Content uploaded by Antoine Elias
Author content
All content in this area was uploaded by Antoine Elias on May 24, 2016
Content may be subject to copyright.
Available via license: CC BY-NC 4.0
Content may be subject to copyright.
Prognostic models in acute pulmonary
embolism: a systematic review and
meta-analysis
Antoine Elias,
1,2
Susan Mallett,
3
Marie Daoud-Elias,
1
Jean-Noël Poggi,
1
Mike Clarke
4
To cite: Elias A, Mallett S,
Daoud-Elias M, et al.
Prognostic models in acute
pulmonary embolism:
a systematic review and
meta-analysis. BMJ Open
2016;6:e010324.
doi:10.1136/bmjopen-2015-
010324
▸Prepublication history and
additional material is
available. To view please visit
the journal (http://dx.doi.org/
10.1136/bmjopen-2015-
010324).
Received 22 October 2015
Revised 18 March 2016
Accepted 22 March 2016
1
Department of Vascular
Medicine, Sainte Musse
Hospital, Toulon La Seyne
Hospital Centre, Toulon,
France
2
DPhil Programme in
Evidence-Based Healthcare,
University of Oxford, Oxford,
UK
3
Department of Primary Care
Health Sciences, University of
Oxford, Oxford, UK
4
Northern Ireland Network for
Trials Methodology Research,
Queen’s University Belfast,
Belfast, UK
Correspondence to
Dr Antoine Elias;
antoine.elias@free.fr
ABSTRACT
Objective: To review the evidence for existing
prognostic models in acute pulmonary embolism (PE)
and determine how valid and useful they are for
predicting patient outcomes.
Design: Systematic review and meta-analysis.
Data sources: OVID MEDLINE and EMBASE, and The
Cochrane Library from inception to July 2014, and
sources of grey literature.
Eligibility criteria: Studies aiming at constructing,
validating, updating or studying the impact of prognostic
models to predict all-cause death, PE-related death or
venous thromboembolic events up to a 3-month
follow-up in patients with an acute symptomatic PE.
Data extraction: Study characteristics and study
quality using prognostic criteria. Studies were selected
and data extracted by 2 reviewers.
Data analysis: Summary estimates (95% CI) for
proportion of risk groups and event rates within risk
groups, and accuracy.
Results: We included 71 studies (44 298 patients).
Among them, 17 were model construction studies
specific to PE prognosis. The most validated models
were the PE Severity Index (PESI) and its simplified
version (sPESI). The overall 30-day mortality rate was
2.3% (1.7% to 2.9%) in the low-risk group and 11.4%
(9.9% to 13.1%) in the high-risk group for PESI (9
studies), and 1.5% (0.9% to 2.5%) in the low-risk
group and 10.7% (8.8% to12.9%) in the high-risk
group for sPESI (11 studies). PESI has proved
clinically useful in an impact study. Shifting the cut-off
or using novel and updated models specifically
developed for normotensive PE improves the ability for
identifying patients at lower risk for early death or
adverse outcome (0.5–1%) and those at higher risk
(up to 20–29% of event rate).
Conclusions: We provide evidence-based information
about the validity and utility of the existing prognostic
models in acute PE that may be helpful for identifying
patients at low risk. Novel models seem attractive for
the high-risk normotensive PE but need to be
externally validated then be assessed in impact studies.
INTRODUCTION
Venous thromboembolism (VTE), including
pulmonary embolism (PE) and deep vein
thrombosis (DVT), is a common and poten-
tially fatal disorder, despite improvements
in its management. The main short-term
complications of PE are all-cause death,
PE-related death, VTE events and bleeding.
In acute PE, there is a real clinical ques-
tioning and interest on how to choose the
appropriate management for a specific
patient.
1–4
Usual care in the early phase is to
treat patients in hospital and to use
Strengths and limitations of this study
▪Comprehensive systematic review and
meta-analysis of prognostic models in acute pul-
monary embolism (PE) that was not restricted to
only clinical prediction rules and derivation or
validation studies, but was expanded to all avail-
able prediction/predictive models including
update and impact studies to inform clinical
decisions, with broad search strategy and prede-
fined selection criteria, and no data or language
restriction.
▪Study quality assessed by using prognostic cri-
teria more appropriate than diagnostic tools for
prognostic studies and by using a domain
approach, and full details on study character-
istics provided.
▪Quantitative analyses performed for ‘both stable
and unstable PE’and for ‘stable PE’, for each
outcome/time point separately, and for every
available risk cut-off for a model to assess how a
model performs along the risk scale, and results
provided in terms of summary estimates of pro-
portion of risk groups and event rates within risk
groups in ‘absolute risk’more meaningful for
clinicians and more appropriate for the study of
prognosis, and summary estimates of sensitivity
and specificity (accuracy).
▪Not included in the systematic review because
they deserve specific reviews, studies performed
in selective populations such as in asymptomatic
PE, unstable PE, patients with cancer, elderly
patients, or studies restricted to only a single
risk group (low-risk or high-risk group) with the
exception of impact studies.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 1
Open Access Research
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
anticoagulants in normotensive patients or rescue
thrombolysis in patients with hypotension or cardiogenic
shock. In normotensive patients, other options are avail-
able that might depend on the estimated risk for
adverse outcomes for the individual patient and include
home treatment for patients at low risk, or closer moni-
toring and escalating therapy for patients at high risk.
Prognostic tools seek to classify patients by their risk of
adverse outcomes to help select the appropriate man-
agement for an individual patient. The European
Society of Cardiology 2008 guidelines
1
suggest using the
haemodynamic status based on systolic blood pressure
(SBP) and signs of right ventricular dysfunction (RVD)
and ischaemia. The American Heart Association (AHA)
Scientific Statement
2
considers that patients with low-risk
stable PE with no prognostic markers of RVD/ischaemia
‘may still have significant rates of morbidity and mortal-
ity that are functions of older age and comorbidities’.
Systematic reviews of individual prognostic variables
5–9
have shown inconsistent results across studies, leading to
the development of prognostic models. The large
number of studies of existing prognostic models and the
differences between them suggest the need for a system-
atic review that could assess study quality, synthesise find-
ings across studies, assess the validity of the original
models and provide the best estimate from the model.
Determining which prognostic models work, and which
work best, may impact on clinical decisions and on
research.
This systematic review assesses the characteristics and
quality of studies of prognostic models in patients with
acute PE. It investigates the performance and utility of
the prognostic models, along their different phases of
development in construction, validation, update and
impact studies.
METHODS
Selection criteria
Criteria for considering studies for this review were
derived from previous reviews of prognostic models.
10–13
To be eligible, studies had to have developed a prognos-
tic model to predict the outcome of patients with an
acute symptomatic PE, with the specific aim of construct-
ing a new prognostic model or validating, updating or
studying the impact of an existing one. The models
needed to contain a combination of at least two prog-
nostic variables to predict patient outcome, incorporated
from across the following characteristics: demographic/
clinical, biological and imaging related. Patient out-
comes were death, PE-related death or VTE (DVT or
PE) recurrence. Only hospital-based studies were consid-
ered for inclusion.
Prognostic studies validating individual prognostic
indicators, population-based studies, studies performed
in selective populations (asymptomatic PE, unstable PE,
patients with cancer, elderly patients, specific risk group
except when assessed in impact studies) or considering
only surrogate outcomes were not included. As they
include prognostic and practical variables for hospitalisa-
tion, checklists of exclusion criteria from early discharge
or home treatment were not considered strictly prognos-
tic. Thus, studies on safety of early discharge or home
treatment based on these criteria were not included
unless their objective was clearly to derive and validate
the checklist or to compare it with existing prognostic
models. Unpublished articles or those published only as
conference abstracts, discussion articles on prognostic
models or indicators or patient management were not
included. If a study was published more than once, the
first published article or the article combining cumula-
tive results from different studies was included.
Search methods and search strategy
Studies were sought through electronic databases: OVID
versions of MEDLINE and EMBASE, and The Cochrane
Library, from inception to July 2014. Given that there
are many conference presentations in this area, we cor-
responded with authors and further searched in Google
Scholar and other databases to check for pending or
recently published full articles. The search also covered
various sources of ‘grey literature’to identify published,
unpublished and ongoing studies (see online supple-
mentary protocol-S). Further searches included the ref-
erence lists of relevant articles and books, citation
indexes and hand-search of issues in relevant journals.
There were no date or language restrictions on the
searches.
The following Medical Subject Headings (MESH)
terms and text words were used: pulmonary embolism.
tw., exp Pulmonary Embolism/, exp Venous
Thromboembolism/, prognos*.tw., exp Prognosis/,
predict*.tw., cohort stud*.tw, exp Cohort Studies/,
course.tw., exp Incidence/, score.tw., model*.tw., index.
tw., rule.tw., criteria.tw., tool*.tw., severity index*.tw.,
geneva.tw., davies.tw. These were combined as shown in
online supplementary box-s.
Data collection
Selection strategy
Potentially eligible studies were identified by examining
titles and abstracts or other summaries as available. Full
articles were obtained to assess eligibility criteria, before
critical appraisal. Study identification was performed
by two independent and blinded assessors to avoid
selection bias. Disagreements were resolved through
discussion.
Data extraction content
The data extraction form (online supplementary
protocol-S) had six detailed sections: information about
the review and verification of study eligibility, informa-
tion about the study, assessment of study quality, assess-
ment of methods relevant to the development phase of
the model, general findings and results of model
performance.
2Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Data extraction management
An electronic data collection form was created in Excel.
An explanation for the data extraction items was avail-
able in corresponding cells, but there were no coding
instructions. Contents were adjusted throughout the
data extraction process. Data were extracted by one
reviewer and checked by a second reviewer to obtain
reliable information on study findings and study meth-
odology. Disagreements were resolved by discussion.
Dealing with missing data
Study investigators were contacted by email for details
not reported in the published reports. A standard data
collection form was used to assist with this, when
needed.
Assessment of study quality
All studies were assessed for methodological quality
10 13–18
and risk of bias in regard to study design, analysis and
reporting according to predefined criteria for type of
study, sample of patients (proper formation of inception
cohort, description of referral pattern), adequacy of prog-
nostic factors, adequacy of outcome measures, blind assess-
ment of outcomes, completeness of follow-up, sample size,
treatment, missing data and adjustment for all potential
confounders (online supplementary protocol-S).
Summary scores were not used to identify studies of
high quality or low quality,
17 19
but ‘risk of bias’figures
were produced to show the findings graphically.
Data analysis
The descriptive analysis addressed reporting on type and
phase of development of the model, study character-
istics, population and setting patterns, patient character-
istics, prognostic information, prognostic modelling and
other data analyses, and study quality.
Quantitative analyses were performed at each relevant
model risk cut-off level for various outcomes and time
points. These used data from model validation and
model update studies, with and without the results of
the model construction study. Outcomes were all-cause
mortality, PE-related mortality, VTE recurrence, major or
clinically relevant bleeding and composite outcomes.
These were assessed for the duration of hospitalisation,
at 1 and at 3 months. Summary estimates and their 95%
CIs were calculated for the population event rates, for
the proportion of patients in risk categories and the inci-
dence of events within risk categories, and were
obtained as weighted average by the inverse variance
method. When data were available from at least four
studies, summary estimates of sensitivity and specificity
and summary receiver operating characteristic (sROC)
curves were obtained using the bivariate random-effects
model.
20 21
Studies that compared models either within the same
article or in different articles or shared the same cohort
in various model development phases were included
once for the analysis of population event rates but as
often as the number of prognostic models that were
assessed for the analysis of risk group distribution, the
incidence of events within risk groups and the prognos-
tic performance of a model. Homogeneity of study
designs, differences because of the case mix and statis-
tical heterogeneity (Cochran’s Q test, Higgins I
2
statistic)
were assessed to decide whether to combine the results
of individual studies. When results were combined, a
fixed-effects or a random-effects model was used
depending on whether the effect was similar or variable
across studies. The likely influence of the presence or
absence of bias was examined in sensitivity analyses and
funnel plots (observed and imputed studies, Egger’s
regression intercept). Subgroup analyses were per-
formed for ‘stable’and ‘both stable and unstable’PE
patient groups, ‘prospective’or ‘retrospective’studies
and with regard to the phase of development (deriv-
ation, internal validation and external validation/
update) of studies.
Analysis was performed using metandi
22
and
GLLAMM
23
modules in Stata/SE (V.13.1), and using
Comprehensive Meta-Analysis (CMA V.2. 2.064).
Data synthesis
To decide whether a prognostic model would be helpful
for clinical practice, the following factors were required
in external validation studies: direction and size of
effect, and effect consistency across studies in the
meta-analysis; and strength of evidence for the effect
based on study quality and statistical measure of uncer-
tainty.
17
For a valid model to be acceptable, its utility
should ideally have been demonstrated in an impact
study, such as in a one-arm management study using the
model or in a randomised trial comparing outcomes for
patients in whom the model is used to inform decision-
making and those for whom it is not used.
RESULTS
Search results
Figure 1 shows the flow of studies through the review
following Preferred Reporting Items for Systematic
Reviews and Meta-Analyses (PRISMA) guidelines.
24
The
numbers of potentially relevant records identified and
screened in each database were 3127 in MEDLINE,
5819 in EMBASE and 1000 in The Cochrane Library.
Most of these were not eligible because they were
clearly not relevant (2711, 5276 and 945, respectively)
or because they did not meet selection criteria (328,
459 and 51, respectively). From 176 publications that
were retained from these databases (88, 84 and 4,
respectively) and from 10 additional publications from
other sources, 75 were removed as duplicate records
that had been found in more than one database and 40
were excluded for various reasons (figure 1): unmet cri-
teria,
25–48
duplicate,
49–52
selective population,
53–56
journal club,
57–59
comment letter,
60
algorithm,
61
long-term
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 3
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
outcome,
62
reliability study
63
and very old variables in
the model.
64
Finally, 71 studies were included in the review, 62 from
databases
65–126
and 9 from other sources that were
mainly lists of references.
127–135
Among studies included
from databases, most were in MEDLINE and EMBASE,
but four studies were not found in MEDLINE
76 80 95 102
and one study was not in EMBASE.
77
Two studies were
identified in The Cochrane Library,
71 100
but these were
also found in MEDLINE and EMBASE. Some studies
were based on retrospective analysis of prospectively col-
lected data from previous diagnostic studies,
136–141
one
137
of which was used in two model construction
studies
68 69
and two external validation studies.
70 97
In total, among the 71 studies, 64 were found to be
including variables specific to the domain of PE and its
prognosis: 17 were identified as model construction
studies,
65 67–69 85 87 89 90 92 110 118 121–124 129 135
41 as
model external validation or model update
studies
70 72 74–78 80–84 86 88 91 93–97 99 101 102 104 106–109
111–117 119 120 127 130 131 134
and 6 as measuring model
impact.
66 71 100 128 132 133
For the remaining seven studies,
variables in the model were either originally not specific
to PE
98 103 105
or to its prognosis
73
or concerned a hos-
pital checklist
79 125 126
that is a combination of prognostic
and practical variables for outpatient management.
Descriptive analysis
Prognostic models
We identified three types of models related to risk strati-
fication of patients with PE (table 1 and see online sup-
plementary table S1). Type 1 model includes variables
specific to the domain and the prognosis of PE. Type 2
applies models that are used in other domains such as
the Global Registry of Acute Coronary Events
(GRACE)
103
or the Charlson Comorbidity Index,
98 105
or applies diagnostic prediction rules
73
for the assess-
ment of PE prognosis. Type 3 models incorporate hos-
pital checklist variables as exclusion criteria from early
discharge or home treatment.
79 125 126
Figure 1 Flow chart of search
results.
4Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 1 Variables used in the original prognostic models—prognostic variables included in the final models (only type 1 prognostic models presented) and points assigned (in numbers) to each variable when applied. Rule stands for scoring rule, score for
simplified scoring rule or scoring system
Models GPS
No
acronym
(Polish
model) PESI
PE
Prognostic
Algorithm
No
acronym
(Spanish
model)
No
acronym
(Dutch
model) EMEP
No
acronym
( Japanese
model)
No
acronym
(Chinese
model)
Simplified
PESI PREP UPS
No
acronym
Simple Score
FAST SIRS-WCC
No
acronym
eStiMaTe
(PROTECT)
Model construction
studies (derivation
samples)
(Wicki
et al
122
)
(Kostrubiec
et al
90
)
(Aujesky
et al
68
)
(Aujesky
et al
69
)
(Uresandi
et al
118
)
(Agterof
et al
67
)
(Volschan
et al
121
)
(Yamaki
et al
123
)
(Zhu
et al
124
)
( Jimenez
et al
85
)
(Sanchez
et al
110
)
(Agterof
et al
65
)
(Huang
et al
129
)
(Lankeit
et al
92
) and
(Dellas et al
80
)
(Jo
et al
89
)
(Bova
et al
135
)
(Jiménez
et al
87
)
N of candidate variables 14 NR 29 29 >30 7 20 16 NR 11 >30 7 38 NR NR Unclear 13
N of variables in final
model
6 2 11 10 7 2 5 7 8 6 5 4 4 3 4 4 4
Proposed risk
assessment method
Score Rule Score Rule Score Rule Score Score Equation Score Score Rule Score Score None Score Calculator
Type of selected
variables in final model
Total
N*
Demographic and clinical characteristics
Age 6 Age, in
year
x1 3 1 x
Gender (male sex) 2 10 2
Syncope 2 x 1.5
Leg symptoms 1 1
Immobilisation/bed
rest
222
No recent surgery 1 1
Active cancer 8 2 30 x 2 1 6 x x
Metastatic cancer 1 4
Non-metastatic cancer 1 2
Previous DVT/PE 2 1 2
Heart failure 3 1 10 x
Chronic lung disease 3 10 x x
Chronic cor pulmonale 1 4
Heart failure or
chronic lung disease
2 1 x
Chronic renal disease 1 x
Cerebrovascular
disease
1x
Recent bleeding 1 4
Inadequate
anticoagulation
1 1
Clinical findings
Cardiogenic shock 2 6x
HR 12 20 x x 2 x 1 x x 2 x 1 x
SBP 6 2 30 x 1 2x
RR 3 20 2 x
Temperature 2 20 x
Altered mental status 4 60 x 10 x
PaO
2
/SpO
2
5 1 20 x 1 x
Accentuation of P2 1 x
Laboratory findings
WCC 3 xxx
Creatinine 1 3
CPK 1 x
cTnI 1 x
cTnT 1 x
hscTn 0
cTn 1 2
Continued
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 5
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Among the 17 ‘type 1’prognostic models that were
identified in model construction studies,
8
68 69 85 87 92 110 118 122
underwent external validation or
were updated; some of them in many studies.
Three
68 118 122
were assessed in impact studies.
71 100 128
There are two other models that were tested in impact
studies,
66 132
without being previously reported in a con-
struction study. Tabl e 1 shows that the variables that were
mostly included in final models were in descending
order: heart rate, active cancer, age, SBP, peripheral
oxygen saturation and altered mental status.
Most of the studies identified for this review were
external validation or update studies.
Study characteristics
Research objective, type of study and study design,
setting, country, number of participating centres,
patients (age, sex and haemodynamic status), outcomes,
time points and corresponding event rates are displayed
in online supplementary tables S2.1–S2.5.
In model construction studies (n=17) (see online sup-
plementary table S2.1), the objective was usually to risk-
stratify patients (13 studies) rather than to identify
low-risk patients. Nine studies were prospective, seven
reported on stable PE subgroup and seven were pro-
spective and concerned stable PE as well. In 8 studies,
patients were recruited in emergency departments
(EDs) and in 13 studies, inception started on admission.
Nine were multicenter studies. The PE Severity Index
(PESI)
68
and Prognostic Algorithm
69
cohort included
the largest derivation and internal validation sets, but
the study was retrospective, had missing data and lack of
information on PE diagnosis and on the type of treat-
ment received. Four studies
87 110 121 135
were sound
methodologically and clinically relevant and present the
following characteristics: prospective multicenter study,
inception on admission, patients from ED, patients with
stable PE reported, adequate early outcome and time
point.
In model external validation studies (n=36) (see
online supplementary table S2.2), the main objective
was to identify low-risk patients. The other objectives
were to assess added value of prognostic variables,
head-to-head comparison or identification of patients
with stable PE at high risk. Among the prospective
studies (n=9) in stable and unstable PE, one study
102
validated the European Society of Cardiology (ESC)
model in terms of 30-day mortality and the other one
compared Hestia checklist criteria with simplified PESI
(sPESI) in terms of 7-day, 30-day and 3-month death; in
stable PE, Geneva prognostic score (GPS),
74
PESI,
104 111
sPESI,
80 101
ESC,
72 80
FAST
80
and eStiMaTe
87
models
were validated in seven studies, of which one study
80
compared three models FAST, sPESI and ESC. Reliability
was also assessed in two studies.
75 106
Model update studies (n=15) (see online supplemen-
tary table S2.3) assessed the value of adding a two-test or
a three-test strategy to PESI, based on a combination of
Table 1 Continued
Models GPS
No
acronym
(Polish
model) PESI
PE
Prognostic
Algorithm
No
acronym
(Spanish
model)
No
acronym
(Dutch
model) EMEP
No
acronym
( Japanese
model)
No
acronym
(Chinese
model)
Simplified
PESI PREP UPS
No
acronym
Simple Score
FAST SIRS-WCC
No
acronym
eStiMaTe
(PROTECT)
Model construction
studies (derivation
samples)
(Wicki
et al
122
)
(Kostrubiec
et al
90
)
(Aujesky
et al
68
)
(Aujesky
et al
69
)
(Uresandi
et al
118
)
(Agterof
et al
67
)
(Volschan
et al
121
)
(Yamaki
et al
123
)
(Zhu
et al
124
)
( Jimenez
et al
85
)
(Sanchez
et al
110
)
(Agterof
et al
65
)
(Huang
et al
129
)
(Lankeit
et al
92
) and
(Dellas et al
80
)
(Jo
et al
89
)
(Bova
et al
135
)
(Jiménez
et al
87
)
BNP 1 x x
NT-proBNP 2 x x
D-dimer 3 x xx
H-FABP 1 1.5
Imaging findings
US-DVT 3 1 1 x
US-proximal DVT 1 1
TTE-RVD 1 x
TTE-RV/LV ratio 2 xx
TTE-SPAP 1 x
CT-RVD 1 x
RVD (TTE or CT) 1 2
*Total N is the number of times a variable has been included across studies.
BNP, brain natriuretic peptide; CK, creatine kinase; cTnI, cardiac troponin I; cTnT, cardiac troponin T; DVT, deep vein thrombosis; EMEP, Estudo Multicêntrico de Embolia Pulmonar; GPS, Geneva prognostic score; H-FABP, heart-type fatty acid-binding
protein; hscTnT, high-sensitive cardiac troponin; NR, not reported; NT-proBNP, N-terminal proBNP; PE, pulmonary embolism; PESI, PE Severity Index; PREP, Facteurs PRonostiques dans l’Embolie Pulmonaire; RR, relative risk; RVD, right ventricular
dysfunction; SBP, systolic blood pressure; SIRS-WCC, systemic inflammatory response system—WCC; SPAP, systolic pulmonary artery pressure; TTE, transthoracic echocardiography; UPS, Utrecht Prediction Score; US-DVT, ultrasound-detected DVT;
WCC, white cell count.
6Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
RVD, cardiac troponin I (cTnI) and ultrasound
(US)-detected DVT, to identify patients with high-risk
PE;
86
assessed the additive value of CT-RVD to Hestia,
126
the additive value of transthoracic echocardiography
(TTE)-RVD,
111
brain natriuretic peptide (BNP)
111
or
cTnI to PESI,
104 111
or TTE-RVD and BNP to Facteurs
PRonostiques dans l’Embolie Pulmonaire (PREP),
110
or
the additive value of cardiac troponin T (cTnT),
101
high-
sensitive cTnT
94 101
or N-terminal proBNP
(NT-proBNP)
95
to sPESI, in order to identify patients at
lower risk or at higher risk; or assessed the effect of
cause of death classification on sPESI whether combined
or not to cTnI.
109
Eight studies were prospective and spe-
cific to stable PE. Retrospective studies assessed the addi-
tive value of CT-RVD,
112
of white cell count (WCC) and
systemic inflammatory response syndrome (SIRS),
89
of
NT-proBNP, cTnI or CT-RVD
77
to PESI; the additive
value of cTn to PESI
96
or to sPESI;
115
and the additive
value of TTE-RVD to Shock Index.
117
Six studies measured model impact (see online supple-
mentary table S2.4). Four were performed in patients
identified as low risk by the model, assessing the safety of
early discharge or outpatient treatment. This was done
in a one-arm management study for GPS
128
and
NT-proBNP,
66
and in a randomised controlled trial
(RCT) where outpatient management was compared
with standard hospitalisation (inpatient treatment)
either by using the Spanish model
100
or by using PESI in
the Outpatient Treatment of Pulmonary Embolism
(OTPE) study.
71
The Pulmonary EmbolIsm
THrOmbolysis (PEITHO) randomised double-blind
trial
132
assessed the role of fibrinolytic therapy tenecte-
plase in normotensive patients with PE considered at
intermediate–high risk by TTE-RVD or CT-RVD, and
myocardial injury as indicated by a positive test for cTnI
or cTnT. Another study assessed the use and the impact
of model-based risk stratification on treatment decisions
and outcome in clinical practice.
133
All studies were mul-
ticenter and prospective except the latter, which was a
single-centre and retrospective study.
Study quality
Quality of individual studies and a summary of study
quality at different model development phases are
shown in online supplementary figures S1.1 to S6.2.
Among model construction studies (see online supple-
mentary figures S1.1 and S1.2), PREP,
110
esTiMaTe
87
and Bova’s study
135
models satisfied most of the study
quality criteria. Criteria that were often met (in at least
70% of the included studies) are: inclusion criteria def-
inition, characteristics description, length of follow-up
and criteria about the outcome (objectiveness, full def-
inition, appropriateness, proportion of patients with
known outcome). Criteria that were least often met (in
<30% of the included studies) are: sample completeness
(selection bias), reasons for lost to follow-up reporting,
similarity of outcome assessment for all study partici-
pants, differences with participants who did not
complete the study, availability of prognostic variables
and justification of sample size.
In model validation studies (see online supplementary
figures S2.1 and S2.2), including those that also updated
the model, the most recent validation studies
87 101 102 111
satisfied most of the study quality criteria. The most
often met criteria concerned specification of inclusion
criteria, sample selection, population characteristics
description, follow-up, outcome (objective, fully defined,
appropriate, known for a high proportion of patients),
and full definition of prognostic variables and descrip-
tion of treatment. Major concerns were about sample
completeness, reporting on reasons for lost to follow-up
and differences with participants who did not complete
the study, justification of sample size and statistical
analysis.
In model update studies (see online supplementary
figures s3.1 and s3.2), three studies
95 101 111
satisfied
most of the criteria. The summary of study quality shows
that most of the criteria were often met, but there are
concerns about sample completeness and reporting on
reasons for lost to follow-up and differences with partici-
pants who did not end the study, justification of sample
size and how lost to follow-up treated.
Model impact studies (see online supplementary
figures S4.1 and S4.2) had the best compliance with
study quality criteria, with most of the key factors
provided.
66 71 100 132
For type 2 and 3 studies, the number of studies is too
small to draw general conclusions. In type 2 studies (see
online supplementary figures S5.1 and S5.2), although
inclusion was well defined, there are problems with
sample selection and sample completeness in
all
98 103 105
but one study,
73
in justification of sample size
and data analysis. In type 3 studies (see online supple-
mentary figures S6.1 and S6.2), the main concerns are
about a lack of information on reasons for lost to
follow-up, justification of sample size and data analysis
(missing data, lost to follow-up, statistical adjustment).
Quantitative analysis
Population event rates
There were a total of 44 298 patients if patients in the
cohorts account once, only regardless of how many
studies reported those patients, and if impact studies
that involved only low-risk or intermediate-risk patients
are excluded.
Analysis based on inverse-variance method with
random-effects model (see online supplementary table
S3) showed overall in-hospital, 30-day and 3-month point
estimate to be 6.5% (95% CI 4.9% to 8.5%) (20
studies), 7.4% (6.5% to 8.5%) (32 studies) and 6.8%
(5.7% to 8.1%) (11 studies), respectively, for mortality;
3.1% (2.6% to 3.8%) (7 studies), 4.0% (3.2% to 5.0%)
(15 studies) and 2.9% (2.2% to 3.7%) (6 studies),
respectively, for PE-related mortality; and 0.2% (0.0% to
1.5%) (3 studies), 1.1% (0.7% to 1.8%) (6 studies) and
2.6% (1.6% to 4.4%) (4 studies), respectively, for VTE
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 7
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
recurrence. Composite event rates, major and non-major
clinically relevant bleeding and fatal bleeding are also
displayed.
Risk groups
All-cause mortality
The distribution of risk groups for a model (proportion
of patients in low-risk or in high-risk groups) according
to specified cut-off levels and the incidence of all-cause
mortality within risk groups, expressed in % (95% CI),
are shown in detail (table 2). The following paragraphs
summarise the findings.
For GPS3vsGPS≤2 cut-off, mortality was reported
only in a single external validation study; the prevalence
of the low-risk group is 82% (77.3% to 85.9%), and the
incidence of 30-day mortality in the low-risk group is
9.8% (3.2% to 26.6%). The prevalence of high-risk
group is 18% (14.1% to 22.7%) and the incidence of
30-day mortality is 20.1% (11.4% to 33%).
For PESIIII vs PESI≤II cut-off, the overall prevalence
of low-risk group is 43.1% (39.6% to 46.6%) and the
overall in-hospital, 30-day and 3-month mortality rates
within low-risk group are 1.4% (1.2% to 1.8%) (5
studies), 2.3% (1.7% to 2.9%) (9 studies) and 1% (0.5%
to 1.8%) (6 validation studies), respectively. The overall
prevalence of high-risk patients is 56.9% (53.4% to
60.4%) and the overall in-hospital, 30-day and 3-month
mortality rates are 9.3% (8.4% to 10.4%), 11.4% (9.9%
to 13.1%) and 13.0% (8.7% to 18.9%), respectively.
Summary estimate of the prevalence and the mortality
rates from external validation studies are consistent with
those of the derivation and internal validation samples.
For sPESI, the overall prevalence of low-risk group is
36.3% (33.3% to 39.4%). The overall in-hospital, 30-day
and 3-month mortality rates are 0.3% (0% to 2.3%) (2
external validation studies), 1.5% (0.9% to 2.5%) (11
studies) and 0.8% (0.3% to 2.2%) (3 external validation
studies), respectively. The overall proportion of high-risk
patients is 63.7% (60.6% to 66.7%) and the overall
in-hospital, 30-day and 3-month mortality rates are 3.2%
(1.6% to 6.4%), 10.7% (8.8% to 12.9%) and 13.6%
(8.8% to 20.3%), respectively. Results are consistent
across derivation, internal validation and external valid-
ation samples.
With the algorithm (4 studies), 22% (19.3% to 24.9%)
of patients are classified as low risk. The in-hospital and
30-day mortality rates are 0.6% (0.4% to 1%) and 1.1%
(0.5% to 2.1%), respectively. The results in the valid-
ation and the derivation samples are similar.
For the PREP model (mortality reported in 1 external
validation study), 67.2% (61.7% to 72.3%) of patients
are classified as low risk. The 30-day and 3-month mor-
tality rates are 1% (0.2% to 3.9%) and 1.5% (0.5% to
4.5%), respectively, in the low-risk group, and 7.1%
(3.4% to 14.1%) and 9.1 (4.8 to 16.6), respectively, in
the high-risk group. In the original study, the incidence
of a 30-day composite outcome is 2.5% (1.2% to 4.9%),
37.3% (33.2% to 41.5%) of patients are classified as
high risk and the incidence of events is 17.7% (12.9% to
23.8%).
Shock Index classifies 80.3% (68% to 88.7%) of
patients in the low-risk group, but with a high 30-day
mortality rate of 10.7% (6.1% to 17.9%) and 24.1%
(13.6% to 39.1%) in the high-risk group.
With ESC, the low-risk group prevalence is 89.8%
(72.7% to 96.7%); the in-hospital and 30-day mortality
rates are 5% (3.6% to 7.1%) (3 studies) and 8.9% (4.7%
to 16.5%) (1 study), respectively, in the low-risk group,
and 41.1% (22.7% to 62.5%) and 26.5% (16.1% to
40.4%), respectively, in the high-risk group. These
results apply for ‘both stable and unstable’patients. For
the stable PE patient subgroup, for ‘intermediate–high’
risk cut-off (intermediate–high-risk patients have RVD
and myocardial injury, and the lower risk that includes
intermediate–low-risk and low-risk groups either one of
them or none), the proportion of the lower risk group is
74.4% (52.4% to 88.5%) (5 studies) and the in-hospital
mortality rate is 3.1% (1.4% to 6.8%) in the lower risk
group and 7.7% (4.7% to 12.2%) in the intermediate–
high-risk group.
Given the high event rate within the ‘low-risk’groups
from Shock Index and ESC, these would not qualify as
suitable for the identification of a low-risk but of a ‘high-
risk’patient group.
The performance of other models recently con-
structed is discussed in model update section.
Outcomes other than all-cause mortality
Tables 3 and 4show summary estimates with results
from studies that report outcomes in terms of PE-related
death, adverse outcome (clinical deterioration, haemo-
dynamic collapse), VTE and their combinations in ‘both
stable and unstable’(table 3), and separately in ‘stable’
(table 4) patient groups. Although many studies were
retrospective or were single studies, the results confirm
the predictive ability for PESI and sPESI at the optimal
cut-offs or less, preferably in combination with biomar-
kers, for identifying patients at low risk for these specific
outcomes as well, mainly in stable PE. They also confirm
the ability for some models at higher cut-offs for identi-
fying patients at very high risk (30-day event rate over
20%) such as PREP—ClinIII without or with BNP—
TTE-RVD, PESIIV, PESI V without or with TTE-RVD,
ESC high, Shock Index, eStiMaTe high, FAST 3, Bova
stage III. Most of the models combine biomarkers or
imaging-based modalities to clinical variables or to pre-
existing models. In most of the studies, the proportion
of patients at very low or at very high risk is low.
Type 2 and type 3 model studies
Table 5 shows the results of type 2 and type 3 model
studies. Revised Geneva score
73
(RGS) and simplified
Geneva score
73
(SGS) originally dedicated for the diag-
nosis of PE may provide at low cut-offs a small 3-month
mortality rate (0.9% (0.1% to 12.3%)) in patients at low
risk. Similarly, at the lowest cut-offs, GRACE ACS
103
and
8Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 2 In-hospital, 30-day and 3-month all-cause mortality, and distribution of risk groups for relevant risk cut-offs
Outcome time point
Prognostic model
Cut-off—risk groups
Prevalence of risk groups
N (%) (95% CI)
Incidence of events within risk groups
N (%) (95% CI)
Derivation Internal validation External validation Overall Derivation Internal validation External validation Overall
In-hospital mortality GPS-3 1 1
High (18.9) (14.1 to 24.9) 7.9 (2.6 to 21.8)
Low (81.1) (75.1 to 85.9) 0.6 (0.1 to 4.2)
PESI-III 11351135
High (59.1) (58.1 to 60) (59.3) (57.9 to 60.6) (55.8) (41 to 69.7) (58.4) (55.1 to 61.6) (8.8) (8.1 to 9.5) (9.8) (8.8 to 10.9) (9.8) (6.5 to 14.4) (9.3) (8.4 to 10.4)
Low (40.9) (40 to 41.9) (40.7) (39.4 to 42.1) (44.2) (30.3 to 59) (41.6) (38.4 to 44.9) (1.3) (1 to 1.7) (1.5) (1.1 to 2.1) (2.2) (1 to 4.7) (1.4) (1.2 to 1.8)
sPESI 2 2
High (53.9) (29.2 to 76.8) (3.2) (1.6 to 6.4)
Low (46.1) (23.2 to 70.8) (0.3) (0 to 2.3)
Algorithm 11241124
High (78.4) (77.6 to 79.2) (78.4) (77.3 to 79.5) (76.1) (55.1 to 89.2) (78) (75.1 to 80.7) (5.2) (4.7 to 5.7) (6.1) (5.4 to 6.9) (3.6) (0.7 to 16.2) (5.8) (4.8 to 7)
Low (21.6) (20.8 to 22.4) (21.6) (20.5 to 22.7) (23.9) (10.8 to 44.9) (22) (19.3 to 24.9) (0.4) (0.2 to 0.8) (0.9) (0.5 to 1.7) (0.6) (0.1 to 3.1) (0.6) (0.4 to 1)
ESC 3 3
High (10.2) (3.3 to 27.3) (41.1) (22.7 to 62.5)
Low (89.8) (72.7 to 96.7) (5) (3.6 to 7.1)
Shock Index 1 1
High (17.4) (12.2 to 24.2) (22.2) (10.3 to 41.4)
Low (82.6) (75.8 to 87.8) (3.9) (1.6 to 9)
30-day mortality GPS-3 1 1
High (18) (14.1 to 22.7) (20.1) (11.4 to 33)
Low (82) (77.3 to 85.9) (9.8) (3.2 to 26.6)
PESI-III 11791179
High (59.1) (58.1 to 60) (59.3) (57.9 to 60.6) (56.1) (48.6 to 63.3) (56.9) (53.4 to 60.4) (14) (13.1 to 14.9) (14.2) (13.1 to 15.5) (9.8) (8.0 to 11.9) (11.4) (9.9 to 13.1)
Low (40.9) (40 to 41.9) (40.7) (39.4 to 42.1) (43.9) (36.7 to 51.4) (43.1) (39.6 to 46.6) (2.1) (1.8 to 2.6) (2.6) (2 to 3.3) (1.8) (1.0 to 3.2) (2.3) (1.7 to 2.9)
sPESI 1 10111 1011
High (69.3) (66.4 to 72.1) (63.1) (59.8 to 66.3) (63.7) (60.6 to 66.7) (10.9) (8.8 to 13.4) (10.6) (8.6 to 13) (10.7) (8.8 to 12.9)
Low (30.7) (27.9 to 33.6) (36.9) (33.7 to 40.2) (36.3) (33.3 to 39.4) (1) (0.3 to 3) (1.6) (0.9 to 2.7) (1.5) (0.9 to 2.5)
PREP Clin—II 1 1
High (32.8) (27.7 to 38.3) (7.1) (3.4 to 14.1)
Low (67.2) (61.7 to 72.3) (1) (0.2 to 3.9)
Algorithm 11241124
High (78.4) (77.6 to 79.2) (78.4) (77.3 to 79.5) (76.1) (55.1 to 89.2) (78) (75.1 to 80.7) (11.5) (10.8 to 12.2) (11.7) (10.7 to 12.7) (7.9) (2.5 to 22.4) ( 11.6) (10.4 to 12.9)
Low (21.6) (20.8 to 22.4) (21.6) (20.5 to 22.7) (23.9) (10.8 to 44.9) (22) (19.3 to 24.9) (0.6) (0.3 to 1) (1.5) (0.9 to 2.4) (1.6) (0.6 to 4.5) (1.1) (0.5 to 2.1)
ESC 1 1
High (10.1) (8 to 12.6) (26.5) (16.1 to 40.4)
Low (89.9) (87.4 to 92) (8.9) (4.7 to 16.5)
Shock Index 2 2
High (19.7) (11.3 to 32) (24.1) (13.6 to 39.1)
Low (80.3) (68 to 88.7) (10.7) (6.1 to 17.9)
3-month mortality GPS-3 22
High (18) (14.1 to 22.7) 23.6 (14.2 to 36.6)
Low (82) (77.3 to 85.9) 4.9 (2.8 to 8.4)
PESI-III 6 6
High (49.6) (43.3 to 56.0) (13.0) (8.7 to 18.9)
Low (50.4) (44.0 to 56.7) (1) (0.5 to 1.8)
sPESI 3 3
High (53.8) (39.4 to 67.6) (13.6) (8.8 to 20.3)
Low (46.2) (32.4 to 60.6) (0.8) (0.3 to 2.2)
PREP Clin=II 1 1
High (32.8) (27.7 to 38.3) (9.1) (4.8 to 16.6)
Low (67.2) (61.7 to 72.3) (1.5) (0.5 to 4.5)
ESC, European Society of Cardiology; GPS, Geneva prognostic score; N, number of studies; PESI, Pulmonary Embolism Severity Index; PREP, Facteurs PRonostiques dans l’Embolie
Pulmonaire; sPESI, simplified PESI.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 9
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 3 PE-related death, adverse outcome (death, clinical deterioration or haemodynamic collapse), VTE, major bleeding and their combinations in ‘both stable and
unstable’PE patient risk groups
Risk cut-off (author, year)
Development
phase
N
studies
Type of
study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
Algorithm1
( Jakobsson et al 2010)
130
External
validation
1 Retrospective Not reported PE-related death 30 days 83.7 (81.3 to 85.9)
16.3 (14.1 to 18.7)
6.6 (5.1 to 8.5)
0.6 (0.1 to 4.3)
Agterof (Herat rate—Ddimer)
(Agterof et al 2009)
67
Derivation 1 Retrospective Both stable and
unstable
Death VTE major
bleeding
In-hospital 67.1 (61.7 to 72.1)
32.9 (27.9 to 38.3)
5.3 (3.0 to 9.3)
0.5 (0 to 7.3)
ESC high
(Vanni et al 2011)
119
External
validation
1 Retrospective Both stable and
unstable
PE-related death In-hospital 5.6 (3.8 to 8)
94.4 (92 to 96.2)
29.6 (15.6 to 49)
4.1 (2.7 to 6.4)
ESC intermediate–high
(Vanni et al 2011)
119
External
validation
1 Retrospective Both stable and
unstable
PE-related death In-hospital 60 (55.6 to 64.3)
40 (35.7 to 44.4)
8.6 (5.9 to 12.4)
1 (0.3 to 4)
GPS1
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 80.6 (75.4 to 84.9)
19.4 (15.1 to 24.6)
12.5 (8.7 to 17.6)
0.9 (0.1 to 13.4)
(Nendaz et al 2004)
97
External
validation
1 Retrospective Unclear Death VTE major
bleeding
3 months 71.4 (64.7 to 77.2)
28.6 (22.8 to 35.3)
12.0 (7.6 to 18.4)
3.5 (0.9 to 13.0)
Overall 2 Death VTE major
bleeding
3 months 76.3 (66.1 to 84.2)
23.7 (15.8 to 33.9)
12.3 (9.3 to 16.1)
2.7 (0.8 to 8.9)
GPS2
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 51.1 (45.1 to 57.1)
48.9 (42.9 to 54.9)
18.2 (12.6 to 25.6)
1.5 (0.4 to 5.9)
(Nendaz et al 2004)
97
External
validation
1 Retrospective Unclear Death VTE major
bleeding
3 months 44.2 (37.5 to 51.2)
55.8 (48.8 to 62.5)
18.2 (11.4 to 27.6)
2.7 (0.9 to 8.0)
Overall 2 Death VTE major
bleeding
3 months 47.9 (41.2 to 54.7)
52.1 (45.3 to 58.8)
18.2 (13.7 to 23.8)
2.1 (0.9 to 51)
GPS3
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 32.8 (27.5 to 38.7)
67.2 (61.3 to 72.5)
26.1 (18.0 to 36.3)
2.2 (0.8 to 5.8)
(Nendaz et al 2004)
97
External
validation
Retrospective Unclear Death VTE major
bleeding
3 months 20.1 (15.1 to 26.2)
79.9 (73.8 to 84.9)
27.5 (15.9 to 43.2)
5.0 (2.5 to 9.7)
Overall 2 Death VTE major
bleeding
3 months 26.2 (15.6 to 40.5)
73.8 (59.5 to 84.4)
26.6 (19.6 to 34.9)
3.6 (1.6 to 7.8)
GPS4
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 11.9 (8.6 to 16.4)
88.1 (83.6 to 91.4)
40.6 (25.3 to 58.1)
5.9 (3.5 to 9.8)
(Nendaz et al 2004)
97
External
validation
1 Retrospective Unclear Death VTE major
bleeding
3 months 8.0 (5.0 to 12.7)
92.0 (87.3 to 95.0)
31.3 (13.6 to 56.7)
7.7 (4.6 to 12.5)
Continued
10 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 3 Continued
Risk cut-off (author, year)
Development
phase
N
studies
Type of
study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
Overall 2 Death VTE major
bleeding
3 months 10.2 (6.9 to 14.7)
89.8 (85.3 to 93.1)
37.6 (25.1 to 52)
6.7 (4.7 to 9.6)
GPS5
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 3.7 (2.0 to 6.8)
96.3 (93.2 to 98.0)
70.0 (37.6 to 90.0)
7.8 (5.1 to 11.7)
(Nendaz et al 2004)
97
External
validation
1 Retrospective Unclear Death VTE major
bleeding
3 months 2.5 (1.0 to 5.9)
97.5 (94.1 to 99.0)
80.0 (30.9 to 97.3)
7.7 (4.7 to 12.4)
Overall 2 Death VTE major
bleeding
3 months 3.3 (2.0 to 5.4)
96.7 (94.6 to 98.0)
73.0 (46.1 to 89.5)
7.7 (5.6 to 10.6)
GPS=6
(Wicki et al 2000)
122
Derivation 1 Retrospective Not reported Death VTE major
bleeding
3 months 1 (0.4 to 3.4)
98.9 (96.6 to 99.6)
87.5 (26.6 to 99.3)
9.1 (6.1 to 13.2)
(Nendaz et al 2004)
97
External
validation
1 Retrospective Unclear Death VTE major
bleeding
3 months 1 0.5 (0.1 to 3.5)
99.5 (96.5 to 99.9)
NA
9.6 (6.2 to 14.6)
Overall 2 Death VTE major
bleeding
3 months 0.9 (0.3 to 2.4)
99.1 (97.6 to 99.7)
87.5 (26.6 to 99.3)
9.3 (7.0 to 12.3)
PESIII
(Aujesky et al 2007)
70
External
validation
1 Retrospective Both stable and
unstable
PE-related death 3 months 79.5 (76.8 to 82)
20.5 (18 to 23.2)
2.9 (1.9 to 4.5)
0.3 (0 to 4.2)
PESIIII
(Vanni et al 2011)
119
External
validation
1 Retrospective Both stable and
unstable
PE-related death In-hospital 68.7 (64.3 to 72.7)
31.3 (27.3 to 35.7)
7.5 (5.1 to 11)
0.7 (0.1 to 4.7)
(Aujesky et al 2007)
70
External
validation
1 Retrospective Both stable and
unstable
PE-related death 3 months 52.6 (49.3 to 55.9)
47.4 (44.1 to 50.7)
3.8 (2.4 to 6)
0.7 (0.2 to 2.2)
PESIIV
(Choi et al 2014)
77
External
validation
1 Retrospective Both stable and
unstable
Adverse outcome In-hospital 18 (15.3 to 21.2)
82 (78.8 to 84.7)
24.8 (17.8 to 33.4)
7.3 (5.4 to 9.9)
(Aujesky et al 2007)
70
External
validation
1 Retrospective Both stable and
unstable
PE-related death 3 months 25 (22.3 to 28)
75 (72 to 77.7)
6.7 (4.1 to 10.8)
0.9 (0.4 to 2)
PESI=V
(Aujesky et al 2007)
70
External
validation
1 Retrospective Both stable and
unstable
PE-related death 3 months 10.3 (8.5 to 12.5)
89.7 (87.5 to 91.5)
7.5 (3.6 to 15)
1.7 (1 to 2.9)
PESIIV and other prognostic factors*
(Choi et al 2014)
77
External
validation
1 Retrospective Both stable and
unstable
Adverse outcome In-hospital
NT-proBNP1136 pg/mL 34.2 (29.7 to 39.0)
65.8 (61.0 to 70.3)
22.4 16.1 to 30.2)
4.7 (2.7 to 8.0)
cTnI0.05 ng/mL 39.1 (34.8 to 43.6)
60.9 (56.4 to 65.2)
21.6 (16.3 to 28.1)
5.2 (3.2 to 8.5)
Continued
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 11
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 3 Continued
Risk cut-off (author, year)
Development
phase
N
studies
Type of
study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
CT RV/LV diameter ratio1 36.1 (32.5 to 39.9)
63.9 (60.1 to 67.5)
20.1 (15.4 to 25.7)
5.1 (3.3 to 7.7)
PESIIV—NT-proBNP1136 pg/mL 6.2 (4.2 to 9.1)
93.8 (90.9 to 95.8)
50.0 (31.0 to 69.0)
8.0 (5.6 to 11.3)
PESIIV—cTnI0.05 ng/mL 10.3 (7.8 to 13.4)
89.7 (86.6 to 92.2)
41.7 (28.7 to 55.9)
8.1 (5.9 to 11.1)
PESIIV—CT RV/LV ratio1 7.9 (6.0 to 10.2)
92.1 (89.8 to 94.0)
43.1 (30.4 to 56.9)
7.7 (5.8 to 10.2)
PESIIV—NT-proBNP—cTnI 4.6 (2.9 to 7.3)
95.4 (92.7 to 97.1)
64.7 (40.4 to 83.2)
8.2 (5.8 to 11.6)
PESIIV—NT-proBNP—RV/LV 4.2 (2.6 to 6.7)
95.8 (93.3 to 97.4)
62.5 (37.7 to 82.1)
8.4 (6.0 to 11.7)
PESIIV—cTnI—RV/LV 6.0 (4.2 to 8.6)
94.0 (91.4 to 95.8)
60.7 (42.0 to 76.7)
8.5 (6.2 to 11.5)
PESIIV—NT-proBNP—cTnI—
RV/LV
3.8 (2.3 to 6.3)
96.2 (93.7 to 97.7)
71.4 (43.9 to 88.9)
8.5 (6.0 to 11.9)
PREP
(Sanchez et al 2010)
110
Derivation 1 Prospective Both stable and
unstable
Death VTE 30 days
PREP—ClinII 37.3 (33.2 to 41.5)
62.7 (58.5 to 66.8)
17.7 (12.9 to 23.8)
1.9 (0.8 to 4.1)
PREP—ClinIII 7.2 (5.2 to 9.8)
92.8 (90.2 to 94.8)
43.2 (28.4 to 59.4)
5.4 (3.7 to 7.9)
PREP—ClinII—BNP—RVD 34.8 (30.8 to 39)
65.2 (61 to 69.2)
18.4 (13.4 to 24.8)
2.1 (1 to 4.3)
PREP—ClinIII—BNP—RVD 5.6 (3.9 to 8)
94.4 (92 to 96.1)
44.8 (28.1 to 62.8)
5.6 (3.8 to 8)
Shock Index
(Toosi et al 2008)
117
External
validation
1 Retrospective Not reported Adverse outcome In-hospital 17.6 (12.4 to 24.5)
82.4 (75.5 to 87.6)
25.9 (12.9 to 45.3)
10.3 (6.1 to 17)
sPESI1
(Righini et al 2011)
106
External
validation
1 Retrospective Unclear PE-related death 30 days 53.8 (48.6 to 58.9)
46.2 (41.1 to 51.4)
3.1 (1.4 to 6.8)
0.6 (0.1 to 4.2)
(Spirk et al 2011)
115
External
validation
1 Retrospective Both stable and
unstable
Death VTE 30 days 70.8 (65.7 to 75.4)
29.2 (24.6 to 34.3)
8.1 (5.2 to 12.3)
1 (0.1 to 7)
Continued
12 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
GRACE PE
103
risk scores show 1.3% (0.1% to 17.1%)
and 1.1% (0.1% to 15.1%) 30-day mortality rate,
respectively, within the low-risk patient groups. Of note
is a large 95% CI for Geneva and GRACE scores due
the small number of patients in the low-risk groups.
Interestingly, the high event rates (23–30%) in the high-
risk groups of GRACE scores whatever the cut-offs with
a high proportion (49–81%) of patients in these risk
groups. These models need to be validated in prospect-
ive studies with larger sample sizes in stable PE. Davies
79
and Hestia
125 126
checklists (table 5 and online supple-
mentary table S2.5) show a high proportion of patients
at low risk (42–55%) with a small 3-month mortality
rate: 1.9% (0.4% to 5.5%) and 1.2% (0.4% to 3.7%),
respectively. These models need to be compared with
PESI/sPESI models in impact studies.
Incidence of events along the risk scale
As expected, the event rates for different outcomes and
time points increase along the risk scale. Tables 3–5
and online supplementary table S4 show how the pre-
dicted outcomes in the derivation samples compare
with the observed outcomes in the validation samples
(internal, external) at different cut-offs for GPS and
PESI. Shifting the cut-off to a lower level or to a higher
level provides more appropriate event rates in low-risk
groups and in high-risk groups, respectively, but at the
expense of a decrease in the proportion of patients
within these risk groups. Thus, improvement in efficacy
is associated with a decline in efficiency.
Prognostic accuracy
Summary estimates are displayed (see online supple-
mentary tables S5 and S6) for different cut-offs as iden-
tified in studies to discriminate between low-risk and
high-risk groups. They are provided only when at least
four studies are identified for a specific cut-off, and a
specific outcome and time point. The highest overall
sensitivity estimates were obtained with the Prognostic
Algorithm (4 studies) for predicting in-hospital mortal-
ity (97% (96% to 98%)) and 30-day mortality (98%
(96% to 99%)), followed by PESI (94% (89% to 97%))
for 3-month mortality (6 external validation studies)
and by sPESI (93% (90% to 95%)) for 30-day (11
studies) and for 3-month mortality (10 external valid-
ation studies) (see online supplementary table S5). All
results are consistent and similar in external validation
studies. The same applies for LR negative estimates.
The specificity is at best 54% (46% to 62%) for
PESIIII cut-off for 3-month mortality. Shifting the
cut-off along the risk scale provides higher values for
sensitivity or for specificity (see online supplementary
table S6). In figures 2 and 3, sROC plots showing test
accuracy of the two most validated models at relevant
cut-offs PESIIII and sPESI1 associated with 30-day
all-cause death for all studies and for external validation
studies are given separately.
Table 3 Continued
Risk cut-off (author, year)
Development
phase
N
studies
Type of
study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
(Righini et al 2011)
106
External
validation
1 Retrospective Unclear PE-related death 3 months 53.8 (48.6 to 58.9)
46.2 (41.1 to 51.4)
4.2 (2.1 to 8.1)
0.6 (0.1 to 4.2)
sPESI1—cTn
(Spirk et al 2011)
115
Update 1 Retrospective Both stable and
unstable
Death VTE 30 days
cTn 28.3 (65.7 to 75.4)
71.7 (66.6 to 76.3)
12.8 (7.4 to 21.1)
3.4 (1.7 to 6.6)
sPESI1—cTn 24.4 (20.1 to 29.3)
75.6 (70.7 to 79.9)
13.6 (7.7 to 22.9)
3.6 (1.9 to 6.7)
In the first column, the cut-off is indicated by the corresponding higher risk group for the cut-off.
*Information from authors.
BNP, brain natriuretic peptide; cTnI, cardiac troponin I; ESC, European Society of Cardiology; GPS, Geneva prognostic score; LV, left ventricle; NT-proBNP, N-terminal proBNP; PE, pulmonary
embolism; PESI, Pulmonary Embolism Severity Index; PREP, Facteurs PRonostiques dans l’Embolie Pulmonaire; RV, right ventricle; RVD, right ventricular dysfunction; sPESI, simplified PESI;
VTE, venous thromboembolism.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 13
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 4 PE-related death, adverse outcome (death, clinical deterioration or haemodynamic collapse), VTE, major bleeding and their combinations in ‘stable’PE patient
risk groups
Risk cut-off
(author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
Bova et al risk score
(Bova et al 2014)
135
Derivation and
internal validation
6 (IPD) Both Stable
Stage III PE-related death,
collapse, VTE
In-hospital 5.8 28
Stage II 18.6 9.7
Stage I 75.5 3.6
Stage III PE-related death,
collapse, VTE
30 days 5.8 29.2
Stage II 18.6 10.8
Stage I 75.5 4.2
Stage III PE-related death 30 days 5.8 15.5
Stage II 18.6 5.0
Stage I 75.5 1.7
ESC intermediate
(Vanni et al 2011)
119
External validation 1 Retrospective Stable PE-related death In-hospital 57.6 (53.1 to 62.1)
42.4 (37.9 to 46.9)
6.4 (4.0 to 10.1)
1.0 (0.3 to 4.0)
(Becattini et al 2013)
72
External validation 1 Prospective Stable PE-related death In-hospital 78.3 (75.4 to 80.9)
21.7 (19.1 to 24.6)
1.8 (1.0 to 3.1)
0.3 (0.00 to 4.1)
Overall 1 Both Stable PE-related death In-hospital 68.9 (46.1 to 85.2)
31.1 (14.8 to 53.9)
3.4 (0.9 to 11.7)
0.8 (0.2 to 2.7)
(Dellas et al 2014)
80
External validation 1 Prospective Stable Adverse outcome 30 days 78.1 (72.2 to 83.1)
21.9 (16.9 to 27.8)
10.9 (7 to 16.4)
1 (0.1 to 14.1)
ESC intermediate–high
(Vanni et al 2011)
119
External validation 1 Retrospective Stable PE-related death In-hospital 14.8 (11.9 to 18.4)
85.2 (81.6 to 88.1)
11.8 (6.0 to 21.8)
2.8 (1.6 to 5.0)
(Becattini et al 2013)
72
External validation 1 Prospective Stable PE-related death In-hospital 41.8 (38.5 to 45.1)
58.2 (54.9 to 61.5)
1.9 (0.9 to 4.0)
1.0 (0.4 to 2.4)
Overall 1 Stable PE-related death In-hospital 26.2 (8.2 to 58.7)
73.8 (41.3 to 91.8)
4.9 (0.8 to 25.1)
1.8 (0.6 to 4.8)
(Becattini et al 2013)
72
External validation 1 Prospective Stable Death adverse
outcome
In-hospital 41.8 (38.5 to 45.1)
58.2 (54.9 to 61.5)
8.8 (6.3 to 12.2)
3.2 (1.9 to 5.1)
(Becattini et al 2013)
72
External validation 1 Prospective Stable PE-related death
adverse outcome
In-hospital 41.8 (38.5 to 45.1)
58.2 (54.9 to 61.5)
5 (3.1 to 7.7)
2.4 (1.4 to 4.1)
eStiMaTe high
( Jiménez et al 2014)
87
Derivation 1 Prospective Stable Death adverse
outcome VTE
30 days 3.7 (2.6 to 5.2)
96.3 (94.8 to 97.4)
25.8 (13.5 to 43.7)
3.4 (2.1 to 5.4)
( Jiménez et al 2014)
87
External validation 1 Prospective Stable Death adverse
outcome
30 days 6.2 (4.5 to 8.6)
93.8 (91.4 to 95.5)
21.2 (10.5 to 38.3)
6.7 (5.2 to 8.7)
Continued
14 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 4 Continued
Risk cut-off
(author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
( Jiménezet al 2014)
87
Overall 2 Prospective Stable Death adverse
outcome VTE
30 days 4.8 (2.8 to 8)
95.2 (92 to 97.2)
23.5 (14.7 to 35.4)
5 (2.5 to 9.4)
eStiMaTe high–intermediate
( Jiménez et al 2014)
87
Derivation 1 Prospective Stable Death adverse
outcome VTE
30 days 74.5 (71.5 to 77.3)
25.5 (22.7 to 28.5)
9.7 (7.6 to 12.2)
0.9 (0.2 to 3.6)
( Jiménez et al 2014)
87
External validation 1 Prospective Stable Death adverse
outcome
30 days 63.5 (59.3 to 37.5)
36.5 (32.5 to 40.7)
7.1 (4.8 to 10.4)
0.3 (0 to 4)
( Jiménez et al 2014)
87
Overall 2 Prospective Stable Death adverse
outcome VTE
30 days 69.3 (57.6 to 79)
30.7 (21 to 42.4)
8.6 (6.5 to 11.4)
0.7 (0.2 to 2.4)
FAST=3
(Dellas et al 2014)
80
External validation 1 Prospective Stable Adverse outcome 30 days 28.4 (23.4 to 34.1)
71.6 (65.9 to 76.6)
22.1 (14.2 to 32.7)
1.5 (0.5 to 4.7)
GPS3
(Bova et al 2009)
74
External validation 1 Prospective Stable PE-related death In-hospital 18.9 (14.1 to 24.9)
81.1 (75.1 to 85.9)
2.6 (0.4 to 16.5)
0.3 (0.0 to 4.7)
Kostrubiec et al (NT-proBNP—cTnT)
(Kostrubiec et al
2005)
90
Derivation 1 Prospective Stable
Intermediate Death 30 days 72.0 (62.4 to 79.9)
28.0 (20.1 to 37.6)
20.8 (13.0 to 31.7)
1.7 (0.1 to 22.3)
Intermediate PE-related death 30 days 69.9 (59.8 to 78.3)
30.1 (21.7 to 40.2)
12.3 (6.3 to 22.7)
1.7 (0.1 to 22.3)
Intermediate–high Death 30 days 18.0 (11.6 to 26.8)
82.0 (73.2 to 88.4)
50.0 (28.4 to 71.6)
7.3 (3.3 to 15.3)
Intermediate–high PE-related death 30 days 16.1 (1.0 to 25.0)
83.9 (75.0 to 90.0)
40.0 (19.2 to 65.2)
2.6 (0.6 to 9.7)
PESIII
(Sanchez et al 2013)
111
External validation 1 Retrospective
(for PESI)
Stable Adverse outcome 30 days 75.2 (71.4 to 78.7)
24.8 (21.3 to 28.6)
6 (4.1 to 8.8)
0.8 (0 to 5.2)
PESIIII
(Palmieri et al 2008)
104
External validation 1 Prospective Stable Death adverse
outcome
In-hospital 69.7 (59.4 to 78.3)
30.3 (21.7 to 40.6)
53.2 (40.9 to 65.2)
11.1 (3.6 to 29.3)
(Sanchez et al 2013)
111
External validation 1 Retrospective
(for PESI)
Stable Adverse outcome 30 days 37.8 (33.8 to 42)
62.2 (58 to 66.2)
9 (5.7 to 13.8)
2.1 (1 to 4.4)
(Vanni et al 2011)
119
External validation 1 Retrospective Stable PE-related death In-hospital 67 (62.3 to 71.3)
33 (28.7 to 37.7)
6.1 (3.8 to 9.5)
0.7 (0.1 to 5)
PESIIV
(Sanchez et al 2013)
111
External validation 1 Retrospective
(for PESI)
Stable Adverse outcome 30 days 17.2 (14.2 to 20.7)
82.8 (79.3 to 85.8)
8.8 (4.5 to 16.6)
3.9 (2.4 to 6.2)
Continued
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 15
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 4 Continued
Risk cut-off
(author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
PESI=V
(Sanchez et al 2013)
111
External validation 1 Retrospective
(for PESI)
Stable Adverse outcome 30 days 4 (2.6 to 6)
96 (94 to 97.4)
14.3 (4.7 to 36.1)
4.3 (2.9 to 6.5)
PESI+biomarkers
(Sanchez et al 2013)
111
External validation 1 Retrospective
(for PESI)
Stable Adverse outcome 30 days
PESI I–II—BNP 32.7 (27.8 to 38.0)
67.3 (62.0 to 72.2)
4.7 (2.0 to 10.8)
0.9 (0.2 to 3.6)
PESI I–II—cTnI 13.7 (10.3 to 17.9)
86.3 (82.1 to 89.7)
9.1 (3.5 to 21.8)
1.1 (0.3 to 3.3)
PESI I–II—TTE-RVD 12.3 (9.0 to 16.5)
87.7 (83.5 to 91.0)
10.8 (4.1 to 25.5)
1.1 (0.4 to 3.5)
PESI III–IV—BNP 56.3 (48.8 to 63.4)
43.8 (36.6 to 51.2)
10.1 (5.5 to 17.8)
6.5 (2.7 to 14.7)
PESI III–IV—cTnI 24.0 (18.2 to 30.9)
76.0 (69.1 to 81.8)
16.7 (8.2 to 31.0)
5.3 (2.5 to 10.6)
PESI III–IV—TTE-RVD 22.8 (17.0 to 29.9)
77.2 (70.1 to 83.0)
10.8 (4.1 to 25.5)
8.8 (4.9 to 15.2)
PESI V—BNP 76.2 (54.0 to 89.7)
23.8 (10.3 to 46.0)
18.8 (6.2 to 44.7)
8.3 (0.5 to 62.2)
PESI V—cTnI 30.0 (14.1 to 52.7)
70.0 (47.3 to 85.9)
16.7 (2.3 to 63.1)
14.3 (3.6 to 42.7)
PESI V—TTE-RVD 23.8 (10.3 to 46.0)
76.2 (54.0 to 89.7)
20.0 (2.7 to 69.1)
12.5 (3.1 to 38.6)
PESI+biomarkers
( Jimenez et al 2011)
86
Update 1 Unclear
Prospective
Stable PE-related death 30 days
cTnI 32.1 (28.5 to 36.0)
67.9 (64.0 to 71.5)
10.5 (6.9 to 15.8)
4.2 (2.7 to 6.7)
TTE-RVD 20.3 (17.3 to 23.7)
79.7 (76.3 to 82.7)
11.7 (7.0 to 18.7)
4.9 (3.3 to 7.2)
US-DVT 38.6 (34.7 to 42.6)
61.4 (57.4 to 65.3)
9.6 (6.4 to 14.2)
4.1 (2.5 to 6.7)
TTE-RVD—cTnI 10.0 (7.8 to 12.7)
90.0 (87.3 to 92.2)
15.3 (8.1 to 26.8)
5.3 (3.7 to 7.5)
PESIIV—TTE-RVD—
cTnI
12.7 (9.0 to 17.7)
87.3 (82.3 to 91.0)
20.7 (9.6 to 39.0)
8.0 (5.0 to 12.7)
cTnI—US-DVT 13.9 (11.3 to 16.9)
86.1 (83.1 to 88.7)
17.1 (10.4 to 26.8)
4.5 (3.0 to 6.7)
Continued
16 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 4 Continued
Risk cut-off
(author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
PESIIV—cTnI—
US-DVT
18.0 (13.5 to 23.5)
82.0 (76.5 to 86.5)
24.4 (13.7 to 39.7)
6.4 (3.7 to 11.0)
TTE-RVD—US-DVT 8.6 (6.6 to 11.2)
91.4 (88.8 to 93.4)
19.6 (10.9 to 32.7)
5.0 (3.5 to 7.2)
TTE-RVD—cTnI—
US-DVT
4.1 (2.7 to 6.0)
95.9 (94.0 to 97.3)
20.8 (8.9 to 41.3)
5.6 (4.0 to 7.9)
PESIIV—TTE-RVD—
US-DVT
10.5 (7.2 to 15.2)
89.5 (84.8 to 92.8)
25.0 (11.7 to 45.6)
7.8 (4.9 to 12.4)
PREP—ClinII
(Sanchez et al 2010)
110
Derivation 1 Prospective Stable Death VTE 30 days 32.3 (28.2 to 36.6)
67.7 (63.4 to 71.8)
12.3 (8 to 18.5)
2.5 (1.2 to 4.9)
PREP—ClinIII
(Sanchez et al 2010)
110
Derivation 1 Prospective Stable Death VTE 30 days 1.9 (1 to 3.6)
98.1 (96.4 to 99)
22.2 (5.6 to 57.9)
4.9 (3.3 to 7.3)
sPESI1
(Dellas et al 2014)
80
External validation 1 Prospective Stable Adverse outcome 30 days 66.1 (60.2 to 71.4)
33.9 (28.6 to 39.8)
11.2 (7.3 to 16.7)
0.5 (0 to 8)
(Lankeit et al 2011)
94
External validation 1 Prospective Stable Death adverse
outcome
30 days 62.4 (58.1 to 66.4)
37.6 (33.6 to 41.9)
8.8 (6.2 to 12.4)
1 (0.3 to 3.9)
sPESI1-hscTnT
(Lankeit et al 2011)
94
External validation 1 Prospective Stable Death adverse
outcome
30 days
hscTnT 59.3 (55.1 to 63.4)
40.7 (36.6 to 44.9)
8.7 (6.0 to 12.3)
1.9 (0.7 to 4.9)
sPESI1-hscTnT 75.9 (72.0 to 79.3)
24.1 (20.7 to 28.0)
7.8 (5.5 to 10.8)
0.4 (0.0 to 5.9)
In the first column, the cut-off is indicated by the corresponding higher risk group for the cut-off except for the Bova et al
135
and Sanchez et al
111
(PESI+biomarkers) studies which are displayed
in risk categories. Bova’s study combined IPD from six studies.
BNP, brain natriuretic peptide; cTnI, cardiac troponin I; cTnT, cardiac troponin T; DVT, deep vein thrombosis; GPS, Geneva prognostic score; hscTnT, high-sensitive cardiac troponin T; IPD,
individual patient data; NT-proBNP, N-terminal proBNP; PE, pulmonary embolism; PESI, Pulmonary Embolism Severity Index; PREP, Facteurs PRonostiques dans l’Embolie Pulmonaire; sPESI,
simplified PESI; TTE-RVD, transthoracic echocardiography-right ventricular dysfunction; US-DVT, ultrasound-detected DVT; VTE, venous thromboembolism.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 17
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 5 Studies reporting on derivation and external validation of models originally not specific to the prognosis or the domain of PE (type 2 model studies), and studies
reporting on hospital criteria checklists (type 3 model studies)
Risk cut-off (author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
RGS
(Bertoletti et al 2013)
73
External validation 1 Prospective for
RGS
Not reported All-cause death 3 months
RGS high 11.8 (8.8 to 15.5)
88.2 (84.5 to 91.2)
14.3 (6.6 to 28.3)
4.8 (2.9 to 7.7)
RGS high–intermediate 84.0 (79.9 to 87.5)
16.0 (12.5 to 20.1)
7.0 (4.6 to 10.5)
0.9 (0.1 to 12.3)
SGS
(Bertoletti et al 2013)
73
External validation 1 Retrospective
for SGS
Not reported All-cause death 3 months
SGS high 7.8 (5.5 to 11.1)
92.2 (88.9 to 94.5)
17.9 (7.6 to 36.4)
4.9 (3.0 to 7.8)
SGS high–intermediate 84.0 (79.9 to 87.5)
16.0 (12.5 to 20.1)
7.0 (4.6 to 10.5)
0.9 (0.1 to 12.3)
GRACE—ACS risk score
(Paiva et al 2013)
103
External validation 1 Retrospective Both stable and
unstable
All-cause death 30 days
GRACE—ACS high 56.8 (49.9 to 63.4)
43.2 (36.6 to 50.1)
25.6 (18.5 to 34.3)
10.1 (5.3 to 18.3)
GRACE—ACS high–
intermediate
81.1 (75.1 to 85.9)
18.9 (14.1 to 24.9)
23.4 (17.6 to 30.4)
1.3 (0.1 to 17.1)
GRACE—PE risk score
(Paiva et al 2013)
103
External validation 1 Retrospective Both stable and
unstable
All-cause death 30 days
GRACE—PE high 49.0 (42.3 to 55.8)
51.0 (44.2 to 57.7)
29.7 (21.6 to 39.3)
8.6 (4.5 to 15.7)
GRACE—PE high–
intermediate
78.2 (72.0 to 83.3)
21.8 (16.7 to 28.0)
24.2 (18.2 to 31.4)
1.1 (0.1 to 15.1)
Davies checklist
(Davies et al 2007)
79
Derivation 1 Prospective Not reported All-cause death 3 months
Unsuitable vs suitable 57.9 (51.0 to 64.5)
42.1 (35.5 to 49.0)
5.1 (2.3 to 10.9)
3.5 (1.1 to 10.4)
Continued
18 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 5 Continued
Risk cut-off (author, year)
Development
phase
N
studies Type of study
Haemodynamic
status Outcome
Time
point
Proportion
(95% CI)
High-risk group
Low-risk group
Event rate
(95% CI)
High-risk group
Low-risk group
Davies checklist
(Davies et al 2007)
79
External validation
and impact
(management
study)
1 Prospective Stable meeting
inclusion criteria
(suitable) for
home treatment
All-cause death 3 months Not reported 1.9 (0.4 to 5.5)
PE-related None
VTE None
Major/fatal bleeding None
Hestia checklist
(Zondag et al 2013)
125
External validation 1 Prospective Both stable and
unstable
All-cause death 30 days 46.9 (42.4 to 51.4)
53.1 (48.6 to 57.6)
4.1 (2.2 to 7.7)
0.8 (0.2 to 3.2)
3 months 46.9 (42.4 to 51.4)
53.1 (48.6 to 57.6)
9.6 (6.4 to 14.3)
1.2 (0.4 to 3.7)
7 days 46.9 (42.4 to 51.4)
53.1 (48.6 to 57.6)
1.8 (0.7 to 4.8)
0.2 (0.0 to 3.1)
Hestia checklist
(Zondag et al 2013)
126
External validation
(and update)
1 Prospective Both stable and
unstable
All-cause death 30 days 44.6 (40.2 to 49.0)
55.4 (51.0 to 59.8)
4.1 (2.1 to 7.6)
0.7 (0.2 to 2.9)
Both stable and
unstable
Death adverse
outcome
30 days 44.6 (40.2 to 49.0)
55.4 (51.0 to 59.8)
19.9 (15.2 to 25.7)
0.7 (0.2 to 2.9)
Stable All-cause death 30 days 41.5 (37.1 to 46.0)
58.5 (54.0 to 62.9)
3.6 (1.7 to 7.3)
0.7 (0.2 to 2.9)
Stable Death adverse
outcome
30 days 41.5 (37.1 to 46.0)
58.5 (54.0 to 62.9)
10.3 (6.7 to 15.4)
0.7 (0.2 to 2.9)
ACS, acute coronary syndrome; GRACE, Global Registry of Acute Coronary Events; PE, pulmonary embolism; RGS, revised Geneva score; SGS, simplified Geneva score; VTE, venous
thromboembolism.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 19
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Model updating
For better identification of lower and higher risk groups,
some studies assessed the effect on outcomes of adding
one or more prognostic variables to an existing model.
There is increasing evidence about the greater effective-
ness of these ‘new models’, even though the existing
ones such as PESI, sPESI and PREP are already highly
effective.
In the study of Moores, the addition of non-elevated
cTnI to low-risk PESI did not improve the negative pre-
dictive value compared with either test alone.
96
In the
Lankeit study,
94
the combination of a negative high-
sensitive cTnT (hscTnT<14 pg/mL) and a sPESI value
of 0 at baseline improved risk classification and reduced
the risk of mortality. The same applies for the combin-
ation of NT-proBNP (<600 pg/mL) and low-risk sPESI.
95
In the study of Ozsu, hscTnT combined with sPESI pro-
vided better predictive information than cTnT.
101
Adding cTn to the PESI
113
and the sPESI
115
models
resulted in a higher area under the curve (AUC) value
with no additive value for Ddimer to PESI and cTn.
113
To sum up, there is an increased value in identifying
lower risk patients by adding NT-proBNP and hscTnT to
PESI or sPESI.
Figure 2 sROC plots showing test accuracy of PESIIII cut-off associated with 30-day all-cause death: all studies (A) and
external validation studies (B). HSROC, hierarchical summary receiver operating characteristic; PESI, Pulmonary Embolism
Severity Index; sROC, summary receiver operating characteristic.
Figure 3 sROC plots showing test accuracy of sPESI1 cut-off associated with 30-day all-cause death: all studies (A) and
external validation studies (B). HSROC, hierarchical summary receiver operating characteristic; PESI, Pulmonary Embolism
Severity Index; sROC, summary receiver operating characteristic.
20 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Other update studies tried to improve the identifica-
tion of high-risk patients. Adding BNP and TTE-RVD
to the PREP clinical model resulted in a significant dif-
ference in AUC.
110
A significant increase in ‘NRI: Net
Reclassification Improvement’,asdefined by Pencina et
al,
142
was obtained in the original study which updated
PREP model by adding BNP and TTE-RVD to the clin-
ical model.
110
Using the PREP cohort in patients with
normotensive PE, Sanchez et al showed that biomarkers
(cTnI, BNP) and echocardiography provided additional
prognostic information to PESI.
111
In the Palmieri
study
104
which included highly selected ‘non-massive
central PE’, increased cTnI contributed to identifying
patients with increased risk of development of haemo-
dynamic instability, which was independent of, and in
addition, to PESI. Novel models
80 87 92 135
specifically
developed for patients with normotensive PE and
integrating biomarkers (cTn, NT-proBNP, heart-type
fatty acid-binding protein (H-FABP)), CT or TTE-RVD
and US-detected DVT with clinical variables (SBP, HR)
or with sPESI, showed a high ability to identify
patients at lower risk as well as at higher risk of early
death or adverse outcome at the expense of a
lower proportion of patients in these risk groups
(tables 3 and 4).
Tests of heterogeneity, investigation for publication bias,
subgroup analysis and sensitivity analysis
Results of tests for heterogeneity using Cochran’s Q test
and Higghins I
2
statistic, and results of subgroup ana-
lyses show major problems of heterogeneity. Comparing
results between studies reporting on 30-day all-cause
death for PESI optimal cut-offs (PESIIII, sPESI1), all
conditions for comparison being optimal (all studies are
retrospective and external validation studies, and
include ‘both stable and unstable’patients with PE), we
found important between-study heterogeneity (for
instance for PESI: Q value=39.69, df 5, p=0.000, I
2
=87)
that might be related to only patient selection in studies.
A high proportion of patients at low risk in study popula-
tion (or a lower proportion of patients at high risk)
results in less event rates in the population sample and
less event rates within risk groups as shown on forest
plots (see online supplementary figures S7–S12).
Heterogeneity (I
2
>50%) was observed in most of the
analyses for 30-day all-cause death with PESIIII and
sPESI. At best I
2
test for heterogeneity was 39% (all
studies combined) and 50% (external validation studies
only) in low-risk PESI group.
Findings from the investigation for publication bias
using the ‘Funnel Plot’of SE and precision (=1/SE) by
Logit event rate (with comparison of plots with observed
studies against plots with observed and imputed
studies), and the Egger’s test of regression intercept
show no evidence of publication bias except for studies
using PESIIII and PESI=V cut-offs for 30-day all-cause
death (see online supplementary figures S13.1 and
S13.2).
Model impact
The clinical utility of a model is assessed by its effect on
clinical decision-making and subsequent patient out-
comes. However, few such studies have been performed
of the PE models (see online supplementary table S2.4).
The safety of treating patients with low-risk PE as out-
patients was examined in two management studies
66 128
and in two randomised trials.
71 100
The first management
study
128
was a feasibility study among low-risk patients
based on the GPS model in a very small number of
patients. It showed the use of the GPS model to be safe.
In the other one-arm management study, patients with
stable PE with NT-proBNP<500 pg/mL were treated at
home. Seven (4.6%) patients were readmitted within the
first 10 days, but there were no deaths, no VTE events
and no clinically relevant or major bleeding at 3-month
follow-up. The first of the two randomised trials
100
did
not demonstrate the clinical utility of Uresandi’s
model
118
for outpatient management. The rate of short-
term mortality was unexpectedly high in both manage-
ment groups, as was the rate of VTE events and major
bleeding. The second randomised trial
71
was a non-
inferiority study in highly selected patients with low-risk
PE. It showed that outpatient care based on the PESI
model can safely and effectively be used in place of
inpatient care: 1 (0.6%) death in each group, with 1
(0.6%) VTE and 2 (1.2%) major bleeding events in the
outpatient group at a 3-month follow-up. Owing to other
eligibility criteria for outpatient management and study
design, only 56% (152/271) with low NT-proBNP level,
66
13% (132/1016) with low-risk Uresandi’s model,
100
18%
(43/244) with low-risk GPS
128
and 44% (344/783) with
PESI I-II
71
could be treated as outpatients.
The PEITHO study
132
randomised patients with
normotensive PE at ‘intermediate–high’risk (with RVD
and myocardial injury) to receive tenecteplase (fibrino-
lytics) or placebo. Death or hemodynamic decompensa-
tion occurred in 13/506 patients (2.6%) in the
tenecteplase group as compared with 28/499 (5.6%) in
the placebo group. Fibrinolytic therapy did not prevent
in-hospital death and increased the risk of major haem-
orrhage and stroke. Models are less specific for predict-
ing poor outcome in normotensive high-risk patients
and might benefit from combining prognostic variables
for better selection of patients, in particular those at risk
of death from PE, as long as patients who are likely to
respond to therapy are identified.
Finally, risk stratification was found to be frequently
performed in patients admitted with acute PE and found
to be stable during a 5-year period.
133
Its use was asso-
ciated with assignment to higher levels of care and more
intense treatment, but did not improve the outcome.
133
DISCUSSION
Summary of main results
This systematic review identified 17 prognostic models in
acute PE and other types of models that were not
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 21
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
originally specific to PE or exclusively to its prognosis.
Eight models were validated or updated and three were
assessed in impact studies. Overall, PESI and sPESI are
the models that have been most widely validated and
updated, and PESI is the only one that has been assessed
and found useful in a randomised trial for treating
patients with low-risk PE as outpatients (level 1 of hier-
archy of evidence according to McGinn et al
143
). As
expected, the event rates for different outcomes and
time points increase along the risk scale in PESI model.
The attempt to simplify PESI is attractive and successful
in validation studies, but needs to be evaluated in
impact studies. Agreement between the simplified PESI
and the original PESI is fair.
106
PESI and sPESI models
are now included in the risk stratification of patients
with PE in the 2014 ESC guidelines.
3
This new strategy
needs to be validated and compared to the existing
models. Other prognostic models have shown improve-
ment in identifying low-risk and high-risk groups either
on their own (algorithm for the low risk—Shock Index,
2008 ESC high, PREP—Clin III, PESIIV, PESI V,
eStiMaTe high, FAST 3, Bova stage III for the high risk),
or by the addition of one or more clinical, biological
and imaging-based markers of RVD and myocardial
injury to the existing models or by incorporating these
markers into new models (NT-proBNP and hscTn for
the low risk—SBP, BNP, cTN, H-FABP, venous US and
imaging-based RVD for the high risk). Large validation
and impact studies are needed to assess these new and
updated models.
Overall completeness and applicability of evidence
Most of the models are effective and provide a low event
rate in low-risk groups, and appear reproducible and
robust. However, using the model for identifying the
low-risk groups may not be worthwhile if the aim is to
make a choice to send the patients home; because the
patients might still have an unacceptable high risk of a
serious adverse outcome. It is important to define the
incidence limit for a specific outcome, which should be
clinically relevant and if models go beyond this they
should not be viewed as of high performance. In agree-
ment with the review of Vinson et al,
144
careful selection
is needed for low-risk patients with acute PE who will be
managed as outpatients. Broad implementation of this
management strategy is controversial and varies across
countries and across organisation of patient care and
patient pathways. Furthermore, decisions about manage-
ment might be modified by the availability of new oral
anticoagulants being studied in large-scale clinical
trials.
144
Hospital checklist criteria
79 125
should also be
helpful for patient selection and seem to be competitive
with prognostic models.
Differences in predictive performance of the models
may be due to differences in prognostic criteria and
threshold for risk groups regardless of study design,
development phase, population, outcome and time
point. In two model construction studies
68 69
using the
same cohort and the same outcome, a change in criteria
and threshold resulted in change of risk group propor-
tions and event rates within these risk groups. Shifting
the cut-off to reduce the incidence of events in the
low-risk group led to a smaller proportion of patients in
the low-risk group, and selecting patients with lower risk
PE makes the model more effective but less efficient.
This also applies to the high-risk patient groups: the
greater the proportion of patients classified as high risk,
the lower the incidence of outcomes in this group and
vice versa. Adding biomarker data may help to identify
higher or lower risk patients, but the findings on the
additive value of biomarkers are inconsistent across
studies. This may be explained by the higher predictive
performance of some models, heterogeneity in the
populations arising from study design
66 71 100
or a
higher sensitivity of some biomarkers.
94
Quality of the evidence
In our review, we used a comprehensive search strategy,
assessed the quality of the included studies to allow us
to focus on those of good quality, and used statistical
techniques to analyse their results and combine and
compare the findings. However, the included studies
contain some methodological flaws in the design,
conduct and reporting. Many are retrospective or use
prospectively collected data for diagnosis purposes, and
one of the most common limitations is the lack of
reporting of the case-ascertainment strategy used in the
study. These problems might lead to selection bias that
could have affected the event rates and influence the
variables included in the final model. The time of
inception of the cohorts (ie, whether it was at diagnosis
or on admission) was ill defined and patient selection
may have been different across studies. Other common
methodological issues are the lack of a justification for
the sample size, the absence of a full definition of prog-
nostic variables and the lack of standardisation of treat-
ments which might have been dictated by the
prognostic variables that were assessed (and, therefore,
may have affected the likelihood of certain outcomes),
the lack of blinded assessment of the outcomes (to the
prognostic variables), and possible problems with data
analysis related to the statistical methods for missing
data.
Potential biases in the review process
Our search was performed up to July 2014. Publications
after that date were not included in the review. We
believe there are no biases in our searches within the
search period and decisions on study eligibility.
Although we used a highly sensitive search to minimise
the influence of reporting biases related to duplicate
and cumulative full publication, time lag, language, loca-
tion and reference list citation,
17
it is not possible to
know how a study’sfindings might influence the
researchers’decision to publish it; unlike in randomised
trials where positive results for new treatments are more
22 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
Table 6 Comparison of our study with existing systematic reviews and meta-analyses of clinical prediction rules
Author, year
Squizatto et al,
2012
146
Zhou et al, 2012
148
Kohn et al, 2015
147
Our study
Study design Systematic review and
meta-analysis
Meta-analysis Systematic review and
meta-analysis
Systematic review and meta-analysis
Years of search Until August 2011 Up to June 2012 January 2000–March 2014 Inception to July 2014
Databases for search MEDLINE, EMBASE MEDLINE, EMBASE MEDLINE, EMBASE MEDLINE, EMBASE, The Cochrane Library
Language included No language restriction English English No language restriction
Number of studies in review 33 21 40 71
Type of studies included Derivation, validation Validation (external) Derivation, validation Derivation, validation, update and impact
Number of models in review 9 2 11 17 (+7 other types of models*)
Type of models included Clinical prediction rules PESI and sPESI Clinical prediction rules All prognostic models
Quality appraisal method used 3-point score for cohort
studies
Newcastle-Ottawa Scale QUADAS2 (diagnostic tool)
Domain approach
Prognostic criteria
Domain approach
Analysis of stable PE patients
separately
No No No Yes
Outcome All-cause mortality,
adverse outcomes
All-cause mortality, PE-related
mortality, adverse outcomes
All-cause mortality All-cause mortality, PE-related mortality,
adverse outcomes, VTE, major bleeding
Analysis of different outcomes
separately
Yes Yes Not applicable Yes
Time point 14 days, 30 days,
3 months
Not reported In-hospital, 7 days, 30 days,
3 months
In-hospital/7 days, 30 days, 3 months
Analysis of different time
points separately
No No No Yes
Risk groups Low-risk group Low-risk group vs high-risk
group
Low-risk group Low-risk group and high-risk group
Number of cut-offs accounted
for
Single cut-off Single cut-off Single cut-off All available cut-offs (analysis along the risk
scale)
Summary estimate in
meta-analysis
Event rate in patients at
low risk for all models
combined
Prognostic accuracy,
†
OR Prognostic accuracy
†
Prognostic accuracy,
†
proportion of patients in
each risk group, event rates within each risk
group (absolute risk), population event rate
Overall summary estimate
from combined derivation and
validation studies
Yes No Yes Yes
Summary estimate from
external validation studies
No Yes No Yes
Analysis of update studies No No No Yes
Analysis of impact studies No No No Yes
Continued
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 23
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
likely to lead to full publication. To assess publication
bias for proportion and event rates, we applied statistical
tests to detect funnel plot asymmetry, but we found little
evidence of publication bias. According to Macaskill
et al,
145
‘applying such tests in systematic review of diag-
nostic test accuracy is likely to result in publication bias
being incorrectly indicated by the test far too often’.
Data extraction was performed by two reviewers. The
second reviewer was not blinded to the data extracted.
To ensure quality, data were rechecked by the first
reviewer. We assessed study quality using a domain
approach
17
rather than a scale. This is more transparent
to the reader, by showing how each study is performed
in each aspect of conduct. As shown by Juni et al,
19
‘the
use of summary scores to identify trials of high quality is
problematic and relevant methodological aspects should
be assessed individually and their influence on effect
sizes explored’.
Our analyses were performed for distinct models, out-
comes/time points, with subgroup and sensitivity ana-
lyses to deal with heterogeneity. We used multiple
statistical techniques, which could introduce bias in the
selection of results to present, but no discrepancies were
found across the results and all analyses are available.
We used absolute risk rather than relative risk because
this is more appropriate for estimating the risks for indi-
vidual patients.
Comparison with other reviews
This study was not limited to only clinical prediction
rules and to our knowledge, it is the first broad system-
atic review and meta-analysis of prognostic models in
patients with acute PE. There are two published system-
atic reviews on clinical prediction rules
146 147
and a
meta-analysis of PESI models,
148
but, as discussed below,
our systematic review has important differences to all of
these, expands their scope considerably and provides
more comprehensive and up-to-date data (table 6).
We believe the key strengths of our review are well-
defined selection criteria, broad search strategy, and
presentation of results of all available predictive models
with full details on study characteristics, population and
prognostic information and study quality. We used a
domain approach and prognostic criteria for the assess-
ment of study quality. Quantitative analyses are provided
to answer each aspect of model development in accord-
ance with the recommendations of recently published
guidelines for prognostic model research,
149
for differ-
ent outcomes and time points in ‘both stable and
unstable’as well as in ‘stable’PE, and at various
cut-offs, with the aim of improving management of
patients with acute PE who are either at low risk or at
high risk. For the analysis of performance, we provide a
summary estimate of sensitivity and specificity and also
a summary estimate of event rates (‘absolute risk’
149
)
within risk groups, which is more meaningful for clini-
cians and seems more appropriate for the study of
prognosis.
Table 6 Continued
Author, year
Squizatto et al,
2012
146
Zhou et al, 2012
148
Kohn et al, 2015
147
Our study
Conclusion of the review ‘Prognostic CPRs
efficiently identify PE
patients at a low risk of
mortality’
‘PESI has discriminative
power to predict the
short-term death and adverse
outcome events in patients
with acute pulmonary
embolism, the PESI and the
sPESI have similar accuracy’
‘Numerous clinical prediction
rules for prognosticating early
mortality in patients with PE
are available, but not all
demonstrate the high
sensitivity needed to reassure
clinicians’
‘We provide evidence-based information about
the validity and utility of the existing prognostic
models in acute PE that may be helpful for
identifying patients at low risk. Novel models
seem attractive for the high-risk normotensive
PE but need to be externally validated than be
assessed in impact studies’
*Seven other types of models include five models whose variables are originally not specific to PE or to its prognosis and two hospital criteria checklists.
†Prognostic accuracy (summary sensitivity, summary specificity and summary receiver operating characteristic).
CPR, cardiopulmonary resuscitation; PE, pulmonary embolism; PESI, Pulmonary Embolism Severity Index; sPESI, simplified PESI; VTE, venous thromboembolism.
24 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
CONCLUSIONS
Our systematic review provides useful information on
the prediction ability and the utility of existing prognos-
tic models in acute PE that might help clinicians and
researchers for the identification of patients at low risk
of events for safe early discharge or outpatient manage-
ment and those at high risk who may need closer moni-
toring or more aggressive therapy. It shows the potential
for improving the selection of lower risk and higher risk
groups in patients with normotensive PE with novel and
updated models that integrate biomarkers (cTn, BNP,
NT-proBNP, H-FABP), CT or TTE-RVD and US-detected
DVT into existing models or with other clinical variables
(systolic blood pressure, heart rate). These findings
provide a good direction for future research in valid-
ation and impact studies.
Acknowledgements The authors wish to thank Nicola PEARCE-SMITH at the
Department of Knowledge and Information Science, Oxford, UK, and Anne
BRICE at James Lind Initiative, Oxford, UK for their helpful advice on search
strategy.
Contributors AE and MC conceived the study. AE, SM, and MC designed the
study. AE, MD-E and J-NP undertook the literature search and extracted data.
AE did data analysis. AE, SM and MC interpreted data. AE developed the first
draft. AE and MC contributed to the writing of the manuscript. All authors
provided critical comments and approved the final version. AE had full access
to all the data in the study and had final responsibility for the decision to
submit for publication. All researchers had access to all the data. AE as
guarantor accepts full responsibility for the work and the conduct of the
study. AE affirms that the manuscript is an honest, accurate, and transparent
account of the study being reported; that no important aspects of the study
have been omitted; and that any discrepancies from the study as planned
have been explained.
Funding This research received no specific grant from any funding agency in
the public, commercial or not-for-profit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement No additional data are available.
Open Access This is an Open Access article distributed in accordance with
the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license,
which permits others to distribute, remix, adapt, build upon this work non-
commercially, and license their derivative works on different terms, provided
the original work is properly cited and the use is non-commercial. See: http://
creativecommons.org/licenses/by-nc/4.0/
REFERENCES
1. Torbicki A, Perrier A, Konstantinides S, 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
J2008;29:2276–315.
2. Jaff MR, McMurtry MS, Archer SL, et al. Management of massive
and submassive pulmonary embolism, iliofemoral deep vein
thrombosis, and chronic thromboembolic pulmonary hypertension:
a scientific statement from the American Heart Association.
Circulation 2011;123:1788–830.
3. Konstantinides SV, Torbicki A, Agnelli G, et al.2014
ESC guidelines on the diagnosis and management of
acute pulmonary embolism. Eur Heart J 2014;35:3033–69,
69a-69k.
4. Kearon C, Akl EA, Ornelas J, et al. Antithrombotic therapy for VTE
disease: CHEST guideline and expert panel report. Chest
2016;149:315–52.
5. Becattini C, Vedovati MC, Agnelli G. Prognostic value of troponins
in acute pulmonary embolism: a meta-analysis. Circulation
2007;116:427–33.
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;178:425–30.
7. Sanchez O, Trinquart L, Colombet I, et al. Prognostic value of right
ventricular dysfunction in patients with haemodynamically stable
pulmonary embolism: a systematic review. Eur Heart J
2008;29:1569–77.
8. Jimenez D, Uresandi F, Otero R, et al. Troponin-based risk
stratification of patients with acute nonmassive pulmonary
embolism: systematic review and metaanalysis. Chest
2009;136:974–82.
9. Lega JC, Lacasse Y, Lakhal L, et al. Natriuretic peptides and
troponins in pulmonary embolism: a meta-analysis. Thorax
2009;64:869–75.
10. Altman DG. Prognostic models: a methodological framework and
review of models for breast cancer. Cancer Invest 2009;27:235–43.
11. Mallett S, Royston P, Dutton S, et al. Reporting methods in studies
developing prognostic models in cancer: a review. BMC Med
2010;8:20.
12. Mallett S, Royston P, Waters R, et al. Reporting performance of
prognostic models in cancer: a review. BMC Med 2010;8:21.
13. Williams C, Brunskill S, Altman D, et al. Cost-effectiveness of using
prognostic information to select women with breast cancer for
adjuvant systemic therapy. Health Technol Assess 2006;10:iii–iv,
ix-xi, 1–204.
14. Altman DG. Systematic reviews of evaluations of prognostic
variables. BMJ 2001;323:224–8.
15. Egger M, Juni P, Bartlett C, et al. How important are
comprehensive literature searches and the assessment of trial
quality in systematic reviews? Empirical study. Health Technol
Assess 2003;7:1–76.
16. Hayden JA, Cote P, Bombardier C. Evaluation of the quality of
prognosis studies in systematic reviews. Ann Intern Med
2006;144:427–37.
17. Higgins JPT, Green SE. Cochrane handbook for systematic reviews
of interventions version 5.1.0 [updated March 2011]. The Cochrane
Collaboration, 2011. http://www.cochrane-handbook.org
18. Schulz KF, Grimes DA, eds. Handbook of essential concepts in
clinical research. Elsevier Ltd, 2006.
19. Juni P, Witschi A, Bloch R, et al. The hazards of scoring the quality
of clinical trials for meta-analysis. JAMA 1999;282:1054–60.
20. Harbord RM, Deeks JJ, Egger M, et al. A unification of models for
meta-analysis of diagnostic accuracy studies. Biostatistics
2007;8:239–51.
21. Harbord RM, Whiting P. Metandi: meta-analysis of diagnostic
accuracy using hierarchical logistic regression. Stata J
2009;2:211–29.
22. Harbord R. Metandi: Stata module to perform meta-analysis of
diagnostic accuracy. Bristol, UK: MRC Health Services Research
Collaboration, University of Bristol, 2006.
23. Rabe-Hesketh S, Skrondal A, Pickles A. GLLAMM Manual. U.C.
Berkeley Division of Biostatistics Working Paper Series. Working
Paper 160, 2004. http://www.bepress.com/ucbbiostat/paper160
24. Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for
systematic reviews and meta-analyses: the PRISMA statement.
PLoS Med 2009;6:e1000097.
25. Al Otair HA, Al-Boukai AA, Ibrahim GF, et al. Outcome of
pulmonary embolism and clinico-radiological predictors of mortality:
experience from a university hospital in Saudi Arabia. Ann Thorac
Med 2014;9:18–22.
26. Almahameed A, Carman TL. Outpatient management of stable
acute pulmonary embolism: proposed accelerated pathway for risk
stratification. Am J Med 2007;120(Suppl 2):S18–25.
27. Bircan A, Karadeniz N, Ozden A, et al. A simple clinical model
composed of ECG, shock index, and arterial blood gas analysis for
predicting severe pulmonary embolism. Clin Appl Thromb Hemost
2011;17:188–96.
28. Bledsoe J, Hamilton D, Bess E, et al. Treatment of low-risk
pulmonary embolism patients in a chest pain unit. Crit Pathw
Cardiol 2010;9:212–15.
29. Budaj-Fidecka A, Kurzyna M, Fija⍰kowska A, et al. In-hospital
major bleeding predicts mortality in patients with pulmonary
embolism: an analysis of ZATPOL Registry data. Int J Cardiol
2013;168:3543–9.
30. Cameron A, Ogilvie C, Teckchandani S, et al. Outpatient imaging
for pulmonary embolism may only be suitable for a minority. Scott
Med J 2012;57:14–17.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 25
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
31. Gallotta G, Palmieri V, Piedimonte V, et al. Increased troponin I
predicts in-hospital occurrence of hemodynamic instability in
patients with sub-massive or non-massive pulmonary embolism
independent to clinical, echocardiographic and laboratory
information. Int J Cardiol 2008;124:351–7.
32. Gul EE, Can I, Guler I, et al. Association of pulmonary artery
obstruction index with elevated heart-type fatty acid binding protein
and short-term mortality in patients with pulmonary embolism at
intermediate risk. Diagn Interv Radiol 2012;18:531–6.
33. Lankeit M, Dellas C, Benz V, et al. The predictive value of
heart-type fatty acid-binding protein is independent from symptom
duration in normotensive patients with pulmonary embolism.
Thromb Res 2013;132:543–7.
34. McCabe A, Hassan T, Doyle M, et al. Identification of patients with
low-risk pulmonary embolism suitable for outpatient treatment using
the pulmonary embolism severity index (PESI). Ir J Med Sci
2013;182:291–5.
35. Paiva L, Providencia R, Faustino A, et al. Manchester triage in
acute pulmonary embolism: can it unmask the grand impersonator?
Emerg Med J 2012;29:e6.
36. Palla A, Ribas C, Rossi G, et al. The clinical course of pulmonary
embolism patients anticoagulated for 1 year: results of a
prospective, observational, cohort study. J Thromb Haemost
2010;8:68–74.
37. Rydman R, Söderberg M, Larsen F, et al. d-Dimer and simplified
pulmonary embolism severity index in relation to right ventricular
function. Am J Emerg Med 2013;31:482–6.
38. Spirk D, Willenberg T, Aujesky D, et al. Use of biomarkers or
echocardiography in pulmonary embolism: the Swiss Venous
Thromboembolism Registry. QJM 2012;105:1163–9.
39. Taylor RA, Davis J, Liu R, et al. Point-of-care focused cardiac
ultrasound for prediction of pulmonary embolism adverse
outcomes. J Emerg Med 2013;45:392–9.
40. Tsai J, Abe K, Boulet SL, et al. Predictive accuracy of
29-comorbidity index for in-hospital deaths in US adult
hospitalizations with a diagnosis of venous thromboembolism.
PLoS ONE 2013;8:e70061.
41. George E, Kumamaru KK, Ghosh N, et al. Computed tomography
and echocardiography in patients with acute pulmonary
embolism: part 2: prognostic value. J Thorac Imaging 2014;29:
W7–12.
42. Goldhaber SZ, Visani L, De Rosa M. Acute pulmonary embolism:
clinical outcomes in the International Cooperative Pulmonary
Embolism Registry (ICOPER). Lancet 1999;353:1386–9.
43. Kanbay A, Kokturk N, Kaya MG, et al. Electrocardiography and
Wells scoring in predicting the anatomic severity of pulmonary
embolism. Respir Med 2007;101:1171–6.
44. Kasper W, Konstantinides S, Geibel A, et al. Management
strategies and determinants of outcome in acute major pulmonary
embolism: results of a multicenter registry. JAm Coll Cardiol
1997;30:1165–71.
45. Konstantinides S, Geibel A, Olschewski M, et al. Association
between thrombolytic treatment and the prognosis of
hemodynamically stable patients with major pulmonary embolism:
results of a multicenter registry. Circulation 1997;96:882–8.
46. Kucher N, Wallmann D, Carone A, et al. Incremental prognostic
value of troponin I and echocardiography in patients with acute
pulmonary embolism. Eur Heart J 2003;24:1651–6.
47. Rodríguez-Cerrillo M, Alvarez-Arcaya A, Fernández-Díaz E, et al.
A prospective study of the management of non-massive
pulmonary embolism in the home. Eur J Intern Med 2009;20:
598–600.
48. Sen HS, Abakay O, Tanrikulu AC, et al. Is a complete blood cell
count useful in determining the prognosis of pulmonary embolism?
Wien Klin Wochenschr 2014;126:347–54.
49. Hariharan P, Takayesu JK, Kabrhel C. Erratum to: Association
between the Pulmonary Embolism Severity Index (PESI) and
short-term clinical deterioration (Thromb Haemost 2011; 105: 706–
711). Thromb Haemost 2012;107:599–9.
50. Aujesky D, Roy PM, Le Manach CP, et al. Validation of a model to
predict adverse outcomes in patients with pulmonary embolism.
Eur Heart J 2006;27:476–81.
51. Zondag W, Hiddinga BI, Crobach MJT, et al. Hestia criteria can
discriminate high- from low-risk patients with pulmonary embolism.
Eur Respir J 2013;41:588–92.
52. Zondag W, Mos ICM, Creemers-Schild D, et al. Outpatient
treatment in patients with acute pulmonary embolism: the Hestia
Study. J Thromb Haemost 2011;9:1500–7.
53. Barra S, Paiva L, Providencia R, et al. LR-PED rule: low risk
pulmonary embolism decision rule—a new decision score for low
risk pulmonary embolism. Thromb Res 2012;130:327–33.
54. Moores L, Zamarro C, Gómez V, et al. Changes in PESI scores
predict mortality in intermediate-risk patients with acute pulmonary
embolism. Eur Respir J 2013;41:354–9.
55. Otero R, Trujillo-Santos J, Cayuela A, et al. Haemodynamically
unstable pulmonary embolism in the RIETE Registry: systolic blood
pressure or shock index? Eur Respir J 2007;30:1111–16.
56. Stuck AK, Stone RA, Pugh N, et al. Comparison of the Pulmonary
Embolism Severity Index to a simplified version: classification and
clinical outcomes. Thromb Res 2014;133:515–16.
57. Dunn AS. ACP Journal Club. The simplified Pulmonary Embolism
Severity Index and high-sensitivity troponin T predicted prognosis
in PE. Ann Intern Med 2012;156:JC4–12.
58. Gould MK. A clinical prediction model was sensitive but not specific
for predicting mortality in pulmonary embolism. ACP J Club
2006;145:53.
59. O’Flaherty M, Gandara E. A simple prediction rule identified
patients with pulmonary embolism at low risk of short term death.
Evid Based Med 2006;11:121.
60. Golpe R, Perez-de-Llano LA, Castro-Anon O. Prognostic value of
the Charlson comorbidity index in pulmonary embolism. Respiration
2013;85:438.
61. Ahmad N, Srinivasan K, Moudgil H. Risk stratification in pulmonary
embolism: an algorithmic tool approach. Thorax 2011;66:1098–9.
62. Chow V, Ng AC, Chung T, et al. Right atrial to left atrial area ratio
on early echocardiography predicts long-term survival after acute
pulmonary embolism. Cardiovasc Ultrasound 2013;11:17.
63. Fraga M, Taffe P, Mean M, et al. The inter-rater reliability of the
Pulmonary Embolism Severity Index. Thromb Haemost
2010;104:1258–62.
64. Sequeira JF, Duarte PC, Madruga IM, et al. [Pulmonary
embolism--mortality risk]. Rev Port Cardiol 1992;11:749–57.
65. Agterof MJ, Schutgens RE, Moumli N, et al. A prognostic model for
short term adverse events in normotensive patients with pulmonary
embolism. Am J Hematol 2011;86:646–9.
66. Agterof MJ, Schutgens REG, Snijder RJ, et al. Out of hospital
treatment of acute pulmonary embolism in patients with a low
NT-proBNP level. J Thromb Haemost 2010;8:1235–41.
67. Agterof MJ, Van Bladel ER, Schutgens REG, et al. Risk
stratification of patients with pulmonary embolism based on pulse
rate and D-dimer concentration. Thromb Haemost 2009;102:683–7.
68. Aujesky D, Obrosky DS, Stone RA, et al. Derivation and validation
of a prognostic model for pulmonary embolism. Am J Respir Crit
Care Med 2005;172:1041–6.
69. Aujesky D, Obrosky DS, Stone RA, et al. A prediction rule to
identify low-risk patients with pulmonary embolism. Arch Intern Med
2006;166:169–75.
70. Aujesky D, Perrier A, Roy PM, et al. Validation of a clinical
prognostic model to identify low-risk patients with pulmonary
embolism. J Intern Med 2007;261:597–604.
71. Aujesky D, Roy PM, Verschuren F, et al. Outpatient versus
inpatient treatment for patients with acute pulmonary embolism: an
international, open-label, randomised, non-inferiority trial. Lancet
2011;378:41–8.
72. Becattini C, Casazza F, Forgione C, et al. Acute pulmonary
embolism: external validation of an integrated risk stratification
model. Chest 2013;144:1539–45.
73. Bertoletti L, Le Gal G, Aujesky D, et al. Prognostic value of the
Geneva prediction rule in patients with pulmonary embolism.
Thromb Res 2013;132:32–6.
74. Bova C, Pesavento R, Marchiori A, et al. Risk stratification and
outcomes in hemodynamically stable patients with acute pulmonary
embolism: a prospective, multicentre, cohort study with three
months of follow-up. J Thromb Haemost 2009;7:938–44.
75. Chan CM, Woods C, Shorr AF. The validation and reproducibility of
the pulmonary embolism severity index. J Thromb Haemost
2010;8:1509–14.
76. Chan CM, Woods CJ, Shorr AF. Comparing the pulmonary
embolism severity index and the prognosis in pulmonary embolism
scores as risk stratification tools. J Hosp Med 2012;7:22–7.
77. Choi KJ, Cha SI, Shin KM, et al. Prognostic implications of
computed tomographic right ventricular dilation in patients with
acute pulmonary embolism. Thromb Res 2014;133:182–6.
78. Choi WH, Kwon SU, Jwa YJ, et al. The pulmonary embolism
severity index in predicting the prognosis of patients with
pulmonary embolism. Korean J Intern Med 2009;24:123–7.
79. Davies CWH, Wimperis J, Green ES, et al. Early discharge of
patients with pulmonary embolism: a two-phase observational
study. Eur Respir J 2007;30:708–14.
80. Dellas C, Tschepe M, Seeber V, et al. A novel H-FABP assay and
a fast prognostic score for risk assessment of normotensive
pulmonary embolism. Thromb Haemost 2014;111:996–1003.
26 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
81. Donze J, Le Gal G, Fine MJ, et al. Prospective validation of the
Pulmonary Embolism Severity Index: a clinical prognostic model for
pulmonary embolism. Thromb Haemost 2008;100:943–8.
82. Erkens PMG, Gandara E, Wells PS, et al. Does the Pulmonary
Embolism Severity Index accurately identify low risk patients
eligible for outpatient treatment? Thromb Res 2012;129:710–14.
83. Ferrer M, Morillo R, Elias T, et al. Validation of two clinical
prognostic models in patients with acute symptomatic pulmonary
embolism. Arch Bronconeumol 2013;49:427–31.
84. Hariharan P, Takayesu JK, Kabrhel C. Association between the
pulmonary embolism severity index (PESI) and short-term clinical
deterioration. Thromb Haemost 2011;105:706–11.
85. Jimenez D, Aujesky D, Moores L, et al. Simplification of the
pulmonary embolism severity index for prognostication in patients
with acute symptomatic pulmonary embolism. Arch Intern Med
2010;170:1383–9.
86. Jimenez D, Aujesky D, Moores L, et al. Combinations of prognostic
tools for identification of high-risk normotensive patients with acute
symptomatic pulmonary embolism. Thorax 2011;66:75–81.
87. Jiménez D, Kopecna D, Tapson V, et al. Derivation and validation
of multimarker prognostication for normotensive patients with acute
symptomatic pulmonary embolism. Am J Respir Crit Care Med
2014;189:718–26.
88. Jimenez D, Yusen RD, Otero R, et al. Prognostic models for
selecting patients with acute pulmonary embolism for initial
outpatient therapy. Chest 2007;132:24–30.
89. Jo JY, Lee MY, Lee JW, et al. Leukocytes and systemic
inflammatory response syndrome as prognostic factors in
pulmonary embolism patients. BMC Pulm Med 2013;13:74.
90. Kostrubiec M, Pruszczyk P, Bochowicz A, et al. Biomarker-based
risk assessment model in acute pulmonary embolism. Eur Heart J
2005;26:2166–72.
91. Kukla P, Bryniarski L, Dlugopolski R, et al. Acute pulmonary
embolism registry in the Malopolska region—clinical course. Kardiol
Pol 2009;67:735–41.
92. Lankeit M, Friesen D, Schafer K, et al. A simple score for rapid risk
assessment of non-high-risk pulmonary embolism. Clin
2013;102:73–80.
93. Lankeit M, Gomez V, Wagner C, et al. A strategy combining
imaging and laboratory biomarkers in comparison with a simplified
clinical score for risk stratification of patients with acute pulmonary
embolism. Chest2012;141:916–22.
94. Lankeit M, Jimenez D, Kostrubiec M, et al. Predictive value of the
high-sensitivity troponin T assay and the simplified pulmonary
embolism severity index in hemodynamically stable patients with
acute pulmonary embolism: a prospective validation study.
Circulation 2011;124:2716–24.
95. Lankeit M, Jiménez D, Kostrubiec M, et al. Validation of N-terminal
pro-brain natriuretic peptide cut-off values for risk stratification of
pulmonary embolism. Eur Respir J 2014;43:1669–77.
96. Moores L, Aujesky D, Jimenez D, et al. Pulmonary Embolism
Severity Index and troponin testing for the selection of low-risk
patients with acute symptomatic pulmonary embolism. J Thromb
Haemost 2010;8:517–22.
97. Nendaz MR, Bandelier P, Aujesky D, et al. Validation of a risk score
identifying patients with acute pulmonary embolism, who are at low
risk of clinical adverse outcome. Thromb Haemost 2004;91:1232–6.
98. Ng AC, Chow V, Yong AS, et al. Prognostic impact of the Charlson
comorbidity index on mortality following acute pulmonary embolism.
Respiration 2013;85:408–16.
99. Nordenholz K, Ryan J, Atwood B, et al. Pulmonary embolism risk
stratification: pulse oximetry and pulmonary embolism severity
index. J Emerg Med 2011;40:95–102.
100. Otero R, Uresandi F, Jimenez D, et al. Home treatment in
pulmonary embolism. Thromb Res 2010;126:e1–5.
101. Ozsu S, Abul Y, Orem A, et al. Predictive value of troponins and
simplified pulmonary embolism severity index in patients with
normotensive pulmonary embolism. Multidiscip Respir Med
2013;8:34.
102. Ozsu S, Ozlu T, Sentürk A, et al. Combination and comparison of
two models in prognosis of pulmonary embolism: results from
Turkey Pulmonary Embolism Group (TUPEG) study. Thromb Res
2014;133:1006–10.
103. Paiva LV, Providencia RC, Barra SN, et al. Cardiovascular risk
assessment of pulmonary embolism with the GRACE risk score.
Am J Cardiol 2013;111:425–31.
104. Palmieri V, Gallotta G, Rendina D, et al. Troponin I and right
ventricular dysfunction for risk assessment in patients with
nonmassive pulmonary embolism in the Emergency Department in
combination with clinically based risk score. Intern Emerg Med
2008;3:131–8.
105. Papaioannou AI, Kastanis A, Malli F, et al. Predicting short-term
mortality in patients with pulmonary embolism: a simple model.
Am J Med Sci 2013;345:462–9.
106. Righini M, Roy PM, Meyer G, et al.The Simplified Pulmonary
Embolism Severity Index (PESI): validation of a clinical
prognostic model for pulmonary embolism. J Thromb Haemost
2011;9:2115–17.
107. Roca IC, Aursulesei V, Roca M, et al. [Comparative evaluation of
prognostic scores in pulmonary embolism]. Rev Med Chir Soc Med
Nat Iasi 2012;116:69–74.
108. Sam A, Sanchez D, Gomez V, et al. The shock index and the
simplified PESI for identification of low-risk patients with acute
pulmonary embolism. Eur Respir J 2011;37:762–6.
109. Sanchez D, De Miguel J, Sam A, et al. The effects of cause of
death classification on prognostic assessment of patients with
pulmonary embolism. J Thromb Haemost 2011;9:2201–7.
110. Sanchez O, Trinquart L, Caille V, et al. Prognostic factors for
pulmonary embolism: the PREP study, a prospective multicenter
cohort study. Am J Respir Crit Care Med 2010;181:168–73.
111. Sanchez O, Trinquart L, Planquette B, et al. Echocardiography and
pulmonary embolism severity index have independent prognostic
roles in pulmonary embolism. Eur Respir J 2013;42:681–8.
112. Singanayagam A, Chalmers JD, Scally C, et al. Right ventricular
dilation on CT pulmonary angiogram independently predicts
mortality in pulmonary embolism. Respir Med 2010;104:1057–62.
113. Singanayagam A, Scally C, Al-Khairalla MZ, et al. Are biomarkers
additive to pulmonary embolism severity index for severity
assessment in normotensive patients with acute pulmonary
embolism? QJM 2011;104:125–31.
114. Soares TH, de Bastos M, de Carvalho BV, et al. Prognostic value
of computed tomographic pulmonary angiography and the
pulmonary embolism severity index in patients with acute
pulmonary embolism. Blood Coagul Fibrinolysis 2013;24:
64–70.
115. Spirk D, Aujesky D, Husmann M, et al. Cardiac troponin testing
and the simplified pulmonary embolism severity index the Swiss
venous thromboembolism registry (SWIVTER). Thromb Haemost
2011;106:978–84.
116. Subramaniam RM, Mandrekar J, Blair D, et al. The Geneva
prognostic score and mortality in patients diagnosed with
pulmonary embolism by CT pulmonary angiogram. J Med Imaging
Radiat Oncol 2009;53:361–5.
117. Toosi MS, Merlino JD, Leeper KV. Prognostic value of the shock
index along with transthoracic echocardiography in risk stratification
of Patients With Acute Pulmonary Embolism. Am J Cardiol
2008;101:700–5.
118. Uresandi F, Otero R, Cayuela A, et al. A clinical prediction rule for
identifying short-term risk of adverse events in patients with
pulmonary thromboembolism. [Spanish]. Arch Bronconeumol
2007;43:617–22.
119. Vanni S, Nazerian P, Pepe G, et al. Comparison of two prognostic
models for acute pulmonary embolism: clinical vs. right ventricular
dysfunction-guided approach. J Thromb Haemost 2011;9:1916–23.
120. Venetz C, Jimenez D, Mean M, et al. A comparison of the original
and simplified pulmonary embolism severity index. Thromb
Haemost 2011;106:423–8.
121. Volschan A, Albuquerque D, Tura BR, et al. Predictors of hospital
mortality in hemodynamically stable patients with pulmonary
embolism. [Portuguese, Spanish, English]. Arq Bras Cardiol
2009;93:135–40.
122. Wicki J, Perrier A, Perneger TV, et al. Predicting adverse outcome
in patients with acute pulmonary embolism: q risk score. Thromb
Haemost 2000;84:548–52.
123. Yamaki T, Nozaki M, Sakurai H, et al. Presence of lower limb deep
vein thrombosis and prognosis in patients with symptomatic
pulmonary embolism: preliminary report. Eur J Vasc Endovasc
Surg 2009;37:225–31.
124. Zhu L, Wang C, Yang YH, et al. [Prognostic value of right
ventricular dysfunction and derivation of a prognostic model for
patients with acute pulmonary thromboembolism]. Zhonghua Liu
Xing Bing Xue Za Zhi 2009;30:184–8.
125. Zondag W, den Exter PL, Crobach MJT, et al. Comparison of two
methods for selection of out of hospital treatment in patients with
acute pulmonary embolism. Thromb Haemost 2013;109:47–52.
126. Zondag W, Vingerhoets LMA, Durian MF, et al. Hestia criteria can
safely select patients with pulmonary embolism for outpatient
treatment irrespective of right ventricular function. J Thromb
Haemost 2013;11:686–92.
127. Barra SNC, Paiva LV, Providência R, et al. Atrial fibrillation in acute
pulmonary embolism: prognostic considerations. Emerg Med J
2014;31:308–12.
Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324 27
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
128. Beer JH, Burger M, Gretener S, et al. Outpatient treatment of
pulmonary embolism is feasible and safe in a substantial proportion
of patients. J Thromb Haemost 2003;1:186–7.
129. Huang CM, Lin YC, Lin YJ, et al. Risk stratification and clinical
outcomes in patients with acute pulmonary embolism.
Clin Biochem 2011;44:1110–15.
130. Jakobsson C, Jimenez D, Gomez V, et al. Validation of a clinical
algorithm to identify low-risk patients with pulmonary embolism.
J Thromb Haemost 2010;8:1242–7.
131. Labyk A, Ciurzynski M, Jankowski K, et al. Acute pulmonary
embolism: analysis of consecutive 353 patients hospitalised in a
single centre. A 3-year experience. Kardiologia polska
2012;70:15–22.
132. Meyer G, Vicaut E, Danays T, et al. Fibrinolysis for patients with
intermediate-risk pulmonary embolism. N Engl J Med
2014;370:1402–11.
133. Stamm JA, Long JL, Kirchner HL, et al. Risk Stratification in acute
pulmonary embolism: frequency and impact on treatment decisions
and outcomes. South Med J 2014;107:72–8.
134. Vuilleumier N, Le Gal G, Cornily JC, et al. Is N-terminal pro-brain
natriuretic peptide superior to clinical scores for risk stratification in
non-massive pulmonary embolism? J Thromb Haemost
2010;8:1433–5.
135. Bova C, Sanchez O, Prandoni P, et al. Identification of
intermediate-risk patients with acute symptomatic pulmonary
embolism. Eur Respir J 2014;44:694–703.
136. Perrier A, Howarth N, Didier D, et al. Performance of helical
computed tomography in unselected outpatients with suspected
pulmonary embolism. Ann Intern Med 2001;135:88–97.
137. Perrier A, Roy PM, Aujesky D, et al. Diagnosing pulmonary
embolism in outpatients with clinical assessment, D-dimer
measurement, venous ultrasound, and helical computed
tomography: a multicenter management study. Am J Med
2004;116:291–9.
138. Perrier A, Roy PM, Sanchez O, et al. Multidetector-row computed
tomography in suspected pulmonary embolism. N Engl J Med
2005;352:1760–8.
139. Righini M, Le Gal G, Aujesky D, et al. Diagnosis of pulmonary
embolism by multidetector CT alone or combined with venous
ultrasonography of the leg: a randomised non-inferiority trial.
Lancet 2008;371:1343–52.
140. Roy PM, Meyer G, Vielle B, et al. Appropriateness of diagnostic
management and outcomes of suspected pulmonary embolism.
Ann Intern Med 2006;144:157–64.
141. Vuilleumier N, Le Gal G, Verschuren F, et al. Cardiac biomarkers
for risk stratification in non-massive pulmonary embolism:
a multicenter prospective study. J Thromb Haemost 2009;7:391–8.
142. Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, et al. Evaluating
the added predictive ability of a new marker: from area under the
ROC curve to reclassification and beyond. Stat Med
2008;27:157–72; discussion 207–12.
143. McGinn TG, Guyatt GH, Wyer PC, et al. Users’guides to the
medical literature: XXII: how to use articles about clinical decision
rules. Evidence-Based Medicine Working Group. JAMA
2000;284:79–84.
144. Vinson DR, Zehtabchi S, Yealy DM. Can selected patients with
newly diagnosed pulmonary embolism be safely treated without
hospitalization? A systematic review. Ann Emerg Med
2012;60:651–62.e4.
145. Macaskill P, Gatsonis C, Deeks JJ, et al. Chapter 10: analysing
and presenting results. In: Deeks JJ, Bossuyt PM, Gatsonis C, eds.
Cochrane handbook for systematic reviews of diagnostic test
accuracy version 1.0. The Cochrane Collaboration, 2010. http://
srdta.cochrane.org/
146. Squizzato A, Donadini MP, Galli L, et al. Prognostic clinical
prediction rules to identify a low-risk pulmonary embolism:
a systematic review and meta-analysis. J Thromb Haemost
2012;10:1276–90.
147. Kohn CG, Mearns ES, Parker MW, et al. Prognostic accuracy of
clinical prediction rules for early post-pulmonary embolism
all-cause mortality: a bivariate meta-analysis. Chest
2015;147:1043–62.
148. Zhou XY, Ben SQ, Chen HL, et al. The prognostic value of
pulmonary embolism severity index in acute pulmonary embolism:
a meta-analysis. Respir Res 2012;13:111.
149. Steyerberg EW, Moons KG, van der Windt DA, et al. Prognosis
Research Strategy (PROGRESS) 3: prognostic model research.
PLoS Med 2013;10:e1001381.
28 Elias A, et al.BMJ Open 2016;6:e010324. doi:10.1136/bmjopen-2015-010324
Open Access
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from
meta-analysis
embolism: a systematic review and
Prognostic models in acute pulmonary
Mike Clarke
Antoine Elias, Susan Mallett, Marie Daoud-Elias, Jean-Noël Poggi and
doi: 10.1136/bmjopen-2015-010324
2016 6: BMJ Open
http://bmjopen.bmj.com/content/6/4/e010324
Updated information and services can be found at:
These include:
References #BIBLhttp://bmjopen.bmj.com/content/6/4/e010324
This article cites 144 articles, 30 of which you can access for free at:
Open Access
http://creativecommons.org/licenses/by-nc/4.0/non-commercial. See:
provided the original work is properly cited and the use is
non-commercially, and license their derivative works on different terms,
permits others to distribute, remix, adapt, build upon this work
Commons Attribution Non Commercial (CC BY-NC 4.0) license, which
This is an Open Access article distributed in accordance with the Creative
service
Email alerting box at the top right corner of the online article.
Receive free email alerts when new articles cite this article. Sign up in the
Collections
Topic Articles on similar topics can be found in the following collections
(488)Evidence based practice (568)Cardiovascular medicine
Notes
http://group.bmj.com/group/rights-licensing/permissions
To request permissions go to:
http://journals.bmj.com/cgi/reprintform
To order reprints go to:
http://group.bmj.com/subscribe/
To subscribe to BMJ go to:
group.bmj.com on May 16, 2016 - Published by http://bmjopen.bmj.com/Downloaded from