Bedside investigation of pulmonary embolism (PE) diagnosis predictors. A, Seven-variable clinical model (≤4.0 points, negative; and >4.0 points, positive). B, D-dimer assay (negative or positive). C, Alveolar dead-space fraction (≤0.15, negative; and >0.15, positive). If at least 2 predictors are negative, PE is excluded; if at least 2 predictors are positive, PE is possible and further investigation is required. DVT indicates deep vein thrombosis; solid circle, blood clot; and PETCO2, end-tidal PCO2 level.
Study algorithm. Bedside investigation of pulmonary embolism diagnosis (BIOPED) negative represents at least 2 negative results on 3 bedside tests; BIOPED positive represents at least 2 positive results on 3 bedside tests. Treatment failure represents patients taking anticoagulant agents after initial investigation for suspected pulmonary embolism (PE) who had venous thromboembolism (VTE) during follow-up. V/Q indicates ventilation-perfusion.
Study participant flow chart. BIOPED indicates bedside investigation of pulmonary embolism diagnosis; V/Q, ventilation-perfusion; and VTE, venous thromboembolism.
Rodger MA, Bredeson CN, Jones G, Rasuli P, Raymond F, Clement AM, Karovitch A, Brunette H, Makropoulos D, Reardon M, Stiell I, Nair R, Wells PS. The Bedside Investigation of Pulmonary Embolism Diagnosis StudyA Double-blind Randomized Controlled Trial Comparing Combinations of 3 Bedside Tests vs Ventilation-Perfusion Scan for the Initial Investigation of Suspected Pulmonary Embolism. Arch Intern Med. 2006;166(2):181-187. doi:10.1001/archinte.166.2.181
We sought to determine whether using combinations of 3 bedside tests (7-variable clinical model, non–enzyme-linked immunosorbent assay D-dimer test, and alveolar dead-space fraction) to exclude pulmonary embolism (PE) before diagnostic imaging was as safe as a standard strategy of starting with ventilation-perfusion (V/Q) scan.
In this double-blind, randomized, controlled equivalency trial, patients were randomized to initial bedside tests or to initial V/Q scan without bedside tests. Patients assigned to the bedside test group had a sham V/Q scan performed if at least 2 of 3 bedside test results were negative; otherwise, they underwent an actual V/Q scan. Further diagnostic management was determined by a blinded physician after V/Q scan. The primary outcome measure was recurrent venous thromboembolic events during 3 months among patients who were not taking anticoagulant agents after the initial investigations were completed.
Four hundred fifty-eight consecutive adults with suspected PE were eligible for the study; 398 of 399 consenting and randomized patients completed the study. The follow-up venous thromboembolic event rate was 2.4% in the bedside test group vs 3.0% in the V/Q scan group (P = .76). Pulmonary embolism was excluded in 34% (67/199) of the bedside test group patients with at least 2 negative results on 3 bedside tests vs 18% (35/199) excluded using only the 7-variable clinical model and the D-dimer test.
Excluding PE with at least 2 negative results on 3 bedside tests safely eliminates the need for diagnostic imaging in 34% of patients with suspected PE.
Pulmonary embolism (PE) is a common potentially lethal yet treatable condition.1- 3 Pulmonary embolism is considered in the differential diagnosis of many clinical presentations, yet fewer than 35% of patients suspected of having PE actually have PE.4,5 Diagnostic imaging for suspected PE is generally available during limited hours and only in larger centers, complicating the diagnostic approach for many patients with suspected PE. Ventilation-perfusion (V/Q) scan is often recommended for the investigation of suspected PE, but most V/Q scans for suspected PE are nondiagnostic. Options after a nondiagnostic V/Q scan include pulmonary angiography, serial noninvasive leg studies for deep vein thrombosis, and helical computed tomography. Each of these options has limitations, including cost, inaccessibility, inconvenience, and potential complications.
Recent investigations have focused on using bedside tools to exclude PE without diagnostic imaging, as well as on refining the diagnostic management algorithm to minimize the need for subsequent imaging after V/Q scan.5- 10 It has been previously demonstrated that, by using a simple 7-variable clinical model, patients with suspected PE can be categorized into pretest probability subgroups (low, intermediate, or high probability of PE and unlikely or likely PE) (Figure 1).5- 7 A low pretest probability (based on the 7-variable clinical model) and a negative non–enzyme-linked immunosorbent assay D-dimer test result safely exclude PE without the need for diagnostic imaging.5 Alveolar dead-space fraction (AVDSf) represents ventilation of those alveoli that are not involved in gas exchange (ie, alveoli that are not perfused or are poorly perfused, as occurs in PE).8 It has recently been demonstrated that a low AVDSf combined with a negative D-dimer test result in ambulatory patients with suspected PE has a high negative-predictive value.8,9
We evaluated combinations of the 7-variable clinical model by Wells et al,10 non–enzyme-linked immunosorbent assay D-dimer test, and AVDSf in a single bedside investigation of pulmonary embolism diagnosis (BIOPED method) (with at least 2 negative results on 3 bedside tests excluding PE). Compared with using 2 of the predictors alone, we sought to determine whether the BIOPED method would be safer, was less resource intensive, and could exclude a larger proportion of patients with suspected PE.
The study was a double-blind, randomized, controlled equivalency trial comparing the BIOPED method with V/Q scan as the initial investigation in patients with suspected PE. The study was conducted from October 1, 1998, to February 1, 2002, at Ottawa Hospital, Ottawa, Ontario, a large academic tertiary care center serving a local population of more than 1 million, and was approved by the institutional ethics review board.
Consecutive patients with suspected PE who were referred to our nuclear medicine department at Ottawa Hospital (General Campus [October 1, 1998, to February 1, 2002] and Civic Campus [February 15, 2000, to July 31, 2001]) for V/Q scan were eligible for the study. Patients were ineligible if they were younger than 18 years, were unable to give informed consent, were expected to survive less than 3 months, were receiving ventilatory assistance, were previously enrolled in the BIOPED study, were known to have chronic PE or had PE diagnosed in the last 3 months, were taking full-dose anticoagulant agents, or had vena caval interruption for confirmed venous thromboembolism (VTE).
After written informed consent was obtained for each patient, the next consecutively numbered sealed opaque envelope with a randomly assigned allocation (BIOPED group or V/Q scan group) was opened. Patients were managed as shown in Figure 2. All study patients, regardless of randomization allocation, had 3 BIOPED tests performed, the results of which were kept blinded from the investigators. All study patients were evaluated by their referring physicians using a standardized patient assessment form that included 7 variables in the clinical model by Wells et al.10 All patients had volumetric capnograms (continuous carbon dioxide vs volume tracings) measured at bedside by respiratory therapists, with the patient breathing through a mouthpiece attached to an airway adapter with a mainstream carbon dioxide volume sensor (CosmoPlus; Respironics Novametrix, Wallingford, Conn). Calibration was verified before each use with gases of known carbon dioxide concentration. Once the patient was breathing at a stable respiratory rate (±2 breaths per minute during 2 minutes) and had a stable end-tidal PCO2 (±1 mm Hg during 2 minutes), a blood sample was obtained for arterial blood gas analysis. The AVDSf was subsequently calculated as follows: AVDSf = (PaCO2 − end-tidal PCO2)/PaCO2.8 The AVDSf measurements were reported to unblinded study personnel as being 0.15 or lower or being higher than 0.15 but were not disclosed to the patients or to their physicians. If AVDSf measurements could not be obtained (as was the case in 133 of 399 patients), they were assumed to be positive (primary reasons why the AVDSf could not be obtained were an inability to achieve a stable respiratory rate or a stable end-tidal PCO2 [n = 64] and an inability to obtain an arterial blood gas level [n = 25]). All study patients had D-dimer analysis using the SimpliRED whole-blood agglutination D-dimer test (AGEN Biomedical, Ltd, Brisbane, Australia) or the Accuclot latex agglutination D-dimer test (Sigma Diagnostics, St Louis, Mo) if the SimpliRED test was unavailable (n = 51). These assays have been demonstrated to have similar accuracy.11 Among 12 patients in whom results were missing, the D-dimer test results were assumed to be positive.
Based on the results of the 7-variable clinical model, D-dimer test, and AVDSf, unblinded study personnel assigned all patients a BIOPED diagnosis. A BIOPED-negative diagnosis was assigned if at least 2 of 3 bedside test results were negative (7-variable clinical model score ≤4.0 points, negative D-dimer assay, and AVDSf ≤0.15). A BIOPED-positive diagnosis was assigned if at least 2 of 3 bedside test results were positive (7-variable clinical model score >4.0 points, positive D-dimer assay, and AVDSf >0.15).
Patients randomized to the BIOPED group of the study who had BIOPED-negative results (in whom PE was considered excluded) received a sham V/Q scan in the nuclear medicine department, which was subsequently reported as showing no evidence of PE. Patients randomized to the BIOPED group who had BIOPED-positive results (in whom PE was considered possible) underwent an actual V/Q scan. Patients randomized to the V/Q scan group (BIOPED-negative or BIOPED-positive result) underwent an actual V/Q scan.
Results of the V/Q scans were reported according to Prospective Investigation of Pulmonary Embolism Diagnosis study4 criteria, except for normal or near-normal results, which werereported as showing no evidence of PE. After the nuclear medicine department reports were released (indicating no evidence of PE or low, intermediate, or high probability of PE), the blinded treating physician determined whether further testing was required for suspected PE. Study personnel documented any further testing that was completed and whether patients were taking anticoagulant agents at the time of discharge.
At discharge, patients were given written instructions to contact the blinded study physicians if they experienced any symptoms of recurrent VTE. All patients reporting leg symptoms were evaluated for deep vein thrombosis by compression ultrasonography or by venography. All patients with suspected PE recurrence were investigated by V/Q scan. Pulmonary embolism was considered excluded if the V/Q scan was normal. However, all patients with nonnormal V/Q scans were further tested by pulmonary angiography or by spiral computed tomography. Patients were seen 3 months after discharge to document any recurrent events or bleeding episodes that were previously unreported and to complete an assessment of patient and study physician blinding.
The primary outcome measure was recurrent VTE events during 3 months in patients not taking anticoagulant agents or vena caval interruption after initial investigations for suspected PE. The following criteria were used to diagnose a VTE event (PE or deep vein thrombosis) on follow-up: (1) compression ultrasonography revealing new noncompressibility above the trifurcation of the popliteal vein, (2) venography demonstrating a new constant intraluminal filling defect above the trifurcation of the popliteal vein, (3) pulmonary angiography demonstrating a constant intraluminal filling defect or a cutoff of a vessel greater than 2.5 mm in diameter, (4) spiral computed tomography demonstrating an intraluminal filling defect in a segmental or greater-sized pulmonary artery, and (5) PE discovered at autopsy.
Secondary outcome measures were the following: (1) all-cause mortality, (2) major bleeding episodes during the 3 months, and (3) number of diagnostic tests used in the investigation for PE. Major bleeding complications were defined as any overt hemorrhage resulting in a decrease in hemoglobin level greater than 20 g/L or requiring transfusion of at least 2 U of packed red blood cells. All suspected outcome events were adjudicated by 2 blinded physicians (M.A.R. and P.S.W.).
All analyses were based on intent to test by 2 of us (M.A.R. and R.N.). A priori, we believed that the minimum clinically important difference (MCID) in the primary outcome measure (3-month recurrent VTE rate) was 3%, accepting a 10% chance of falsely rejecting the null hypothesis (ie, 1-tailed α = .10) and a 10% chance of falsely not rejecting the hypothesis (ie, 90% power); therefore, we would require a sample size of 570. Because of funding limitations, the trial was terminated early after having recruited 399 patients. The null hypothesis was tested by examining the 95% confidence interval surrounding the difference in event rates between the 2 study groups among patients who were not taking anticoagulant agents. Given an MCID of 3% if the 95% confidence interval around the difference in event rates excluded a 3% difference, we would conclude equivalence. Diagnostic accuracy of the bedside tests was determined by calculating sensitivity, specificity, negative-predictive value, positive-predictive value, and likelihood ratios for each bedside test individually and in combinations of 2 and 3. For the diagnostic accuracy analyses, 2 × 2 tables were constructed using all study patients (regardless of randomization group). This was made possible by performing all 3 bedside tests in all study patients regardless of randomization allocation (Figure 2). Patients were considered PE negative if they were not taking anticoagulant agents after initial testing for PE and had no confirmed VTE at the 3-month follow-up. Patients were considered PE positive if they were not taking anticoagulant agents after initial testing for PE or had a confirmed VTE event at the 3-month follow-up. t Test was used to compare the mean number of tests performed after V/Q scan in each study group.
Eight hundred twenty-four patients were screened, 458 of whom were eligible for the study (Figure 3). The most common reasons for ineligibility were an inability to provide informed consent (30%), current anticoagulation therapy for proven deep vein thrombosis (25%), and life expectancy less than 3 months (15%). Three hundred ninety-nine patients consented to participate, 1 patient withdrew consent (who preferred not to wait for investigation of suspected PE), and no patients were lost to follow-up. Baseline characteristics were comparable between the 2 study groups except that patients randomized to the BIOPED group were more likely to have a malignant neoplasm but were less likely to be tachypneic (Table 1).
In the primary outcome analysis, 4 (2.4%) of 165 patients not taking anticoagulant agents in the BIOPED group had a VTE event during the 3-month follow-up period vs 5 (3.0%) of 169 patients not taking anticoagulant agents in the V/Q scan group who had a VTE event during the 3-month follow-up period (P=.76). The 95% confidence interval around the −0.6% difference in favor of the BIOPED group was −4.1% to 2.9% and excluded our a priori assigned MCID of 3% (Table 2). Of 199 patients randomized to the BIOPED group, 67 had at least 2 negative results on 3 bedside tests (which excluded PE in 34% of patients). Only 18% (35/199) would have been excluded using just the 7-variable clinical model and the D-dimer test.
Fifty-seven VTE events were confirmed during the initial workups of suspected PE (ie, the index visits) (Table 3). Of 65 patients taking anticoagulant agents at the time of their index visits, 57 had confirmed PE and 6 had probable PE; the latter 6 had V/Q scans indicating an intermediate probability of PE, with 5 of them showing normal results on subsequent ultrasonography. The remaining 2 patients were taking anticoagulant agents for atrial fibrillation after their index visits.
Adding AVDSf to the 7-variable clinical model and the D-dimer test in the BIOPED method more than doubled the proportion of patients excluded from 16% to 32%, with comparable negative-predictive values (95.4% and 94.7%) (Table 4). No significant difference was noted in the results of the BIOPED method using the SimpliRED D-dimer test (sensitivity, 89.8%; and specificity, 40.1%) vs the Accuclot D-dimer test (sensitivity, 88.9%; and specificity, 23.8%).
Six patients experienced bleeding episodes during the 3-month follow-up period; 3 were adjudicated as major bleeding episodes (2 in the BIOPED group and 1 in the V/Q scan group) and 3 as minor bleeding episodes (2 in the BIOPED group and 1 in the V/Q scan group). There was no statistically significant difference in total bleeding episodes, major bleeding episodes, minor bleeding episodes, or all-cause mortality between the 2 study groups (P = .45). The mean number of diagnostic imaging tests performed to investigate suspected PE was 1.36 per patient in the BIOPED group vs 1.90 per patient in the V/Q scan group (P<.001) (Table 3).
This study establishes that using a combination of 3 bedside tests to exclude PE is as safe as using a V/Q scan as the initial investigation for patients with suspected PE. Using 2 of 3 bedside predictors of PE resulted in a rate of subsequent VTE during 3 months that was comparable to that associated with a standard approach of initial investigation with a V/Q scan (without the use of bedside tests).
The addition of AVDSf to the 7-variable clinical model and the D-dimer test doubled the number of patients in whom PE could be safely excluded compared with the use of only the 7-variable clinical model and the D-dimer test. This doubling was achieved despite the fact that AVDSf measurements were indeterminate in one third of patients tested. Efforts to increase the proportion of patients with conclusive AVDSf results, such as using less stringent criteria for stability of breathing at the time of AVDSf measurement, will need to be evaluated. Hyperventilation at the time of arterial blood gas measurement can lower the PaCO2 level and falsely normalize AVDSf results in patients with PE. Furthermore, we were obligated for patient safety to assume that a bedside test result was positive if it was unavailable. This may have biased the results in favor of finding a higher sensitivity and a lower specificity for AVDSf. Six of 8 patients who had VTEs on follow-up had high AVDSf measurements at baseline (Table 2). This raises the possibility that AVDSf not only is useful in the diagnostic management of PE but also is a predictor for risk of development of subsequent symptomatic PE.
On average, the use of initial bedside tests resulted in 0.6 fewer tests being conducted per patient compared with the standard approach. Given the bedside nature of these tests, this approach can be adopted in smaller facilities without access to diagnostic imaging rather than immediately transferring patients with suspected PE to larger centers for diagnostic imaging.
To our knowledge, our study is the first randomized trial comparing 2 diagnostic strategies for suspected PE. Randomized trials of competing diagnostic strategies eliminate selection bias that may be seen in cohort studies examining new diagnostic tests. If a new diagnostic strategy is compared with the gold standard in a low-risk population, high negative-predictive values can be demonstrated by virtue of the low prevalence of disease in the population studied (eg, ambulatory outpatients in PE diagnostic studies). Examining the same diagnostic strategy against the standard diagnostic strategy in a randomized trial permits a direct comparison of 2 competing strategies without selection bias.
Our study used clinical outcome as a primary outcome measure, which is the ultimate test of a diagnostic intervention. We demonstrated no significant difference in follow-up VTE event rates among patients initially tested using our bedside strategy vs patients tested with standard V/Q scan. The 3-month VTE event rate in the BIOPED group was 2.4%, which was 0.6% lower than that in the V/Q scan group, with a 95% confidence interval that excluded our MCID of 3%. Some may argue that a 3% MCID is too high. However, if we assume that 15% of recurrences are fatal, this 3% difference in recurrence rates translates into a mortality difference of 0.45% at 3 months,12 which is comparable to the 0.4% mortality risk that was associated with the use of pulmonary angiography to diagnose PE among 1400 subjects.13 Furthermore, the MCID used in our study was also used in a recent large trial of the treatment of patients with VTE with low-molecular-weight heparin sodium.12
The accuracy analysis demonstrated a negative-predictive value (≤95%) that was lower than expected with the combinations of the 3 bedside tests. Had we designed a prospective cohort study with initial and follow-up VTE as the clinical outcome measure, we may have been reluctant to conclude that the combinations of bedside tests were safe enough to exclude PE. This highlights the importance of conducting randomized controlled trials of competing diagnostic strategies to eliminate the risk of false conclusions caused by selection bias (eg, among patient populations with varying risks of VTE recurrence). The lower than expected negative-predictive value was likely a consequence of the high risk of recurrence in our study population (2.7% during 3 months among patients not taking anticoagulant agents). For example, our study population had higher proportions of hospitalized patients, patients with malignant neoplasms, and patients with PE compared with those in another recent investigation of PE diagnostic management.14
A diagnostic management strategy using at least 2 negative results on 3 bedside tests (7-variable clinical model, D-dimer test, and AVDSf) to exclude PE before V/Q scan is as safe as using initial V/Q scan among patients with suspected PE. The addition of AVDSf to the other 2 bedside tests doubles the proportion of patients who can be safely excluded at the bedside and eliminates the need for diagnostic imaging in 34% of patients with suspected PE.
Correspondence: Marc A. Rodger, MD, MSc, Ottawa Hospital, General Campus, 1812-E Box 201, 501 Smyth Rd, Ottawa, Ontario, Canada K1H 8L6.
Accepted for Publication: June 20, 2004.
Financial Disclosure: None.
Funding/Support: This study was funded by grant 4508 from the Heart and Stroke Foundation of Canada, Ottawa. Dr Rodger is the recipient of a new investigator award from the Heart and Stroke Foundation of Ontario, Ottawa. Dr Wells is the recipient of a research chair from Canada Research Chairs, Ottawa. Dr Stiell is the recipient of a Distinguished Investigator award from the Canadian Institutes of Health Research, Ottawa.
Role of the Sponsor: The Heart and Stroke Foundation of Canada and the Canadian Institutes of Health Research had no role in the design and conduct of the study; in the collection, management, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
Acknowledgment: We are grateful to the many busy clinicians who completed data collection forms; to the staff of the Nuclear Medicine Department for their cooperation; to Denise Blanchette, RRT, and the many respiratory therapists who participated and provided valuable input in the AVDSf analysis; to Jane Browing and Ted Tabor from Respironics Novametrix for valuable input and support with AVDSf measurement; to Bill Kerr, BA, and Greg Ralston, BA, for editorial support; to Julie Beck, BSc, for data collection, data entry, and data analysis; and to Michèle Willson for assistance in preparing the manuscript.