Comparison of 3 Clinical Models for Predicting the
Probability of Pulmonary Embolism
Massimo Miniati, MD, PhD, Matteo Bottai, ScD, and Simonetta Monti, MD, PhD
Abstract: Two clinical models have been described to predict the
probability of pulmonary embolism: the Canadian (or Wells) model,
and the Geneva model. A third model has been developed recently
at our institution (the Pisa model). We compared the performance of
the 3 models in 215 consecutive patients with suspected pulmonary
embolism. The clinical probability predicted by the models was
categorized as low, intermediate, or high. In all patients, pulmonary
angiography was used as the reference diagnostic standard. In
patients with pulmonary embolism, the extent of pulmonary
embolization was assessed on the lung scan as an index of disease
severity.
The prevalence of pulmonary embolism was 43.3%, and the
median extent of pulmonary embolization at diagnosis was 39.8%
(range, 4.5%–75.3%). The proportions of patients categorized as
having low, intermediate, or high probability were, respectively:
12%, 60%, and 28%, for the Geneva model; 30%, 55%, and 15%,
for the Wells model; 37%, 37%, and 26% for the Pisa model. The
frequencies of pulmonary embolism in the low, intermediate, and
high probability categories were, respectively: 50%, 39%, and 49%
for the Geneva model; 12%, 54%, and 64% for the Wells model;
5%, 42%, and 98% for the Pisa model. Among patients with
pulmonary embolism, there was a strong, positive relation between
clinical probability predicted by the Pisa model and the extent of
pulmonary embolization. The Pisa model proved more accurate
than the 2 other models. It may be useful to physicians in defining
precisely the pretest probability of pulmonary embolism.
(Medicine 2005;84:107–114)
INTRODUCTION
P ulmonary embolism is a challenging diagnostic problem3.
Pulmonary angiography remains the reference diagnostic
From Istituto di Fisiologia Clinica (MM, SM), and Istituto di Scienza
e Tecnologie dell’Informazione (MB), Consiglio Nazionale delle
Ricerche, 56124 Pisa, Italy.
Dr. Matteo Bottai’s current affiliation is Arnold School of Public Health,
University of South Carolina, Columbia, South Carolina.
Address reprint requests to: Massimo Miniati, MD, PhD, Istituto di
Fisiologia Clinica del CNR, Via G. Moruzzi 1, 56124 Pisa, Italy. Fax:
50-315-2166; e-mail: miniati@ifc.cnr.it.
This work was supported in part by the Italian Ministry of Health. The
funding source had no role in the study design, the collection, analysis,
or interpretation of data, the writing of the report, or the decision to
submit the paper for publication.
Copyright n 2005 by Lippincott Williams & Wilkins
ISSN: 0025-7974/05/8402-0107
DOI: 10.1097/01.md.0000158793.32512.37
Medicine ! Volume 84, Number 2, March 2005
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
standard, but its use is limited by the high cost, invasiveness,
and logistic constraints3. Spiral computed tomographic an-
giography was introduced some 10 years ago as a newer
imaging modality for suspected pulmonary embolism. In
spite of initial enthusiasm, the sensitivity of the technique
is still too low for pulmonary embolism to be diagnosed or
excluded without additional tests10 –12.
Clinical evaluation is a fundamental step in the
diagnostic workup of patients with suspected pulmonary
embolism. Although the diagnostic yield of individual
symptoms, signs, and common laboratory tests is limited,
the combination of these variables can be used to express a
clinical probability of pulmonary embolism. This may serve
as the pretest probability in estimating the likelihood of the
disease after further objective testing.
Since 1998, 2 prediction models have been introduced
to improve the diagnostic approach to patients suspected of
having pulmonary embolism14,15,17. In 20036, we described a
structured clinical model that proved accurate in predicting
the probability of pulmonary embolism.
In the present study, we compare the performance of
the 3 clinical models in a sample of both in- and outpatients
referred to our institution for suspected pulmonary embo-
lism. In all patients, the diagnosis or exclusion of the disease
was based on the results of pulmonary angiography.
Patient Sample
PATIENTS AND METHODS
The sample consisted of 248 consecutive patients who
were referred to our institution for suspected pulmonary
embolism between June 2000 and July 2001. Thirty-three
patients (13%) were excluded from the study because of
inability to obtain an informed consent, contraindication to
pulmonary angiography, or anticoagulant therapy for more
than 3 days. The 215 other patients were examined uniformly
according to a diagnostic protocol including assessment of
the clinical probability, perfusion lung scanning, and pul-
monary angiography. None of the patients had undergone
any objective testing for pulmonary embolism before study
entry. All the procedures (including pulmonary angiography)
were carried out in a dedicated diagnostic unit at our
107
Miniati et al
institution. The protocol was approved by the local ethics
committee.
Clinical Evaluation
Upon study entry, patients were examined by 1 of 12
chest physicians. Evaluation included clinical history, phys-
ical examination, interpretation of the electrocardiogram and
chest radiograph, and measurement of arterial blood gases.
All clinical and laboratory data were recorded by the phy-
sicians on a standard form before any further objective
testing8.
On interviewing the patients, care was taken to identify
risk factors for pulmonary embolism and preexisting diseases
that might mimic the clinical presentation of pulmonary
embolism. In evaluating dyspnea, attention was paid to
whether it was sudden or gradual in onset, and whether it was
associated with orthopnea. Chest pain was categorized as
pleuritic or substernal. Unilateral leg swelling with tender-
ness and redness of the skin was considered suggestive of
deep vein thrombosis.
Electrocardiograms obtained within 24 hours before
study entry were considered for evaluation. The following
abnormalities were regarded as suggestive of right ventric-
ular overload: S-wave in lead I and Q-wave in lead III each
of amplitude >1.5 mm, with or without T-wave inversion in
lead III, S-waves in lead I, II, and III each of amplitude >1.5
mm, T-wave inversion in right precordial leads, transient
right bundle branch block, and pseudoinfarction13. If any
of the above abnormalities were present in electrocardio-
grams taken before the onset of symptoms, they were
disregarded.
Analogic 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. With patients who
were unable to stand upright or seated (20% in the present
study), anteroposterior chest radiographs were obtained in
the supine or semirecumbent position.
Chest radiographs were examined by the physician on-
call according to a reading table that included the following
items: size and shape of the heart and hilar arteries, posi-
tion of the diaphragm, presence or absence of pulmonary
parenchymal abnormalities (consolidation, atelectasis, olige-
mia, edema), and pleural effusion. On evaluating the hilar
arteries, attention was paid to the presence of abrupt vascular
amputation, which gives the hilum a ‘‘plump’’ appearance4.
Pulmonary consolidations were considered suggestive of
infarction if they had a semicircular or half-spindle shape
and were arranged peripherally along the pleural surface (so-
called Hampton hump)4. Oligemia was considered to be
present if, in a given lung region, the pulmonary vasculature
was greatly diminished with concomitant hyperlucency of
the lung parenchyma4.
108
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Medicine ! Volume 84, Number 2, March 2005
Arterial blood samples were obtained upon study entry
in all patients while they were breathing room air.
Assessment of the Clinical Probability of
Pulmonary Embolism
The clinical probability of pulmonary embolism was
estimated according to 3 logistic regression models, the
characteristics of which are given in Table 1. According to
the Wells model15, the clinical probability was rated low in
patients with a score <2, intermediate in patients with a score
of 2–6, and high in those with a score >6. According to the
Geneva model17, the clinical probability was rated low with
a score " 4, intermediate with a score of 5–8, and high with a
score # 9.
With the Pisa model6, the probability of pulmonary
embolism was calculated directly from the algebraic sum
of the regression coefficients, as indicated in the footnote
of Table 1. A quicker way to estimate the probability of
pulmonary embolism is provided in Figure 1, which dis-
plays the relation between the sum of coefficients and the
predicted probability of pulmonary embolism. For the sake
of the comparison with the 2 other models, the clinical
probability was categorized as low (" 10%), intermediate
(>10% to " 90%), or high (>90%).
Physicians estimated the clinical probability of pul-
monary embolism according to the Wells and Pisa models at
the time of patients’ enrollment in the study. The Geneva
score was calculated retrospectively as all the items neces-
sary for estimating the clinical probability had been recorded
by the physicians.
Perfusion Lung Scanning
Perfusion lung scans were obtained after intravenous
injection of human serum albumin microspheres labeled with
99m
Technetium (1.8 $ 108 Bq), taking care to inject the
radioactive bolus with the patient held as closely as possible
to the sitting position in order to preserve the effect of
gravity on the regional distribution of pulmonary blood flow.
Lung scans were acquired by means of a large field gamma
camera equipped with a high resolution, parallel-hole
collimator, using a 20% symmetric window set over the
140 KeV photopeak. Images consisted of anterior, posterior,
both lateral, and both posterior oblique views, with 500,000
counts per image.
Lung scans were independently attributed to 1 of 4
predetermined categories7: normal (no perfusion defects);
near-normal (impressions caused by enlarged heart, hila,
or mediastinum are seen on an otherwise normal scan);
abnormal, suggestive of pulmonary embolism (single or
multiple wedge-shaped perfusion defects); abnormal, not
suggestive of pulmonary embolism (single or multiple per-
fusion defects other than wedge-shaped).
n 2005 Lippincott Williams & Wilkins
Medicine ! Volume 84, Number 2, March 2005 Clinical Probability of Pulmonary Embolism

TABLE 1. Clinical Models for Predicting the Probability of Pulmonary Embolism

Wells Model Geneva Model Pisa Model*
Characteristic Score Characteristic Score Characteristic Coefficient
Previous PE or DVT 1.5 Previous PE or DVT 2 Male sex 0.81
Heart rate >100 beats/min 1.5 Heart rate >100 beats/min 1 Age (yr)
Recent surgery or 1.5 Recent surgery 3 63–72 0.59
immobilization
Clinical signs of DVT 3 Age (yr) # 73 0.92
Alternative diagnosis 3 60–79 1 Preexisting cardiovascular disease % 0.56
less likely than PE
Hemoptysis 1 # 80 2 Preexisting pulmonary disease % 0.97
Cancer 1 Blood gases History of thrombophlebitis 0.69
PaCO2 (mmHg) Dyspnea (sudden onset) 1.29
<36 2 Chest pain 0.64
36–38.9 1 Hemoptysis 0.89
PaO2 (mmHg) Fever >388C % 1.17
<48.7 4 ECG signs of acute right 1.53
ventricular overload
48.7–59.9 3 Chest radiograph
60–71.2 2 Oligemia 3.86
71.3–82.4 1 Amputation of the hilar artery 3.92
Chest radiograph Consolidation (infarction) 3.55
Plate-like atelectasis 1 Consolidation (no infarction) % 1.23
Elevated hemidiaphragm 1 Pulmonary edema % 2.83
Constant % 3.26
Abbreviations: PE = pulmonary embolism; DVT = deep vein thrombosis; PaO2 = partial pressure of oxygen, arterial; PaCO2 = partial pressure of carbon
dioxide, arterial; ECG = electrocardiogram.
*To estimate the probability of PE, add all of the regression coefficients that apply to a particular patient to the constant. The probability of PE then equals
1/[1+exp(% sum)].

Pulmonary Angiography followed by oral anticoagulation for 1 year. Anticoagulant

In all patients, pulmonary cineangiograms were
obtained according to standardized procedures within 24
hours of study entry. Before angiography, an informed
written consent was obtained. Initial filming was in the
anteroposterior view, after having advanced the catheter into
the main pulmonary artery of the lung that showed the
greatest perfusion abnormalities on lung scanning. Whenever
necessary, injections were repeated using lateral and/or
oblique views. If there was a doubt about the presence of
filling defects, the appropriate vessel was selectively entered
with a balloon-tipped catheter and angiograms were repeated
by manual injection of contrast material. Pulmonary angio-
grams were examined by experienced physicians who were
blind to clinical information. Angiographic criteria for diag-
nosing pulmonary embolism included the identification of
an embolus obstructing a vessel or the outline of an embolus
within a vessel. A normal pulmonary angiogram ruled out
pulmonary embolism.
Patients with pulmonary embolism received anticoag-
ulant therapy consisting of a 1-week heparin infusion
treatment was withheld or withdrawn in patients in whom
pulmonary embolism was excluded.
Outcomes
We measured the proportion of patients and the
frequency of pulmonary embolism in the 3 clinical prob-
ability categories for each prediction model.
In patients with pulmonary embolism, we estimated
the extent of pulmonary vascular obstruction on the lung
scan as an index of the disease severity. This analysis was
carried out by a nuclear medicine specialist, who was blind
to clinical information, according to a method previously
described5. Briefly, each lobe was attributed a weight
according to regional blood flow5 as follows: right upper
lobe: 0.18; right middle lobe: 0.12; right lower lobe: 0.25;
left upper lobe: 0.13; lingula: 0.12; left lower lobe: 0.20. The
perfusion of each lobe was estimated by means of a 5-point
scale—0, 0.25, 0.5, 0.75, 1.0 —where 0 is ‘‘not perfused’’
and 1 is ‘‘normally perfused.’’ Each lobar perfusion score
was obtained by multiplying the weight assigned to the lobe

n 2005 Lippincott Williams & Wilkins 109

Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Miniati et al
FIGURE 1. Relation between sum of regression coefficients
(Pisa model) and probability of pulmonary embolism. To
estimate the probability of pulmonary embolism, take the
algebraic sum of the regression coefficients that apply to a
given patient (without adding the constant). Then, read the
corresponding probability value on the Y-axis. Vertical line on
the left indicates the sum of coefficients corresponding to the
10% probability cutoff. Vertical line on the right indicates the
sum of coefficients corresponding to the 90% probability
cutoff.
by the estimated perfusion of that lobe. The overall perfusion
score was the sum of the perfusion scores of the 6 lobes, and
the percentage of pulmonary vascular obstruction was
calculated5 as follows: (1 - overall perfusion score) $ 100.
The percentage of pulmonary vascular obstruction was
estimated on the lung scans obtained at the time of diagnosis
and 1 year later (see below).
Follow-Up
In patients with confirmed pulmonary embolism, a
clinical and scintigraphic follow-up was pursued at 1 week, 1
month, and 1 year after study entry to assess the efficacy of
anticoagulant therapy. Patients who were diagnosed as not
having pulmonary embolism were followed over a period of
1 year by a team of 4 physicians in charge of follow-up
studies. Critical events recorded by the physicians were
recurrent episodes of pulmonary embolism, hospital read-
missions for any cause, and death.
Recurrent pulmonary embolism was diagnosed if
patients had symptoms that prompted a new investigation
and met any of the following criteria: new wedge-shaped
perfusion defects on the lung scan, a positive angiogram, or
evidence of pulmonary emboli at autopsy.
The 1-year mortality rate was assessed in patients with
and without pulmonary embolism. The cause of death was
established by reviewing autopsy findings, clinical files, or
death certificates.
110
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Medicine ! Volume 84, Number 2, March 2005
Statistical Analysis
Patient characteristics were compared across the 2
diagnostic groups by contingency tables for categorical
variables. For the continuous variables, differences between
the 2 diagnostic groups were tested for by Mann-Whitney or
Wilcoxon rank-sum nonparametric procedures. The predic-
tive accuracy of the 3 clinical models was compared by
means of receiver operating characteristic (ROC) curve
analysis2.
The relation between the clinical probability of pul-
monary embolism and the extent of pulmonary vascular
obstruction was assessed by the Kruskal-Wallis nonpara-
metric test. Two-tailed p values < 0.05 were considered
statistically significant throughout. Ninety-five percent con-
fidence intervals (95% CI) were calculated according to the
binomial distribution.
RESULTS
The prevalence of pulmonary embolism was 43.3%
(93/215), and the median extent of pulmonary vascular
obstruction at diagnosis was 39.8% (range, 4.5%–75.3%).
The patients’ baseline characteristics are summarized in
Table 2.
Performance of the 3 Clinical Models
Results of ROC curve analysis are displayed in
Figure 2. The areas under the ROC curves for the 3 clinical
models were as follows: Geneva model 0.54 (95% CI, 0.46–
0.62); Wells model 0.75 (95% CI, 0.69–0.81); Pisa model
0.94 (95% CI, 0.91–0.97). In pair-wise comparisons, the
differences between the areas under the ROC curves were all
highly significant (p < 0.0001).
Table 3 shows the breakdown of patients according to
the clinical probability of pulmonary embolism, and the
frequency of pulmonary embolism in the 3 clinical
probability categories for each prediction model. With the
Pisa model, fewer patients were categorized as having
intermediate probability than with the Geneva or Wells
models. The positive predictive values of the high probabil-
ity category for the 3 clinical models were as follows:
Geneva model 49% (95% CI, 36%–62%); Wells model 64%
(95% CI, 45%–80%); Pisa model 98% (95% CI, 91%–
100%). The negative predictive values of the low probability
category were as follows: Geneva model 50% (95% CI,
30%–70%); Wells model 88% (95% CI, 77%–94%); Pisa
model 95% (95% CI, 88%–99%).
There was a highly significant relation (p < 0.0001)
between the clinical probability predicted by the Pisa model
and the severity of pulmonary embolism (Figure 3). Pul-
monary vascular obstruction was highest in patients who
had been rated a high clinical probability, and lowest in
those rated a low clinical probability. The relation between
clinical probability and extent of pulmonary embolization
n 2005 Lippincott Williams & Wilkins
Medicine ! Volume 84, Number 2, March 2005 Clinical Probability of Pulmonary Embolism

TABLE 2. Baseline Characteristics of Study Sample

All Patients Pulmonary Embolism No Pulmonary Embolism
(n = 215) (n = 93) (n = 122)
Characteristic No. (%) or Median (range)
Age (yr) 70 (17–90) 71 (24–90) 69 (17–89)
Inpatient 168 (78) 72 (77) 96 (79)
Male sex 98 (46) 45 (48) 53 (43)
Preexisting disease
Cardiovascular 100 (47) 35 (38) 65 (53)z
Pulmonary 39 (18) 15 (16) 24 (20)
Neoplastic* 36 (17) 14 (15) 22 (18)
Endocrine 23 (11) 7 (8) 16 (13)
Risk factors
Immobilization (>3 days)y 110 (51) 47 (51) 63 (52)
Thrombophlebitis (ever) 51 (24) 28 (30) 23 (19)
Recent surgeryy 73 (34) 32 (34) 41 (34)
Recent traumay 35 (16) 8 (9) 27 (22)z
Postpartum* 2 (1) 0 (0) 2 (2)
Estrogen use* 5 (2) 3 (3) 2 (2)
Symptoms and signs
Dyspnea (sudden onset) 120 (56) 74 (80) 46 (38)z
Chest pain 85 (40) 45 (48) 40 (33)z
Hemoptysis 8 (4) 5 (5) 3 (2)
Signs of DVT 29 (13) 15 (16) 14 (11)
Fever >388C 34 (16) 8 (9) 26 (21)z
Heart rate >100 beats/min 54 (25) 23 (25) 31 (25)
Electrocardiogram
Signs of acute RV overload 53 (25) 40 (43) 13 (11)z
Chest radiograph
Oligemia 35 (16) 34 (37) 1 (1)z
Amputation of hilar artery 24 (11) 24 (26) 0 (0)z
Consolidation (infarction) 16 (7) 15 (16) 1 (1)z
Consolidation (no infarction) 36 (17) 3 (3) 33 (27)z
Pulmonary edema 10 (5) 0 (0) 10 (8)z
Plate-like atelectasis 44 (20) 20 (22) 24 (20)
Elevated hemidiaphragm 100 (47) 47 (51) 53 (43)
Blood gases
PaO2 (mm Hg) 65 (33–107) 67 (33–107) 64 (35–94)
PaCO2 (mm Hg) 33 (14–64) 34 (23–42) 33 (14–64)
Abbreviations: See Table 1. RV = right ventricle.
*In the past 3 months.
y
In the past 4 weeks.
z
Significantly different from patients with pulmonary embolism.

was borderline significant for the Wells model (p = 0.05),
and not significant for the Geneva model (p = 0.42).
Follow-Up
Follow-up was completed in all patients. Among the
93 patients diagnosed as having pulmonary embolism, 9
(9.7%; 95% CI, 4.5%–7.6%) had recurrent episodes of
pulmonary embolism. None of the 122 patients with normal
angiograms had an acute embolic event during follow-up.
Sixteen patients with pulmonary embolism (17%; 95%
CI, 10%–26%) died within 1 year of study entry, as did 32
of those without pulmonary embolism (26%; 95% CI,
18%–34%). Among patients with pulmonary embolism, the
causes of death were pulmonary embolism (6 patients),

n 2005 Lippincott Williams & Wilkins 111

Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Miniati et al Medicine ! Volume 84, Number 2, March 2005

FIGURE 2. Receiver operating characteristic curves for the 3 clinical models. Score cutoffs for the Geneva model are (from right
to left): 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14. Score cutoffs for the Wells model are (from right to left): 0, 1, 1.5, 2.5, 3, 4, 5,
5.5, 6, 7, 8, 9, 10, 10.5. Probability cutoffs (%) for the Pisa model are (from right to left): 1, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90,
95, 100. Solid oblique line: line of useless test.

cancer (3 patients), congestive heart failure (2 patients), and
(1 each) stroke, acute myocardial infarction, intestinal
infarction, sepsis, and liver failure. For patients without
pulmonary embolism, the causes of death were: cancer (16
patients), congestive heart failure (6 patients), stroke (3
patients), chronic obstructive lung disease (2 patients),
pneumonia (2 patients), and (1 each) hypovolemic shock,
vasculitis, and head trauma.
Of the 77 patients who survived a full year after pul-
monary embolism, 76 (99%) completed the 1-year scinti-
graphic follow-up. In these patients, the median pulmonary
vascular obstruction at diagnosis was 39% (range, 4.5%–
75.3%). After 1 year of anticoagulant therapy, the residual
pulmonary vascular obstruction was " 8% in 90% of the
patients, and " 3% in 75%. In 51 (67%) of 76 patients, the
perfusion lung scan was rated normal.
DISCUSSION
The present study was designed to assess the
performance of 3 structured models in predicting the clinical
probability of pulmonary embolism. As opposed to other
studies1,14,16,17, pulmonary angiography was relied upon as
the reference diagnostic standard. Pulmonary angiogram
results were further supported by follow-up data. Most
patients with angiographically diagnosed pulmonary embo-
lism showed a nearly complete restoration of pulmonary
perfusion in response to anticoagulant therapy, and none of
those with negative angiograms had symptomatic episodes
of pulmonary embolism during follow-up. In addition, the
patients with pulmonary embolism described here were
representative of the full spectrum of the disease in terms
of clinical setting at onset (the hospital and the community),
and extent of pulmonary vascular obstruction (from minor
to massive).
The 3 clinical models performed quite differently as
regards their predictive accuracy. In the present study, the
Geneva model turned out to have no predictive value. This
model was derived from a database of outpatients presenting
to the emergency ward with suspected pulmonary embolism.
In the original report, patients with pulmonary embolism

TABLE 3. Proportion of Patients and Frequency of Pulmonary Embolism in the 3 Clinical Probability Categories According to Each
Prediction Model
Geneva Model Wells Model Pisa Model
Patients Patients With PE Patients Patients With PE Patients Patients With PE
Clinical Probability (%) (%, 95% CI) (%) (%, 95% CI) (%) (%, 95% CI)
Low 26 (12)* 13 (50, 30–70)* 64 (30) 8 (12, 6–23) 79 (37) 4 (5, 1–12)
Intermediate 128 (60) 50 (39, 31–48) 118 (55) 64 (54, 45–63) 79 (37)y 33 (42, 31–53)
High 61 (28) 30 (49, 36–62) 33 (15)z 21 (64, 45–80) 57 (26) 56 (98, 91–100)y
Abbreviations: See previous tables. CI = confidence interval.
*Significantly different (p < 0.001) from Wells and Pisa models.
y
Significantly different (p < 0.001) from Wells and Geneva models.
z
Significantly different (p < 0.005) from Geneva and Pisa models.

112 n 2005 Lippincott Williams & Wilkins

Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Medicine ! Volume 84, Number 2, March 2005
FIGURE 3. Box-and-whisker plot of the extent of scinti-
graphically detectable pulmonary vascular obstruction as a
function of the clinical probability in 93 patients with
pulmonary embolism. Line in box: 50th percentile (median);
limits of box: 25th and 75th percentile; whiskers: 10th and
90th percentile.
featured older age, higher prevalence of predisposing risk
factors, and more severe arterial blood gas abnormalities
than those without embolism17. In our study sample, instead,
there was no difference between the 2 diagnostic groups as to
age, prevalence of risk factors, and arterial blood gas profile.
Radiographic abnormalities such as band atelectasis or
elevated hemidiaphragm, which are included in the Geneva
model, were found with nearly equal frequency in patients
with and without pulmonary embolism. The Geneva model
therefore seems better suited for evaluating outpatients who
have a low prevalence of risk factors, such as recent surgery
or trauma, and of comorbid conditions that may affect
pulmonary gas exchange.
n 2005 Lippincott Williams & Wilkins
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Clinical Probability of Pulmonary Embolism
The Canadian model introduced by Wells et al14 was
derived from a cohort of both in- and outpatients. As
opposed to the Geneva model, it does not require measuring
arterial blood gases. This model heavily depends on
subjective judgment as to whether an alternative diagnosis
is as likely as or more likely than pulmonary embolism, and,
as such, it can hardly be standardized. Nevertheless, in our
patient population, it performed better than the Geneva
model. With this model, however, fewer patients were
categorized as having a high probability than with the 2 other
models, and only 64% of them were found to have pul-
monary embolism. The frequency of the disease in the low
probability category was somewhat higher than that reported
by Wells14. This is likely the consequence of the higher
prevalence of pulmonary embolism in our study as compared
to Wells’ study. The Wells model, therefore, is better suited
to rule out rather than to rule in the diagnosis of pulmonary
embolism, and its performance is likely to be better in
a clinical setting where the prevalence of the disease is
expected to be low16.
The Pisa model rests on the evaluation of relevant
clinical symptoms and signs, and the interpretation of the
electrocardiogram and the chest radiograph. In contrast to
the other models, it includes variables associated with a
decreased likelihood of pulmonary embolism. This gives the
model much greater flexibility. However, it is undoubtedly
more complex than the Geneva and Wells models, and
requires medical expertise in identifying electrocardiograph-
ic and chest radiographic abnormalities. It should be noted,
however, that identification of the electrocardiographic
abnormalities was based on carefully defined criteria that
have been known for many years13. Similarly, the descrip-
tion of the chest radiographic features of pulmonary em-
bolism included in the model can be traced back as far as 40
years ago4.
The quality of the chest film may be a limiting factor in
evaluating the radiographic abnormalities included in the
Pisa model. For example, an overexposed film would make
the assessment of oligemia nearly impossible. We do believe,
however, that digital chest radiography, which is now avail-
able in most hospitals, may overcome the above limitation
as it allows image contrast enhancement and magnification
of regions of interest within the lung fields.
With the Pisa model, substantially fewer patients fell
in the intermediate probability category, far fewer were found
to have pulmonary embolism among those who had been
rated a low probability, and far more had the diagnosis con-
firmed among those with a high probability than with either
the Geneva or Wells model. In patients with pulmonary
embolism, there was a strong relation between clinical prob-
ability predicted by the model and severity of the disease,
the extent of embolization being highest in patients with
high probability and lowest in those with low probability.
113
Miniati et al
The prevalence of pulmonary embolism in our sample
(43%) was substantially higher than that reported by
others1,14,16,17. Such high prevalence increases the preva-
lence of the disease in patients who are rated as having high
clinical probability. At the 90% probability cutoff, the Pisa
model yielded a 60% sensitivity, a 99% specificity, and a
98% positive predictive value for pulmonary embolism (see
Figure 2, Table 3). By applying the above sensitivity and
specificity values to a population with a 20% prevalence of
pulmonary embolism, the positive predictive value decreases
to 94%, a value still much higher than that attained by the
Geneva and Wells models.
The Pisa model may prove useful in assisting
physicians to define precisely the pretest probability of
pulmonary embolism. This can be used in turn to estimate
the posterior probability of the disease after appropriate
objective testing, by means of the Bayes rule of conditional
probability. For example, a pretest (clinical) probability
>50% associated with a perfusion scan suggestive of
pulmonary embolism yields a posterior probability suffi-
ciently high (93%–100%) to justify the institution of
anticoagulant therapy6. By contrast, a low (<10%) pretest
probability paired with a perfusion scan not suggestive of
pulmonary embolism makes the diagnosis of pulmonary
embolism very unlikely (posterior probability <2%)6. In our
experience, pulmonary angiography is required in only a
minority of patients (& 15%), those in whom clinical
probability and lung scan findings are discordant9. Thus,
incorporating a standardized assessment of the clinical
probability with lung scan results may effectively reduce
the cost of diagnostic procedures for pulmonary embolism,
and minimize the radiation burden to the patients.
ACKNOWLEDGMENTS
The authors thank the following physicians who took
part in the study: Carolina Bauleo, Laura Carrozzi, Giosué
Catapano, Alba Dainelli, Giorgio Di Ricco, Bruno Formichi,
Carlo Marini, Renato Prediletto, Elvio Scoscia, and Lucia
Tonelli.
REFERENCES
1. Chagnon I, Bounameaux H, Aujesky D, Roy PM, Gourdier AL, Cornuz
J, Perneger T, Perrier A. Comparison of two prediction rules and
implicit assessment among patients with suspected pulmonary embo-
lism. Am J Med. 2002;113:269–275.
2. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas
114
Copyr ight © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Medicine ! Volume 84, Number 2, March 2005
under two or more correlated receiver operating curves: a nonparametric
approach. Biometrics. 1988;44:837–845.
3. Fedullo PF, Tapson VF. The evaluation of suspected pulmonary
embolism. N Engl J Med. 2003;349:1247–1256.
4. Fleischner FG. Pulmonary embolism. Clin Radiol. 1962;13:169–182.
5. Meyer G, Collignon MA, Guinet F, Jeffrey AA, Barritault L, Sors H.
Comparison of perfusion lung scanning and angiography in the
estimation of vascular obstruction in acute pulmonary embolism. Eur
J Nucl Med. 1990;17:315–319.
6. Miniati M, Monti S, Bottai M. A structured clinical model for pre-
dicting the probability of pulmonary embolism. Am J Med. 2003;114:
173–179.
7. Miniati M, Pistolesi M, Marini C, Di Ricco G, Formichi B, Prediletto R,
Allescia G, Tonelli L, Sostman DH, Giuntini C (The PISA-PED
Investigators). 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. 1996;154:1387–1393.
8. Miniati M, Prediletto R, Formichi B, Marini C, Di Ricco G, Tonelli L,
Allescia G, Pistolesi M. Accuracy of clinical assessment in the diag-
nosis of pulmonary embolism. Am J Respir Crit Care Med. 1999;159:
864–871.
9. Miniati M, Monti S, Bauleo, C, Scoscia E, Tonelli L, Dainelli A,
Catapano G, Formichi B, Di Ricco G, Prediletto R, Carrozzi L, Marini
C. A diagnostic strategy for pulmonary embolism based on standardized
pretest probability and perfusion lung scanning—A management study.
Eur J Nucl Med Mol Imaging. 2003;30:1450–1456.
10. Mullins MD, Becker DM, Hagspiel KD, Philbrick JT. The role of
volumetric computed tomography in the diagnosis of pulmonary
embolism. Arch Intern Med. 2000;160:293–298.
11. Musset D, Parent F, Meyer G, Maitre S, Girard P, Leroyer C, Revel MP,
Carette MF, Laurent M, Charbonnier B, Laurent F, Mal H, Nonent M,
Lancar R, Grenier P, Simonneau G, for the Evaluation du Scanner
Spirale dans l’Embolie Pulmonaire study group. Diagnostic strategy for
patients with suspected pulmonary embolism: a prospective multicentre
outcome study. Lancet. 2002;360:1914–1920.
12. Rathbun SW, Raskob GE, Whitsett TL. Sensitivity and specificity of
helical computed tomography in the diagnosis of pulmonary embolism:
a systematic overview. Ann Intern Med. 2000;132:227–232.
13. Stein PD, Dalen JE, McIntyre KM, Sasahara AA, Wenger NK, Willis
PW III. The electrocardiogram in acute pulmonary embolism. Prog
Cardiovasc Dis. 1975;17:247–257.
14. Wells PS, Ginsberg JS, Anderson DR, Kearon C, Gent M, Turpie AG,
Bormanis J, Weitz J, Chamberlain M, Bowie D, Barnes D, Hirsh J. Use
of a clinical model for safe management of patients with suspected
pulmonary embolism. Ann Intern Med. 1998;129:995–1005.
15. Wells PS, Anderson DR, Rodger M, Ginsberg JS, Kearon C, Gent M,
Turpie AG, Bormanis J, Weitz J, Chamberlain M, Bowie D, Barnes D,
Hirsh J. Derivation of a simple clinical model to categorize patients
probability of pulmonary embolism: increasing model utility with the
SimpliRED D-dimer. Thromb Haemost. 2000;83:416–420.
16. Wells PS, Anderson DR, Rodger M, Stiell I, Dreyer JF, Barnes D,
Forgie M, Kovacs G, Ward J, Kovacs MJ. Excluding pulmonary
embolism at the bedside without diagnostic imaging: management of
patients with suspected pulmonary embolism presenting to the
emergency department by using a simple clinical model and D-dimer.
Ann Intern Med. 2001;135:98–107.
17. Wicki J, Perneger TV, Junod AF, Bounameaux H, Perrier A. Assessing
clinical probability of pulmonary embolism in the emergency ward. A
simple score. Arch Intern Med. 2001;161:92–97.
n 2005 Lippincott Williams & Wilkins