Potential Impact of Adjusting the Threshold of the Quantitative D-dimer Based on Pretest Probability of Acute Pulmonary Embolism

Authors

  • Christopher Kabrhel MD, MPH,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • D. Mark Courtney MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Carlos A. Camargo Jr MD, DrPH,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Christopher L. Moore MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Peter B. Richman MD, MBA,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Michael C. Plewa MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Kristen E. Nordenholtz MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Howard A Smithline MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Daren M. Beam MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Michael D. Brown MD,

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author
  • Jeffrey A. Kline MD

    1. From the Department of Emergency Services, Massachusetts General Hospital, Harvard Medical School (CK, CAC), Boston, MA; the Department of Emergency Medicine, Northwestern University Medical Center (DMC), Chicago, IL; the Department of Emergency Medicine, Yale University Medical Center (CLM), New Haven, CT; the Department of Emergency Medicine, Mayo Clinic Arizona (PBR), Scottsdale, AZ; the Department of Surgery, St. Vincent Mercy Medical Center, Medical College of Ohio (MCP), Toledo, OH; the Division of Emergency Medicine, Department of Surgery, University of Colorado (KEN), Denver, CO; the Department of Emergency Medicine, Baystate Medical Center (HAS), Springfield, MA; the Pitt County Memorial Hospital, East Carolina University School of Medicine (DMB), Greenville, NC; the Department of Emergency Medicine, Michigan State University (MDB), Grand Rapids, MI; and the Department of Emergency Medicine, Carolinas Medical Center (JAK), Charlotte, NC.
    Search for more papers by this author

  • Presented at the Society for Academic Emergency Medicine (SAEM) Annual Meeting, Washington, DC, 2008.

  • Jeffrey Kline owns stock in CP Diagnostics LLC.

Address for correspondence and reprints: Jeffrey A. Kline, MD; e-mail: Jeffery.kline@carolinashealthcare.org.

Abstract

Objectives:  The utility of D-dimer testing for suspected pulmonary embolism (PE) can be limited by test specificity. The authors tested if the threshold of the quantitative D-dimer can be varied according to pretest probability (PTP) of PE to increase specificity while maintaining a negative predictive value (NPV) of >99%.

Methods:  This was a prospective, observational multicenter study of emergency department (ED) patients in the United States. Eligible patients had a diagnostic study ordered to evaluate possible PE. PTP was determined by the clinician’s unstructured estimate and the Wells score. Five different D-dimer assays were used. D-dimer test performance was measured using 1) standard thresholds and 2) variable threshold values: twice (for low PTP patients), equal (intermediate PTP patients), or half (high PTP patients) of standard threshold. Venous thromboembolism (VTE) within 45 days required positive imaging plus decision to treat.

Results:  The authors enrolled 7,940 patients tested for PE, and clinicians ordered a quantitative D-dimer for 4,357 (55%) patients who had PTPs distributed as follows: low (74%), moderate (21%), or high (4%). At standard cutoffs, across all PTP strata, quantitative D-dimer testing had a test sensitivity of 94% (95% confidence interval [CI] = 91% to 97%), specificity of 58% (95% CI = 56% to 60%), and NPV of 99.5% (95% CI = 99.1% to 99.7%). If variable cutoffs had been used the overall sensitivity would have been 88% (95% CI = 83% to 92%), specificity 75% (95% CI = 74% to 76%), and NPV 99.1% (95% CI = 98.7% to 99.4%).

Conclusions:  This large multicenter observational sample demonstrates that emergency medicine clinicians currently order a D-dimer in the majority of patients tested for PE, including a large proportion with intermediate PTP and high PTP. Varying the D-dimer’s cutoff according to PTP can increase specificity with no measurable decrease in NPV.

Quantitative D-dimer testing is commonly used to rule out acute pulmonary embolism (PE). Previous analyses of the quantitative D-dimer test have found it to be highly sensitive for PE, but limited to some degree by low specificity. Independent meta-analyses have demonstrated that automated quantitative D-dimer assays, including immunoturbidimetric and enzyme-linked immunosorbent assays, have approximate sensitivity of 95% and specificity of 45% at thresholds that correspond to 1,000 fibrinogen equivalent units, leading to a negative likelihood ratio of approximately 0.1.1–4

Published clinical guidelines and Food and Drug Administration labeling recommend the use of pretest probability (PTP) assessment in combination with a singular D-dimer threshold to arrive at an acceptably low posttest probability as a method to rule out PE.5,6 Available data support using a negative D-dimer to rule out PE in patients with low PTP. Data from European studies suggest that D-dimer testing may also be appropriate for intermediate PTP patients, but data from high PTP patients are sparse.7–10

The practice of limiting D-dimer use to low PTP patients assumes that the D-dimer has the same diagnostic performance across the spectrum of patients tested. Adjusting the positive/negative threshold of the D-dimer level according to the patient’s PTP could allow the D-dimer to safely exclude PE in a larger number of patients, including those with higher PTP. This would increase the fraction of patients with a low PTP who could have PE excluded without ionizing radiation. This adjustment strategy was evaluated by Linkins et al.,11 who showed in a retrospective analysis that adjusting the D-dimer’s threshold could increase the D-dimer’s specificity to 60% without compromising sensitivity (95%), or negative predictive value (NPV; >98%).11 A similar strategy was also analyzed retrospectively by Righini et al.,12 who demonstrated a marginal increase in the D-dimer’s usefulness.

The aim of this study was to evaluate the quantitative D-dimer in patients using data obtained from a large multicenter study of emergency department (ED) patients evaluated for PE. We hypothesized that an approach using a variable positive/negative threshold would increase the specificity of quantitative D-dimer testing while maintaining the high NPV of the test.

Methods

Study Design

Data for this article were collected as part of a prospective, multicenter, observational study of ED patients undergoing testing for possible PE from May 1, 2003, to March 31, 2007. The institutional review board of each participating institution approved the protocol for the enrollment of patients.

Study Setting and Population

This study analyzed final data from participating centers in the United States, comprising 10 academic medical centers and two community hospitals. Patients were eligible for enrollment if the treating clinician had sufficient suspicion to order any of the following tests to diagnose possible PE: D-dimer, computed tomography (CT) pulmonary angiography, ventilation/perfusion lung scan (V/Q), or extremity venous ultrasound. We required that the treating physician confirm that the diagnostic test was ordered as part of an evaluation for acute PE, so that studies ordered to evaluate deep vein thrombosis (DVT) without PE, or for other reasons, did not trigger enrollment. Sites enrolled patients either 1) during randomly selected ED shifts representative of all day, evening, night, weekday, and weekend shifts at the institution or 2) consecutively, defined as a minimum of 85% capture of eligible patients. Details of study enrollment are described elsewhere.13

Study Protocol

After a diagnostic test for PE was ordered, but before results were known, data were collected based upon knowledge of the clinician who ordered the qualifying diagnostic study for PE. To enhance generalizability, we characterized PTP using both an unstructured approach and the more structured scoring system developed by Wells et al.14–17 When using the unstructured estimate, we defined low PTP as <15%, intermediate PTP as 15%–40%, and high PTP as >40%. When using the Wells score, we defined low PTP as <2 points, intermediate PTP as 2–6 points, and high PTP as >6 points.14,18 Both the unstructured estimate and data for the Wells score were collected prospectively at the time of enrollment.

For the purpose of follow-up, patients were also asked whether they would return to the enrolling institution after discharge if their symptoms returned or worsened or if they required additional care. Either the treating clinician or a study coordinator uploaded data into a Web-based, secure, electronic data collection form, accessible from any computer in the ED that had Internet access.19

The decision to order a D-dimer was at the discretion of the evaluating physician. D-dimer assays were those used for routine clinical care at participating institutions. We analyzed patients who had one of the following quantitative D-dimer assays ordered in the ED, prior to pulmonary vascular imaging: VIDAS (bioMerieux SA, Marcy-Etoile, France), Liatest (Diagnostica Stago, Asnières sur Seine, France), Hemosil (Dade Behring, Marburg, Germany), Advanced D-dimer (Dade Behring), Biopool Minutex (diaPharma, West Chester, OH), or MDA (Organon Teknika Corporation, Durham, NC). Using the standard definitions of negative, Liatest, VIDAS, and MDA D-dimers were considered negative at concentrations of <500 ng/ml, Biopool Minutex at <250 ng/mL, Hemosil at <244 ng/mL, and the Advanced D-dimer at <1.6 μg/mL.

When using variable D-dimer cutoffs, we defined a negative D-dimer as either twice (for low PTP patients) or half (for high PTP patients) the standard definition and no change for patients with intermediate PTP. We chose to use twice/half the standard values, rather than the values used by Linkins et al.,11 for two main reasons: 1) we favored a simple system that clinicians could remember, even at the expense of statistical efficiency, and 2) it was unclear whether the cutoffs Linkins et al. used (0.2, 0.5, and 2.1 μg FEU mL−1, MDA D-dimer [bioMerieux, Inc., Durham, NC]) should be applied proportionately across different D-dimer assays. We preferred to confirm the usefulness of predefined cutoffs rather than reestablish novel cutoffs (e.g., using receiver operating characteristic [ROC] curves), which would then need to be validated. The levels we chose were decided a priori, and no other levels were tested.

Outcome Measures

The outcome of interest was a diagnosis of acute PE during the ED visit or within 45 days of the patient’s ED evaluation. We considered patients to have PE if they were evaluated for possible PE in the ED, if they had radiologic confirmation of the diagnosis of either PE or DVT during the index visit or within 45 days of the index visit, or if they died of PE during the 45-day follow-up period. Confirmatory imaging included CT angiography or conventional angiography showing a pulmonary arterial or deep venous filling defect interpreted as positive for PE or DVT, high-probability V/Q scan, or positive venous ultrasound consistent with DVT in the proximal or distal vasculature of the upper or lower extremities. All imaging results were based on the dictated report of board-certified attending radiologists not affiliated with the study. We followed patients for 45 days using a previously validated, published methodology that included chart review and telephone follow-up.13,19

Data Analysis

We used SAS v 9.l, (SAS Institute, Cary, NC) for all statistical calculations. Baseline characteristics are reported as simple proportions, means with standard deviations (±SD), and medians with interquartile ranges (IQR). Differences between groups were tested using chi-squared analysis. Test characteristics were calculated from 2 × 2 tables using standard techniques and are reported with exact 95% confidence intervals (CI).

Results

We prospectively enrolled 7,940 patients (Figure 1), of whom 545 (6.9%) had PE and 4,357 (55%) had a quantitative D-dimer performed. The Liatest was the most commonly performed assay (34%), followed by the VIDAS (29%). There were 10 patients for whom the type of D-dimer assay performed could not be ascertained. These were included in the total number of patients, but excluded from all analyses related to D-dimer. Among patients who had a quantitative D-dimer performed, 8 had unexpected electronic data transfer errors that precluded analysis of clinician’s unstructured assessment of PTP. All patients had the data necessary to calculate the Wells score. The D-dimer concentrations were normal in the majority of patients tested (n = 2,454; 56%). Two-hundred thirty-four (5.4%) patients who had a quantitative D-dimer performed were diagnosed with PE. Table 1 shows the baseline characteristics of enrolled patients and patients who had a D-dimer as part of their diagnostic evaluation.

Figure 1.

 Patients evaluated for possible PE. PE = pulmonary embolism; PTP = pretest probability. *Eight patients had unexpected electronic errors that precluded clinician’s unstructured assessment of PTP. §Using standard cutoffs.

Table 1. 
Baseline Characteristics of Enrolled Patients
 Entire CohortQuantitative D-dimer Available
DoneNot DoneQuantitative D-dimer Not Available
  1. PE = pulmonary embolism.

Enrolled7,9404,3673,029544
Age, years (mean ±SD)49 ± 1748 ± 1751 ± 1849 ± 18
Female, n (%)5,328 (67)2,937 (67)2,006 (66)385 (70.8)
Race/ethnicity, n (%)
 White4,541 (57)2,507 (57)1,722 (57)312 (57)
 African American2,704 (34)1,478 (34)1,081 (35)145 (27)
 Hispanic482 (6)238 (5)170 (6)74 (14)
 Asian74 (1)53 (1)17 (1)4 (1%)
 Native American8 (<1)6 (<1)2 (<1)0
 Other131 (2)85 (2)37 (1)9 (2)
Primary insurance, n (%)
 Private3,482 (44)2,063 (47)1,137 (38)282 (52)
 Medicare1,128 (14)548 (13)453 (15)127 (23)
 Medicaid906 (11)459 (11)377 (12)70 (13)
 Self-pay693 (9)376 (9)256 (9)61 (11)
 Military68 (1)36 (1)32 (1)0
 Unknown1,528 (19)755 (17)769 (25)4 (1)
Medical history, n (%)
 Malignancy489 (6)161 (4)302 (10)26 (5)
 PE858 (11)378 (9)416 (14)64 (12)
 Surgery (within 1 month)520 (7)193 (4)294 (10)33 (6)

We compared patients who had a D-dimer ordered to patients who had pulmonary vascular imaging ordered as the first test in centers where a quantitative D-dimer was available. Patients with a D-dimer done had slightly lower mean age (48 years vs. 51 years, p < 0.0001), were similarly likely to be female (67% vs. 66%, p = 0.35), and had similar racial background, although the overall distribution was slightly different across groups (57% vs. 57% white; 34% vs. 35% African American; 5% vs. 6% Hispanic; p = 0.006). The median Wells score was similar in patients undergoing D-dimer testing when compared with patients who did not (1.5 [IQR 0.0–2.5] vs. 1.5 [IQR 0.0–3.0]). Patients who underwent D-dimer testing were less likely to have had prior PE (9% vs. 14%, p < 0.0001), a history of cancer (4% vs. 10%, p < 0.0001), or recent surgery (4% vs. 10%, p < 0.0001). Compared with patients who had pulmonary vascular imaging without a preceding D-dimer, patients who underwent D-dimer testing were less likely to be diagnosed with venous thromboembolism (VTE; 5% vs. 9%, p < 0.0001).

Overall, the imaging test most frequently performed to evaluate for PE was CT angiography, which was performed in 4,237 (53%) patients. Venous ultrasound was performed in 987 (12%) patients, and V/Q scanning was performed for 468 (6%) patients. Conventional pulmonary angiography was performed in 180 (2.2%) enrolled patients. Among 4,357 patients who had quantitative D-dimer testing, CT angiography was performed in 2,068 (47%), venous ultrasound was performed in 458 (11%), V/Q scanning was performed in 210 (5%), and conventional pulmonary angiography was performed in 47 (1%).

In general, unstructured PTP assessment and the Wells score categorized patients as low, intermediate, and high PTP in similar proportions (Table 2). Within these groups, the outcome rates of PE were similar. In both low and intermediate PTP groups, the prevalence of PE was ≤13%.

Table 2. 
Prevalence of PE According to PTP
PTP*Patients in CategoryPatients with PE
  1. All values expressed as n (%).

  2. PE = pulmonary embolism; PTP = pretest probability.

  3. *Assessed using unstructured clinical estimates (for percentage values) or using the system described by Wells et al.18

Low
 <15%5357 (68)167 (3)
 Wells score <25482 (69)173 (3)
Intermediate
 15%–40%2087 (26)215 (10)
 Wells score 2–62201 (28)280 (13)
High
 >40%488 (6)163 (33)
 Wells score >6257 (3)92 (36)

Among patients who had a quantitative D-dimer performed, 191 (4%) patients had an unstructured clinician estimate of PTP > 40%, and 58 (30%) of these had PE (Table 3). The D-dimer had a sensitivity of 93% and NPV of 93%. Ninety-five (2%) patients had a Wells score of >6, and 32 (34%) of these patients had PE. The D-dimer had a sensitivity of 94% and NPV of 91%. However, in both subgroups of high PTP patients, CIs were wide due to small sample size. There were two deaths among patients with negative D-dimer results (using standard cutoffs). Neither of these patients had PE. These patients were both categorized as intermediate PTP by unstructured estimate and both had Wells scores between 2 and 6.

Table 3. 
Test Characteristics of Quantitative D-dimer using Standard and Variable Cutoffs
PTPSensitivity, % (95% CI)Specificity, % (95% CI) LR(–)Prevalence of PE,*%NPV, % (95% CI)
  1. LR(–) = negative likelihood ratio, defined as (1 – sensitivity)/(specificity); NPV = negative predictive value; PE = pulmonary embolism.

  2. *Prevalence of PE in patients who underwent quantitative D-dimer testing.

  3. †Standard cutoff defined as: D-dimer considered negative if <500 ng/mL for VIDAS and Liatest and <1.6 μg/mL for Advanced D-dimer.

  4. ‡D-dimer cutoff at twice standard value when PTP < 15%, standard value when PTP 15%–40%, and twice standard value when PTP > 40%

  5. §D-dimer cutoff at twice standard value.

  6. ||D-dimer cutoff at standard value.

  7. ¶D-dimer cutoff at half standard value.

  8. **D-dimer cutoff at twice standard value when Wells score <2, standard value when Wells score 2–6, and twice standard value when Wells score >6.

Unstructured PTP Score
All, n = 4,349   5.4 
 Standard cutoff†94 (91, 97)58 (56, 60)0.10 (0.06–0.16) 99.5 (99.1, 99.7)
 Variable cutoff‡88 (83, 92)75 (74, 76)0.17 (0.12–0.23) 99.1 (98.7, 99.4)
<15%, n = 3,224   2.7 
 Standard cutoff†94 (87, 98)63 (61, 65)0.09 (0.04–0.22) 99.7 (99.4, 99.9)
 Variable cutoff§72 (61, 81)86 (85, 88)0.32 (0.23–0.45) 99.1 (98.7, 99.4)
15%–40%, n = 934   9.7 
 Standard cutoff†96 (89, 99)42 (39, 46)0.11 (0.04–0.28) 98.9 (97.2, 99.7)
 Variable cutoff||96 (89, 99)42 (39, 46)0.11 (0.04–0.28) 98.9 (97.2, 99.7)
>40%, n = 191   30.4 
 Standard cutoff†93 (83, 98)38 (29, 46)0.18 (0.07–0.48) 92.6 (82.1, 97.9)
 Variable cutoff¶98 (91, 100)13 (8, 20)0.14 (0.02–0.99) 94.4 (72.7, 99.9)
Wells Score
All, n = 4,357   5.4 
 Standard cutoff†94 (91, 97)58 (56, 60)0.10 (0.06–0.16) 99.5 (99.1, 99.7)
 Variable cutoff**85 (80, 89)75 (74, 77)0.20 (0.15–0.27) 98.9 (98.5, 99.2)
<2, n = 3,228   2.9 
 Standard cutoff†93 (85, 97)63 (61, 65)0.12 (0.06–0.24) 99.6 (99.3, 99.9)
 Variable cutoff§68 (58, 78)86 (85, 87)0.37 (0.27–0.49) 98.9 (98.4, 99.3)
2–6, n = 1,034   10.4 
 Standard cutoff†96 (91, 99)43 (40, 46)0.09 (0.03–0.23) 99.0 (97.5, 99.7)
 Variable cutoff||96 (91, 99)43 (40, 46)0.09 (0.03–0.23) 99.0 (97.5, 99.7)
>6, n = 95   33.4 
 Standard cutoff†94 (79, 99)33 (22, 46)0.19 (0.05–0.75) 91.3 (72.0, 98.9)
 Variable cutoff¶97 (84, 100)14 (7, 25)0.22 (0.03–1.65) 90.0 (55.5, 99.7)

Using the standard cutoffs across all levels of PTP, the quantitative D-dimer had a sensitivity of 94%, specificity of 58%, and NPV of 99.5% (Table 3). In a hypothetical analysis, had the proposed variable D-dimer cutoffs been used, the overall sensitivity would have decreased to 88% (using PTP) and 85% (using Wells), but the specificity would have increased to 75% (using either Wells or PTP). As would be expected in each PTP subgroup, when variable cutoffs resulted in decreased sensitivity, there was an associated increase in specificity (and vice versa). The hypothetical decrease in sensitivity was most significant for patients with low PTP, such that using the D-dimer at twice threshold would have yielded a sensitivity of 72% in patients with unstructured PTP of <15% and 68% in patients with a Wells score of <2. Simultaneously, however, the specificity would have increased to 86%, for unstructured PTP or Wells score. When coupled with the 3% prevalence of PE in this subgroup, the NPV would remain high (99%), with the lower limit of the 95% CI if >98%.

We performed subgroup analyses to assess whether this strategy could be applied to patients in whom D-dimer testing tends to be less useful due to low specificity: patients who have malignancy, are older, or are pregnant. In 161 patients with cancer, standard and variable cutoffs had similar sensitivity (92% vs. 92%) and NPV (95% vs. 96%), but the specificity using variable cutoffs was 39% compared to 30% using standard cutoffs. In 972 patients more than 60 years old, variable cutoffs resulted in sensitivity of only 86%, compared to 97% using standard cutoffs, although NPV remained high for both (98% vs. 99%). Using variable cutoffs, specificity was 56%, compared to only 36% using standard cutoffs. In 75 pregnant patients, sensitivity (86% vs. 86%) and NPV (100% vs. 100%) were again similar, but variable cutoffs yielded specificity of 53%, compared to only 24% with standard cutoffs.

In Table 4, we combined raw numbers from 2 × 2 tables to estimate the potential effect of changing to a variable D-dimer cutoff on testing. We undertook this analysis with these assumptions: that no patient with a negative D-dimer undergoes diagnostic imaging, all patients with a positive D-dimer undergo diagnostic imaging, and imaging results in no false-positive or false-negative diagnoses. In this scenario, we estimate that using variable cutoffs would result in an additional 12–18 false-negative D-dimers, or “missed PE” diagnoses (depending on whether the Wells score or unstructured PTP estimate is used). This corresponds to 2.7–4.1 per 1,000 patients evaluated for PE. The vast majority of these false-negative D-dimers would occur in low-PTP patients. We combined the total number of negative D-dimers to estimate the number of imaging tests that could be avoided and determined that using a variable cutoff would obviate 662–690 imaging tests, or 152–158 per 1,000 patients evaluated for PE. Thus, using our assumptions, variable cutoff would obviate 38–55 imaging tests for each missed PE diagnosis.

Table 4. 
Estimated Number of False and True Negatives Using Standard and Variable D-dimer Cutoffs
 Standard CutoffVariable Cutoff
False NegativesTrue NegativesTotal NegativesFalse NegativesTrue NegativesTotal Negatives
  1. Estimates are based on raw numbers from 2 × 2 tables, assuming no additional workup after a negative D-dimer, complete diagnostic imaging after a positive D-dimer, and no false-negative or false-positive imaging results.

  2. PTP = pretest probability.

Wells score
 <282,0042,012302,7012,731
 2–674104174399403
 >6223251910
 Total172,4372,454353,1093,144
PTP
 <15%52,0122,017242,7112,735
 15%–40%63683744355359
 >40%6535911718
 Total172,4332,450293,0833,112

Discussion

The current analysis was undertaken to determine the current standard of practice regarding the use of the quantitative D-dimer in a large representative sample of ED patients and to determine if D-dimer efficiency could be safely increased by varying the numeric cutoff defining a negative test according to PTP.

We submit that the present standard of care supports using a negative D-dimer to rule out PE in patients with low PTP. Our analysis confirms the low prevalence (3%) of PE in patients with low PTP evaluated in the United States. Previous analyses of European data suggest that the sensitivity of the quantitative D-dimer may be sufficiently high to rule out PE in patients with intermediate PTP as well.8–10 In our sample, we found that in patients with intermediate PTP, the prevalence of PE was <13%. Thus, conservatively assuming that the quantitative D-dimer in use at a given institution has a negative likelihood ratio of 0.13 (consistent with our study and prior meta-analyses),1,2,4 our analysis implies that a negative quantitative D-dimer can exclude PE with a posttest probability of <2% in intermediate PTP patients. This result compares favorably to other criterion standards used to exclude PE, including the Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED-II) study20 (low clinical probability and negative CT angiogram: NPV = 96%). To our knowledge, this is the largest prospective study to date that has examined actual clinical practice and shown that D-dimer is often used to rule out PE in intermediate PTP patients.

Based upon our previous work, we were not surprised that low and intermediate PTP patients comprised the vast majority of all patients tested for PE, regardless of the first test ordered (i.e., D-dimer or CT angiography).21 Clinicians did order a D-dimer on approximately 40% of patients with a high PTP, resulting in about 3%–6% of all patients with a D-dimer ordered having a high PTP. Our data suggest that the D-dimer at standard cutoff yielded an unacceptably low NPV (approximately 90%–92%), and this was not significantly improved at the half-standard cutoff. Taken together, our data do not support the use of D-dimer at any threshold to rule out PE in high-PTP patients.

A key objective of this work was to assess the concept proposed in a previous work that showed that varying the positive/negative cutoff of the quantitative D-dimer according to the patient’s PTP increased specificity with little decrease in sensitivity or NPV.11,12 We believed that the cutoff Linkins et al. proposed for low PTP patients (more than four times the standard cutoff, based on ROC curve analysis) would represent a sharp change from current practice, so we decided by consensus to test the effect of a simple doubling the threshold low PTP patients. This adjustment would have allowed the NPV of the quantitative D-dimer to remain 99%. It must be emphasized that the hypothetical increase in D-dimer cutoff for low-PTP patients would have substantially decreased the test’s sensitivity and that the observed high overall NPV is dependent on a low prevalence of disease in low-PTP patients. We estimate that using a variable cutoff would result in approximately 3–4 additional missed PEs per 1,000 patients, although it would have afforded a 15% decrease in the rate of pulmonary vascular imaging.

Limitations

The results of this study must be interpreted within the context of its design. This was a large, multicenter study, performed in academic and community centers, resulting in a heterogeneous population with a variety of D-dimer assays used, although all were quantitative and automated in format. We did not explore differences between specific assays. Our study was observational and noninterventional, such that we believe the results represent the real world and probably should not be compared or contrasted to studies that purport to follow a rigid study protocol. The diagnostic criterion standard for this study was clinically significant PE (or DVT) within 45 days of the index visit that was detected by standard care processes. We did not have resources to perform radiologic testing to monitor asymptomatic patients for VTE. It remains possible that a few patients had a PE or DVT and went undiagnosed during the follow-up period, and these patients were incorrectly classified as true-negatives.

Our observational design may also have engendered some selection bias. We attempted to minimize this issue by enrolling either from a random sample of ED shifts or consecutively and by collecting PTP data contemporaneously with the ED workup. There were small differences in the baseline characteristics of patients who underwent D-dimer testing and the overall study population. The rate of PE was also lower in patients undergoing D-dimer testing than the overall study population. This may be because high-PTP patients who went directly to imaging have a higher prevalence of PE than those for whom the clinician sent a D-dimer. Clinicians may have also avoided D-dimer testing on patients known to have conditions that elevate the D-dimer (e.g., malignancy). However, the data we present describing the prevalence of PE across levels of PTP apply to all patients enrolled, including those who did not undergo D-dimer testing. Furthermore, our PTP estimates were collected prospectively, so while there may have been selection bias in terms of which patients underwent D-dimer testing, the relationship between PTP and D-dimer results should still be accurate.

Our population of ED patients had a lower prevalence of PE than some previously published studies. While we feel that this percentage is consistent with clinical practice in U.S. EDs, our results may not translate to other countries and would likely not apply to inpatient populations. Despite the fact that we enrolled a large number of patients, we found relatively few patients with high PTP. This is consistent with observations from our prior work.21 Unfortunately, the relatively small number of high PTP patients limited our ability to confidently assess the test characteristics of the quantitative D-dimer in high-PTP patients. A substantially larger data set will be required to accumulate an unbiased sample of high-PTP patients sufficiently large to have narrow CIs. Finally, we did not assess whether clinicians would actually be willing to use variable cutoffs in their practice. Studying the implementation of variable cutoffs, or a study that allows clinicians to choose cutoffs based on their own comfort level, may be worthwhile.

Conclusions

We found a sufficiently low prevalence of PE in a large, multicenter sample of ED patients with both low and intermediate PTP to conclude that all patients with a PTP less than high are appropriate for quantitative D-dimer testing. Varying the D-dimer’s positive/negative cutoff according to the patient’s PTP could decrease sensitivity, increase specificity, and maintain a NPV of ≥99%. Future work is needed to better quantify the potential health risks and benefits that would occur if different D-dimer thresholds were adopted in a management protocol.

Ancillary