Accuracy of Clinical Assessment in the Diagnosis of Pulmonary Embolism

Transcription

1 Accuracy of Clinical Assessment in the Diagnosis of Pulmonary Embolism MASSIMO MINIATI, RENATO PREDILETTO, BRUNO FORMICHI, CARLO MARINI, GIORGIO DI RICCO, LUCIA TONELLI, GERMANA ALLESCIA, and MASSIMO PISTOLESI Istituto di Fisiologia Clinica del Consiglio Nazionale delle Ricerche (CNR), Pisa, Italy To provide clinical diagnostic criteria for pulmonary embolism (PE), we evaluated 750 consecutive patients with suspected PE who were enrolled in the Prospective Investigative Study of Acute Pulmonary Embolism Diagnosis (PISA-PED). Prior to perfusion lung scanning, patients were examined independently by six pulmonologists according to a standardized diagnostic protocol. Study design required pulmonary angiography in all patients with abnormal scans. Patients are reported as two distinct groups: a first group of 500, whose data were analyzed to derive a clinical diagnostic algorithm for PE, and a second group of 250 in whom the diagnostic algorithm was validated. PE was diagnosed by angiography in 202 (40%) of the 500 patients in the first group. A diagnostic algorithm was developed that includes the identification of three symptoms (sudden onset dyspnea, chest pain, and fainting) and their association with one or more of the following abnormalities: electrocardiographic signs of right ventricular overload, radiographic signs of oligemia, amputation of hilar artery, and pulmonary consolidations compatible with infarction. The above three symptoms (singly or in some combination) were associated with at least one of the above electrocardiographic and radiographic abnormalities in 164 (81%) of 202 patients with confirmed PE and in only 22 (7%) of 298 patients without PE. The rate of correct clinical classification was 88% (440/500). In the validation group of 250 patients the prevalence of PE was 42% (104/250). In this group, the sensitivity and specificity of the clinical diagnostic algorithm for PE were 84% (95% CI: 77 to 91%) and 95% (95% CI: 91 to 99%), respectively. The rate of correct clinical classification was 90% (225/250). Combining clinical estimates of PE, derived from the diagnostic algorithm, with independent interpretation of perfusion lung scans helps restrict the need for angiography to a minority of patients with suspected PE. Miniati M, Prediletto R, Formichi B, Marini C, Di Ricco G, Tonelli L, Allescia G, Pistolesi M. Accuracy of clinical assessment in the diagnosis of pulmonary embolism. AM J RESPIR CRIT CARE MED 1999;159: The clinical diagnosis of pulmonary embolism (PE) is thought to be unreliable because symptoms, signs, and laboratory data to support the diagnosis are often deceivingly nonspecific (1 3). However, some early studies recognized that clinical findings such as unexplained dyspnea, tachypnea, or chest pain, although nonspecific for PE, are useful to select patients for further diagnostic testing (4 6). The importance of clinical assessment is further supported by the results of two broad prospective studies on the diagnosis of PE, the PIOPED (7) and the PISA-PED study (8). In the PIOPED study (7), a multicenter trial designed to establish the diagnostic value of the ventilation-perfusion (Received in original form June 22, 1998 and in revised form August 11, 1998) Supported by the Ministry of Health and the Ministry of University and Scientific and Technological Research of Italy. Correspondence and requests for reprints should be addressed to Massimo Miniati, M.D., Reparto Polmonare, Istituto di Fisiologia Clinica del CNR, Via P. Savi 8, Pisa, Italy. Am J Respir Crit Care Med Vol 159. pp , 1999 Internet address: scan, a clinical probability of PE was rated, prior to lung scanning, as low (0 19%), intermediate (20 79%), or high (80 100%). PE was diagnosed in 68% of the patients with high clinical probability and in only 9% of those who were assigned a low clinical probability. Even more impressive results were obtained in the PISA- PED study (8), an investigation aimed at assessing the value of the perfusion lung scan (without ventilation imaging) in the diagnosis of PE. In that study, patients referred for lung scanning were examined independently by six pulmonary specialists according to a standardized diagnostic protocol (8). Prior to lung scanning, a clinical probability of PE was assigned out of three alternatives: very likely (90%), possible (50%), or unlikely (10%). PE was diagnosed in only 9% of the patients with an unlikely clinical presentation, and in 91% of those in whom the disease was considered very likely on clinical grounds. In those patients in whom PE was rated as possible, the prevalence of the disease was 47%. Thus, it appears that physicians estimates of the clinical likelihood of PE, even if based on empirical assessment, do have predictive value. The characteristics of these clinical estimates, however, have not been described in the reported stud-

2 Miniati, Prediletto, Formichi, et al.: Clinical Diagnosis of Pulmonary Embolism 865 ies (7, 8), so there is no way of knowing whether these estimates can be replicated by others (9). The aim of the present study was to provide clinical diagnostic criteria for PE that may be used as guidelines for nonexperts in the field. Furthermore, we wanted to establish whether a diagnostic strategy, based on combining well-characterized clinical estimates of PE with perfusion lung scan interpretation, may help to forego pulmonary angiography in patients suspected of having PE. To this end, we analyzed clinical, electrocardiographic, radiographic, and arterial blood gas data pertinent to patients enrolled in the PISA-PED study (8). METHODS Patient Population The study population consisted of 750 consecutive patients referred to our Institute for suspected PE; they were enrolled in the PISA-PED study, and a definite diagnosis (confirmation or exclusion of the disease) was established for each. For the sake of the present study, patients are considered as two distinct groups: a first group of 500 patients, whose data were analyzed to derive a clinical diagnostic algorithm for PE, and a second group of 250 patients in whom the accuracy of the diagnostic algorithm was assessed prospectively. All patients were examined uniformly according to a standardized protocol described below. Study Protocol Clinical evaluation. Upon study entry, patients were examined by one of six pulmonologists who served as on-call physicians one day a week. All of the pulmonary specialists who took part in the study had experience of the diagnostic procedures for PE. Clinical evaluation included detailed clinical history, physical examination, interpretation of the electrocardiogram and the chest radiograph, and measurement of the partial pressure of oxygen (Pa O2 ) and carbon dioxide (Pa CO2 ) in arterial blood. All clinical and laboratory data were recorded by the on-call physician on a standard form prior to any further objective testing (i.e., prior to lung scanning and, if required, pulmonary angiography). On interviewing the patients, care was taken to identify risk factors for PE and preexisting disease, which may mimic the clinical presentation of PE. In evaluating dyspnea, attention was paid to establish whether it was sudden or gradual in onset, and whether it was associated with orthopnea. Arterial blood samples were obtained upon study entry in all patients while they were breathing room air. Electrocardiograms obtained within 24 h of study entry were considered for evaluation by the on-call physician. The following abnormalities were regarded as suggestive of right ventricular (RV) overload: S1Q3 pattern (with or without T-wave inversion in lead III), S1S2S3 pattern, T-wave inversion in right precordial leads, transient right bundle branch block (RBBB), and pseudoinfarction. Carefully defined criteria were used to identify these electrocardiographic abnormalities (10). If any of the above abnormalities were present in electrocardiograms taken long before the onset of symptoms, they were disregarded. Chest radiographs were obtained in all patients at the time of study entry using a stationary x-ray unit. In most patients, posteroanterior and lateral views were taken in the upright or seated position at a 2-m focus to film distance. With patients who were unable to stand upright or seated (30% in our study), anteroposterior chest radiographs were obtained in the supine or semirecumbent position. Chest radiographs were examined by the on-call physician according to a reading table attached to the standard form. The reading table included the following items: size and shape of the heart and hilar arteries, position of the diaphragm, presence or absence of pulmonary parenchymal abnormalities (e.g., consolidation, atelectasis, oligemia, edema), presence or absence of pleural effusion. On evaluating the hilar arteries, attention was paid to the presence of abrupt vascular amputation, which gives the hilum a plump appearance (11, 12). In retrospective studies, this radiographic abnormality was found to be strongly associated with PE (13, 14). Pulmonary consolidations were considered compatible with infarction if they had a semicircular or half-spindle shape and were arranged peripherally along the pleural surface (so-called Hampton s hump) (15, 16). Oligemia was considered to be present if, in a given lung region, the pulmonary vasculature was greatly diminished with concomitant hyperlucency of the lung parenchyma (17). Perfusion lung scanning. All patients underwent perfusion lung scanning (without ventilation imaging) according to standard techniques (8). Perfusion lung scans were independently classified into one of four predetermined categories (8): (1) normal (no perfusion defects), (2) near-normal (impressions caused by enlarged heart, hila, or mediastinum are seen on an otherwise normal scan), (3) abnormal, compatible with PE or PE scan (single or multiple wedge-shaped perfusion defects), and (4) abnormal, not compatible with PE or PE scan (single or multiple perfusion defects other than wedge-shaped). By study design, pulmonary angiography was required in all patients with abnormal scans (8). Angiography was omitted in patients with normal or near-normal scans because available data indicate that such a scintigraphic pattern rules out clinically significant PE (18, 19). Pulmonary angiography. Pulmonary angiograms were performed according to standardized procedures described elsewhere (8), and were obtained within 24 h of the lung scan. Prior to angiography, all patients had to sign an informed consent form. Angiographic criteria for diagnosing PE included the identification of an embolus obstructing a vessel or the outline of an embolus within a vessel (filling defect) (8). Data Analysis Clinical and laboratory data, recorded by the on-call physicians, were stored in a computer to create a data base from which the prevalence and distributions of the chosen variables were derived. The chi-square test with Yates correction or Fisher s exact test were used in comparing the prevalence of variables assessed in a dichotomous fashion (present or absent). The unpaired t test was used in comparing continuous variable means. Statistical significance was set at a p value of less than 0.05; unless stated otherwise, continuous variables are reported in the text as mean SD. Sensitivity is defined as the proportion of patients classified as having PE among those with angiographically proven PE. Specificity is the proportion of patients classified as not having PE among those in whom the disease was excluded. Positive predictive value is the proportion of patients with confirmed PE among those classified as having PE. Negative predictive value is the proportion of patients without PE among those classified as not having PE. Ninety-five percent confidence intervals (95% CI) were calculated according to the binomial distribution. In patients with confirmed PE, a semiquantitative assessment of the severity of the disease was accomplished by counting the number of unperfused lung segments on the lung scan (20). This analysis was carried out by a nuclear medicine specialist who was blind to clinical information. Bayes theorem (21) was used to calculate the probability of PE conditioned by a given lung scan result (posterior or post-test probability) as a function of the clinical probability of PE (pretest probability). Calculations are given in the APPENDIX. RESULTS Patient Characteristics and Prevalence of PE The 500 patients, who had been referred for lung scanning with clinically suspected PE, were aged yr (range, 15 to 91 yr); 243 of them (49%) were male. Most (437/500, or 85%) were hospitalized at the time of study entry. The time lapse between onset of symptoms and enrollment in the study averaged 4.5 d (range, 1 40 d). Perfusion lung scans were normal or near-normal in 105 (21%) of 500 patients and abnormal in 395. Pulmonary angiography was performed in 391 of 395 patients with abnormal lung scans. Angiograms were positive for PE in 200 patients and negative in 191. In four patients, who died before angiog-

3 866 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL raphy could be done, the diagnosis was established at autopsy. Two of these patients had PE. Thus, based on angiography and autopsy data, the overall prevalence of PE in this patient population was 40% (202/500). Patients with confirmed PE did not differ in age and sex from those without PE (mean age, 63.3 and 64.1 yr, respectively; male 50 and 48%, respectively). Some 20% of the patients in either group were overweight (body mass index greater than 30 kg/m 2 ). There was a significantly higher prevalence of smokers in the group of patients without PE (33 versus 23% of those with PE, p 0.02). Preexisting Disease and Risk Factors for PE As shown in Table 1, there was a significantly higher prevalence of pulmonary disease among patients in whom PE was excluded. Most of the patients with preexisting pulmonary disorders (82%) had chronic obstructive pulmonary disease (COPD). There was no difference in the prevalence of cardiovascular, neoplastic, or endocrine disease in patients with versus those without PE. Ischemic heart disease and hypertension accounted for the majority of cardiovascular disorders. Endocrine disease included diabetes mellitus and thyroid dysfunction. Of the risk factors for PE listed in Table 1, prolonged immobilization, history of thrombophlebitis, and bone fractures of the lower extremities prevailed significantly in patients with confirmed PE. Although there was no difference in the prevalence of recent surgery between the two groups, a higher proportion of patients with PE had undergone orthopedic surgery (24 versus 14% of those without PE, p 0.008). At least one risk factor for PE among those listed in Table 1 was present in 164 (81%) of 202 patients with PE and in 206 (69%) of 298 without PE (p 0.004). TABLE 1 PREEXISTING DISEASE AND RISK FACTORS IN 500 PATIENTS WITH CLINICALLY SUSPECTED PULMONARY EMBOLISM (PE) PE Present (n 202) PE Absent (n 298) n (%) n (%) p Value Preexisting disease Cardiovascular 53 (26) 96 (32) 0.18 Pulmonary 14 (7) 72 (24) Neoplastic* 36 (18) 43 (14) 0.37 Endocrine 21 (10) 32 (11) 0.98 Risk factors Immobilization 3 d 119 (59) 138 (46) Surgery 81 (40) 116 (39) 0.72 Thrombophlebitis (ever) 69 (34) 57 (19) Bone fractures (lower limb) 46 (23) 36 (12) Estrogen use 1 (0.5) 1 (0.3) 1.0 Pregnancy/postpartum 1 (0.5) 4 (1) 0.34 * Clinically active malignancy with pathologic diagnosis within 3 mo of study entry. Within 4 wk of study entry. Symptoms and Signs Clinical symptoms and signs in 500 patients with suspected PE are reported in Table 2. Sudden onset dyspnea was by far the most frequent symptom in patients with PE, being reported by nearly 80% of the patients in this group. Only 2 (1%) of 202 patients with confirmed PE complained of orthopnea. Chest pain (either pleuritic or substernal) and fainting (with or without loss of consciousness) occurred in 60 and 26% of the patients with PE, respectively. Hemoptysis, a symptom often regarded as suggestive of PE, occurred in only a minority of the patients in whom the disease was diagnosed. Sudden onset dyspnea, chest pain, and fainting (singly or in combination) were present in 194 (96%) of 202 patients with PE and in 175 (59%) of 298 patients without PE (p ). There was no significant difference between the two groups of patients as to the prevalence of most of the physical signs reported in Table 2. Unilateral leg swelling (suggestive of deep vein thrombosis) prevailed in patients with PE, whereas fever and wheezes were more frequent in patients without PE. The second finding is consistent with the higher prevalence of COPD in patients in whom PE was excluded (Table 1). Arterial Blood Gas Data Among patients with PE, the Pa O2 was on average mm Hg (range, 40 to 101 mm Hg), and it was mm Hg (range, 35 to 129 mm Hg) in patients without PE (p 0.02). The Pa CO2 averaged 32 4 mm Hg (range, 20 to 42 mm Hg) in patients with PE and 34 5 mm Hg (range, 21 to 49 mm Hg) in patients without PE (p ). The frequency distributions of Pa O2 and Pa CO2 values in patients with and without PE are displayed in Figure 1. Ninety percent of the patients with PE had Pa O2 lower than 82 mm Hg and Pa CO2 lower than 37 mm Hg. Thus, it appears that PE is usually accompanied by arterial hypoxemia and hypocapnia. However, 50% of the patients in whom the disease was excluded had Pa O2 lower than 66 mm Hg and Pa CO2 lower than 34 mm Hg (Figure 1). Electrocardiographic Findings Electrocardiographic signs of RV overload were present in 100 (49.5%) of 202 patients with PE and in only 35 (12%) of 298 without PE (p ). Findings suggestive of RV overload in patients with PE were, in order of frequency, T-wave inversion in right precordial leads (23%), S1Q3/S1Q3T3 (19%), transient RBBB (9%), pseudoinfarction (6%), S1S2S3 (3%). TABLE 2 SYMPTOMS AND SIGNS IN 500 PATIENTS WITH CLINICALLY SUSPECTED PULMONARY EMBOLISM (PE) PE Present (n 202) PE Absent (n 298) n (%) n (%) p Value Symptoms Dyspnea (sudden onset) 158 (78) 87 (29) Dyspnea (gradual onset) 12 (6) 59 (20) Orthopnea 2 (1) 27 (9) Chest pain (pleuritic) 89 (44) 89 (30) Chest pain (substernal) 33 (16) 29 (10) 0.04 Fainting 53 (26) 38 (13) Hemoptysis 19 (9) 16 (5) 0.12 Cough 22 (11) 45 (15) 0.22 Palpitations 36 (18) 46 (15) 0.56 Signs Tachycardia 100/min 48 (24) 69 (23) 0.96 Cyanosis 33 (16) 44 (15) 0.73 Hypotension 90 mm Hg 6 (3) 5 (2) 0.15 Neck vein distention 25 (12) 28 (9) 0.36 Leg swelling (unilateral) 35 (17) 27 (9) Fever 38 C 14 (7) 63 (21) Crackles 37 (18) 76 (26) 0.08 Wheezes 8 (4) 39 (13) Pleural friction rub 8 (4) 11 (4) 0.93

4 Miniati, Prediletto, Formichi, et al.: Clinical Diagnosis of Pulmonary Embolism 867 TABLE 3 CHEST RADIOGRAPHIC FINDINGS IN 500 PATIENTS WITH CLINICALLY SUSPECTED PULMONARY EMBOLISM (PE) PE Present (n 202) PE Absent (n 298) n (%) n (%) p Value Finding Right heart enlargement 77 (38) 43 (14) Left heart enlargement 10 (5) 30 (10) 0.06 Cardiomegaly 36 (18) 65 (22) 0.33 Amputation of hilar artery 72 (36) 3 (1) Oligemia 91 (45) 3 (1) Consolidation (infarction) 31 (15) 2 (1) Consolidation (no infarction) 12 (6) 54 (18) Platelike atelectasis 65 (32) 70 (23) 0.04 Interstitial edema 1 (0.5) 27 (9) Elevated diaphragm (unilateral) 86 (43) 89 (30) Elevated diaphragm (bilateral) 39 (19) 57 (19) 0.95 Pleural effusion 91 (45) 104 (35) 0.02 Figure 1. Box-whisker plot of the partial pressure of oxygen (Pa O2 ) and carbon dioxide (Pa CO2 ) in arterial blood in patients with (n 202) and without (n 298) pulmonary embolism. Horizontal line in box: 50th percentile (median); limits of box: 25th and 75th percentile; whiskers: 10th and 90th percentile. PE: pulmonary embolism. Twenty-four percent of the patients with PE and 20% of those without PE had sinus tachycardia (heart rate 100 cycles/min). Atrial fibrillation occurred in 8% of the patients with PE and in 6% of those without. Chest Radiographic Findings Among the radiographic findings listed in Table 3, oligemia, amputation of hilar artery and pulmonary consolidations compatible with infarction were present in 45, 36, and 15% of the patients with PE, respectively. These three radiographic abnormalities occurred in only 1% of the patients without PE, respectively. Other radiographic findings such as elevated diaphragm, atelectasis, and pleural effusion often regarded as early hints of PE had a much lower specificity for PE because they occurred in 23 to 35% of the patients in whom the disease was excluded. Pulmonary interstitial edema was present in 27 (9%) of 298 patients without PE and in only one of 202 patients with angiographically proven PE. A Clinical Diagnostic Algorithm for PE Sudden onset dyspnea, chest pain, and fainting (singly or in combination) were present in 96% (194/202) of the patients with PE and in 59% (175/298) of those without. Because of the high sensitivity, the absence of these symptoms had a strong negative predictive value for PE (94%). However, because of their low specificity (41%), the presence of the above symptoms (in various combinations) had a positive predictive value for PE of only 53%. If the presence or absence of these symptoms had been relied on to diagnose or exclude PE, only 317 (63%) of 500 patients would have been correctly classified. Among the electrocardiographic and radiographic findings previously described, there were four abnormalities that turned out to be specific for PE: electrocardiographic signs of RV overload (specificity, 88%), and radiographic signs of oligemia, amputation of hilar artery, and pulmonary consolidations compatible with infarction (specificity, 99%). These four abnormalities (singly or in combination) were present in 165 (82%) of 202 patients with PE and in only 40 (13%) of 298 patients without PE (p ). It could therefore be expected that the search for an association between the above three symptoms and any of the above four elecrocardiographic and radiographic abnormalities would improve the rate of correct clinical classification. As indicated in Table 4, sudden onset dyspnea, chest pain, and fainting (singly or in combination) were associated with at least one of the above electrocardiographic and radiographic abnormalities in 164 (81%) of 202 patients with PE. This association was observed in only 22 (7%) of 298 patients without PE (p ). By relying on the presence or absence of such an association as a criterion to diagnose or exclude PE, as many as 440 (88%) of 500 patients would have been correctly classified. As shown in Table 4, PE was diagnosed in 17 of 33 patients in whom the above three symptoms were associated with electrocardiographic signs of RV overload only (positive predictive value, 52%). When the three symptoms (in various combinations) were associated with any one of the three radiographic abnormalities, or with any two of the four abnormalities (electrocardiographic and/or radiographic) listed in Table 4, the positive predictive value for PE rose to 94% (47/50) and 96% (74/77), respectively. When any three of the four abnormalities listed in Table 4 were present, all patients had PE (positive predictive value, 100%). Validation of the Diagnostic Algorithm The accuracy of the clinical diagnostic algorithm described above was assessed prospectively in 250 patients referred for lung scanning with clinically suspected PE. Patients were yr of age (range, 17 to 88 yr); 44% of them were male, and 80% were inpatients at the time of study entry. Perfusion lung scans were normal or near-normal in 62 (25%) of 250 patients and abnormal in 188. PE was diagnosed by pulmonary angiography in 104 of 188 patients with abnormal scans. Thus, the overall prevalence of PE in this group of patients was 42% (104/250) a figure comparable with that observed in the 500 patients already described. Sudden onset dyspnea, chest pain, and fainting (singly or in combination) occurred in 103 (99%) of 104 patients with PE and in 85 (58%) of 146 without PE (p ). These symptoms were associated with at least one of the four abnormalities (electrocardiographic and radiographic) specific for PE in

5 868 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL TABLE 4 ASSOCIATION OF SYMPTOMS WITH ELECTROCARDIOGRAPHIC AND RADIOGRAPHIC ABNORMALITIES IN 500 PATIENTS WITH CLINICALLY SUSPECTED PULMONARY EMBOLISM (PE) PE Present (n 202) (n) PE Absent (n 298) (n) Symptoms* associated with: Electrocardiographic signs of right ventricular overload (alone) Pulmonary oligemia (alone) 25 1 Amputation of hilar artery (alone) 13 1 Pulmonary infarction (alone) 9 1 Any two of the above four abnormalities 74 3 Any three of the above four abnormalities 24 0 All of the above four abnormalities 2 0 At least one of four abnormalities 164 (81%) 22 (7%) * Symptoms (either singly or in combination) are: sudden onset dyspnea, chest pain (pleuritic or substernal), and fainting. These symptoms are not explained otherwise. 87 (84%) of 104 patients with confirmed PE and in only 8 (5%) of 146 in whom the disease had been excluded (p ). Thus, the sensitivity and specificity of the diagnostic algorithm for PE were 84% (95% CI: 77 to 91%) and 95% (95% CI: 91 to 99%), respectively. Accordingly, a correct diagnosis or exclusion of PE was made in 225 (90%) of 250 patients. Relationship of the Diagnostic Algorithm with the Severity of PE Of the 750 patients included in this study, 306 had angiographically (or autopsy) proven PE (202 in the test group of 500 patients, and 104 in the validation group of 250 patients). In these 306 patients, the number of unperfused lung segments on the lung scan taken as an index of the severity of PE averaged (range, 1 to 14). Thus, in this patient population, PE spanned a wide range of severity, from minor (one unperfused lung segment) to massive (14 unperfused lung segments out of a total of 18). By applying the diagnostic algorithm described above, 251 (82%) of 306 patients with confirmed PE were classified as having PE (true positive). In these 251 patients, the number of unperfused lung segments averaged Fifty percent of these patients had more than eight unperfused lung segments, and 25% had more than 10. By contrast, the 55 other patients with PE who, using the diagnostic algorithm, were rated as not having PE (false negative) had a significantly lower number of unperfused lung segments ( , p ). In 50% of these patients, the number of unperfused lung segments was equal to or less than 5, and in 25% it was equal to or less than 2.5. Clinical Probability of PE Combined with Perfusion Lung Scan Interpretation On the basis of the above results, we may then formally describe the clinical probability of PE as follows. High probability (90%): presence of at least one of three symptoms (sudden onset dyspnea, chest pain, or fainting) not explained otherwise and associated with: (1) any two of the following abnormalities: electrocardiographic signs of RV overload, radiographic signs of oligemia, amputation of hilar artery, or pulmonary consolidations compatible with infarction; (2) any one of the above three radiographic abnormalities. Intermediate probability (50%): presence of at least one of the above symptoms, not explained otherwise, but not associated with the above electrocardiographic and radiographic abnormalities, or associated with electrocardiographic signs of RV overload only. Low probability (10%): absence of the above three symptoms, or identification of an alternative diagnosis that may account for their presence (e.g., exacerbation of COPD, pneumonia, lung edema, myocardial infarction, pneumothorax, and others). The clinical probability of PE defined above was used as pretest probability to calculate the posterior probability of PE, i.e., the probability of PE conditioned by perfusion lung scan results. Posterior (post-test) probability of PE was calculated according to Bayes theorem (21) as described in the APPENDIX. The relationship between pretest (clinical) probability of PE and post-test probability of PE as derived from the application of Bayes theorem to our data can be seen in Figure 2. The upper curve in Figure 2 describes the probability of PE conditioned by an abnormal perfusion scan compatible with PE (PE scan) as a continuous function of the pretest probability of PE, whereas the lower curve describes the probability of PE conditioned by an abnormal perfusion scan not compatible with PE (PE scan). For pretest (clinical) probabilities of PE of 10, 50, and 90%, as defined in the present study, the calculated post-test probabilities of PE conditioned by a lung scan result compatible with PE (PE scan) were 58, 93, and 99%, respectively (closed circles in Figure 2). For pretest (clinical) probabilities of 10, 50, and 90%, the calculated post-test probabilities of PE conditioned by a lung scan result not compatible with PE (PE scan) were 2, 13, and 58%, respectively (open circles in Figure 2). As shown in Table 5, there was a close agreement between probabilistic estimates of PE, derived from Bayes theorem, and actual prevalence of angiographically or autopsy proven PE in 583 patients with abnormal perfusion scans (395 in the test group of 500 patients, and 188 in the validation group of 250 patients) classified according to the six combinations of clinical and lung scan findings. It is worth noting that in most patients (464/583, or 80%) clinical probability of PE and perfusion lung scan results were concordant (high or intermediate clinical probability paired with a PE scan in 277 patients, and low clinical probability paired with a PE scan in 187). DISCUSSION The clinical manifestations of PE have been extensively investigated with the aim of identifying features able to determine the correct diagnosis.

6 Miniati, Prediletto, Formichi, et al.: Clinical Diagnosis of Pulmonary Embolism 869 TABLE 5 COMBINED CLINICAL AND LUNG SCAN EVALUATION: COMPARISON BETWEEN PROBABILISTIC ESTIMATES AND ACTUAL PREVALENCE OF PULMONARY EMBOLISM (PE) Observed PE Clinical Probability Scan Category* Expected PE (%) (PE present/ n of patients) (%) High PE / Intermediate PE 93 59/64 92 Low PE 58 10/17 59 High PE 58 15/21 71 Intermediate PE 13 8/81 10 Low PE 2 2/187 1 * PE abnormal scan compatible with PE; PE abnormal scan not compatible with PE. Probabilistic estimate or post-test probability of PE derived from Bayes theorem. Observed PE: actual prevalence of angiographically or autopsy proven PE in 583 patients with abnormal perfusion scans reported on in the present study. Figure 2. Relationship between pretest (clinical) probability of pulmonary embolism (PE) and post-test probability of PE as predicted by Bayes theorem. Upper curve describes post-test probability of PE conditioned by abnormal perfusion scan compatible with PE (PE scan); lower curve describes post-test probability of PE conditioned by abnormal perfusion scan not compatible with PE (PE scan). Closed circles indicate post-test probabilities of PE conditioned by PE scan for pretest probabilities of 10, 50, and 90%, respectively. Open circles indicate post-test probabilities of PE conditioned by PE scan at the three levels of pretest probability. Early reports focused on the evaluation of clinical data from patients with PE treated with thrombolytic agents (4 6). Study design required that all patients had angiographically proven PE, either massive or submassive. Results of these studies indicated that: (1) no single symptom or sign was, in itself, diagnostic of PE; (2) the classic triad of chest pain, hemoptysis, and dyspnea occurred in only a minority of patients with massive or submassive PE; (3) chest pain, dyspnea, and cough were the most frequent symptoms, and tachypnea and tachycardia the most frequent clinical signs; these symptoms and signs (in various combinations) occurred in greater than 95% of the patients with confirmed PE. The PIOPED study (7) expanded on previous experience by comparing the clinical characteristics of patients with PE with those of patients suspected of having PE, but in whom the disease was subsequently excluded. Among patients with PE, the most frequent symptoms were dyspnea and chest pain, and the most frequent signs were tachypnea and tachycardia (22, 23). However, the prevalence of these symptoms and signs did not differ significantly from that of patients who turned out not to have PE (22, 23). The most common radiographic abnormalities were atelectasis and pulmonary parenchymal consolidation, either or both of which occurred in 69% of the patients with PE and in 58% of those without (23). Dyspnea, tachypnea, and pleuritic chest pain (singly or a combination) were present in 97% of the patients with confirmed PE (23). This combination of clinical symptoms and signs occurred with nearly equal frequency in patients in whom the disease had been excluded (23). It was concluded that: (1) only a minority of patients with PE do not have any of the important clinical manifestations of the disease; (2) although these clinical manifestations are nonspecific, their identification may assist in selecting patients in whom further diagnostic testing is required (22, 23). The present study differs from previous reports because: (1) all patients referred for lung scanning with suspected PE were considered for clinical evaluation independently of how the suspicion of PE had been expressed by the attending physicians; (2) all patients were evaluated directly by a physician according to a standardized diagnostic protocol; (3) unlike other studies, attention was paid to characterize the onset of dyspnea (either sudden or gradual); (4) physicians participating in the study were aware of the importance of the chest radiograph in the evaluation of patients suspected of having PE; every effort was therefore made to obtain good-quality chest radiographs in all patients, especially if they were critically ill; (5) chest films were interpreted according to a reading table, which included the evaluation of radiographic features of PE derived from a review of the literature; (6) in the series of patients reported on in this study, PE ranged over a wide spectrum of severity, from minor to massive. A clinical diagnostic algorithm was developed that includes the identification of three relevant symptoms (sudden onset dyspnea, chest pain, and fainting) and their association with one or more of the following abnormalities: electrocardiographic signs of RV overload, radiographic signs of oligemia, amputation of hilar artery, and pulmonary consolidations compatible with infarction. Sudden onset dyspnea, chest pain, and fainting (singly or in combination) were associated with at least one of the above electrocardiographic and radiographic abnormalities in greater than 80% of the patients with confirmed PE. This association occurred in only a few patients among those in whom the disease had been excluded. The diagnostic algorithm described here does not include the evaluation of risk factors for PE or arterial blood gas abnormalities. This does not imply that risk factors for PE and arterial hypoxemia should be disregarded. In our study, at least one of the known risk factors for PE occurred in greater than 80% of the patients with confirmed PE. Similarly, 90% of the patients with PE also had arterial hypoxemia and hypocapnia. Therefore, in a patient who meets the diagnostic criteria for PE described above, the coexistence of risk factors and arterial hypoxemia with hypocapnia may add further strength to the diagnosis. In our study, however, at least one risk factor for PE was

7 870 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL present in nearly 70% of the patients in whom the disease was excluded. Furthermore, 50% of these patients had conspicuous hypoxemia and hypocapnia. Therefore, physicians should not be led astray by the presence of risk factors or arterial hypoxemia with hypocapnia if other important elements supporting the diagnosis of PE are absent, or if clinical, electrocardiographic, and radiographic findings suggest an alternative diagnosis. Although the diagnostic algorithm turned out to be fairly accurate, it failed to identify some 20% of the patients with confirmed PE. In these patients, mistakenly classified as not having PE, the severity of the disease was significantly less than in those who were correctly rated as having PE. In addition, whenever the three symptoms included in the diagnostic algorithm were not associated with the electrocardiographic and radiographic abnormalities described above, or were associated with electrocardiographic signs of RV overload only, the positive predictive value for PE did not exceed 50%. Thus, it appears that PE cannot be diagnosed or excluded with certainty by clinical assessment only. However, on the basis of the diagnostic algorithm, it was possible to characterize the clinical probability of PE as low (10%), intermediate (50%), or high (90%). These clinical estimates of PE served as pretest probabilities to calculate, after perfusion lung scanning, the post-test probabilities of PE by means of Bayes theorem. Results of this formal analysis indicate that: (1) a high or intermediate clinical probability combined with an abnormal scan compatible with PE (PE scan) make the diagnosis of PE very likely; (2) a low clinical probability paired with an abnormal scan not compatible with PE (PE scan) makes the diagnosis of PE very unlikely; (3) when clinical probability and lung scan results are discordant, the post-test probability of PE is neither sufficiently high nor sufficiently low to render the diagnosis or exclusion of PE reasonably certain; under these circumstances, further diagnostic testing may be required, be this pulmonary angiography or some additional noninvasive procedure such as lower limb compression ultrasonography or D-dimer test (24). In the present study, most patients had concordant clinical and perfusion lung scan findings (Table 5). Thus, it appears that, in most patients, PE could be diagnosed or excluded noninvasively by combining well-defined clinical estimates of PE with perfusion lung scan interpretation (Table 5). In the PIOPED study (7), a high clinical probability combined with a high probability ventilation-perfusion scan had a 96% positive predictive value for PE, whereas a low clinical probability combined with a normal/near-normal or low probability ventilation-perfusion scan had a 97% negative predictive value for PE. However, some 75% of the patients in the PIOPED study did not fit into the above clinicoscintigraphic categories. Other combinations of clinical and lung scan findings were of little diagnostic value (7). Thus, combining clinical probability with the ventilation-perfusion scan improved the predictive accuracy of either alone only for a minority of patients those with clear and concordant clinical and ventilation-perfusion scan findings (7). Because inconclusive ventilation-perfusion studies often lead to improper therapeutic decisions, it has been proposed that ventilation-perfusion scanning be replaced by two newer imaging modalities: spiral (helical) CT angiography, and lower limb B-mode compression ultrasonography (25). In our opinion, the proposal of omitting the lung scan altogether is questionable for at least two reasons: (1) a normal perfusion scan rules out in itself clinically significant PE, and renders further diagnostic testing, particularly pulmonary angiography, unnecessary; (2) an abnormal scan with perfusion defects not strictly suggestive of PE is still useful as a guide to obtain targeted CT reconstructions in those areas of the lung with the greatest perfusion abnormalities. This integrated approach would likely improve the detection of subsegmental emboli that are notoriously difficult to visualize by both conventional (26) and spiral CT angiography (27). In conclusion, the present study indicates that clinical assessment based on the standardized evaluation of anamnestic, electrocardiographic, and radiographic findings is a fundamental step in the diagnostic work-up of PE. Combining well-characterized clinical estimates of PE with independent interpretation of perfusion lung scan results helps restrict the need for angiography to a minority of patients suspected of having PE. Acknowledgment : The writers thank Giuseppe Maltinti, Claudio Mazzotti, Loredana Salis, Armando Perissinotto, Alberto Pollastri, and Nilo Faraoni for their excellent technical assistance; Claudio Michelassi for helping in the statistical analysis; Prof. Anthony L. Johnson for reviewing the manuscript. References 1. Hildner, F. J., and R. S. Ormond Accuracy of the clinical diagnosis of pulmonary embolism. J.A.M.A. 202: Robin, E. D Overdiagnosis and overtreatment of pulmonary embolism: the emperor may have no clothes. Ann. Intern. Med. 87: Moser, K. M Venous thromboembolism. Am. Rev. Respir. Dis. 141: Wenger, N. K., P. D. Stein, and P. W. Willis, III Massive acute pulmonary embolism: the deceivingly non-specific manifestations. J.A.M.A. 220: Bell, W. R., T. L. Simon, and D. L. DeMets The clinical features of submassive and massive pulmonary emboli. Am. J. Med. 62: Stein, P. D., P. W. Willis, III, and D. L. DeMets History and physical examination in acute pulmonary embolism in patients without pre-existing cardiac or pulmonary disease. Am. J. Cardiol. 47: PIOPED Investigators Value of the ventilation-perfusion scan in acute pulmonary embolism: results of the Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED). J.A.M.A. 263: Miniati, M., M. Pistolesi, C. Marini, G. Di Ricco, B. Formichi, R. Prediletto, G. Allescia, L. Tonelli, H. D. Sostman, and C. Giuntini Value of perfusion lung scan in the diagnosis of pulmonary embolism: results of the Prospective Investigative Study of Acute Pulmonary Embolism Diagnosis (PISA-PED). Am. J. Respir. Crit. Care Med. 154: Kelley, M. A., J. L. Carson, H. I. Palevsky, and J. S. Schwartz Diagnosing pulmonary embolism: new facts and strategies. Ann. Intern. Med. 114: Stein, P. D., J. E. Dalen, K. M. McIntyre, A. A. Sasahara, N. K. Wenger, and P. W. Willis, III The electrocardiogram in acute pulmonary embolism. Prog. Cardiovasc. Dis. 17: Hanelin, J., and W. R. Eyler Pulmonary artery thrombosis: roentgen manifestations. Radiology 56: Fleischner, F. G Pulmonary embolism. Clin. Radiol. 13: Kerr, I. H., G. Simon, and G. C. Sutton The value of the plain radiograph in acute massive pulmonary embolism. Br. J. Radiol. 44: Palla, A., V. Donnamaria, S. Petruzzelli, G. Rossi, G. Riccetti, and C. Giuntini Enlargement of right descending pulmonary artery in pulmonary embolism. Am. J. Roentgenol. 141: Hampton, A. O., and B. Castleman Correlation of postmortem chest teleroentgenograms with autopsy findings with special reference to pulmonary embolism and infarction. Am. J. Roentgenol. 43: Fleischner, F. G Roentgenology of pulmonary infarct. Semin. Roentgenol. 2: Westermark, N On roentgen diagnosis of lung embolism. Acta Radiol. 19: Kipper, M. S., K. H. Moser, K. E. Kortman, and W. L. Ashburn Long term follow-up of patients with suspected pulmonary embolism and a normal lung scan. Chest 82: Hull, R. D., G. E. Raskob, G. Coates, and A. A. Panju Clinical va-

8 Miniati, Prediletto, Formichi, et al.: Clinical Diagnosis of Pulmonary Embolism 871 lidity of a normal perfusion scan in patients with suspected pulmonary embolism. Chest 97: Marini, C., G. Di Ricco, A. Palla, G. Susini, G. Maltinti, A. Santolicandro, and C. Giuntini Perfusion scintigraphy versus pulmonary arteriography in the diagnosis of pulmonary embolism. J. Nucl. Med. All. Sci. 22: Diamond, G. A., and J. S. Forrester Analysis of probability as an aid in the clinical diagnosis of coronary-artery disease. N. Engl. J. Med. 300: Stein, P. D., M. L. Terrin, C. A. Hales, H. I. Palevsky, H. A. Saltzman, T. Thompson, and J. G. Weg Clinical, laboratory, roentgenographic and electrocardiographic findings in patients with acute pulmonary embolism and no pre-existing cardiac or pulmonary disease. Chest 100: Stein, P. D., H. A. Saltzman, and J. G. Weg Clinical characteristics of patients with acute pulmonary embolism. Am. J. Cardiol. 68: Perrier, A., H. Bounameaux, A. Morabia, P. de Moerloose, D. Slosman, D. Didier, P. F. Unger, and A. Junod Diagnosis of pulmonary embolism by a decision analysis-based strategy including clinical probability, D-dimer levels, and ultrasonography: a management study. Arch. Intern. Med. 156: Goodman, L. R., and R. J. Lipchik Diagnosis of acute pulmonary embolism: time for a new approach. Radiology 199: Stein, P. D., C. Athanasoulis, A. Alavi, R. H. Greenspan, C. A. Hales, H. A. Saltzman, C. E. Vreim, M. L. Terrin, and J. G. Weg Complications and validity of pulmonary angiography in acute pulmonary embolism. Circulation 85: Gefter, W. B., H. Hatabu, G. A. Holland, K. B. Gupta, C. I. Henschke, and H. I. Palevsky Pulmonary thromboembolism: recent developments in diagnosis with CT and MR imaging. Radiology 197: APPENDIX Calculation of the Posterior Probability of PE The posterior (post-test) probability of a disease D, conditioned by the result of a given diagnostic test T, can be calculated as a function of the pretest probability by means of Bayes theorem of conditional probability (21). For the calculation, it is necessary to assign a value to the following variables: PD ( ) pretest probability of the disease D PT ( D ) sensitivity of the diagnostic test T PT ( D ) specificity of the diagnostic test T The posterior probability of the disease D, conditioned by a positive diagnostic test result (T ), is described by the following equation: PD ( ) PT ( D ) , (1) PD ( ) PT ( D ) + [ 1 P( D ) ] PT ( D ) where P(T /D ) 1 P(T /D ) 1 specificity of the diagnostic test T. The posterior probability of the disease D, conditioned by a negative diagnostic test result (T ), is described by the following equation: PD ( ) PT ( D ) , (2) PD ( ) PT ( D ) + [ 1 P( D ) ] PT ( D ) where P(T /D ) 1 P(T /D ) 1 sensitivity of the diagnostic test T. In our study, P(D ) is the clinical probability of PE, and the diagnostic test T is the perfusion lung scan (without ventilation imaging) which can be either compatible (PE scan) or not compatible with PE (PE scan) (8). For the calculation of the post-test probability of PE, we used sensitivity and specificity values of the PE scan corrected for selection bias in referring patients to pulmonary angiography as described elsewhere (8). Thus: PT ( D ) 86% ( debiased sensitivity of the PE scan) PT ( D ) 93% ( debiased specificity of the PE scan) Examples of Calculation For pretest (clinical) probabilities of 10, 50, and 90%, as defined in the present study, the post-test probabilities of PE, conditioned by a lung scan result compatible with PE (PE scan), are calculated according to Equation 1: PD ( ) 10%, PD ( ) 50%, PD ( ) 90%, ( ) ) + ( ) % ( ) ) + ( ) % ( ) ) + ( ) % For pretest (clinical) probabilities of 10, 50, and 90%, the post-test probabilities of PE, conditioned by a lung scan result not compatible with PE (PE scan), are calculated according to Equation 2: PD ( ) 10%, PD ( ) 50%, PD ( ) 90%, ( ) ) + ( ) % ( ) ) + ( ) % ( ) ) + ( ) %