Diagnostic accuracy of magnetic resonance imaging for an acute pulmonary embolism: results of the ÔIRM-EPÕ study

Transcription

1 Journal of Thrombosis and Haemostasis, 10: DOI: /j x IN FOCUS Diagnostic accuracy of magnetic resonance imaging for an acute pulmonary embolism: results of the ÔIRM-EPÕ study M. P. REVEL,*,O.SANCHEZ, à, S. COUCHON,* B. PLANQUETTE, à, A. HERNIGOU,* R. NIARRA, G. MEYER, à and G. CHATELLIER ** *Department of Radiology, Hôpital Européen Georges Pompidou, APHP, Université Paris Descartes, Sorbonne Paris Cité; àdepartment of Respiratory and Intensive Care, Hôpital Européen Georges Pompidou, APHP; INSERM Unité 765; Clinical Epidemiology, Hôpital Européen Georges Pompidou, APHP; and **INSERM CIC EC E4, Paris, France To cite this article: Revel MP, Sanchez O, Couchon S, Planquette B, Hernigou A, Niarra R, Meyer G, Chatellier G. Diagnostic accuracy of magnetic resonance imaging for an acute pulmonary embolism: results of the ÔIRM-EPÕ study. J Thromb Haemost 2012; 10: See also Huisman MV, Klok FA. Magnetic resonance imaging for diagnosis of acute pulmonary embolism: not yet a suitable alternative to CT-PA. This issue, pp 741 2; and Sodhi KS, Saxena AK. Diagnostic accuracy of computed tomography pulmonary angiography and magnetic resonance imaging for an acute pulmonary embolism. This issue, pp 961. Summary. Background: Magnetic resonance imaging (MRI) has not been validated as an alternative diagnostic test to computed tomography angiography (CTA) in patients with suspicion of a pulmonary embolism (PE). Objectives: To evaluate performance of current MRI technology in diagnosing PE, in reference to a 64-detector CTA. Patients/methods: Prospective investigation including 300 patients with a suspected PE, referred for CTA after assessment of clinical probability and D-dimer testing. MRI protocol included unenhanced, perfusion and angiographic sequences. MRI results were interpreted by two independent readers, to evaluate inter-reader agreement. Sensitivity and specificity were evaluated globally and according to PE location and to clinical probability category. Results: Of 300 enrolled patients, 274 were analyzed and 103 (37.5%) had a PE diagnosed by CTA. For patients with conclusive MRI results (72% for reader 1, 70% for reader 2), sensitivity and specificity were 84.5% (95% confidence interval [CI], %) and 99.1% (95% CI, %), respectively, for reader 1, and 78.7% (95% CI, %) and 100% (95% CI, %) for reader 2. After exclusion of inconclusive MRI results for both readers, inter-reader agreement was excellent (kappa value: 0.93, 95% CI: ). Sensitivity was better for proximal ( %) than for segmental ( %) and sub-segmental ( %) PE (P < ). Sensitivity was similar for both readers within each clinical probability category. Conclusions: Current MRI technology demonstrates high specificity and high sensitivity for Correspondence: Marie-Pierre Revel, Service de radiologie, Hôpital Européen Georges Pompidou, 20 rue Leblanc, Paris, France. Tel.: ; fax: marie-pierre.revel@htd.aphp.fr Received 24 September 2011, accepted 26 January 2012 proximal PE, but still limited sensitivity for distal PE and 30% of inconclusive results. Although a positive result can aid in clinical decision making, MRI cannot be used as a stand-alone test to exclude PE. Keywords: magnetic resonance imaging, perfusion imaging, pulmonary embolism. Introduction Most patients with a suspected pulmonary embolism (PE) undergo either computed tomography angiography (CTA) or ventilation/perfusion scanning. Magnetic resonance imaging (MRI) has recently been evaluated by the PIOPED III investigators, as an alternative technique for PE diagnosis [1]. Indeed, CTA is contraindicated in up to 22% of the patients because of allergy to iodine-based contrast agents or renal insufficiency [2]. Radiation exposure related to the use of CTA is also a concern, especially in young or gravid patients. Two recent publications have addressed the risks of cumulative radiation dosage in patients with a history thromboembolism who have multiple subsequent examinations for suspected recurrence [3,4]. A ventilation/perfusion (V/Q) lung scan is often non-diagnostic or logistically not feasible after-hours. For these reasons, other alternative diagnostic tests such as MRI needed to be evaluated. However, the technique, mainly based on MR angiography (MRA), was reported to have limited sensitivity for PE diagnosis and a high rate of technically inadequate studies [5]. The aim of the present study was evaluate whether current MRI including newer sequences could be used as a reliable alternative diagnostic test for PE diagnosis. The second objective was to evaluate inter-reader agreement, which has not been done previously.

2 744 M. P. Revel et al Patients and methods Design and setting The IRM-EP study was a prospective cohort study, conducted in a French University Hospital from the 8 June 2007 until the 15 June It was designed to evaluate the sensitivity and specificity of MRI in diagnosing PE in reference to CTA results in patients from the emergency department or clinical wards with a suspected PE who were referred for CTA. The study protocol and consent form were approved by our local Ethics Committee (Comité de Protection des Personnes Ile de France II). Patients Inclusion criteria Patients referred for a CTA with suspicion of an acute PE were eligible for inclusion if they were at least 18 years old, had a high clinical probability of PE assessed by the revised Geneva score [6] or had a D-dimer level > 500 lg L )1 on an ELISA-based test (VIDAS; BioMérieux, Lyon, France) and had provided written informed consent. Eligibility was mainly evaluated during normal working hours, although all patients had an assessment of clinical probability and subsequent D-dimer testing before a CTA. Owing to limited availability of the MRI unit for research purposes, we could only include one patient per day: therefore, the first patient fulfilling the inclusion criteria on any one day was included. Exclusion criteria Patients were ineligible if they had signs of a severe PE such as unstable hemodynamics; had been on therapeutic anticoagulation therapy for more than 48 h; had a contraindication to MRI including claustrophobia, a metallic ocular implant or pace maker, or a reported allergy to gadolinium-based contrast agents; or had a glomerular filtration rate under 30 ml min )1 or other contraindications to CTA. A patient weight above 130 kg or a postero-anterior abdominal diameter over 60 cm were added as exclusion criteria beacause of difficulties encountered in performing a MRI in obese patients. Lastly, patients with an inconclusive CTA result were deemed non-analyzable unless they had a normal V/Q scan or an uneventful 3 month follow-up without anticoagulant treatment, in which case they were determined to not have a PE. Magnetic resonance imaging MRI protocol Magnetic resonance imaging was performed within 24 h of CTA on a 1.5 Tesla unit (Signa HDxt system; GE Healthcare, Chalfont St. Giles, UK) with fast gradientecho capability (33 mt/m max gradient field strength). A specialized multichannel phased array coil (HD Cardiac coil) (eight channels) was used for reception of the pulmonary MRI data, whereas the body coil was used for signal transmission. The MRI protocol included three different sequences, all performed using parallel imaging (ASSET) with an acceleration factor of 2. 1 Unenhanced steady-state-free precession sequences Steady-state-free precession (SSFP) sequences were acquired first without ECG-gating or breath-holding in the axial plane in the multiphase cine mode, with six phases per location. The following parameters were used: TR (repetition time): ms, TE (echo time): ms, flip angle 60, matrix, 42 cm field of view, bandwidth 125 khz, single acquisition/number of excitations (0.5 NEX), 4 mm slice thickness and 20 slices. Imaging time was 44 s. SSFP sequences were then performed with ECG-gating and 18-s breath holds, resulting in 2 to 4 slices per apnea depending on the heart rate. The same parameters were used except that slice thickness was increased to 5 mm. The acquisition was repeated to cover two-thirds of the thorax, from the roof of the aorta to the diaphragm. 2 Perfusion sequences Perfusion sequences were acquired in the coronal plane, using 3-dimensional Fast SPGR (spoiled gradient echo) with the following parameters: TR: 2 ms, TE: 0.8 ms, flip angle 20, 160 by 224 matrix, 48 cm field of view and bandwidth kh, single acquisition/number of excitations (0.75 NEX). Eighteen slices with a 6.4-mm slice thickness interpolated to 3.2 mm were acquired every 3 s. A first set of slices was acquired before contrast administration and, during the same apnea, seven consecutive acquisitions were performed beginning 7 s after the start of injection. The whole sequence required an apnea of 31 s. For perfusion sequences, 0.1 ml kg )1 of DOTA-Gd was injected at a rate of 5 ml s )1 followed by an injection of 15 ml of normal saline at 3 ml s )1. Automated mask subtraction was performed to finally obtain two sets of perfusion images, namely source images and post-processed images after unenhancing tissue subtraction. 3 Angiography sequences A pulmonary gradient recalled echo (GRE) (Vasc TOF SPGR, time-of-flight spoiled gradient echo) sequence was performed in the axial plane, with the following parameters: TR: 3.4 ms, TE: 1.2 ms with asymmetric echo, flip angle 25, matrix, 36 cm field of view with phase field of view 0.8, bandwidth 83.3 khz, single acquisition/number of excitations (1 NEX). Slice thickness was 2.4 mm and there were 40 slices per slab. The acquisition was triggered to start when contrast enhancement occurred in the right ventricle. Two consecutive acquisitions were necessary to cover the anatomy. For each acquisition, 0.15 ml kg )1 body weight of DOTA- Gd was injected at a rate of 3 ml s )1 followed by an injection of 15 ml of normal saline at 3 ml s )1. Diagnostic reference standard The results of CTA performed on a 64-detector multislice CT unit served as reference. CTA was interpreted by one expert with a 12-yeasr experience in CTA, blinded to other results and using previously

3 Magnetic resonance imaging for PE 745 described criteria [7]. We considered all positive CTA results, including single sub-segmental PE (SSS-PE) as confirming PE regardless of the pre-test clinical probability (reference, standard 1). Because the clinical significance of a single sub-segmental PE (SSS-PE) is unclear, we considered a second reference standard where SSS-PE was considered as a negative CTA result (reference, standard 2) [8]. When CTA was inconclusive, PE was excluded using a normal lung scintigraphy or an uneventful 3-month follow-up in patients left untreated. Patients with a negative CTA were followed-up for 3 months. MRI readings MRI readings were performed by two independent readers, blinded to CTA results and clinical probability, with 7 and 18 years of experience, respectively, in chest imaging. The readers had all sequences available simultaneously and were free to rate the images as positive, negative or inconclusive if images were technically inadequate. Readings were deferred until all inclusions were completed, so the readers remained blinded to CTA results during the whole study duration. Statistical analysis We calculated that 300 patients were required to estimate MRI sensitivity with a 10% precision, hypothesizing an 80% sensitivity of MRI, a 20% rate of technical failure and a 30% prevalence of PE in our population. According to conventional analysis, we excluded technically inadequate studies from computation of sensitivity and specificity. We obtained exact 95% confidence intervals (CIs) for sensitivity and specificity from the binomial distribution. We calculated likelihood ratios with 95% CIs using the normal distribution approximation of binomial distribution. We determined MRI sensitivity and specificity globally as well as for each category of PE probability using the revised Geneva score [6]. We also determined MRI sensitivity for the different locations of PE (proximal, segmental and subsegmental). We used the chi-square test or FisherÕs exact test to compare categorical variables. Between-radiologist agreement was assessed using the kappa statistic. The kappa values were interpreted on the following basis: kappa < 0.20, poor agreement; kappa = , fair agreement; kappa = , moderate agreement; kappa = , good agreement; kappa = , very good agreement [9]. Statistical significance was considered at the 0.05 level. We used SAS software version 9.2 (SAS Inc., Cary, NC, USA) for all statistical analyzes. The results are reported according to the recommendations of the STARD statement [10]. Results During the study period, 1796 patients with suspicion of a PE were referred for CTA in our institution. Eligibility was not evaluated in 882 patients because of presentation out-of-hours. Of the remaining 914 patients, a further 614 were not included Potentially eligible patients n = 1796 Eligibility evaluated (working-hours presentation) n = 914 Included : n = 300 Final study population n = 274 Fig. 1. Flow chart of the population. Eligibility not evaluated (out of hours presentation) n = 882 Not included: n = Not meeting inclusion criteria: Declined to participate: 49 - Other inclusion on the same day: 43 - Inconclusive CTA (performed first): 75 Excluded: n = 26 - Not completing whole MRI protocol: 25 - Inconclusive CTA, treated: 1 because they did not fulfill the inclusion criteria, declined to participate, had an inconclusive CTA performed before MRI or because one patient had already been included on the same day (Fig. 1, flow chart of the study population). Of the 300 included patients, 277 completed the full MRI protocol. The 23 who did not complete the protocol either withdrew their consent (five patients), could not have a MRI (11 patients, eight claustrophobic and three obese) or interrupted it prematurely (six patients), or experienced extravasation of the injected contrast agent (one patient). MR images not stored in our Pictures Archiving and Communicating System (PACS) were lost for two patients. Of the remaining 275 patients, 11 had an inconclusive CTA result. A PE was excluded in 10 of these patients who had a normal V/Q scan or uneventful 3-month follow-up without anticoagulation therapy, but not in one patient who had received anticoagulation therapy. This patient was thus excluded from the analysis.

4 746 M. P. Revel et al A C D B Fig. 2. Proximal pulmonary embolism (PE) in a 47-year-old woman. Marginal clots within the right interlobar pulmonary artery (white arrow head) demonstrated on an unenhanced magnetic resonance imaging (MRI) sequence (A), MRI angiography sequence (B) with the corresponding image on computed tomography (CT) angiography (C). A large right-sided perfusion defect is demonstrated on the perfusion MRI sequence (D). Table 1 Demographic characteristics, presenting signs and symptoms, coexisting illness and clinical probability of a PE Reference test result (n = 274) Parameters All patients (n = 274) Positive for PE (n = 103) Negative for PE (n = 171) Site, n/n (%) Outpatient 228/274 (83.2) 87/103 (84.5) 141/171 (82.5) Inpatient 43/274 (15.7) 15/103 (14.6) 28/171 (16.4) Demographic characteristic Women, n/n (%) 147/274 (53.6) 52/103 (50.5) 95/171 (55.6) Mean age (SD), years 59.8 (19.0) 59.0 (19.1) 60.2 (19.0) Mean body weight (SD), kg 70.8 (15.2) 74.5 (16.2) 68.6 (14.1) Presenting signs and symptoms, n/n (%) Chest pain 136/274 (49.6) 55/103 (53.4) 81/171 (47.4) Dyspnea 173/274 (63.1) 75/103 (72.8) 98/171 (57.3) Hemoptysis 18/274 (6.6) 7/103 (6.8) 11/171 (6.4) Discomfort, faintness 24/274 (8.8) 9/103 (8.7) 15/171 (8.8) Signs of acute right heart failure 25/274 (9.1) 11/103 (10.7) 14/171 (8.2) Coexisting illness, n/n (%) Congestive heart failure 11/274 (4.0) 1/103 (1.0) 10/171 (5.9) Coronary artery disease 21/274 (7.7) 5/103 (4.9) 16/171 (9.4) Chronic obstructive or restrictive lung disease 35/274 (12.8) 9/103 (8.7) 26/171 (15.2) Previous deep vein thrombosis 45/274 (16.4) 23/103 (22.3) 22/171 (12.9) Previous pulmonary embolism 33/274 (12.0) 14/103 (13.6) 19/171 (11.1) Clinical probability, n/n (%) Low 74/274 (27.0) 24/103 (23.3) 50/171 (29.2) Intermediate 166/274 (60.6) 58/103 (56.3) 108/171 (63.1) High 33/274 (12.0) 21/103 (20.4) 12/171 (7.0) PE, pulmonary embolism; SD, standard deviation. Of the remaining 274 patients, 103 had a PE on CTA, 47 proximal PEs involving main or lobar arteries, 33 segmental and 23 sub-segmental, resulting in a 37.5% prevalence of PE in our study population using reference standard 1 (Fig. 2). Their clinical and demographic characteristics are presented in

5 Magnetic resonance imaging for PE 747 Table 1. Excluding isolated sub-segmental PEs (reference standard 2), PE prevalence was 33.6% (92/274). Although statistically significant, there were only small differences between included and non-included patients: mean age (60 ± 19 vs. 63 ± 19 years old, P = 0.003); male/female distribution (141/300, [47%] vs. 606/1496, [40.5%], P = 0.04); and proportions of high, intermediate and low clinical probability of PE (26.8%, 61.7% and 11.4% vs. 20.0%, 71.5% and 8.5%, respectively, P = 0.004). A larger difference was observed between included and non-included patients in terms of PE prevalence (37.5 vs. 17.2%, P < ). MRI results Mean examination duration (from the first scout view to the last image) was 28.2 ± 7.4 min. Mean dose of gadolinium was 28.4 ± 7.2 ml. Inconclusive results The proportion of inconclusive MRI results was similar for the two readers: 28% (76/274) for reader 1 vs. 30% (83/274) for reader 2. Inconclusive results were mainly as a result of technically inadequate MR examinations showing artifacts (48.7% [37/76] for reader 1 and 62.7% [52/83] for reader 2) or poor opacification on angiographic sequences (35.5% [27/76] for reader 1 and 22.9% [19/83] for reader 2), the remainder being due to isolated perfusion abnormalities, without clot detection within the pulmonary arteries. The proportion of inconclusive MRI results was marginally lower in younger patients for both reader 1 and reader 2 (P =0.07 and P = 0.03, respectively, FisherÕs exact test) (Table 2). The proportions of inconclusive MRI results did not differ from the first to the last semester of the study period for either reader (Table 3). Diagnostic performance of MRI After exclusion of inconclusive MRI examinations, sensitivity was 84.5% and 89.6% for reader 1 and 78.7% and 83.6% for reader 2, according to reference standards 1 and 2, respectively (Table 4). Specificity was 99.1% and 97.4% for reader 1 and 100% and 98.3% for reader 2, according to reference standards 1 and 2, respectively (Table 4). Considering inconclusive MRI as negative, and taking into account all MR examinations (Ôintention to diagnose analysisô), sensitivity was 68.9% (95% CI: 59.1% to 77.7%) for reader 1 for reference standard 1 and 75.0% (95% CI: 64.9% to 83.5%) for reference standard 2. None of the 10 patients with inconclusive CTA, which for PE was excluded by other means, had a positive MRI result. Table 2 Proportion of inconclusive MRI according to the patientõs age Inconclusive MRI result 18 to 45 years (n = 65) 45 to 62 years (n = 67) 62 to 75 years (n = 71) 75 years (n = 71) P-value* Reader 1, N (%) 12 (15.8) 16 (21.1) 27 (35.5) 21 (27.6) 0.07 Reader 2, N (%) 16 (19.3) 15 (18.1) 31 (37.3) 21 (25.3) 0.03 *FisherÕs exact test. This table assesses the proportion of inconclusive magnetic resonance imaging (MRI) results according to the patientõs age, considering four age categories. The proportion of inconclusive MRI results was marginally lower in younger patients for both reader 1 and reader 2(P = 0.07 and P = 0.03, respectively, FisherÕs exact test). Table 3 Proportions of inconclusive MRI in each semester of the study period Number of Included patients Whole period (n = 274) June 7 to December 7 (n = 51) January 8 to June 8 (n = 92) July 9 to December 8 (n = 80) January 9 to June 9 (n = 51) P-value Inconclusive result Reader 1 N (%) 76 (24.7) 17 (22.4) 18 (23.7) 22 (28.9) 19 (25.0) Inconclusive result Reader 2 N (%) 83 (30.3) 15 (18.1) 28 (33.7) 22 (26.5) 18 (21.7) This table assesses the proportion of inconclusive magnetic resonance imaging (MRI) result according to each semester of the study period, considering four periods. The proportion of inconclusive MRI results did not decrease over time (P = 0.10 and P = 0.82, for reader 1 and reader 2, respectively, FisherÕs exact test). Table 4 MRI reading results for reader 1 and 2 Reader 1 Reader 2 Reference standard 1 Reference standard 2 Reference standard 1 Reference standard 2 Sensitivity (%) [95%CI] 84.5 [ ] 89.6 [ ] 78.7 [ ] 83.6 [ ] Specificity [95%CI] 99.1 [ ] 97.4 [ ] [ ] 98.3 [ ] LR+ [95%CI] 94.7 [ ] 35.5 [ ] NC* 48.5 [ ] LR) [95%CI] 0.2 [ ] 0.1 [ ] 0.2 [ ] 0.2 [ ] PPV [95%CI] 98.6 [ ] 95.8 [ ] [ ] 96.8 [ ] NPV [95%CI] 89.5 [ ] 93.5 [ ] 86.5 [ ] 90.5 [ ] *NC, not calculable. LR+, positive likelihood ratio; LR), negative likelihood ratio; PPV, positive predictive value; NPV, negative predictive value. This table shows the MRI reading results for both readers, with the two reference standards. With reference-standard 1, single sub-segmental PE (SSS-PE) was considered as a positive CTA result, whereas with reference-standard 2, SSS-PE was considered as a negative CTA result.

6 748 M. P. Revel et al Reader 1: ; Reader 2 : o PE location Proximal Segmental Subsegmental Clinical probability Low Intermediate High Taking into account all MRI results, including inconclusive MRI results for each reader, inter-reader agreement was moderate (kappa value: 0.59, 95% CI: ). However, after exclusion of inconclusive MRI results for both readers, inter-reader agreement was excellent (kappa value: 0.93, 95% CI: ). Adverse events One patient experienced extravasation of the contrast agent when the gadolinium was injected for the perfusion sequence. This led to interruption of the MRI study. The patient had a spontaneous favorable outcome but was excluded from the analysis because he did not complete the whole MRI protocol Fig. 3. Magnetic resonance imaging (MRI) sensitivity according to pulmonary embolism (PE) location and to clinical probability. MRI sensitivity (%), with 95% confidence intervals (CIs) (error bars), according to PE location (proximal, segmental, subsegmental) and to clinical probability (low, intermediate, high), for reader 1 (d) and reader 2(s). MRI sensitivity ranges from 21.4% to 100% according to PE location and from 68.4% to 100% according to clinical probability. Negative predictive values (NPV) were not significantly different between categories of clinical probability of PE, although they tended to be less in patients with high clinical probability: 94.6%, 89.6%, 70.0% for reader 1 (P = 0.079) and 84.6%, 89.5%, 72.7% (P = 0.289) for reader 2 in patients with low, intermediate and high clinical probability, respectively. Sensitivity was similar for both readers within each clinical probability category: 88.2%, 83.7%, 83.3% for reader 1 (P = 0.925) and 68.4%, 81.4%, 83.3% (P =0.624)forreader 2 in patients with low, intermediate and high clinical probability, respectively (Fig. 3; Table 5). For both readers, sensitivity was greater for proximal than for segmental and sub-segmental PE (100.0%, 91.7%, 21.4% [P < ] and 97.7%, 68.0%, 33.3%; [P < ] for reader 1 and reader 2, respectively) (Fig. 3). Sensitivity was similar between readers except at the segmental level (P = 0.043). Discussion The main results of the present study are the high specificity and good inter-reader agreement that current MRI techniques allow in diagnosing PE. While sensitivity is high for diagnosis of proximal PE, it becomes less so as pulmonary arteries segment and extend distally. The PIOPED III study, conducted between 2006 and 2008, was the first large series evaluating MRI in 371 patients with confirmed or excluded PE [1]. Prior to its publication in 2010, MRI studies on PE included small numbers of patients or were not prospectively conducted [11]. Our study design was different from that of the PIOPED III study in that the proportion of positive test results was 50% by construction in PIOPED III [12], where patients with negative reference test were randomly selected to participate, vs. 37.5% by occurrence in our study; our reference standard was not composite but based on CTA findings. Lastly, our MRI protocol did not include an evaluation of the deep veins, but included unenhanced and perfusion sequences in addition to conventional MR angiography. In spite of these differences, our results confirm the high specificity of the technique (97.4% to 100% compared with 99%). Global sensitivity was also similar in the two studies but proximal PE, the easiest to detect, represented 93% of PE in the PIOPED III study and only 46% in the present study. For proximal PE, sensitivity was actually higher in our study Table 5 Magnetic resonance imaging (MRI) diagnostic performance according to clinical probability Clinical probability Sensivity Specificity NPV PPV Reader I Low 88.2 [ ] [ ] 94.6 [ ] [ ] Intermediate 83.7 [ ] 98.6 [ ] 89.6 [ ] 97.6 [ ] High 83.3 [ ] [ ] 70.0 [ ] [ ] P of trend Reader II Low 68.4 [ ] [ ] 84.6 [ ] [ ] Intermediate 81.4 [ ] [ ] 89.5 [ ] [ ] High 83.3 [ ] [ ] 72.7 [ ] [ ] P of trend * * *Not calculable. NPV, negative predictive value; PPV, positive predictive value.

7 Magnetic resonance imaging for PE 749 ranging from 97.7% to 100% compared with 78% in PIOPED III. The gain in sensitivity may be because of technical improvements allowing thinner thicknesses for MR angiography (2 vs. 3 mm in the PIOPED III study), thus increasing the spatial resolution and the possibility to evaluate not only the pulmonary arteries but also the lung parenchyma perfusion. A perfusion defect demonstration is likely to help clot detection in the supplying pulmonary artery. We found a very good inter-reader agreement when excluding inconclusive results for both readers, meaning readers strongly agreed on positive and negative MRI results. Conversely, agreement was only moderate when inconclusive results were not excluded, meaning readers did not consider the same MRI examinations as inconclusive. Indeed, artifacts and the quality of lung parenchyma enhancement were subjectively estimated in the present study. The proportion of inconclusive results, mainly because of artifacts, remained similar to that of the PIOPED III study. MRI susceptibility to artifacts is well known and may limit its use, if in spite of further technical improvements, the rate of inconclusive result remains as high as 30%. Another limitation in the use of MRI is the risk of nephrogenic systemic fibrosis induced by gadolinium-based contrast agents in patients with renal disease [13 15]. None of the patients included in the present study had renal insufficiency, because one of the inclusion criteria was the ability to have CTA as the reference standard. Moreover, we only used gadoterate meglumine (Gd-DOTA) and not gadodiamide or gadopentetate dimeglumine which have caused most of the cases of nephrogenic systemic fibrosis reported in this setting [16]. Time to complete the examination is another limitation of MRI. The 28 min required in the present study were related to the use of three different sequences in the study protocol and could probably be reduced. This duration also explained why some patients asked to interrupt participation. A last limitation is the low availability of MRI units, especially in emergency, which explains why we could only include one patient per day, but MRI availability may be different in various countries. Strengths and weaknesses The strengths of this study are the following. First, we strictly complied with all of the STARD (Standards for Reporting Diagnostic Accuracy) criteria [10,17]. In addition, the reference standard used in the present study was a 64-detector CTA, the current state-of-the-art multirow detector CT technology and the Õgold-standardÔ technique for PE diagnosis [18]. Furthermore, we compared patients of our study population with eligible patients not included in terms of age, prevalence and clinical probability of PE and demonstrated there were only slight differences, except for PE prevalence. Our study also presents some weaknesses. First, it was a single-center study which could limit the reliability of the results but we considered that a single institution would provide a higher homogeneity of results in view of the complex MRI protocol including three types of sequences. Our study design did not allow us to evaluate the contribution of each sequence to PE diagnosis, because the readers had all three sequences available simultaneously. This will require another reading session to consider each sequence independently in a random order. Second, our patients represent only 17% (300/1794) of those referred for a CTA during the study period as only one patient per day could be included owing to limited MRI availability. PE prevalence was found to be higher in our population than that reported in recent CTA series [4,19]. However, patients with negative D-dimers were excluded, thus leaving only those with a high clinical probability or elevated D-dimers who are reported as having a 28% to 38% prevalence of a PE [20 22]. Moreover, patients with a positive CTA were more likely to be tested for eligibility than patients discharged after a negative CTA and this may also account for the high prevalence of PE in the study population. Third, CTA was evaluated by only one radiologist and this may have led to misclassification of a sub-segmental PE. To take into account the inter-observer variability in the diagnosis of sub-segmental PE on CTA, we used two different reference standards, one considering single sub-segmental PE as a positive CTA result and the second one as a negative CTA result. Clinical applicability The use of MRI avoids exposing patients to ionizing radiation and is therefore particularly relevant during pregnancy and for young patients. Chronic kidney disease is a contra-indication to the use of MRI for PE diagnosis, given the risk of nephrogenic systemic fibrosis in Gadolinium-exposed patients with renal insufficiency. Owing to the high specificity and inter-reader agreement of MRI, anticoagulation therapy may be initiated on the basis of positive findings, whereas additional tests are still required in patients with a negative MRI result. MRI sensitivity in the present study is similar to that previously reported with singleslice helical CT, in which a lower limb ultrasound was required in addition to a negative CTA to rule out a PE [23]. The use of compression ultrasound after a normal MR could overcome the lack of sensitivity. In conclusion, the present study confirms MRI specificity and demonstrates a high sensitivity for proximal PE diagnosis together with high agreement, whereas the sensitivity for segmental and sub-segmental PE remains limited. Continued technical improvement may allow us to overcome these present limitations. In the meantime, MRI cannot be used as a standalone test and needs to be evaluated in combination with other non-invasive techniques. Addendum Concept and design: M.-P. Revel, G. Meyer, G. Chatellier. Analysis and/or interpretation of data: O. Sanchez, S. Couchon, A. Hernigou, R. Niarra, B. Planquette. Critical

8 750 M. P. Revel et al writing: M.-P. Revel, O. Sanchez, G. Meyer Revising the intellectualcontent:m.-p.revel,g.meyer,r.niarra,g. Chatellier. Final approval: all authors. Acknowledgements The study was funded by grant CRC from the Assistance Publique Hoˆ pitaux de Paris. The authors were totally independent of the funder for all scientific aspects of the research. The authors would like to thank Noel Lucas, Yann Guivarch, Romain Chenu, Ine` s Ben Jaballah, Severine Peyrard and Isabelle Sauret from the Clinical Research Unit of Hoˆ pital Européen Georges Pompidou for their help in this study. Marie Pierre Revel had full access to all study data and takes responsibility for the integrity of the data and the accuracy of the data analysis. Disclosure of Conflict of Interest The authors state that they have no conflict of interest. References 1 Stein PD, Chenevert TL, Fowler SE, Goodman LR, Gottschalk A, Hales CA, Hull RD, Jablonski KA, Leeper KV Jr, Naidich DP, Sak DJ, Sostman HD, Tapson VF, Weg JG, Woodard PK. Gadoliniumenhanced magnetic resonance angiography for pulmonary embolism: a multicenter prospective study (PIOPED III). Ann Intern Med 2010; 152: Stein PD, Fowler SE, Goodman LR, Gottschalk A, Hales CA, Hull RD, Leeper KV Jr, Popovich J Jr, Quinn DA, Sos TA, Sostman HD, Tapson VF, Wakefield TW, Weg JG, Woodard PK. Multidetector computed tomography for acute pulmonary embolism. N Engl J Med 2006; 354: Kline JA, Courtney DM, Beam DM, King MC, Steuerwald M. Incidence and predictors of repeated computed tomographic pulmonary angiography in emergency department patients. Ann Emerg Med 2009; 54: Donohoo JH, Mayo-Smith WW, Pezzullo JA, Egglin TK. Utilization patterns and diagnostic yield of 3421 consecutive multidetector row computed tomography pulmonary angiograms in a busy emergency department. J Comput Assist Tomogr 2008; 32: Davidson BL, Lacrampe MJ. Why canõt magnetic resonance imaging reliably diagnose pulmonary embolism? AnnInternMed2010; 152: LeGalG,RighiniM,RoyPM,SanchezO,AujeskyD,Bounameaux H, Perrier A. Prediction of pulmonary embolism in the emergency department: the revised Geneva score. Ann Intern Med 2006; 144: Remy-Jardin M, Remy J, Deschildre F, Artaud D, Beregi JP, Hossein- Foucher C, Marchandise X, Duhamel A. Diagnosis of pulmonary embolism with spiral CT: comparison with pulmonary angiography and scintigraphy. Radiology. 1996; 200: Carrier M, Righini M, Wells PS, Perrier A, Anderson DR, Rodger MA, Pleasance S, Le Gal G. Subsegmental pulmonary embolism diagnosed by computed tomography: incidence and clinical implications. A systematic review and meta-analysis of the management outcome studies. J Thromb Haemost 2010; 8: Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977; 33: BossuytPM,ReitsmaJB,BrunsDE,GatsonisCA,GlasziouPP,Irwig LM, Moher D, Rennie D, de Vet HC, Lijmer JG. The STARD statement for reporting studies of diagnostic accuracy: explanation and elaboration. AnnInternMed2003; 138: W Stein PD, Woodard PK, Hull RD, Kayali F, Weg JG, Olson RE, Fowler SE. Gadolinium-enhanced magnetic resonance angiography for detection of acute pulmonary embolism: an in-depth review. Chest 2003; 124: Stein PD, Gottschalk A, Sostman HD, Chenevert TL, Fowler SE, Goodman LR, Hales CA, Hull RD, Kanal E, Leeper KV Jr, Nadich DP,SakDJ,TapsonVF,WakefieldTW,WegJG,WoodardPK. Methods of Prospective Investigation of Pulmonary Embolism Diagnosis III (PIOPED III). Semin Nucl Med 2008; 38: Bhave G, Lewis JB, Chang SS. Association of gadolinium based magnetic resonance imaging contrast agents and nephrogenic systemic fibrosis. JUrol2008; 180: discussion Othersen JB, Maize JC, Woolson RF, Budisavljevic MN. Nephrogenic systemic fibrosis after exposure to gadolinium in patients with renal failure. Nephrol Dial Transplant 2007; 22: Perazella MA. Gadolinium-contrast toxicity in patients with kidney disease: nephrotoxicity and nephrogenic systemic fibrosis. Curr Drug Saf. 2008; 3: Marckmann P. Nephrogenic systemic fibrosis: epidemiology update. Curr Opin Nephrol Hypertens 2008; 17: BossuytPM,ReitsmaJB,BrunsDE,GatsonisCA,GlasziouPP,Irwig LM, Lijmer JG, Moher D, Rennie D, de Vet HC. Towards complete and accurate reporting of studies of diagnostic accuracy: the STARD Initiative. Ann Intern Med 2003; 138: Wittram C, Waltman AC, Shepard JA, Halpern E, Goodman LR. Discordance between CT and angiography in the PIOPED II study. Radiology. 2007; 244: Weir ID, Drescher F, Cousin D, Fraser ET, Lee R, Berman L, Strauss E, Wang Y, Fine JM. Trends in use and yield of chest computed tomography with angiography for diagnosis of pulmonary embolism in a Connecticut hospital emergency department. Conn Med 2010; 74: Righini M, Le Gal G, Aujesky D, Roy PM, Sanchez O, Verschuren F, Rutschmann O, Nonent M, Cornuz J, Thys F, Le Manach CP, Revel MP, Poletti PA, Meyer G, Mottier D, Perneger T, Bounameaux H, Perrier A. Diagnosis of pulmonary embolism by multidetector CT alone or combined with venous ultrasonography of the leg: a randomised non-inferiority trial. Lancet 2008; 371: Perrier A, Roy PM, Sanchez O, Le Gal G, Meyer G, Gourdier AL, Furber A, Revel MP, Howarth N, Davido A, Bounameaux H. Multidetector-row computed tomography in suspected pulmonary embolism. NEnglJMed2005; 352: van Belle A, Buller HR, Huisman MV, Huisman PM, Kaasjager K, Kamphuisen PW, Kramer MH, Kruip MJ, Kwakkel-van Erp JM, Leebeek FW, Nijkeuter M, Prins MH, Sohne M, Tick LW. Effectiveness of managing suspected pulmonary embolism using an algorithm combining clinical probability, D-dimer testing, and computed tomography. JAMA 2006; 295: Perrier A, Howarth N, Didier D, Loubeyre P, Unger PF, de Moerloose P, Slosman D, Junod A, Bounameaux H. Performance of helical computed tomography in unselected outpatients with suspected pulmonary embolism. Ann Intern Med 2001; 135: