‡ Blind

Demers et al, 199228 20 84 (100) NA 35 300 94 (69 to 100) 52 (40 to 64) A A YesGinsberg et al, 199343 17 150 (100) 54 36 300 100 (87 to 100) 35 (27 to 45) B A YesLenzhofer et al, 199342 19 107 (100) 54 36 250 88 (62 to 98) 79 (69 to 86) B A Node Moerloose et al, 199444 34 150 (100) NA NA 500 100 (93 to 100) 47 (37 to 58) B A Node Moerloose et al, 199645 24 195 (100) 60 44 500§ 100 (92 to 100) 38 (30 to 46) B A YesPerrier et al, 199746 29 671 (100) 62 NA 500 99 (97 to 100) 41 (37 to 46) B A NoMeyer et al, 199848 42 142 (80) 55 41 500 92 (81 to 97) 50 (39 to 61) A A YesTardy et al, 199847 42 96 (100) 81 34 500 100 (91 to 100) 14 (7 to 27) A B YesBarro et al, 199949 58 26 (100) 72 50 500 87 (59 to 98) 27 (7 to 60) A B YesHeit et al, 199927 31 61 (100) 63 44 500 88 (61 to 98) 51 (36 to 66) A B YesPerrier et al, 199929 23 444 (100) 61 47 500§ 100 (97 to 100) 47 (42 to 53) B A No

N, Number of patients suspected of having PE; NA, adequate data not available after author contact.*Demographics are close approximations on the basis of the information available.†Reference standard: grade A, same reference standard regardless of the ELISA D-dimer; Grade B, potential for differential reference standard bias.‡Patient spectrum: grade A, consecutive or random sampling of a typical outpatient population presenting with symptoms suspicious for PE; grade B, potential for spectrum bias.§Used a rapid ELISA D-dimer method.

80100 60 40 20 0

REM

80 1006040200

1110987654321

1110987654321

Barro et al49Heit et al27Perrier et al29Meyer et al48Tardy et al47Perrier et al46de Moerloose et al45de Moerloose et al44Ginsberg et al43Lenzhofer et al42Demers et al28

Study

19991999199919981998199719961994199319931992

Year

1516

1066040

1964651261717

Dis

1145

3388256

47514999

1249067

NoDisSensitivity SpecificityFigure 2. Sensitivity and specificityplot, with the 95% CI dis-played as horizontallines. The circle at thebottom labeled REM is thepooled sensitivity andspecificity with a random-effects model.


in the acute care setting. After an exhaustive search, theapplication of stringent inclusion-exclusion criteria,and rigorous selection methodology, we included 11studies involving 2,126 patients in the study. Theresults demonstrate that the traditional ELISA tech-nique is highly sensitive (94%) yet only moderatelyspecific (45%). A highly sensitive test will assist theclinician in ruling out PE. By using evidenced-basedmedicine terms,22,50 the negative likelihood ratio forthis test would be clinically helpful (0.1), especiallywhen applied in the face of accurate pretest probabilitymodels for PE.3,4 For example, a patient with a pretestprobability of 20% (low) in the face of a negative ELISAD-dimer result would have a posttest probability of 2%;in such a case, further testing is probably not war-


ranted. However, the positive likelihood ratio for thistest is only 1.7 and is of limited value in most clinical sce-narios.

An adequate reference standard (criterion standard)and an appropriate spectrum of patients are key ele-ments of study methodology when considering applica-tion of the results to clinical practice.51 Althoughangiography has long been considered the criterionstandard for the diagnosis of PE, it has been recom-mended that clinicians now use an outcome-based stan-dard.5 As expected, the majority of studies included inthis overview used an outcome-based standard, such assymptom-free follow-up and survival. Most emergencyphysicians would consider the absence of a thrombo-embolic event or death over a 3-month period a validoutcome measure.5 Studies that use a different refer-ence standard, depending on the ELISA D-dimer test

Figure 3. SROC curve analysis of ELISA D-dimer in the diagnosis ofPE. Plotted in each of the SROC graphs are individual studiesdepicted as ellipses. The x and y dimensions of the ellipsesare proportional to the square root of the number of patientsavailable to study the sensitivity and specificity, respectively,within the analysis. Also shown is the unweighted SROCcurve limited to the range in which data are available. X rep-resents the independent random-effects pooling of sensitivityand specificity values of the studies.

Sens

itivi

ty

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.00.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1–Specificity

Area under curveunweighted=0.5631

Table 4. Sensitivity analysis: Random-effects pooled estimates of sen-sitivity and specificity for various study subgroups, includingtesting for heterogeneity.

Sensitivity, Specificity, β ValueStudy Group N % (95% CI) % (95% CI) (95% CI)

Reference standard: 5 90 (83 to 94) 40 (26 to 55) .18 (–.52 to .88)grade A*

Patient spectrum: 8 97 (92 to 99) 48 (41 to 56) .46 (.10 to .81)grade A†‡

Proper blinding§ 7 92 (85 to 96) 39 (30 to 49) .39 (–.16 to .94)Traditional ELISA 9 94 (88 to 97) 45 (36 to 55) .32 (–.18 to .82)Traditional ELISA, 6 95 (87 to 98) 40 (32 to 50) .55 (–.31 to 1.41)

500 ng/mLll

Rapid ELISA, 2 100 44 (40 to 48) NA500 ng/mLll¶

Age≥70 y 1 100 14 (7 to 21) NANo comorbidity‡ 3 89 (75 to 96) 55 (46 to 64) .59 (.04 to 1.14)Duration≥4 d 1 73 (59 to 86) 33 (19 to 48) NA

N, Number of studies included in pooled subgroup-sensitivity analysis; β, regressioncoefficient for the sum (logit TPR+logit FRP), which indicates heterogeneity amongstudies when the value is significantly different from zero; NA, not appropriate orapplicable.*Reference standard: grade A, same reference standard regardless of the ELISA D-dimer.†Patient spectrum: grade A, consecutive or random sampling of a typical outpatientpopulation presenting with symptoms suspicious for PE.‡Statistically significant heterogeneity.§Proper blinding: radiologist interpreting the reference standard was blind to the ELISAD-dimer result.llUsed a threshold of greater than 500 ng/mL as criteria for a positive D-dimer test result.¶Values reported are an average for the 2 studies.


have found that the average effect of inappropriate blind-ing is small,20 and our meta-analysis had similar find-ings. After excluding studies in which the radiologist wasnot blinded, there was not a clinically significant changein the pooled estimates of diagnostic accuracy. In themajority of studies with proper blinding (4/5), the per-sonnel performing clinical follow-up were also blindedto the ELISA D-dimer test results and clinical course.

The traditional ELISA D-dimer has been shown tohave excellent sensitivity but requires specialized equip-ment available only at major academic centers and istoo time consuming to be valuable in most ED set-tings.2,8 Therefore, studies with a rapid ELISA D-dimermethod are important and were consequently reportedas a subgroup. The 2 studies in this subgroup wereamong the largest included in the review and reportedan impressive sensitivity of 100%.29,45 However, bothof these investigations were prone to differential refer-ence standard bias, which might yield an erroneouslyhigh sensitivity.19 For the clinician, the pooled resultsof studies using traditional methods would appear to bea more conservative estimate for test sensitivity (94%)compared with the perfect sensitivity (100%) reportedby these 2 studies using a rapid ELISA D-dimer method.

Despite the rigorous methodology used in this sys-tematic review, there are inherent limitations in thistype of analysis. Typically, the studies included in ameta-analysis should be evaluated for the presence ofpublication bias by using graphic and statistical meth-ods.16 Publication bias is expected to be a much greaterproblem in meta-analysis of diagnostic tests than inmeta-analysis of randomized controlled trials of thera-peutic effects, but there is no agreed on method toassess publication bias for diagnostic meta-analysis.23

Aware of these limitations, we did perform a regressionanalysis, which demonstrated possible evidence of pub-lication bias given the large value of the intercept. How-ever, the small number of studies included limits thestatistical power of the test, and the intercept was notstatistically significant.16 Our search attempted toexamine all literature on this topic, but it is possiblethat some unpublished or other foreign-language stud-ies were missed. In addition, it might be that lower sen-

result, might provide an erroneously high estimate fortest sensitivity.20 As expected, the 5 studies using a refer-ence standard not susceptible to differential referencestandard bias20 (grade A) had a lower pooled sensitivity(90%) and specificity (40%). However, because they hadselected a subgroup of the target outpatient populationsuspected of PE, these studies were predominantly oflower quality with respect to the patient spectrum (gradeB) and susceptible to spectrum bias (or assembly bias).These studies might have been representative of patientpopulations with comorbidity or a longer duration ofsymptoms. For example, Heit et al27 selected a studypopulation referred to the Mayo Clinic for angiography.Of all studies included, this was the only investigationthat performed angiography on all patients and thereforewas the least susceptible to differential reference stan-dard bias. However, the external validity of the resultsmust be questioned because these patients were not typi-cal of the ED patient population suspected of having PE.The study by Barro et al49 was also rated excellent inregard to the reference standard but was rated grade Bwith respect to the patient spectrum. This study includedonly those patients suspected of PE who were subse-quently admitted to the hospital, which explains the veryhigh prevalence (58%) of PE. The specificity of the ELISAD-dimer might also be affected by the patient spectrum,as demonstrated by the low specificity (14%) found inthe study focused only on the elderly.47 In a subsequentpublication, investigators combined data from 2 of thestudies included in the review29,46 and performed a sub-group analysis providing similar results, with a speci-ficity of 17% in those aged older than 70 years.52 The rea-sons for the lower specificity in the elderly is unknownbut might be related to comorbidity that raises the D-dimer levels in settings not confined to thromboem-bolism (eg, cancer, inflammation, recent surgery).

Different forms of interpretation bias are another con-cern in the evaluation of diagnostic studies.19 To limitdiagnostic review bias, the radiologist performing theinterpretation of the reference standard should beblinded to the results of the ELISA D-dimer test.19 Pre-vious investigators who have assessed the effects ofdesign-related bias in studies evaluating diagnostic tests



sitivity results remained unpublished, which wouldinflate the pooled estimates. However, we did includeforeign-language publications and searched extensivelyfor other evidence of unpublished work, and therefore,we believe this bias to be limited.

There is also the possibility that some informationmight have been lost during translation of those studiespublished in a language other than English. For instance,we could not determine the reason for the unusuallyhigh specificity (80%) reported in one study publishedin German,42 and author contact was unsuccessful.Finally, in the original protocol, a subgroup analysiswas planned to determine whether the test characteris-tics might differ in those patients based on varying lev-els of pretest probability. Although attempts were madeto derive this information from the authors, in mostcases, pretest probability estimation was not performed,and therefore, additional information was not available.Consequently, there were insufficient data in the pri-mary studies to comment on this important issue.

Restricting the systematic review to D-dimer teststhat use an ELISA method makes it inappropriate toapply the results to clinical settings where alternativeD-dimer testing methods are used. However, a recentinvestigation has shown that a newer latex-basedmethod had remarkably similar test characteristics(sensitivity 96%, specificity 45%)53 to the summaryestimates derived in this meta-analysis. A normalwhole-blood agglutination D-dimer assay in combina-tion with a normal alveolar dead-space fraction hasalso been shown to have adequate sensitivity to ruleout PE.54 A systematic review by Kline et al7 comparedvarious D-dimer testing methods. However, themethodology of this meta-analysis differed from ourswith respect to the search, the inclusion criteria, andthe quality assessment. In fact, only 4 of the studieswere common to both reviews.43,45-47

Notwithstanding the aforementioned concerns, thismeta-analysis followed widely accepted methodologyfor the selection of studies, the data-extraction process,the analysis of study quality, and the evaluation of sub-groups.55 Overall, the summary data reveal that theELISA D-dimer is highly sensitive but only moderately


specific. Given these results, the D-dimer test might be a

safe and efficient test to reduce the overall costs associ-

ated with rule-out PE assessment in the ED, although

the cost effectiveness of D-dimer testing in the clinical

setting has not been adequately evaluated. Moreover, as

rapid D-dimer tests become more readily available, the

number of patients screened for PE will likely increase,

which could result in more V/Q or CT scans ordered by

physicians.56 Pooling the results of studies using tradi-

tional ELISA methods provides the most valid pooled

estimates. Although the studies using a rapid ELISA

D-dimer were large and would be useful for many

nonacademic medical centers, the sensitivity reported

in these studies appears to be suspiciously high. The

test characteristics of the ELISA D-dimer appear to

change among certain subgroups of patients, such as

those having comorbidity or prolonged symptom dura-

tion. In addition to assisting clinicians, the summary

estimate might assist researchers performing decision

analyses or developing a protocol to test a diagnostic

algorithm for use in the ED.

Author contributions: MDB and BHR conceived the study. MDB,BHR, and MJR designed the meta-analysis protocol. JBM and SZGperformed critical review of the protocol. MDB and BHR were re-sponsible for study selection, quality assessment, and data extrac-tion. JBM assisted with data extraction and translation. MDB, BHR,and MJR performed data analysis and statistics. MDB drafted themanuscript, and all authors contributed substantially to its revisionand final review. MDB takes responsibility for the paper as a whole.

Received for publication August 28, 2001. Revisions receivedJanuary 2, 2002, and February 22, 2002. Accepted for publicationMarch 10, 2002.

Reprints not available from the authors.

Address for correspondence: Michael D. Brown, MD, SpectrumHealth-Butterworth, Emergency Medicine, 100 Michigan Avenue,NE, Grand Rapids, MI 49503; 616-391-3588, fax 616-391-3674; [email protected].

We thank Joseph Lau, MD, New England Medical Center, for pro-viding the Meta-Test statistical software. In addition, we thank thefollowing corresponding authors: Claire Barro, MD; Francois Bonnin,MD; Henri Bounameaux, MD; Philippe de Moerloose, MD; MicheleDuet, MD; Jeffrey Ginsberg, MD; John Heit, MD; Guy Meyer, MD;Arnaud Perrier, MD; Deborah Quinn, MD; Guido Reber, MD; PaulSijens, PhD; Claudine Soria, PhD; and Bernard Tardy, MD.


26. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials.1986;7:177-188.

27. Heit JA, Minor TA, Andrews JC, et al. Determinants of plasma fibrin D-dimer sensi-tivity for acute pulmonary embolism as defined by pulmonary angiography. Arch PatholLab Med. 1999;123:235-240.

28. Demers C, Ginsberg JS, Johnston M, et al. D-Dimer and thrombin-antithrombin IIIcomplexes in patients with clinically suspected pulmonary embolism. Thromb Haemost.1992;67:408-412.

29. Perrier A, Desmarais S, Miron MJ, et al. Non-invasive diagnosis of venous throm-boembolism in outpatients. Lancet. 1999;353:190-195.

30. Sijens PE. Rapid ELISA assay for plasma D-dimer in the diagnosis of segmental andsubsegmental pulmonary embolism: a comparison with pulmonary angiography. ThrombHaemost. 2000;84:156-159.

31. Quinn DA, Fogel RB, Smith CD, et al. D-Dimers in the diagnosis of pulmonary embolism.Am J Respir Crit Care Med. 1999;159:1445-1449.

32. Brimble KS, Ginsberg JS. Evaluation of the combination of a bedside D-dimer assayand enzyme-linked immunosorbent soluble fibrin assay in patients with suspectedvenous thromboembolism. Thromb Res. 1997;88:291-297.

33. Duet M, Benelhadj S, Kedra W, et al. A new quantitative D-dimer assay appropriatein emergency: reliability of the assay for pulmonary embolism exclusion diagnosis.Thromb Res. 1998;91:1-5.

34. Reber G, Vissac AM, de Moerloose P, et al. A new, semi-quantitative and individualELISA for rapid measurement of plasma D-dimer in patients suspected of pulmonaryembolism. Blood Coagul Fibrinolysis. 1995;6:460-463.

35. Bonnin F, Hadjikostova H, Jebrak G, et al. Complementarity of lung scintigraphyand D-dimer test in pulmonary embolism. Eur J Nucl Med. 1997;24:444-447.

36. van Beek EJ, Schenk BE, Michel BC, et al. The role of plasma D-dimers concentra-tion in the exclusion of pulmonary embolism. Br J Haematol. 1996;92:725-732.

37. Rochemaure J, Laaban JP, Achkar A, et al. Value of the determination of D-dimersin the diagnostic approach of venous thrombo-embolic disorders. Bull Acad Natl Med.1995;179:299-314.

38. Flores J, Lancha C, Perez Rodriguez E, et al. Efficacy of D-dimer and total fibrindegradation products evaluation in suspected pulmonary embolism. Respiration.1995;62:258-262.

39. Goldhaber SZ, Simons GR, Elliott CG, et al. Quantitative plasma D-dimer levelsamong patients undergoing pulmonary angiography for suspected pulmonary embolism.JAMA. 1993;270:2819-2822.

40. van Beek EJ, van den Ende B, Berckmans RJ, et al. A comparative analysis of D-dimerassays in patients with clinically suspected pulmonary embolism. Thromb Haemost.1993;70:408-413.

41. Bounameaux H, Schneider PA, Slosman D, et al. Plasma D-dimer in suspected pul-monary embolism: a comparison with pulmonary angiography and ventilation-perfusionscintigraphy. Blood Coagul Fibrinolysis. 1990;1:577-579.

42. Lenzhofer R, Wimmer F, Haydl H, et al. Prospective study of determining the valueof D-dimer in diagnosing pulmonary embolism. Wien Klin Wochenschr. 1993;105:492-496.

43. Ginsberg JS, Brill-Edwards PA, Demers C, et al. D-Dimer in patients with clinicallysuspected pulmonary embolism. Chest. 1993;104:1679-1684.

44. de Moerloose P, Minazio P, Reber G, et al. D-Dimer determination to exclude pul-monary embolism: a two-step approach using latex assay as a screening tool. ThrombHaemost. 1994;72:89-91.

45. de Moerloose P, Desmarais S, Bounameaux H, et al. Contribution of a new, rapid,individual and quantitative automated D-dimer ELISA to exclude pulmonary embolism.Thromb Haemost. 1996;75:11-13.

46. Perrier A, Desmarais S, Goehring C, et al. D-Dimer testing for suspected pulmonaryembolism in outpatients. Am J Respir Crit Care Med. 1997;156:492-496.

47. Tardy B, Tardy-Poncet B, Viallon A, et al. Evaluation of D-dimer ELISA test in elderlypatients with suspected pulmonary embolism. Thromb Haemost. 1998;79:38-41.

R E F E R E N C E S1. Pauker SG, Kassirer JP. The threshold approach to clinical decision making. NEngl J Med. 1980;302:1109-1117.

2. Goldhaber SZ. Pulmonary embolism. N Engl J Med. 1998;339:93-104.

3. Wells PS, Ginsberg JS, Anderson DR, et al. Use of a clinical model for safemanagement of patients with suspected pulmonary embolism. Ann Intern Med.1998;129:997-1005.

4. Wicki J, Perneger TV, Junod AF, et al. Assessing clinical probability of pulmonaryembolism in the emergency ward: a simple score. Arch Intern Med. 2001;161:92-97.

5. Wolfe TR, Hartsell SC. Pulmonary embolism: making sense of the diagnostic evalu-ation. Ann Emerg Med. 2001;37:504-514.

6. Gallagher EJ. Clots in the lung. Ann Emerg Med. 2000;35:181-187.

7. Kline JA, Johns KL, Colucciello SA, et al. New diagnostic tests for pulmonaryembolism. Ann Emerg Med. 2000;35:168-180.

8. Janssen MC, Wollersheim H, Verbruggen B, et al. Rapid D-dimer assays to excludedeep venous thrombosis and pulmonary embolism: current status and new develop-ments. Semin Thromb Hemost. 1998;24:393-400.

9. Farrell S, Hayes T, Shaw M. A negative SimpliRED D-dimer assay result does notexclude the diagnosis of deep vein thrombosis or pulmonary embolus in emergencydepartment patients. Ann Emerg Med. 2000;35:121-125.

10. Becker DM, Philbrick JT, Bachhuber TL, et al. D-Dimer testing and acute venousthromboembolism. A shortcut to accurate diagnosis? Arch Intern Med. 1996;156:939-946.

11. Bounameaux H, de Moerloose P, Perrier A, et al. Plasma measurement of D-dimer asdiagnostic aid in suspected venous thromboembolism: an overview. Thromb Haemost.1994;71:1-6.

12. Thompson SG. Why sources of heterogeneity in meta-analysis should be investi-gated. BMJ. 1994;309:1351-1355.

13. Meade MO, Richardson WS. Selecting and appraising studies for a systematic re-view. Ann Intern Med. 1997;127:531-537.

14. McAuley L, Pham B, Tugwell P, et al. Does the inclusion of grey literature influ-ence estimates of intervention effectiveness reported in meta-analyses? Lancet.2000;356:1228-1231.

15. Galbraith RF. A note on graphical presentation of estimated odds ratios from severalclinical trials. Stat Med. 1988;7:889-894.

16. Egger M, Davey Smith G, Schneider M, et al. Bias in meta-analysis detected by asimple, graphical test. BMJ. 1997;315:629-634.

17. Steichen TJ. sbe19: tests for publication bias in meta-analysis. Stata Tech BullReprints. 1999;7:125-133.

18. Begg CB. Biases in the assessment of diagnostic tests. Stat Med. 1987;6:411-423.

19. Mower WR. Evaluating bias and variability in diagnostic test reports. Ann EmergMed. 1999;33:85-91.

20. Lijmer JG, Mol BW, Heisterkamp S, et al. Empirical evidence of design-related biasin studies of diagnostic tests. JAMA. 1999;282:1061-1066.

21. Irwig L, Tosteson AN, Gatsonis C, et al. Guidelines for meta-analyses evaluatingdiagnostic tests. Ann Intern Med. 1994;120:667-676.

22. Jaeschke R, Guyatt G, Sackett DL. Users’ guides to the medical literature. III. Howto use an article about a diagnostic test. A. Are the results of the study valid? Evidence-Based Medicine Working Group. JAMA. 1994;271:389-391.

23. Irwig L, Macaskill P, Glasziou P, et al. Meta-analytic methods for diagnostic testaccuracy. J Clin Epidemiol. 1995;48:119-132.

24. Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostictest into a summary ROC curve: data-analytic approaches and some additional consid-erations. Stat Med. 1993;12:1293-1316.

25. Lau J, Ioannidis JP, Balk EM, et al. Diagnosing acute cardiac ischemia in the emer-gency department: a systematic review of the accuracy and clinical effect of currenttechnologies. Ann Emerg Med. 2001;37:453-460.



48. Meyer G, Fischer AM, Collignon MA, et al. Diagnostic value of two rapid and indi-vidual D-dimer assays in patients with clinically suspected pulmonary embolism: com-parison with microplate enzyme-linked immunosorbent assay. Blood Coagul Fibrinolysis.1998;9:603-608.

49. Barro C, Bosson JL, Pernod G, et al. Plasma D-dimer testing improves the manage-ment of thromboembolic disease in hospitalized patients. Thromb Res. 1999;95:263-269.

50. Jaeschke R, Guyatt GH, Sackett DL. Users’ guides to the medical literature. III.How to use an article about a diagnostic test. B. What are the results and will they helpme in caring for my patients? The Evidence-Based Medicine Working Group. JAMA.1994;271:703-707.

51. Oxman AD, Cook DJ, Guyatt GH. Users’ guides to the medical literature. VI. How touse an overview. Evidence-Based Medicine Working Group. JAMA. 1994;272:1367-1371.

52. Righini M, Goehring C, Bounameaux H, et al. Effects of age on the performance ofcommon diagnostic tests for pulmonary embolism. Am J Med. 2000;109:357-361.

53. Bates SM, Grand’Maison A, Johnston M, et al. A latex D-dimer reliably excludesvenous thromboembolism. Arch Intern Med. 2001;161:447-453.

54. Kline JA, Israel EG, Michelson EA, et al. Diagnostic accuracy of a bedside D-dimerassay and alveolar dead-space measurement for rapid exclusion of pulmonary embo-lism: a multicenter study. JAMA. 2001;285:761-768.

55. Cook DJ, Sackett DL, Spitzer WO. Methodologic guidelines for systematic reviewsof randomized control trials in health care from the Potsdam Consultation on Meta-Analysis. J Clin Epidemiol. 1995;48:167-171.

56. Goldstein NM, Kollef MH, Ward S, et al. The impact of the introduction of a rapid D-dimer assay on the diagnostic evaluation of suspected pulmonary embolism. ArchIntern Med. 2001;161:567-571.


Documents

The accuracy of the enzyme-linked immunosorbent assay D-dimer test in the diagnosis of pulmonary embolism: A meta-analysis