Prognostic Factors for Pulmonary Embolism The PREP Study, A Prospective Multicenter Cohort Study Olivier Sanchez1*, Ludovic Trinquart2*, Vincent Caille3, Francis Couturaud4, Ge´rard Pacouret5, Nicolas Meneveau6, Franck Verschuren7, Pierre-Marie Roy8, Florence Parent9, Marc Righini10, Arnaud Perrier10, Christine Lorut11, Bernard Tardy12, Marie-Odile Benoit13, Gilles Chatellier2, and Guy Meyer1 1

Service de Pneumologie et Soins Intensifs, and 2Unite´ de Recherche Clinique, INSERM CIE 4, Universite´ Paris Descartes, Assistance Publique ˆ pitaux de Paris, Ho ˆ pital Europe´en Georges Pompidou, Paris, France; 3Ho ˆ pital Ambroise Pare´, Service de Re´animation me´dicale, Assistance Ho ˆ Publique Hopitaux de Paris, Boulogne, France; 4Universite´ Europe´enne de Bretagne, Universite´ de Brest, EA3878, IFR148, and Departement de ˆ pital Trousseau, Service de Cardiologie A, CHRU de Tours, Me´decine Interne et de Pneumologie, CHU de La Cavale Blanche, Brest, France; 5Ho xon; France 7Universite´ Catholique de Louvain, Cliniques Universitaires Saint-Luc, Acute France; 6CHU Jean Minjoz, Service de Cardiologie, Besanc ˆ pital Medicine Department, Accidents and Emergency Unit, Brussels, Belgium; 8CHU d’Angers, Service d’accueil des urgences, Angers, France; 9Ho ˆ pitaux de Paris, Clamart, France; 10Geneva University Hospital, Division of General Antoine Be´cle`re, service de pneumologie, Assistance Publique Ho ˆ tel Dieu de Paris, service de pneumologie, Assistance Publique Ho ˆ pitaux de Paris, Paris, France; 12CHU Internal Medicine, Geneva, Switzerland; 11Ho ˆ pital Europe´en Georges Pompidou, service de biochimie, St-Etienne Bellevue, service des urgences et re´animation me´dicale, St-Etienne, France; 13Ho ˆ pitaux de Paris, Paris, France Assistance Publique Ho

Rationale: The short-term prognosis of pulmonary embolism (PE) depends on hemodynamic status and underlying disease. The prognostic value of right ventricular dysfunction and injury is less well established. Objectives: To evaluate prognostic factors of PE in a multicenter prospective cohort study. Methods: Echocardiography, brain natriuretic peptide (BNP), Nterminal–proBNP and cardiac troponin I measurements were done on admission of 570 consecutive patients with an acute PE. A predictive model was based on independent predictors of 30-day adverse events defined as death, secondary cardiogenic shock, or recurrent venous thromboembolism. Measurements and Main Results: At 30 days, 42 patients (7.4%; 95% confidence interval [CI], 5.5–9.8%) had adverse events. On multivariate analysis, altered mental state (odds ratio [OR] 6.8; 95% confidence interval [CI], 2.0–23.3), shock on admission (OR 2.8; 95% CI, 1.1–7.5), cancer (OR 2.9; 95% CI, 1.2–6.9), BNP (OR 1.3 for an increase of 250 ng/L; 95% CI, 1.1–1.6) and right to left ventricle diameter ratio (OR 1.2 for an increase of 0.1; 95% CI, 1.1–1.4) were associated with 30-days of adverse events. The predictive performance of the model was good (area under receiver operating characteristics curve 0.84 [95% CI, 0.78–0.90]), making it possible to develop a bedside prognostic score. Conclusions: BNP and echocardiography may be useful determinants of the short-term outcome for patients with PE, together with clinical findings. Patients with PE can be stratified according to the initial risk of adverse outcome, using a simple score based on clinical, echocardiographic, and biochemical variables. Keywords: echocardiography; natriuretic peptides; prognosis; pulmonary embolism

Early mortality rates for pulmonary embolism (PE) range from 5% in patients who are clinically stable to 58% in patients with cardiogenic shock (1, 2). Some patients with PE may be con(Received in original form June 27, 2009; accepted in final form November 6, 2009) Supported by a grant of the Chancellerie des Universite´s (Legs Poix). * These two authors contributed equally to this work. Correspondence and requests for reprints should be addressed to Olivier Sanchez, ˆ pital Europe´en Georges M.D., Service de Pneumologie et Soins Intensifs, Ho Pompidou, 20 rue Leblanc 75015 Paris, France. E-mail: olivier.sanchez@egp. aphp.fr This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org Am J Respir Crit Care Med Vol 181. pp 168–173, 2010 Originally Published in Press as DOI: 10.1164/rccm.200906-0970OC on November 12, 2009 Internet address: www.atsjournals.org

AT A GLANCE COMMENTARY Scientific Knowledge on the Subject

Recent guidelines suggest that initial risk stratification of patients with pulmonary embolism (PE) should be based on the presence of shock and hypotension. Further stratification based on imaging or biochemical markers of right ventricular dysfunction or injury has been suggested, but current evidence is insufficient to make definitive recommendations on this point. What This Study Adds to the Field

BNP and echocardiography may be useful determinants of the short-term outcome for patients with PE, together with clinical findings. PE patients may be stratified according to the initial risk of adverse outcome, using a simple score based on clinical, echocardiographic and biochemical variables.

sidered for outpatient treatment, whereas others require hospital admission and even careful monitoring in the intensive care unit. Risk stratification tools may help to distinguish between these different categories of patient. The short-term prognosis of PE has been reported to depend on hemodynamic status and underlying disease (1, 3). Clinical variables have been used to construct a prognostic score that has been validated in different settings (3, 4). It has been reported that right ventricular dysfunction, detected by echocardiography, computed tomography, or high levels of brain natriuretic peptide (BNP) or N-terminal–proBNP (NT-proBNP), and myocardial injury, detected by high levels of cardiac troponins, are associated with a higher risk of short-term death (5–7). These results should be interpreted with caution, given the methodological limitations and the clinical diversity of studies (7). Recent guidelines suggest that initial risk stratification of patients with PE should be based on the presence of shock and hypotension (8). Further stratification based on imaging or biochemical markers of right ventricular dysfunction or injury has been suggested, but current evidence is insufficient to make definitive recommendations on this point (8, 9). We therefore performed this prospective cohort study to assess the additional prognostic value of echocardiography and biomarkers for stratifying patients with PE in different risk categories.

Prognostic Factors for Pulmonary Embolism

METHODS A detailed description of the methods is available in the online supplement.

Study Design, Setting This prospective multicenter observational cohort study took place between January 2006 and May 2007, at 11 academic centers. The study was approved by local ethics committees. All patients provided written informed consent.

Patients Patients were eligible if they were at least 18 years of age and had objectively confirmed PE according to current guidelines (8). Noneligibility criteria were: curative anticoagulant treatment for more than 24 hours, unavailability for follow-up, inability to give informed consent, life-expectancy less than 1 month. During the study period, 981 consecutive patients were admitted for an objectively confirmed PE in the participating centers. Among them, 342 (35%) were not eligible, whereas 47 (5%) refused to participate. Overall, 592 consecutive patients with objectively confirmed PE were included. Finally, we excluded 22 patients with normotensive PE from the principal analysis because they were receiving fibrinolytic treatment. The final study population therefore comprised 570 patients (Figure 1). Cardiogenic shock on admission was defined by at least one of the following criteria: systolic blood pressure (SBP) less than 90 mm Hg, signs of end-organ hypoperfusion, or a need for catecholamine administration to maintain SBP greater than 90 mm Hg. Patients were managed according to the usual practice of each participating center, by physicians blind to the results obtained for cardiac biomarkers. According to the current guidelines, it was suggested that fibrinolytic treatment should not be used in patients with hemodynamically stable PE (9).

Echocardiography Echocardiography was performed, within 24 hours of PE diagnosis, by an echocardiographer blind to the results of cardiac biomarker determinations. The end-diastolic diameters of the right and left ventricles (RV and LV) were measured in the long axis parasternal view and the RV/LV ratio was calculated.

Cardiac Biomarker Determinations On admission, blood samples were collected and were centrifuged at 3,000 3 g for 15 minutes. The resulting plasma was frozen and stored at

169

2808C. At the end of the study, circulating levels of cTnI, NT-proBNP, and BNP were determined centrally by investigators blind to the patients’ baseline characteristics and clinical outcome.

Outcomes Adverse clinical events were defined as all-cause death, secondary cardiogenic shock as previously defined, or objectively confirmed recurrent venous thromboembolism during a 30-day clinical followup. All adverse events and the cause of death (i.e., related or unrelated to PE) were adjudicated by an independent committee of two physicians unaware of the results of the initial clinical examination, echocardiography, and biomarker determinations.

Statistical Methods Model construction. Categorical variables are presented as numbers and percentages, and continuous variables are presented as means 6 SD or medians (25–75% percentile]. Univariate analyses, based on chisquared tests or student’s t tests, were performed. Independent associations with the outcome were assessed by including variables with a significance level of P less than 0.20 on univariate analysis in a multivariate logistic regression model. Variables associated with the outcome at a significance level of P less than 0.05 in backward stepwise regression analysis were retained. In the final model, we estimated, for each variable, the proportion of explained variation (10). Internal validity. We used the area under the receiver operating characteristic curve to quantify the ability of the final model to distinguish high-risk subjects from low-risk subjects and the Brier score to quantify the accuracy of predictions. Both these measures were calculated for the overall population and the group of patients without shock. Bootstrapping was used to estimate the internal validity of the model. Missing data. We performed a multiple imputation analysis to ensure that the analyses were consistent. Reclassification of patients experiencing and not experiencing 30-day adverse events. The greater discriminative value of BNP and RV/LV ratio was further examined with the method described by Pencina and colleagues (11). Bedside tool. We developed a prognostic score by assigning the variables retained in the final multivariate model a weighted number of points proportional to the regression coefficient values (rounded to the nearest integer) (12). The constant of the scoring system, corresponding to one point, was defined as the increase in risk associated with an increase of 0.1 in RV/LV ratio. A risk score was then calculated for each patient.

Figure 1. Flow-chart of the study.

170

AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 181

RESULTS

2010

The characteristics of the study population are shown in Table 1. PE was diagnosed on the basis of spiral computed tomography (CT) in 471 patients (83%), ventilation/perfusion (V/Q) lung scan in 80 patients (14%), proximal deep vein thrombosis, and high clinical suspicion of PE in 15 cases (2.5%), echocardiography and a high clinical probability of PE in 3 patients, and pulmonary angiography in 1 patient. At inclusion, 566 patients (99%) were receiving anticoagulant treatment. Fibrinolytic therapy was administered to 31 patients (5%), due to cardiogenic shock at inclusion in 23 patients and secondary cardiogenic shock in 8 patients. A vena cava filter was inserted in 25 patients (4%).

adverse events: death in 26 patients (4.6%; 95% CI, 3.1– 6.6%), secondary cardiogenic shock in 25 (4.4%) and recurrent venous thromboembolism in 11 patients (1.9%). One patient may have had several events qualifying for complicated outcome. Death was considered to be related to PE in 12 patients (46%). The other causes of death were cancer (n 5 6), sepsis (n 5 2), major bleeding (n 5 2) and other causes (n 5 4). Seventeen of the 41 patients with cardiogenic shock on admission (41.5%; 95% CI, 27.8–56.6%) experienced an adverse event, whereas such events were observed in only 25 of the 529 clinically stable patients (4.7%; 95% CI, 3.2–6.9%; P , 0.0001). The 30-day mortality rate was also higher in patients with shock (26.8%; 95% CI, 14.2–42.9%) than in clinically stable patients (2.8%; 95% CI, 1.7–4.6; P , 0.0001).

Outcome

Echocardiography and Cardiac Biomarkers

The 30-day follow-up was complete for all patients. During follow-up, 42 patients (7.4%; 95% CI, 5.5–9.8%) suffered

Median RV/LV ratio, BNP, NT-proBNP and cTnI levels were significantly higher among patients experiencing 30-d adverse

Patients

TABLE 1. CLINICAL FINDINGS AT DIAGNOSIS FOR THE 570 INCLUDED PATIENTS 30-Day Adverse Event Variable Age, years, median (25th–75th percentiles) Male, n (%) Comorbid conditions, n (%) Chronic congestive heart failure Chronic coronary insufficiency Chronic lung disease Cancer Previous VTE Chronic renal failure Major surgery or trauma within the last month Seated travel .6 hours within the last month Pregnancy or 1 month post-partum Clinical findings, n (%) Acute dyspnea Chest pain Hemoptysis Altered mental state* Syncope Signs of right heart failure Heart rate, median (25th–75th percentiles), bpm Respiratory rate, median (25th–75th percentiles), bpm Systolic blood pressure, median (25th–75th percentiles), mm Hg Cardiogenic shock on admission† Median 8C temperature (25th–75th percentiles) Positive spiral CT‡ Main pulmonary artery Lobar artery Segmental artery Multi-subsegmental arteries Echocardiography and cardiac biomarkers, median (25th–75th percentiles) RV/LV ratio BNP, ng/L NT-proBNP, ng/L Troponin I, mg/L Treatment, n (%) Fibrinolysis Surgical embolectomy Inferior vena cava filter Volemic expansion

Overall (n 5 570)

No (n 5 528)

Yes (n 5 42)

P Value

68 (52–77) 268 (47)

68 (52–77) 247 (47)

70 (57–80) 21 (50)

0.22 0.75

27 39 36 87 142 26 70 60 11

(5) (7) (6) (15) (25) (5) (12) (11) (2)

23 34 31 75 134 21 59 59 10

(4) (6) (6) (14) (25) (4) (11) (11) (2)

4 5 5 12 8 5 11 1 1

0.13 0.20 0.17 0.02 0.46 0.04 0.01 0.11 1.00

461 265 25 18 41 113 89 21

(81) (46) (4) (3) (7) (20) (75–101) (18–26)

423 258 23 9 30 96 88 21

(80) (49) (4) (2) (6) (18) (75–100) (18–26)

38 7 2 9 11 17 100 24

(10) (12) (12) (29) (19) (12) (26) (2) (2) (90) (17) (5) (21) (26) (40) (84–120) (20–28)

0.15 ,0.0001 0.71 ,0.0001 ,0.0001 0.002 0.0005 0.10

132 (120–148)

134 (120–149)

126 (108–139)

41 37.1 471 150 183 120 18

(7) (36.8–37.5) (83) (32) (39) (25) (4)

24 37.1 440 139 171 113 17

(5) (36.8–37.5) (83) (32) (39) (26) (3)

17 37.1 31 11 12 7 1

(40) (36.4–37.5) (74) (35) (39) (23) (3)

,0.0001 0.41

0.7 78 384 0.01

(0.5–0.8) (29–261) (73–2,031) (0–0.08)

0.7 74 349 0.01

(0.5–0.8) (28–204) (69–1,633) (0–0.06)

0.9 286 3,743 0.18

(0.7–1.0) (115–743) (594–8,385) (0.03–1.03)

,0.0001 ,0.0001 ,0.0001 0.0009

(40) (2) (12) (43)

,0.0001 0.02 0.01 ,0.0001

31 2 25 50

(5) (,1) (4) (9)

14 1 20 32

(3) (,1) (4) (6)

17 1 5 18

0.003

1.00

Definition of abbreviations: BNP 5 brain natriuretic peptide; CT 5 computed tomography; LV 5 left ventricle; NT-proBNP 5 N-terminal–proBNP; RV 5 right ventricle; VTE 5 venous thromboembolic disease. * Altered mental state was defined as disorientation, stupor or coma. † Shock was defined as at least one of the following: systolic blood pressure , 90 mmHg, signs of systemic hypoperfusion (at least: clouded sensorium, oliguria, cold and clammy skin), need for hemodynamic support. ‡ Most proximal anatomic level of pulmonary embolism shown by spiral computed tomography.

Prognostic Factors for Pulmonary Embolism

171

TABLE 2. RESULTS OF ECHOCARDIOGRAPHY AND CARDIAC BIOMARKER DETERMINATIONS AS A FUNCTION OF HEMODYNAMIC STATUS AT INCLUSION AND 30-DAY OUTCOME Cardiogenic Shock on Admission Yes (n 5 41)

No (n 5 529)

Available data, n

Overall

No* (n 5 24)

Yes* (n 5 17)

P Value

No* (n 5 504)

Yes* (n 5 25)

P Value

BNP, ng/L

562

Troponin I, mg/L

558

RV/LV ratio

522

735 (142;1,277) 9,381 (1,395;15,464) 0.91 (0.13;2.84) 1.0 (0.7;1.1)

67 (27;190) 287 (63;1,330) 0.01 (0.00;0.05) 0.7 (0.5;0.8)

275 (69;431) 3,117 (537;4,086) 0.01 (0.01;0.34) 0.8 (0.6;1.1)

,0.001

558

340 (141;556) 4,266 (1,964;9,335) 0.56 (0.14;1.52) 1.0 (0.9;1.2)

0.08

NT–pro-BNP, ng/L

79 (29;261) 380 (73;2,031) 0.01 (0.00;0.08) 0.7 (0.5;0.8)

0.38 0.53 0.54

,0.001 0.001 0.005

Definition of abbreviations: BNP 5 brain natriuretic peptide; LV 5 left ventricle; NT–pro-BNP 5 N-terminal–proBNP; RV 5 right ventricle. * 30-Day complicated outcome. Values are median (25th;75th percentile).

events than among those not experiencing such events (Table 1). Results of echocardiography and cardiac biomarkers according to hemodynamic status at inclusion and 30-day adverse events are presented in Table 2. Risk Factors for 30-day Adverse Events

Two multivariate logistic regression models were constructed: a clinical model including clinical variables only and a final model including clinical variables, echographic, and biological variables. For the clinical model, altered mental state, cardiogenic shock, and cancer were independently associated with 30-day adverse events. If cardiac biomarkers and echocardiography were included as candidate variables, altered mental state, cardiogenic shock, cancer, BNP and RV/LV ratio were independently associated with 30-day adverse events in the final multivariate model estimated for the 515 patients for whom data were complete (Table 3). The variables included in the model accounted for 20% of the variation in individual 30-day outcome. Lastly, the variable selection process yielded the same model with multiple imputed data, giving estimates consistent with the complete case analysis (Table 3). The predictive performance of the final model was good. For the total population of 570 patients, the area under the curve was 0.84 (95% CI, 0.78–0.90) and the Brier score was 0.056 (Figure 2). The internal validity of the model was checked by bootstrapping. Optimism, which is the tendency of the model to perform better with the data from which it was constructed than on new data, was low. The internally validated area under the curve and Brier score were 0.80 and 0.066, respectively. In the 529 patients without shock, the AUC and the Brier score were 0.79 (95% CI, 0.68–0.87) and 0.04, respectively.

Reclassifications for patients with and without 30-day adverse events are summarized in Table 4. For patients with 30-day adverse events, classification with the final model was more accurate for 12 patients and less accurate for 1 patient than classification with the clinical model. For patients who did not experience a 30-day adverse event, classification with the final model was less accurate for 69 patients and more accurate for 7 patients than classification with the clinical model. The net reclassification improvement with the final model was estimated at 14.5%, which was highly significant (P , 0.001). The addition of BNP and RV/LV ratio to the clinical model improved classification for 27.5% of patients with events, and moderately overestimated the risk for 13.0% of patients who did not experience an adverse event during follow-up. In a similar analysis on patients without shock at admission, the net reclassification improvement with the final model was estimated at 26.5% (P , 0.001; see Table E1 in the online supplement). Risk Score

We developed a practical prognostic score for use as a bedside tool (Table 5). We assigned patients to three risk classes for 30-day adverse events: class I, low risk (predicted risk ,5% or score ,7); class II, intermediate risk (predicted risk 5–30% or score 7–17); and class III, high risk (predicted risk >30% or score >18). For the total study population, the observed risk class specific230-day adverse event frequency was 2.5% in class I (8/323 patients), 11.6% in class II (18/155 patients), and 43.2% in class III (16/37 patients). In patients without shock, the observed risk class specific230-day adverse event frequency was 1.8% in class I (6/323 patients), 11.7% in class II (17/145 patients), and 22.2% in class III (2/9 patients).

TABLE 3. RISK FACTORS FOR 30-DAY ADVERSE EVENTS IN MULTIVARIATE ANALYSIS Complete Case Analysis (n 5 515) Variable

OR (95% CI) †

Altered mental state Cardiogenic shock on admission Cancer BNP ([ of 250 ng/L) RV/LV ratio ([ of 0.1)

6.8 2.8 2.9 1.3 1.2

(2.0–23.3) (1.1–7.5) (1.2–6.9) (1.1–1.6) (1.1–1.4)

Multiple Imputation Analysis (n 5 570)

P Value

PEV* (%)

Partial PEV* (%)

OR (95% CI)

P Value

,0.01 0.03 0.02 ,0.01 ,0.01

9.0 11.2 0.7 6.3 8.7

3.2 2.7 0.1 2.4 0.3

6.8 (2.0–25.5) 3.5 (1.4–9.0) 3.1 (1.3–7.2) 1.3 (1.1–1.6) 1.2 (1.1–1.4)

,0.01 ,0.01 ,0.01 ,0.01 ,0.01

Definition of abbreviations: BNP 5 brain natriuretic peptide; CI 5 confidence interval; LV 5 left ventricle; OR 5 odds ratio; PEV 5 proportion of explained variation; RV 5 right ventricle. * PEV is the proportion of the total variability of the outcome variable attributable to the variable concerned. Partial PEV measures the decline in explained variation when the prognostic factor is removed from the model containing the other four factors. † Altered mental state was defined as disorientation, stupor, or coma. Hosmer and Lemeshow Goodness-of-Fit Test statistic was used for the complete case analysis: chi-square (8 degrees of freedom) 5 9.11; P value 5 0.33.

172

AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 181

2010

TABLE 4. RECLASSIFICATION OF PATIENTS WHO DID AND DID NOT EXPERIENCE A 30-DAY ADVERSE EVENT (COMPLETE CASE ANALYSIS, N 5 515) Clinical Model: Cardiogenic Shock on Admission 1 Altered Mental State 1 Cancer

Clinical Model 1 BNP 1 RV/LV Ratio ,5%

5–30%

Patients experiencing a 30-d adverse event ,5% 6 10 5–30% 1 10 >30% 0 0 Total 7 20 Patients not experiencing a 30-d adverse event ,5% 325 59 5–30% 4 68 >30% 0 3 Total 329 130

Figure 2. Receiver operating characteristic curves for 30-day adverse events. The curves are based on logistic regression models for risk prediction incorporating: Model 1, cardiogenic shock on admission 1 altered mental state 1 cancer. Area under the curve (AUC) 0.73; 95% CI, 0.65–0.82; Model 1 1 brain natriuretic peptide (BNP). AUC 0.82; 95% CI, 0.75–0.89. P 5 0.002 versus Model 1. Model 1 1 BNP 1 RV/LV ratio. AUC 0.84; 95% CI, 0.78–0.90. P 5 0.003 versus Model 1. P 5 0.21 versus Model 1 1 BNP. RV/LV 5 right ventricle/left ventricle.

DISCUSSION In this large population of patients with acute PE, hemodynamic status (cardiogenic shock and altered mental state as a possible consequence of low cardiac output) and underlying disease (cancer) were the main independent clinical prognostic factors for 30-day major adverse events. Adding right ventricular dysfunction, as assessed by echocardiography and BNP levels, to the clinical model substantially improved the risk stratification for 30day adverse events in the overall population but also in patients with clinically stable PE. Right ventricular dysfunction provided additional prognostic information independent of clinical findings. Our results confirmed previous findings, suggesting that altered mental state, cardiogenic shock, and cancer are associated with adverse outcomes (3). Other studies suggest that right ventricular dysfunction, is of significant prognostic value in patients with PE (5–7). However, these studies are limited by the small number of patients, the lack of adjustment for other important prognostic variables, such as clinical variables and, in most cases, the absence of an independent assessment of outcome and test results. Our study included a large number of patients, and outcomes were adjudicated by an independent committee unaware of the results of the initial clinical examination, echocardiography, and biomarker determinations. Our results do not suggest that markers of myocardial injury, such as troponin, are independently associated with a higher risk of adverse events. Reviewing recent studies, Becattini and colleagues found that high troponin concentrations were associated with a higher mortality rate (5). We found that median cTnI levels were significantly higher in patients experiencing adverse events, but this association did not remain significant after adjustment for other covariables, and there was no adjustment for confounding factors in Becattini’s meta-analysis (5). We chose an objective, unbiased, composite outcome that reflected the severity of the disease. Unlike other studies, we included not only mortality, but also major complications, such as

>30%

Total

1 1 11 13

17 12 11 40

3 7 6 16

387 79 9 475

Definition of abbreviations: BNP 5 brain natriuretic peptide; LV 5 left ventricle; RV 5 right ventricle. The greater discriminative value of BNP and RV/LV ratio was investigated further with the method described by Pencina and colleagues (13) and was based on cross-tabulation of the classifications of the predicted risks obtained with the two models (clinical model and clinical model 1 BNP 1 RV/LV ratio) into three categories (,5, 5–30, and >30% 30-d risk). Patients who did and did not experience adverse events were considered separately. For patients with adverse events, reclassification to a higher risk category (cells in bold italic) indicates more accurate prediction by the model. Conversely, reclassification to a lower risk category (cells in bold) indicates less accurate prediction by the model. For patients without adverse events, the opposite interpretation is applied (reclassification to a higher risk group [cells in bold italic] 5 less accurate prediction; reclassification to a lower risk group [cells in bold] 5 more accurate prediction). The improvement in reclassification was quantified by calculating the difference of two differences: (1) the difference between the proportions of individuals reclassified to a higher risk group and the proportion of individuals reclassified to a lower risk group, for individuals with adverse events ([12–1]/40) and (2) the corresponding difference in proportions for individuals without adverse events ([69–7]/475).

cardiogenic shock and recurrent PE. These endpoints are associated with a poor outcome in patients with PE and preventive measures therefore need to be taken during treatment of the initial episode. Most of the adverse events were consequences of the PE itself (recurrent PE, secondary cardiogenic shock, and death due to PE) rather than due to underlying disease. There are two goals for risk stratification in patients with PE: (1) the selection of patients with a low risk of complications who could potentially be managed as outpatients and (2) the selection of normotensive patients with a high risk of complication not suggested by the usual clinical predictors and who may be considered candidates for thrombolytic treatment according to some authorities (8). The predictive performance of our model remains good in the subgroup of normotensive patients (AUC 0.79; 95% CI, 0.68–0.87). In these patients, the presence of right ventricular dysfunction assessed by echocardiography and BNP increases the risk of complicated outcome from 1.8% in class I patients to 11.7% in class II, and 22.2% in class III patients. Our study has several limitations. First, the number of adverse events was small and the stability of the prognostic model is therefore debatable. However, different analyses with and without cardiac biomarkers and echocardiography variables, and with and without multiple imputation, gave similar results. Second, we excluded from the analysis 22 normotensive patients receiving thrombolytic therapy. The benefit of thrombolytic treatment and its role on prognosis in these patients remains unclear, and current guidelines recommend against the use of thrombolytic therapy for patients with normotensive PE (8, 9). According to these guidelines, our study protocol suggested that fibrinolytic treatment should not be used in patients with normotensive PE. Therefore, we decided a priori

Prognostic Factors for Pulmonary Embolism

173

TABLE 5. RISK SCORE FOR 30-DAY ADVERSE EVENTS Prognostic Factor

Categories

Points

No Yes

0 10

No Yes

0 6

No Yes

0 6

,100 100–249 250–499 500–999 >1,000

0 1 2 4 8

0.2–0.49 0.5–0.74 0.75–1.00 1.00–1.25 >1.25

0 3 5 8 11

Altered mental state *

Cardiogenic shock on admission

Cancer

BNP (ng/L)

RV/LV ratio

See Table 2 for definition of abbreviations. * Altered mental state was defined as disorientation, stupor, or coma. Range of total prognostic score, 0–41. The points assigned correspond to the following risk classes: 18 5 class III, high risk.

to exclude from the analysis patients with normotensive PE who received thrombolytic treatment to be able to analyze properly prognostic factors in these patients without any influence of a treatment on prognosis. However, in our observational cohort study, patients were managed according to the usual practice of each participating center, which accounts for the fact that some normotensive patients received thrombolytic therapy. These patients were recruited at 4 of the 11 participating centers. None of these patients experienced a 30-day adverse event. Their mean RV/LV ratio was significantly higher than that of the 529 clinically stable patients receiving anticoagulant alone (0.99 6 0.2 vs. 0.69 6 0.2; P , 0.0001), suggesting that echocardiography results had probably influenced the decision to initiate thrombolytic therapy in these patients. However, the results of multivariate logistic regression analysis were similar with and without the inclusion of these 22 normotensive patients on fibrinolytic treatment (see Table E2 in the online supplement). Third, our risk score requires an echocardiography to be performed, which is not always available or feasible due to technical reasons. However, the RV/LV ratio can also be measured on spiral CT used to diagnose PE. As previously demonstrated, this measurement is associated with increased risk of early death in patients with PE (7). Lastly, the small number of outcome events precluded assessment of the risk score in an internal validation cohort. Our risk stratification score therefore requires external validation in an independent cohort. In conclusion, the results of this study suggest that BNP and echocardiography may be useful determinants of the short-term outcome of patients with PE, together with clinical findings. PE patients can be stratified according to the initial risk of adverse outcomes, using a simple score based on clinical, echocardiographic, and biochemical variables. If validated in independent cohorts, this risk stratification score may be an attractive tool for selecting patients for outpatient treatment and identifying patients requiring more careful follow-up at hospital, for whom additional treatment may be necessary. Conflict of Interest Statement: O.S. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. L.T. does not have a financial relationship with a commercial entity that has an

interest in the subject of this manuscript. V.C. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. F.C. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. G.P. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. N.M. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. F.V. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. P-M.R. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. F.P. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. M.R. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. A.P. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. C.L. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. B.T. received $10,001–$50,000 from Organon France (research grants NCT00748839) and $50,001–$100,000 (research grants NCT00748839) from Sanofi in industrysponsored grants. M-O.B. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. G.C. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. G.M. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. Acknowledgment: The authors thank Daniel Pontal, Noe¨l Lucas, Jean-Franc xois Leforestier, and Dominique Mariole in Paris for their assistance with data collection and their invaluable collaboration throughout the study. The authors also thank the members of adjudication committee Joseph Emmerich and Philippe Girard for their important contribution.

References 1. Goldhaber SZ, Visani L, De Rosa M. Acute pulmonary embolism: clinical outcomes in the International Cooperative Pulmonary Embolism Registry (ICOPER). Lancet 1999;353:1386–1389. 2. Quinlan DJ, McQuillan A, Eikelboom JW. Low-molecular-weight heparin compared with intravenous unfractionated heparin for treatment of pulmonary embolism: a meta-analysis of randomized, controlled trials. Ann Intern Med 2004;140:175–183. 3. Aujesky D, Obrosky DS, Stone RA, Auble TE, Perrier A, Cornuz J, Roy PM, Fine MJ. Derivation and validation of a prognostic model for pulmonary embolism. Am J Respir Crit Care Med 2005;172:1041–1046. 4. Donze J, Le Gal G, Fine MJ, Roy PM, Sanchez O, Verschuren F, Cornuz J, Meyer G, Perrier A, Righini M, et al. Prospective validation of the pulmonary embolism severity index. A clinical prognostic model for pulmonary embolism. Thromb Haemost 2008;100:943–948. 5. Becattini C, Vedovati MC, Agnelli G. Prognostic value of troponins in acute pulmonary embolism: a meta-analysis. Circulation 2007;116:427–433. 6. Klok FA, Mos IC, Huisman MV. Brain-type natriuretic peptide levels in the prediction of adverse outcome in patients with pulmonary embolism: a systematic review and meta-analysis. Am J Respir Crit Care Med 2008;178:425–430. 7. Sanchez O, Trinquart L, Colombet I, Durieux P, Huisman MV, Chatellier G, Meyer G. Prognostic value of right ventricular dysfunction in patients with haemodynamically stable pulmonary embolism: a systematic review. Eur Heart J 2008;29:1569–1577. 8. Torbicki A, Perrier A, Konstantinides S, Agnelli G, Galie N, Pruszczyk P, Bengel F, Brady AJ, Ferreira D, Janssens U, et al. Guidelines on the diagnosis and management of acute pulmonary embolism: the Task Force for the Diagnosis and Management of Acute Pulmonary Embolism of the European Society of Cardiology (ESC). Eur Heart J 2008;29:2276–2315. 9. Kearon C, Kahn SR, Agnelli G, Goldhaber SZ, Raskob G, Comerota J. Antithrombotic therapy for venous thromboembolic disease: American College of Chest Physicians evidence-based clinical practice guidelines (8th edition). Chest 2008;133:454–545. 10. Heinze G, Schemper M. Comparing the importance of prognostic factors in Cox and logistic regression using SAS. Comput Methods Programs Biomed 2003;71:155–163. 11. Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008; 27:157–172, discussion 207–212. 12. Sullivan LM, Massaro JM, D’Agostino RB Sr. Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med 2004;23:1631–1660. 13. Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008; 27:157–172.