M. J. L. van Strijen, St Antonius Ziekenhuis, Department of Radiology, Postbus 2500, 3430 EM Nieuwegein, the Netherlands. Tel.: +31 30 609 9111; fax: +31 30 609 2561; e-mail: email@example.com
Summary. Background: Spiral computed tomography (CT) has emerged as a potentially conclusive diagnostic test to exclude pulmonary embolism (PE) in patients with non-high probability scintigraphy and is already widely used—sometimes as the sole primary diagnostic test in the diagnosis of suspected PE. Its true sensitivity and specificity has, however, not been evaluated previously in a large cohort of consecutive patients. Methods: In a multicenter prospective study 627 consecutive patients with clinically suspected PE were studied. Patients with normal perfusion scintigraphy were excluded from further analysis. Single-detector spiral CT scanning and ventilation scintigraphy were then performed in all patients to diagnose PE, while pulmonary angiography was performed as the gold standard. The only exceptions were those patients who had both a high-probability VQ scan and a CT scan positive for PE: these patients were considered to have PE and did not undergo additional pulmonary angiography. All imaging tests were read by independent expert panels. Results: Five hundred and seventeen patients were available for complete analysis. The prevalence of PE was 32%. Spiral CT correctly identified 88 of 128 patients with PE, and 92 of 109 patients without PE, for a sensitivity and specificity of 69%[95% confidence interval (CI) 63–75] and 84% (95% CI 80–89), respectively. The sensitivity of spiral CT was 86% (95% CI 80–92) for segmental or larger PE and 21% (95% CI 14–29) in the group of patients with subsegmental PE. Conclusion: The overall sensitivity of spiral CT for PE is too low to endorse its use as the sole test to exclude PE. This holds true even if one limits the discussion to patients with larger PE in segmental or larger pulmonary artery branches. We conclude that, in patients with clinically suspected PE and an abnormal perfusion scintigraphy, single-slice detector spiral CT is not sensitive enough to be used as the sole test to exclude PE.
Objective and rapid diagnosis of pulmonary embolism (PE) is mandatory because it is a potentially lethal disease . While anticoagulant treatment is highly effective in the prevention of recurrent PE [2,3], it should not be used unnecessarily, since this treatment carries a considerable risk of hemorrhage.
The clinical diagnosis of PE is not specific, because it has been shown that only 20–30% of patients presenting with signs and symptoms suggestive of PE have the disease confirmed by objective diagnostic tests . The currently available diagnostic tests either share this non-specificity or have practical problems . A normal perfusion lung scintigraphy virtually rules out PE [6,7], and a segmental or larger ventilation-perfusion mismatch has a high probability of 85–90% for PE . However, up to 50–60% of patients have non-diagnostic lung scans . In addition, ventilation scintigraphy, using the highly unstable radioactive 81mKrypton, is often not available on a daily basis. Pulmonary angiography (PA) is generally considered the gold standard diagnostic test [10–13] but it is often not used when indicated, probably because it is often perceived as a dangerous and cumbersome procedure. It may be associated with cardiac and pulmonary morbidity in 2% of patients [14,15].
Some years ago, single-detector helical computed tomography (CT) or spiral CT was promoted as a new and effective test in the diagnostic work-up of patients with suspected PE. Using spiral CT, a volumetric two-dimensional image of a patient is made using a rotating detector. Pulmonary emboli are visible as filling defects, which can be located centrally, mural or eccentric, and may totally or partially occlude a pulmonary artery. The reported sensitivity of single-detector spiral CT for the diagnosis of PE has ranged from 53 to 100% with specificities varying from 81 to 100%[16–19]. In two recent articles, however, the available literature has been critically reviewed and it was concluded that there currently is no large study available that has met all criteria for adequately assessing the sensitivity and specificity within this setting [20,21]. Consequently, the true accuracy of spiral CT remains unclear.
To address this issue, we performed a prospective cohort study in which the accuracy of single-detector spiral CT was evaluated as a part of a fixed diagnostic algorithm in consecutive patients with clinically suspected PE. Special emphasis was placed upon determining the sensitivity and specificity of spiral CT in all patients with abnormal perfusion scans as well as the sensitivity and specificity of spiral CT separately for PE in segmental or greater arteries as opposed to PE in subsegmental arteries.
Materials and methods
All consecutive in- and outpatients, seen because of clinically suspected PE in the six participating hospitals (four academic and two general community teaching hospitals) over a 1-year period, were eligible. The hospital size varies from 400 to 1000 beds (mean 840 beds), estimated emergency department visits on average 90 000 patients per year per hospital and an estimated maximum of 30 000 admissions per year per hospital. Inclusion criteria were clinical suspicion of PE as judged by the referring physician based on history and physical examination as well as written informed consent. Exclusion criteria were: anticoagulant treatment for more than 24 h, diagnostic tests for pulmonary emboli already performed, < 18 years of age, and possible pregnancy. Renal failure and allergy to contrast agents were not absolute contraindications for participation in the study. These patients are part of a normal population of patients suspected of PE. In patients with renal impairment use of contrast agents was possible with appropriate precautions. In rare cases of possible allergic reactions in the past an anaphylaxis preparation protocol was used prior to CT scanning. The initial tests had to be performed within 24 h of presentation of PE, and all diagnostic tests had to be completed within 48 h. The protocol was approved by the six local ethics review boards. During the study period the characteristics of a sample of consecutive patients who refused or were ineligible for study entry during a 2-month period were evaluated for comparison with study patients.
Perfusion scintigraphy was performed as the first test in all patients. Patients with a normal perfusion scintigram were considered not to have PE and were excluded from further analysis. Ventilation scintigraphy was obtained in patients with segmental or larger perfusion defects (see Image analysis; using Revised PIOPED criteria in conjunction with chest radiographs). All other patients did not undergo ventilation scintigraphy, as there was no possibility of these scans having a high-probability result. Subsequently, all patients with perfusion defects underwent spiral CT scanning. If a patient had a high-probability lung scan defect and spiral CT scan showed PE, the patient was considered to have PE and PA was not indicated. In the case of a high-probability lung scan, where the spiral CT scan did not show PE, additional PA had to be performed. If the result of a spiral CT scan after a non-high-probability lung scan had not been verified by PA, the patient was excluded from analysis. Since this study was performed in a clinical setting the result of the local reading of the perfusion scintigraphy or the combined result of perfusion-ventilation scintigraphy (V/Q scan) determined how patients proceeded to subsequent levels in the algorithm as indicated in Fig. 1, but the later performed central reading was used for analysis.
Afterwards the scintigraphic lung scans, the spiral CT scans and the PAs were reviewed by independent expert panels consisting of nuclear medicine specialists (lung scans) and radiologists (CT and PAs) blinded to the results of different tests and the started treatment. The scintigraphic scans were read by two readers in consensus. The CT scans and PAs were read by two independent radiologists. Discrepant readings were resolved by a third independent reader whenever necessary. The readers were unaware of the results of other diagnostic tests or the patients' clinical symptoms. The reference criteria for the presence of PE according to central reading were a positive PA or a high-probability ventilation-perfusion scan. The reference criterion for the absence of PE was a negative PA. To be included in the assessment of accuracy of spiral CT, patients had to have a V/Q scan that was interpreted as non-high probability in both the local and central reading and a PA or a high-probability V/Q scan result and a positive CT scan result.
Chest radiographs were made erect and in two directions (posterior-anterior and left lateral). If this was not possible an anterior-posterior radiograph in bed was made.
Perfusion scintigraphy was performed after intravenous administration of 0.03 mCi kg−1 Tc-99m-labeled macro-aggregated serum albumin (Mallinckrodt, Petten, the Netherlands). Perfusion images were acquired with a minimum of 200 000 counts per view in at least four views: anterior, posterior, left anterior oblique and right anterior oblique views. Ventilation studies were performed using 81mKrypton (Mallinckrodt) via a Rb-Kr generator (Cygne-Amersham, Eindhoven, The Netherlands) during a single breath. A minimum of 200 000 counts was required for views in the same four directions as the perfusion scintigraphies. Gamma-camera and collimator depended on local availability (Siemens Orbiter 3700 Digitrac, Siemens Medical Systems, Iselin, NJ, USA; ADAC Vertex and Argus, ADAC Laboratories Europe B.V., Maarsen, the Netherlands; Toshiba GCA501 and GCA9015a, Toshiba Medical Systems Division, Tokyo, Japan; Varicam, Elscint Inc., Hackensack, NJ, USA). Collimators used were medium-energy all purpose (MEAP), low-energy high-resolution (LEHR), medium-energy high-resolution (MEHR) and low-energy general purpose (LEGP).
The spiral CT scanners used in this study were all of the first generation, with a single detector and capable of scanning at least 16 cm (Siemens Somatom plus 4 and Somatom plus S, Siemens Medical Systems, Erlangen, Germany; Elscint, Elscint, Haifa, Israel; Philips SR7000 and SR8000, Philips Medical Systems, Best, the Netherlands). Scanning was performed using a 5-mm table speed and slice thickness. First the level of the aortic arch was determined on the scout view. Using this as a reference point the starting point was identified 16 cm caudally. The scanning process was then started from this position in a caudocephalad direction, 20 s after intravenous injection of 900 mg s−1 of iodine for 40 s, either by injection of 100 mL of non-ionic contrast agent with 350 mgI mL−1 iodine content (Iomeron 350; Bracco Byk Gulden, Konstanz, Germany) at an injection rate of 2.5 mL s−1 or by injection of 120 mL of non-ionic contrast agent 300 mgI mL−1 (Ultravist 300; Schering, Berlin, Germany) at an injection rate of 3.0 mL s−1. The scan was performed during a single breath hold, although shallow breathing was allowed in very dyspnoeic patients. Overlapping images were reconstructed at 2-mm intervals.
PA was performed by using a 7-F Grollmann catheter (Cook, Eindhoven, the Netherlands) via a femoral vein under local anesthesia. The left and right main pulmonary artery were selectively catheterized in turn, and angiographic images were obtained in antero-posterior and left anterior oblique views. Depending on the patient and the expected cardiac output, a contrast injection protocol of either 15 mL s−1 (total 30 mL) or 20 mL s−1 (40 mL total) was used. If considered necessary, additional images were obtained in areas of special interest according to the perfusion images.
V/Q scintigrams were interpreted in conjunction with the chest radiographs using the revised PIOPED criteria . The perfusion-ventilation-scintigraphies were classified as normal, non-diagnostic (non-high), or high probability.
For interpretation of the CT scans in the central reading a workstation allowing cine mode viewing with various window and level settings was always used. Standard settings were window width 350 Hounsfield units (HU), window level 50 HU for mediastinal structures and pulmonary vasculature and window width 1500 HU, window level − 500 HU for comparison of opacified lung vasculature and anatomical relation to bronchi and lung parenchyma.
To detect PE on the CT scan the criteria previously described  were used. PE was considered to be present if in case of a well-opacified scan there was an intraluminal filling defect on more than one slice and no artefacts were present. A filling defect could be seen as a complete occlusion of the vessel, an eccentric partial filling defect or a partial central filling defect surrounded by the contrast agent. A CT scan was considered normal if in case of a good quality scan no filling defects could be seen. A scan was considered to be inconclusive if there was insufficient opacification of the vessels or in case of imaging artefacts that could lead to false-positive results.
Interpretation of the full angiographic runs was done on a viewing station. The criteria used for determination of the presence of PE as described by Sagel and Greenspan [23,24] were used to diagnose PE.
In all patients with the final diagnosis of PE the largest involved branch of the pulmonary vasculature was noted and categorized in five categories: main pulmonary stem, left or right main pulmonary artery, lobar artery, segmental artery or subsegmental artery. These categories were used to make a distinction between PE in segmental or larger arteries and PE in subsegmental arteries. This reading allowed only for assessment of the most proximally involved branch; no specification was made to determine whether patients had isolated or multiple subsegmental emboli.
Data management and statistical analysis were done by an independent group of biostatisticians and epidemiologists. Sensitivity and specificity (including their 95% CIs) for spiral CT were calculated. In addition, the sensitivity for spiral CT was calculated separately for both the subsegmental and segmental or larger PEs. Statistical analyses were performed using statistical software (Statistical Product and Service Solutions (SPSS) Inc., Chicago, IL, USA).
During the inclusion period of the study, of 1162 patients that presented at the participating centers with clinically suspected PE, 179 (15%) did not meet the inclusion criteria. Sixteen patients were below the age of 18, 11 were pregnant, six needed immediate treatment with thrombolytic therapy at the time of screening, and 43 patients had had diagnostic tests performed prior to start of the study protocol. Another 104 patients could not be included in the study because the diagnostic work-up could not be started within 24 h after presentation due to logistic or technical reasons (e.g. weekends and apparatus failure). Informed consent was obtained from 627 (64%) of the remaining 983 eligible patients. The majority of 356 patients refused informed consent because they did not want to participate in a study or were not able to give informed consent within the 24 h required to perform the initial imaging studies. The characteristics of the sample of consecutive patients who refused or were ineligible for study entry were not different from the characteristics of study patients.
A total of 110 patients had to be excluded for analysis (Table 1). A group of 44 patients had a disparity between locally assessed V/Q scan result (30 with normal, 14 with high-probability scan) and consensus V/Q scan result (all 44 non-diagnostic V/Q scan). In another 66 patients either ventilation scintigraphy, spiral CT or PA was not performed due to: (i) logistic problems or hardware down time (n = 22); (ii) withdrawal by the treating physician due to an apparent alternative diagnosis (n = 21); (iii) various medical reasons (n = 13); or (iv) withdrawal of informed consent (n = 10).
Table 1. Patients screened for the study
Age < 18 years
Indication for immediate thrombolytic therapy
Objective diagnostic work-up already started elsewhere
Logistic reasons (weekend, holidays)
Study protocol violators
Discrepancy between local and consensus V/Q scan result
Ultimately 517 patients with a mean age of 51.1 years were analyzed. Fifty-eight percent were male, 42% female; 19% were inpatients and 81% were outpatients. Seventy-three patients (14%) had had a previous venous thromboembolic event. The median duration of symptoms of PE was 3 days. Further demographic characteristics and risk factors for PE are shown in Table 2 and compared with the group of eligible patients and the excluded patients.
Table 2. Comparison of clinical and demographic characteristics of 627 eligible patients, 110 excluded patients and 517 analyzed study patients
All patients with suspected PE
Analyzed study patients
VTE, Venous thromboembolism; COPD, chronic obstructive pulmonary disease. *Period of immobilization, surgery or trauma in period of 3 months before presentation.
Of the 517 patients, 235 (45%) had a normal perfusion scintigraphy. No PE was considered to be present, and according to the protocol no spiral CT scan or PA was performed.
In 282 patients (55%) the perfusion scan showed abnormalities, of which 227 were segmental. In all of these patients additional ventilation scintigraphy was performed.
High-probability result of V/Q scintigraphy
The perfusion defects were not matched by ventilation defects in 153 patients and thus called high-probability. In 133 patients with these high-probability scintigraphy results a spiral CT scan was performed. Spiral CT was not performed in 20 patients, because of refusal by patient or treating physician in five patients, because the maximum amount of contrast agent had been used during PA in four patients, because of deterioration of the clinical situation in two patients, because a spiral CT scan was not possible (halotraction, claustrophobia) in four patients and because of an unknown reason in another five patients. In eight of the 133 patients no central reading of the spiral CT scan was available, while in four patients no consensus among observers could be obtained. Therefore 121 patients with CT scan results were available for analysis (Fig. 2).
Non-high-probability result of V/Q scintigraphy
In 74 of 227 patients with segmental perfusion defects there were also ventilation defects and thus the result of the combined ventilation-perfusion scintigraphy was non-high-probability or non-diagnostic. In 47 of the 274 patients with abnormal perfusion scan the defect was subsegmental.
In eight patients no central reading of the ventilation-perfusion scintigraphy was available, but in the local reading the scans were all with subsegmental perfusion defects (non-high-probability) and it was decided that further testing was indicated in these patients. Therefore, in total there were 129 patients with a non-diagnostic result of ventilation-perfusion scintigraphy. In all but 13 of these patients a spiral CT scan was available with a central reading and all of these patients had PA performed. In all of these patients a final conclusion could be made based on PA as the strongest source of evidence (Fig. 2).
Sensitivity and specificity in the overall population
Of the 517 patients available for analysis, 235 local interpreted normal perfusion scans were excluded for the comparison, and 274 patients had abnormal perfusion scintigrams. In 237 of these 274 patients a spiral CT scan was performed, 121 in patients with a high-probability scan, 109 in patients with a non-high-probability scan, and seven in V/Q without a central reading. All these 237 scans were available for central reading and received a diagnosis according to the study protocol. Spiral CT scans identified correctly 88 of 128 patients with proven PE (positive PA in 31 patients and high-probability ventilation-perfusion scintigraphy in 97 patients), for a calculated sensitivity of 69% (95% CI 63–75). Spiral CT scans showed no signs of PE in 92 of 109 patients with normal angiograms, for a calculated specificity of 84% (95% CI 80–89) (Table 3). Calculated sensitivity and specificity were therefore 69% (95% CI 63–75) and 84% (95% CI 80–89), respectively. The prevalence of PE in the total group of 517 patients was 32%.
Table 3. Patients with conclusion according to protocol, spiral computed tomography (CT) indicated and scan available for central reading
Sensitivity 69%, specificity 84%. *Diagnostic result of pulmonary angiography or a positive CT scan result in case of a high-probability ventilation perfusion scintigraphy.
Spiral CT scan result
Subgroup analysis of results; larger PE vs. isolated subsegmental PE
In 121 of 128 patients with a final diagnosis of PE a segmental analysis of the largest involved branch of the pulmonary vasculature was also performed. In two patients the readers were uncertain, and in five patients this analysis was missing in the case record form. This analysis was made on the result of the PA, if available. In all other patients with PE this was based on the result of the spiral CT scan. The largest involved branch was the main pulmonary stem in 10 patients (8.3%), the left or right pulmonary branch in 18 patients (14.9%), a lobar artery in 37 patients (30.6%), a segmental artery in 28 patients (23.1%) and a subsegmental artery in 28 patients (23.1%). In the group of patients with segmental or larger PE the spiral CT scan did not identify PE in 13 of 93 patients (14.0%). In contrast, spiral CT missed 22 out of 28 patients with PE (78.6%) in the subsegmental group. Therefore, sensitivity was 86% (95% CI 80–92) in the segmental or larger PE group and 21% (95% CI 14–29) in the group with subsegmental PE.
In this prospective study we have evaluated the accuracy of single-slice detector spiral CT in patients with clinical suspected PE and an abnormal perfusion scintigraphy result. The overall sensitivity of spiral CT was 69% and the specificity was 84%.
When looking only at PE in segmental and larger arteries the sensitivity and specificity were 88% and 85%, respectively. This is too low to withhold anticoagulant treatment on the basis of a normal spiral CT result alone. Furthermore, the sensitivity for subsegmental thrombi was a mere 26%.
Earlier studies, performed in a varying number of patients, reported different overall outcomes with sensitivities ranging from 64 to 100% and specificities ranging from 89 to 100%[16–19]. The reported results for segmental PE vary between 83 and 100%[21, 24] and for subsegmental emboli between 0% and 50%[17,18,25]. Our overall results, as well as the results of segmental and subsegmental emboli, fall well within these ranges and are similar to the results of a recent study by Perrier et al. .
The high percentage of normal lung scans has been reported before . This is not caused by liberalization of the criteria for normal scans, but is probably a reflection of a representative population suspected for PE in the participating hospitals.
Compared with the previous studies the overall sensitivity and specificity of spiral CT scan in the present study were relatively in the lower range. Three important reasons may account for this: the prospective inclusion of all patients suspected of having PE instead of including selective patients; the independent reading of all diagnostic tests; and the decision to take PA as the reference gold standard test. Indeed, the observed prevalence of PE in patients in the present study (32%) was fully comparable to the prevalence found in the large prospective PIOPED study (28%) .
Evaluation of PA in patients with a clinical suspicion of PE has shown a considerable interobserver variability in blinded readings up to 35%. An animal study has shown that PA and CT angiography might in fact have comparable results for sensitivity and specificity . Taking PA as a gold standard reference would therefore automatically result in a lower sensitivity and specificity for CT angiography in case of different test results for both tests.
Some methodological issues should be further discussed. First, spiral CT scanning was not performed in case of a normal perfusion scintigraphy, because there is solid evidence to safely exclude PE and withhold treatment with anticoagulants in case of a normal perfusion scintigram . In our study it is therefore impossible to assess the sensitivity and specificity of spiral CT in the group of patients with a normal perfusion scan. Second, it was decided not to perform PA when patients had concordant results on both V/Q scans and spiral CT scans in case of double evidence of PE based on a high-probability lung scan in conjunction with PE seen with spiral CT, due to medico-ethical reasons.
Finally, two recent reviews pointed out the essential requirements for prospective studies to evaluate definitively the sensitivity and specificity of spiral CT in the diagnosis of suspected PE [20,21]. Nearly all of the major criteria mentioned in this review were fulfilled by the design of the present study. During the study period all consecutive patients suspected of PE, irrespective of the severity of symptoms or comorbid conditions, were included according to preset inclusion criteria. All patients had a spiral CT scan and PA performed in case of perfusion defects. The CT technique used was a generally accepted technique for evaluating PE and was fully comparable to the techniques described in earlier studies. Recently, newer acquisition protocols have been advocated using 3-mm collimation, 5-mm tablefeed resulting in a pitch of 1.7, and the recently introduced multidetector CT scanners permit shorter imaging time, thinner imaging sections, and more extensive coverage of the thorax. Most study centers, however, still use single-detector CT equipment capable of scanning with the parameters described in this study. The study protocol was followed as rigidly as possible, and there was a thorough registration of all excluded patients and the reasons for not completing the study protocol. Possible bias due to diagnostic suspicion was avoided by accepting only a blinded interpretation of all available V/Q scans, spiral CT scans and PAs. Finally, in this study both spiral CT scans and PAs were evaluated for segmental and subsegmental levels of the pulmonary vasculature.
The results of this study challenge the clinical practice of withholding anticoagulant therapy based only on a normal spiral CT. Recent studies support that patients with a normal CT have a low likelihood of subsequent PE [29–31] and can be followed safely without anticoagulation. However, there are limitations in both the methods and follow-up of these studies. Particularly in the first study a considerable number of patients did not complete follow-up.
In patients with inadequate cardiorespiratory reserve a helical CT alone may be insufficient to rule out PE since even subsegmental emboli may be important. However, in this study only stable patients (able to give written informed consent) were included. The final clinical outcome was not a part of our study but should be investigated along with the result of possible further investigations in patients with a normal spiral CT scan. For this there are several options. The first possibility is to involve a clinical prediction rule, in combination with D-dimer testing . A second option is the use of serial compression ultrasonography or impedance plethysmography to detect possible deep vein thrombosis as a source of pulmonary emboli [7,8]. Indeed, after this study was completed, three studies, including one of our group with a different patient population, have been published that have evaluated algorithms in which normal helical CT was combined with compression ultrasonography [33–35]. In the first study by Musset et al. low to intermediate probability patients with negative findings on CT and US were left untreated with recurrent thromboembolism in 1.8% (95% CI 0.8–3.3) . The results of the more recent studies by Perrier and van Strijen found recurrent thromboembolism in case of negative findings in 1.0% (95% CI 0.5–2.1)  and 0.8% (95% CI 1.6–2.3) , respectively. Both Musset  and Perrier  stress the importance of additional ultrasonography of the legs in case of negative CT findings to withhold anticoagulant treatment safely. A somewhat limited value of additional ultrasonography was found in our own study. A more recent study found considerably more deep vein thrombosis in suspected patients with a complete lower limb US when compared with the more commonly performed limited two point evaluation test .
We conclude that based on the results of this study, in patients with clinically suspected PE and an abnormal perfusion scintigraphy single-detector spiral CT alone is not sensitive enough to be used as the sole test to exclude PE.
Financial support for this study was provided by the Dutch National Health Insurance Council (Ziekenfondsraad), grant no. D94-090.