Sensitivity and Specificity of a Rapid Whole-Blood Assay for D-Dimer in the Diagnosis of Pulmonary Embolism

Methods

The study was performed from September 1993 to May 1996 and was approved by the institutional review boards of each participating hospital.

Patients

We included most patients from a recent management study that evaluated a standardized clinical model of pretest probability and developed a management strategy involving serial compression ultrasonography in most patients with nondiagnostic lung scans [21]. Thus, our sample comprised consecutive patients 18 years of age or older with clinically suspected acute pulmonary embolism who were referred to thromboembolism consultants at one of the participating tertiary-care hospitals: Chedoke-McMaster Hospitals and Hamilton Civic Hospitals, Hamilton, Ontario, Canada; Ottawa Civic Hospital, Ottawa, Ontario, Canada; and Queen Elizabeth II Health Sciences Centre, Halifax, Nova Scotia, Canada.

Patients were excluded from the study if they met any of the following criteria: 1) suspected upper-extremity deep venous thrombosis, 2) no symptoms within 48 hours of presentation, 3) treatment with anticoagulants for 72 hours or more, 4) life expectancy less than 3 months, 5) contraindication to contrast media, or 6) geographic inaccessibility. Informed consent was obtained from all patients in the study.

Clinical Assessment

Patients with clinically suspected pulmonary embolism were seen by a physician or nurse-practitioner (or both) on the health center's thromboembolism service. A history was taken and physical examination was done. Independent of diagnostic testing, all patients had clinical assessment of pretest probability, which was categorized as high, moderate, or low, as described elsewhere [21]. This evaluation involved assessment of 1) presenting symptoms and signs, 2) risk factors for venous thromboembolism, and 3) presence or absence of an alternative diagnosis at least as likely as pulmonary embolism.

Objective Testing for Pulmonary Embolism

The management approach used in the patient population is summarized in Figure 1. Ventilation-perfusion lung scanning was done in all patients within 24 hours of presentation by using a technique described elsewhere [22]. Scans were classified as normal (that is, the perfusion scan was normal), high probability (segmental or larger perfusion defects with normal ventilation) or nondiagnostic (perfusion defects not meeting criteria for a high-probability scan) [2]. All patients also had bilateral compression ultrasonography from the common femoral vein to the calf trifurcation within 24 hours of presentation, which was performed and interpreted according to a technique described elsewhere [23].

Patients with nondiagnostic lung scans, low or moderate pretest probability, and a normal initial compression ultrasonogram had repeated compression ultrasonography on days 3 to 5, 6 to 8, and 13 to 15. Anticoagulation was withheld provided that the results of compression ultrasonography remained normal. If the initial or serial ultrasonogram was abnormal, pulmonary embolism was diagnosed. Patients with nondiagnostic scans, high pretest probability, and normal initial compression ultrasonograms underwent venography; if the results of this test were normal, pulmonary angiography was done.

If the perfusion lung scan was normal, compression ultrasonography was performed; if ultrasonography results were normal, no further testing was done and pulmonary embolism was considered excluded. If the results were abnormal, pulmonary embolism was diagnosed. Patients with a high-probability lung scan and a high or moderate pretest probability were considered to have pulmonary embolism regardless of compression ultrasonography results. Patients with a high-probability scan, low pretest probability, and normal initial compression ultrasonograms had venography; if venograms were normal, these patients also underwent pulmonary angiography. All patients who did not receive anticoagulants were followed-up for 3 months for the presence or absence of symptomatic venous thromboembolism by clinical evaluation. This consisted of a careful history to elicit symptoms of pulmonary embolism or deep venous thrombosis and appropriate investigation if such symptoms occurred.

The final classification of patients as positive or negative for pulmonary embolism was based on a heterogeneous group of outcome measures. Patients were classified as positive if one or more of the following occurred: positive pulmonary angiogram; positive compression ultrasonogram (at any time) or positive contrast venogram; high-probability perfusion lung scan plus moderate or high pretest probability; or symptomatic, objectively confirmed venous thromboembolism during the 3-month follow-up. All other patients were classified as negative.

Patients considered positive for pulmonary embolism received full-dose intravenous heparin or subcutaneous low-molecular-weight heparin followed by at least 3 months of oral anticoagulation. Patients who were considered negative for pulmonary embolism did not receive anticoagulant therapy.

Measurement of D-Dimer Levels

Blood was taken and processed by a research assistant for quantitation of d-dimer by the SimpliRED assay at the time of referral to the thromboembolism consultants. Results were categorized as normal or abnormal on the basis of the absence (normal) or presence (abnormal) of erythrocyte agglutination. This corresponds to the interpretation of d-dimer assay results used in our previous studies [17, 19, 20]. The method for the performance of the assay has been described comprehensively elsewhere [18]. The results of the d-dimer assay were not disclosed to caregivers and were obtained independently of the pretest probability assessment and results of other diagnostic tests. This assay has been shown to have excellent interobserver agreement (κ = 0.95 [95% CI, 0.88 to 1.0]), between-assay agreement (κ = 0.96 [CI, 0.90 to 1.0]), and reproducibility (97%) [24].

Statistical Analysis

Negative predictive values, likelihood ratios, and posterior probabilities and their corresponding exact 95% CIs were calculated by using the binomial distribution [25].

Role of Industry Sponsor

Agen Biomedical, Ltd., donated the d-dimer kits but had no role in the design or conduct of the study or the decision to submit this paper for publication.

Results

During the study, 1881 patients were screened for eligibility. Of these, 484 were excluded because of prolonged anticoagulant therapy (n = 158), expected survival less than 3 months (n = 89), geographic inaccessibility (n = 68), contraindication to contrast medium (n = 60), attending physician refusal (n = 57), pregnancy (n = 23), suspected upper-extremity deep venous thrombosis (n = 17), lack of acute symptoms within 72 hours (n = 7), or age younger than 18 years (n = 5). Of the 1397 eligible patients, 1250 (89%) agreed to enter the study. Of the consenting patients, 2 had inadequate lung scans, 58 could not undergo d-dimer assays because of assay unavailability, and 13 were lost to follow-up and could not be included. Therefore, 1177 patients (59% women; mean age, 53.4 years [range, 20 to 94 years]) were included in the final analysis.

Of the 1177 patients, 197 (17%) were classified by the end of the study as positive for pulmonary embolism. None of the 980 patients classified as negative for pulmonary embolism after initial testing died of pulmonary embolism during follow-up. Overall, the d-dimer assay showed a sensitivity of 84.8%, a specificity of 68.4%, a likelihood ratio of a positive result of 2.7 (CI, 2.3 to 3.0), and a likelihood ratio of a negative result of 0.22 (CI, 0.16 to 0.31).

Subgroup Analyses by Pretest Probability

Of 703 patients (60% of the cohort) who were classified on the basis of clinical assessment alone as having a low pretest probability, 24 (3.4%) had pulmonary embolism. Of these 703 patients, 521 had a normal d-dimer test result; 5 of the 521 patients were classified as positive for pulmonary embolism. Therefore, in the subgroup of patients with a low pretest probability, the likelihood ratio of a negative d-dimer test result was 0.27. Applying this likelihood in the low-probability group resulted in a posterior probability of pulmonary embolism of 1.0% (CI, 0.3% to 2.2%) (Table 1).

Of 382 patients (32% of the cohort) who had a moderate pretest probability of pulmonary embolism according to clinical assessment, 101 (26.4%) were classified as positive for pulmonary embolism. The likelihood ratio of a negative result on the d-dimer test was 0.38, which produced a posterior probability of pulmonary embolism of 12.1% (CI, 7.6% to 18.1%) in this group (Table 1).

Of 92 patients (8% of the cohort) who had a high pretest probability on the basis of clinical assessment, 72 (78.3%) were classified as positive for pulmonary embolism. The likelihood ratios for negative and positive d-dimer test results were 0.15 and 1.7, respectively, resulting in posterior probabilities of 35.7% and 85.9%, respectively, in this group.

Subgroup Analyses by Ventilation-Perfusion Scanning Results

Of 698 patients (59% of the cohort) with a nondiagnostic lung scan, 52 (7.4%) had pulmonary embolism (Table 2). The d-dimer test result was normal in 431 (62%) of these patients, of whom 12 were classified as positive for pulmonary embolism. Therefore, in the subgroup with a nondiagnostic lung scan, the likelihood ratio of a negative d-dimer test result was 0.36; this produced a posterior probability of pulmonary embolism of 2.8% (CI, 1.4% to 4.8%). Of the 12 patients with false-negative d-dimer results, 3 had abnormal compression ultrasonograms obtained on the day of presentation and 6 had abnormal compression ultrasonograms after day 1 (4 on day 3, 1 on day 7, and 1 on day 14). Therefore, the negative predictive value of a normal d-dimer assay result plus a normal compression ultrasonogram obtained on the day of presentation in patients with a nondiagnostic lung scan increased marginally from 97.2% to 97.9% (CI, 96.1% to 99.0%). The d-dimer test result was abnormal in 267 (38%) of patients with a nondiagnostic lung scan; 40 (15%) of these patients were classified as positive for pulmonary embolism (Table 2), reflecting the moderate specificity of the test in this group. A summary of negative predictive values obtained by using various combinations of compression ultrasonography and d-dimer testing in patients with nondiagnostic lung scans is shown in Table 3.

Of 319 patients who had normal perfusion scans, 4 (1.3%) were positive for pulmonary embolism. In this subgroup, the likelihood ratio of a negative result was 0.15; this resulted in a posterior probability of pulmonary embolism of 0.4% (CI, 0.0% to 2.3%). Of 160 patients who had high-probability lung scans, 141 (88%) were categorized as positive for pulmonary embolism. In this subgroup, the likelihood ratios associated with positive and negative d-dimer results were 1.7 and 0.26, respectively; accordingly, the posterior probabilities in this group were 92.5% and 65.4%, respectively. None of the patients with a low pretest probability, high-probability lung scan, and normal d-dimer test result proved to have pulmonary embolism.

Discussion

In this study of more than 1000 patients with suspected pulmonary embolism, we made two important observations that may markedly simplify the diagnostic testing for this condition. First, in a subgroup analysis, we observed that a low pretest probability and a normal d-dimer test result occurred in 44% of the study sample and had a negative predictive value of 99%. This observation is potentially important because it implies that this sizable subgroup of patients with suspected pulmonary embolism can be managed without the need for ventilation-perfusion lung scanning. In support of the high negative predictive value of the combination of a normal d-dimer test result and low pretest probability and the hypothesis that such patients do not require lung scanning, none of the 9 patients with a normal d-dimer test result, low pretest probability, and high-probability lung scans had pulmonary embolism. It is well known that of patients with suspected pulmonary embolism and high-probability lung scans, approximately 10% have false-positive scans and do not have pulmonary embolism [3].

In addition, the observed negative predictive value of a nondiagnostic lung scan and a normal d-dimer test result at presentation was 97.2%. This implies that if patients with nondiagnostic scans and a normal d-dimer test result are discharged without further study, about 3% might return with symptomatic venous thromboembolism. This is probably an overestimate because if these patients had been left untreated rather than having serial compression ultrasonography, symptomatic recurrence would have been unlikely in all of the 3% of those classified as positive for pulmonary embolism because they had a later abnormal compression ultrasonogram.

Our results are consistent with those of several other investigations and show that d-dimer testing has high sensitivity and moderate specificity for pulmonary embolism [6, 9, 10-12, 20]. The results are also consistent with our recent study evaluating d-dimer testing in patients with suspected deep venous thrombosis, a disease that shares a common pathophysiology and epidemiology with pulmonary embolism [19]. In that study, we showed that further testing and anticoagulant therapy could be obviated if the d-dimer test result was negative in patients with suspected deep venous thrombosis and either a low pretest probability or a normal initial result on impedance plethysmography [19]. Although other studies evaluating d-dimer testing in patients with pulmonary embolism have been published, no study this large has shown that an assay suitable for point-of-care testing has such clinical utility. However, clinicians must realize that the d-dimer assay cannot be used as a stand-alone test, nor can it reliably exclude pulmonary embolism in patients whose pretest probability is not low unless it is combined with other objective tests, such as lung scanning.

It is important to address the potential limitations of our conclusions, particularly with respect to the clinical classification of patients. There are two methods of evaluating the d-dimer assay in patients with clinically suspected pulmonary embolism. The first is to perform pulmonary angiography, the reference standard for the diagnosis of pulmonary embolism, either in all patients or in patients whose perfusion lung scans are abnormal. However, this technique is expensive and invasive. The second approach is to use a clinical management strategy that tests the long-term safety of withholding anticoagulant therapy in patients classified as negative for pulmonary embolism. Inherent in this approach is the likelihood that pulmonary embolism may have been the cause of presenting symptoms in some patients whom we classified as negative for pulmonary embolism. Our classification of patients with normal perfusion scans or normal pulmonary angiography as negative for pulmonary embolism is valid on the basis of the results of published studies [3, 4]. On the other hand, two studies recently demonstrated that most patients (approximately 90%) with high-probability lung scans have pulmonary embolism as a cause of their symptoms [2, 3].

To summarize, our results strongly suggest that further investigation and anticoagulant therapy can be obviated in two subgroups of patients with suspected pulmonary embolism and a normal d-dimer result: 1) those with a low pretest probability as determined by a clinical assessment protocol and 2) those with a nondiagnostic lung scan. Our findings should be verified in large clinical management studies by other investigators.

Drs. Wells and Bormanis: Ottawa Civic Hospital, Civic Parkdale Clinic, Fourth Floor, 1053 Carling, Ottawa, Ontario K1Y 4E9, Canada.

Drs. Kearon and Crowther: Hamilton Health Sciences Corp., Henderson General Division, 711 Concession Street, Hamilton, Ontario L8V 1C3, Canada.

Dr. Anderson: Centre for Clinical Research, West Wing, McKenzie Building, Room 117B, 5788 University Avenue, Halifax, Nova Scotia B3H 1Y8, Canada.

Drs. Weitz, MacKinnon, Gent, and Hirsh: Hamilton Health Sciences Research Centre, Henderson General Division, 711 Concession Street, Hamilton, Ontario L8V 1C3, Canada.

Dr. Brill-Edwards: McMaster University Medical Centre, 1200 Main Street West, HSC-3W11, Hamilton, Ontario L8N 3Z5, Canada.

Dr. Turpie: Hamilton Health Sciences Corp., Hamilton General Division, 237 Barton Street East, Hamilton, Ontario L8L 2X2, Canada.