| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Reviews |
Departments of1 Family Medicine, 2 Internal Medicine, and 3 Health Evaluation Sciences, University of Virginia Health System, Charlottesville, VA.
aAddress correspondence to this author at: University of Virginia Health System, Department of Family Medicine, PO Box 800729, Charlottesville, VA 22908. Fax 434-243-4800; e-mail heims{at}virginia.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Methods: We identified eligible studies, using MEDLINE entries from February 1995 through October 2003, supplemented by a review of bibliographies of relevant articles. Studies reporting accuracy evaluations comparing D-dimer test results with lower extremity ultrasound or venography in symptomatic patients with suspected acute DVT were selected for review. Two reviewers critically appraised each study independently according to previously established methodologic standards for diagnostic test research. Those studies judged to be of highest quality were designated Level 1.
Results: The 23 Level 1 studies reported data on 21 different D-dimer assays. There was wide variation in assay sensitivity, specificity, and negative predictive values, and major differences in methodology of reviewed studies. A multivariate analysis of assay performance, controlling for sample size, DVT prevalence, reference standard, and patient mix, found few differences among the assays in effect on test performance as measured by diagnostic odds ratio. Increasing prevalence of DVT was associated with poorer test performance (P = 0.01), whereas the choice of venography as the reference standard was associated with better test performance (P <0.005).
Conclusions: Explanations for the wide variation in assay performance include differences in biochemical and technical characteristics of the assays, heterogeneity and small size of patient groups, and bias introduced by choice of reference standards. Assay sensitivity and negative predictive value were frequently <90%, uncharacteristic of a good rule-out test. General use of D-dimer assays as a stand-alone test for the diagnosis of DVT is not supported by the literature.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The measurement of D-dimer, degradation products of circulating cross-linked fibrin formed during activation of the coagulation system, has been studied extensively as an adjuvant test in the diagnosis of DVT. D-Dimer testing has become rapid, simple, and inexpensive, and it has the potential to detect thrombosis in any part of the venous system. If the sensitivity of the D-dimer test for DVT is consistently very high, its negative predictive value will also be high and reliably exclude the presence of disease. These are characteristics of a good "rule-out" test. As such, the use of D-dimer assays has been suggested as an initial test to rule out DVT to reduce the number of patients requiring diagnostic imaging. In 1996, our systematic review did not find sufficient evidence to support the use of D-dimer as a diagnostic test for DVT (1). However, since that time several new D-dimer assays have been introduced, and >50 studies and 8 reviews on the subject have been published (2)(3)(4)(5)(6)(7)(8)(9). Because none of the reviews critically appraised the primary studies or discussed the potential for bias in their results, we undertook this metaanalysis of the D-dimer accuracy literature published since our last systematic review to clarify the role of the test in the diagnosis of lower extremity DVT.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
inclusion criteria and assessment of study quality
We included for review studies that calculated sensitivities and specificities or provided paired test data for at least one D-dimer assay. We excluded abstracts, subgroup analyses of previously published studies, and studies reporting combined data of patients with suspected DVT and pulmonary embolism (PE).
Two of the authors (S.W.H. and J.T.P.) reviewed each study independently according to previously established methodologic standards for diagnostic test research (Table 1
) (10)(11). A "Level 1" designation was given to those studies judged to be of higher quality, fulfilling standards 1, 3, 6, and 7. Standards 1 and 7 were designed to exclude studies with potential for work-up and incorporation bias (12). Standards 3 and 6 were designed to exclude studies with poorly described and biased methods of patient assembly (13).
|
Our search strategy identified 52 studies. Twenty-three studies were designated Level 1, and their results form the basis for this review (14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36). We excluded those studies that failed to adequately describe the method of patient selection (Standard 3) (37)(38)(39)(40)(41)(42)(43)(44)(45)(46)(47)(48)(49)(50)(51), studied a population that did not represent the full spectrum of disease (Standard 6) (40)(42)(46)(52)(53), or based the performance of the reference standard on the results of the D-dimer testing (Standard 7) (40)(41)(44)(46)(48)(51)(52)(54)(55)(56).
data abstraction
Test characteristics of D-dimer assays from the Level 1 studies were abstracted by both reviewers, and discrepancies were resolved by consensus. In addition, we recorded details of the study population, including setting, participant recruitment, and patient demographics, as well as details of the study methods, including reference standard and assay performance. All studies used as reference standard ultrasound, venography, or a combination of the two. Because the algorithm for combining/sequencing the two modalities was often poorly described, for analysis purposes we contrasted studies in which venography was performed on all patients vs studies utilizing ultrasound or an ultrasound/venography combination. "Patient mix" was defined as outpatient if the study specified that all study participants were enrolled either from outpatient offices or through emergency departments. The patient mix was classified as "mixed" if it included both outpatients and inpatients. Because the number of purely inpatient studies was small, they were combined with the mixed category for analysis purposes. "Study setting" was defined as the method of assembly of the patient cohort and classified as inpatient, outpatient, emergency department, or referred. The "referred setting" included study cohorts assembled on presentation to radiology departments, specialized venous thromboembolism (VTE) services, and vascular medicine departments.
statistical analysis
A marginal logistic regression with exchangeable correlation matrix was used to model the diagnostic odds ratio (DOR) as a function of D-dimer assay while controlling for the reference standard, sample size, overall prevalence of disease, and patient mix of each study (57)(58). The study setting was not clearly distinct from patient mix and was not included in the model. These covariates were chosen a priori. Because each study evaluated from 1 to 13 different assays on the same sample, a repeated-measures approach was used with each study defining a cluster of nonindependent assay measures. The actual model is illustrated below with the parameter subscripts indicating the indexing of assay (a) within study (s):
 |
Our model assessed the independent effect of each covariate on diagnostic discrimination as assessed by the DOR, the ratio of the positive and negative likelihood ratios. The DOR is a summary measure of the ability of the diagnostic test to discriminate between diseased and nondiseased individuals and reflects the trade-off between sensitivity and specificity. It is closely related to the ROC curve and the area under the curve (AUC). A DOR of 1 is equivalent to an AUC of 50%. The bigger the DOR, the bigger the AUC, and as the AUC grows to 100%, the DOR grows to infinity. It is an accepted and commonly used single measure of diagnostic accuracy in the comparison of diagnostic tests (60)(61). As an aggregate measure, it has utility as a dependent variable in the analysis of predictors of test performance and diminishes heterogeneity attributable to test threshold effects or other study factors that might inflate sensitivity or specificity. Similar to other metaanalytic methods (62)(63), this statistical modeling approach can derive summary ROC curves, test hypotheses, and provide effect estimates. However, it requires fewer assumptions, uses all available observations, and takes into account data clustering. The criterion used for statistical significance in this metaanalysis was a two-tailed 
of 0.05. SAS (Ver. 8.01) was used to fit the statistical model (PROC GENMOD).
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
. These 23 studies were performed in 11 countries. Ultrasound was used as the reference standard in eight studies, venography in seven, and a combination of ultrasound and venography in eight. Many of the studies were small; the median number of patients was 132, and the range was 30–474 patients.
|
Exclusion criteria varied, with 8 studies excluding patients with a prior episodes of VTE, 13 studies excluding patients on anticoagulants, and 5 studies excluding patients with suspected PE. Only five studies explicitly examined D-dimer assays in patients with no previous history of VTE and not taking anticoagulants. Only one study did not adequately describe methods for independent, blinded test performance and comparison (Standard 2) (14). Although all Level 1 studies described the method of patient selection (Standard 3), three did not report simple demographic descriptions of their patients (Standard 4) (14)(20)(24). Results of eight studies were stratified by proximal and distal DVT (Standard 5) (19)(20)(25)(28)(29)(30)(34)(36). All 23 studies described methods (Standard 8) and reported results (Standard 9) for at least one D-dimer assay, but only 3 reported data on the reproducibility of the test results (Standard 10) (18)(26)(27).
Fourteen of 23 studies were performed exclusively in outpatient populations. The prevalence of DVT in these 14 studies ranged from 20% to 68%, with a median of 35%. We classified the other nine studies as mixed populations, with four exclusively inpatient, three including both inpatients and outpatients, and two not reporting the patient mix. The four studies performed exclusively on inpatient populations had a prevalence range of 28–69% and a median of 60%.
The studies evaluated 21 different D-dimer assays. Fourteen of the assays were included in more than one study, with 6 assays examined by four or more studies. The data on D-dimer assay performance are listed in Table 3
. Only 27% of these evaluations were performed on samples with DVT prevalence <40%, and only two assays (SimpliRED and VIDAS) were examined by more than one such lower prevalence study.
|
|
The results of the multivariate analysis are presented in Table 4
. The VIDAS assay DOR was used as the reference for test performance comparison. The multivariate analysis identified three assays with DORs that were significantly (P <0.05) different (all lower) from the VIDAS assay. The DORs of the other 17 assays were not significantly different from the VIDAS assay DOR, although most trended toward lower discriminant ability. Increasing prevalence of DVT in the study population was independently associated with poorer assay performance (P = 0.01), whereas the choice of venography as the reference standard was associated with better assay performance (P <0.005). Our model found no independent effect on assay performance by sample size or patient mix (outpatient or other).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
differences among d-dimer assays
To be useful for clinical decision-making, D-dimer assays must be rapid, accurate, and reliable in the result range used for DVT diagnosis. Not all assays have these characteristics because of important differences in both assay methodology and in the monoclonal antibodies used (5)(9).
Assay methodology.
The 21 D-dimer assays can be divided into one of six different method categories (Table 3
), each of which has advantages and disadvantages. The microplate ELISA tests, because of their accuracy, quantitative results, and ability to measure low concentrations of D-dimer, have long been considered the standard against which newer assays should be compared. They have the disadvantage of taking 2–4 h to perform, limiting their clinical usefulness in the evaluation of thromboembolic diseases. In addition, they are labor-intensive and must be run in batches rather than on single samples.
The first-generation latex agglutination assays are based on the visible agglutination of antibody-coated latex particles. Positive samples may be serially diluted to provide a semiquantitative estimate of D-dimer concentrations. Although these assays are rapid and easy to perform, the results are qualitative, observer-dependent, and limited in their ability to detect minimally increased D-dimer concentrations. The second-generation latex immunoassays use the same basic technique as the first-generation but use a photometric analyzer to provide a quantitative measure of D-dimer and are generally able to reproducibly measure lower concentrations of D-dimer.
The membrane ELISAs use a monoclonal antibody that is chemically tagged to produce a visible color change in the presence of increased plasma concentrations of D-dimer. Both the NycoCard and the Instant I.A. assays can be interpreted manually, but an automated reflectometer can be used with the NycoCard assay to provide a quantitative result. These assays are usually performed in a centralized laboratory and can provide results within 20 min.
The SimpliRED D-dimer assay is an erythrocyte agglutination assay performed on whole blood. Similar to the first-generation latex assays, the qualitative presence or absence of visible agglutination is determined by an observer. Because the SimpliRED assay requires no plasma separation or specialized equipment, it can be performed at the point of care with results available in less than 2 min.
The VIDAS D-dimer assay is a fully automated rapid enzyme-linked fluorescent assay (ELFA) that produces quantitative results within 35 min. This assay uses the same biochemical principles as the microplate ELISA tests, but also has a single-sample cartridge system.
Assays of all method categories are capable of correctly identifying samples with very high concentrations of D-dimer and can also correctly give "negative" results for samples with very low or "normal" concentrations of D-dimer. However, a substantial proportion of patients with DVT have D-dimer concentrations that are only slightly increased (17)(20)(31)(36)(64). Many of the assays perform less well when required to accurately and reproducibly measure these slightly increased D-dimer concentrations (65), providing lower sensitivities and poor inter- and intraassay reproducibility (66)(67). This is especially true for those tests that require an observer to determine the presence of agglutination (17)(66).
Monoclonal antibodies.
Although several latex agglutination assays use a common monoclonal antibody, most D-dimer assays use a unique monoclonal antibody. Each monoclonal antibody has a unique affinity for D-dimer fragments and unique testing properties. The affinity of the monoclonal antibody in each assay for D-dimer fragments directly determines the assay result, and each assay is affected to various degrees by factors in individual patient samples, such as the presence of anticoagulants (51)(68). The quantitative results of one D-dimer assay cannot be directly compared with those of a different assay. Consequently, an assay with a cut point of 300 µg/L fibrinogen equivalent units (FEU) is not necessarily more sensitive than an assay with a cut point of 500 µg/L FEU. Additionally, the units of measurement are not standardized across assays, with some assay results expressed as FEU and others as D-dimer units (one FEU is approximately one-half of a D-dimer unit).
There is a lack of consensus on cut points for individual assays (Table 3
). Our review found results for the quantitative assays frequently reported at different cut points. In several studies the cut point was not determined a priori but instead was chosen retrospectively to maximize the sensitivity of the assay (19)(25)(26)(27).
Recommended assays.
Although the second-generation latex, the membrane ELISA, and the automated rapid ELFA methods are all rapid, simple, and quantitative, no single method is clearly best. Within each method category, individual assays demonstrate wide variability in sensitivity and specificity. Our multivariate analysis identified the VIDAS assay as statistically significantly better than three of the other assays, but it was indistinguishable from assays representing all of the other five method categories. Interestingly, the three assays that the VIDAS assay "outperformed" were from a mixture of methods: a membrane ELISA, a first-generation latex agglutination assay, and a second-generation latex agglutination assay. Unfortunately, our analysis lacked power to detect small to moderate differences among assays. Our metaanalysis has not led to a definite recommendation with regard to specific assay.
heterogeneity of patient groups
Our review shows that there were major differences among the patient groups. This is not surprising considering the variety of countries, settings, and exclusion criteria used in the 23 studies. The variation in DVT prevalence is the most striking measure of this clinical heterogeneity. Prevalence ranged from 20% to 69%, with a median of 37%. Although this prevalence is similar to that reported in previous research on DVT diagnosis with ultrasound testing (69)(70), it is considerably higher than other studies of outpatients referred for evaluation of suspected DVT (16)(71)(72)(73). Test performance often varies with patient characteristics and severity of disease. Patients with more severe disease are more likely to have positive tests than those with mild or early disease; healthy patients are more likely to have negative tests than those with significant comorbidities, particularly those with conditions known to cause false-positive tests (12)(74). In the case of D-dimer and DVT, studies in which patients with more extensive clots are enrolled are likely to find higher D-dimer concentrations and to report higher sensitivities. This was confirmed in our review by the eight studies that stratified results by extent of disease (19)(20)(25)(28)(29)(30)(34)(36). All reported higher sensitivities for patients with clot extending to the thigh than for clot limited to the calf (75).
D-Dimer assays are also often positive in common conditions other than DVT, including cancer, recent trauma, or surgery (76). Studies in which many patients with these and other conditions associated with increased D-dimer concentrations are enrolled are likely to find higher D-dimer concentrations in patients classified as free of DVT and to report lower specificities. In our metaanalysis, prevalence of DVT, a proxy for severity of illness and comorbidity of the patient population, was an independent predictor of test performance, with the relative DOR decreasing by 22% for every 10% increase in prevalence of DVT. D-Dimer concentrations also vary with duration of symptoms (77), use of anticoagulants (78)(79), and age of the patient (80). However, we were unable to explore the effect of clinical heterogeneity on test discrimination because few studies stratified their results by patient characteristics and none provided disaggregated patient-level data. This variation in test performance across population subgroups, or "spectrum effect", makes it difficult to predict D-dimer assay performance in clinical settings with a broad disease spectrum (81).
random variation
A rule-out test must have a high sensitivity. However, estimates of sensitivity and specificity can be unstable with associated wide confidence intervals. For example, assuming point estimates of D-dimer assay sensitivity of 95% and specificity of 80%, a study would need a sample size of 395 individuals to ensure that the sensitivity 95% confidence interval lower limit would be >90%. Only three of the reviewed studies enrolled this number of patients (16)(20)(29).
reference standard bias
Our metaanalysis found that the choice of reference standard biased study results. In research on lower extremity DVT, the use of ultrasound has disadvantages as a reference standard because of its lower sensitivity in detecting calf DVT (69)(70). A reference standard, such as venography, that fully images both calf and thigh DVT will enable all patients with DVT to be properly counted as diseased. To the extent that ultrasound fails to image the calf veins, some diseased patients will be counted in the nondiseased category. Because the sensitivity of D-dimer for thigh DVT is higher than for isolated calf DVT, studies using ultrasound as the reference standard tend to have higher sensitivities, lower specificities, and higher negative predictive values than studies using venography (75). Our metaanalysis confirmed this bias. Reference standard was an independent predictor of test performance, with use of venography alone associated with a 2.7-fold higher DOR than reference standards using ultrasound. In future research, investigators must choose the reference standard with care.
To minimize heterogeneity of study results from multiple reference standards, we compared D-dimer test results with a single ultrasound and/or venogram, all tests performed at the same time. We did not adjust results when serial ultrasounds were done nor for added information from clinical outcomes on follow-up. These strategies, more often used in management trials, can partly compensate for not using venography by revealing the presence of previously undiagnosed calf DVT.
|
Conclusion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Almost one-half of the assays had reported specificities of <0.50; the resulting number of false-positive D-dimer tests therefore often exceeded the total number of negative tests. Although the intent of D-dimer testing is to rule out DVT, this situation can lead to increased rather than decreased diagnostic evaluation (82). The easy availability of the D-dimer test may induce clinicians to order it more frequently than they would have ordered diagnostic imaging. Increased utilization of ultrasonography and/or venography may then result from the obligation to pursue the work-up of patients with positive D-dimer tests.
More research is needed to determine the place of D-dimer testing in clinical practice. This could be undertaken in two ways. The first is additional accuracy studies that, similar to the reports included in this review, compare selected D-dimer assay results with the best possible reference standard. Although we believe that venography is the preferable reference standard, it has been replaced in clinical practice by ultrasonography. Strategies to improve the performance of ultrasound as a reference standard, such as serial ultrasounds or clinical follow-up, are better than a single ultrasound but are likely to uncover only a minority of calf DVT. Future accuracy studies should collect sufficient clinical information about enrolled patients to allow for stratification by risk of DVT as well as for conditions known to cause an increased D-dimer test result other than DVT. Adequate numbers of patients should be enrolled to achieve sufficient statistical power for the primary analysis as well as subgroup analysis. This is the approach that was successful in establishing the role of the ventilation-perfusion lung scan for the diagnosis of PE (83).
The second approach is the use of management trials, the type of research that established the safety of the serial noninvasive testing approach to patients with suspected lower extremity DVT (84)(85)(86). In management trials, patients are evaluated through diagnostic algorithms with outcome assessment to determine safety and cost-effectiveness. This approach has the advantage of avoiding the difficulty of performing venography on all patients. Accuracy studies and management trials are not mutually exclusive. The two types of research are usually done in sequence, with management trials being performed once promising results have been obtained from accuracy studies. To date, most management trials for suspected DVT have begun with a probability assessment followed by D-dimer testing, with the results determining whether ultrasonography would be performed (55)(87)(88)(89)(90)(91). These studies show that outpatients or emergency department patients with a low pretest clinical probability of DVT as determined by a validated clinical prediction rule (73) and a normal D-dimer test result can be safely managed without diagnostic imaging and without anticoagulation. Other management trials suggest that repeated ultrasound examinations may be safely avoided in patients with an initially negative ultrasound and a negative D-dimer test result (92)(93). Interestingly, these studies have used various D-dimer assays from several different method categories, including second-generation latex agglutination assays (87)(91), membrane ELISA assays (94), and the erythrocyte agglutination assay (55)(90)(92)(93). To firmly establish the safety of these and other approaches, large trials are needed that have adequate numbers of patients at the ends of the branching pathways (95).
Although much progress has been made in our knowledge of D-dimer testing since our last systematic review in 1996, many of the same questions that we raised then remain unanswered. Published estimates of the accuracy of D-dimer assays have limited generalizability to clinical practice, and no one D-dimer assay has emerged clearly as the best. Management studies have established a role for several D-dimer assays in an outpatient or emergency department population when coupled with a validated clinical prediction rule. The clinical utility of D-dimer testing for the diagnosis of DVT in most other clinical situations is still uncertain.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  C. Legnani, G. Palareti, B. Cosmi, M. Cini, A. Tosetto, A. Tripodi, and for the PROLONG Investigators (on behalf of FCSA a Different cut-off values of quantitative D-dimer methods to predict the risk of venous thromboembolism recurrence: a post-hoc analysis of the PROLONG study Haematologica, June 1, 2008; 93(6): 900 - 907. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. B. Segal, J. Eng, L. J. Tamariz, and E. B. Bass Review of the Evidence on Diagnosis of Deep Venous Thrombosis and Pulmonary Embolism Ann. Fam. Med, January 1, 2007; 5(1): 63 - 73. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. S. Wells, C. Owen, S. Doucette, D. Fergusson, and H. Tran Does This Patient Have Deep Vein Thrombosis? JAMA, January 11, 2006; 295(2): 199 - 207. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Perrier Review: several factors are associated with the performance of D-dimer assays for detecting deep venous thrombosis Evid. Based Med., November 1, 2004; 9(6): 185 - 185. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
