Vascular Medicine
Risk Assessment of Recurrence in Patients With
Unprovoked Deep Vein Thrombosis or
Pulmonary Embolism
The Vienna Prediction Model
Sabine Eichinger, MD*; Georg Heinze, PhD*; Lisanne M. Jandeck, MSc; Paul A. Kyrle, MD
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
Background—Predicting the risk of recurrent venous thromboembolism (VTE) in an individual patient is often not
feasible. We aimed to develop a simple risk assessment model that improves prediction of the recurrence risk.
Methods and Results—In a prospective cohort study, 929 patients with a first unprovoked VTE were followed up for a
median of 43.3 months after discontinuation of anticoagulation. We excluded patients with a strong thrombophilic defect
such as a natural inhibitor deficiency, the lupus anticoagulant, and homozygous or combined defects. A total of 176
patients (18.9%) had recurrent VTE. Preselected clinical and laboratory variables (age, sex, location of VTE, body mass
index, factor V Leiden, prothrombin G20210A mutation, D-dimer, and in vitro thrombin generation) were analyzed in
a Cox proportional hazards model, and those variables that were significantly associated with recurrence were used to
compute risk scores. Male sex (hazard ratio versus female sex 1.90, 95% confidence interval 1.31 to 2.75), proximal
deep vein thrombosis (hazard ratio versus distal 2.08, 95% confidence interval 1.16 to 3.74), pulmonary embolism (hazard
ratio versus distal thrombosis 2.60, 95% confidence interval 1.49 to 4.53), and elevated levels of D-dimer (hazard ratio per
doubling 1.27, 95% confidence interval 1.08 to 1.51) were related to a higher recurrence risk. Using these variables, we
developed a nomogram that can be used to calculate risk scores and to estimate the cumulative probability of recurrence in
an individual patient. The model was cross validated, and patients were assigned to different risk categories based on their risk
score. Recurrence rates corresponded well with the different risk categories.
Conclusions—By use of a simple scoring system, the assessment of the recurrence risk in patients with a first unprovoked
VTE and without strong thrombophilic defects can be improved. (Circulation. 2010;121:1630-1636.)
Key Words: venous thrombosis 䡲 recurrence 䡲 risk 䡲 D-dimer
V
enous thrombosis is a chronic disease, and recurrent
events are fatal in approximately 5% to 9% of patients.1
Predicting the likelihood of recurrence in an individual
patient is of utmost importance, because most recurrences can
be prevented by antithrombotic therapy. The presence or
absence of certain clinical and laboratory patient characteristics determines a low or high recurrence risk. The risk is low
among patients with venous thromboembolism (VTE) provoked by surgery, trauma, immobilization, pregnancy, or
female hormone intake, whereas it is higher among those with
unprovoked thrombosis.2 Stratification of patients with unprovoked VTE according to their recurrence risk can be
achieved on the basis of clinical risk factors including
patient’s sex, comorbidities, or overweight or by measuring
laboratory markers of thrombophilia such as factor V Leiden,
the prothrombin mutation, natural coagulation inhibitor deficiencies, elevated coagulation factors, and antiphospholipid
antibodies.3 A novel approach to assess the recurrence risk is
the use of global coagulation markers, including D-dimer,4 or
in vitro thrombin generation.5– 8
Clinical Perspective on p 1636
Despite substantial progress in identifying the determinants
of recurrence risk, its prediction in an individual patient is
often not feasible in daily routine care. VTE is a disease with
many causes, and the combined effect of clinical and laboratory characteristics on the risk of recurrence is unknown.
The determination of some laboratory risk factors is costly,
lacks standardization, or is too elaborate for routine purposes.
To overcome these limitations, we aimed to develop a simple
Continuing medical education (CME) credit is available for this article. Go to http://cme.ahajournals.org to take the quiz.
Received November 22, 2009; accepted February 9, 2010.
From the Department of Medicine I (S.E., P.A.K.), the Karl Landsteiner-Institut für Klinische Thromboseforschung (S.E., P.A.K.), and the Core Unit
for Medical Statistics and Informatics (G.H., L.M.J.), Medical University of Vienna, Vienna, Austria.
*Drs Eichinger and Heinze contributed equally to this article.
The online-only Data Supplement is available with this article at http://circ.ahajournals.org/cgi/content/full/CIRCULATIONAHA.109.925214/DC1.
Correspondence to Sabine Eichinger, MD, Department of Internal Medicine I, Waehringer Guertel 18-20, 1090 Wien, Austria. E-mail sabine.
[email protected]
© 2010 American Heart Association, Inc.
Circulation is available at http://circ.ahajournals.org
DOI: 10.1161/CIRCULATIONAHA.109.925214
1630
Eichinger et al
Risk Assessment Model to Predict Recurrent VTE
risk model that improves prediction of the recurrence risk in
patients with unprovoked VTE.
Methods
Patients
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
Patients were recruited from 4 thrombosis centers in Vienna, Austria,
between July 1992 and August 2008. Consecutive patients older than
18 years with a first VTE who had been treated with oral anticoagulants for at least 3 months were included. Patients were excluded if
they had VTE provoked by surgery, trauma, pregnancy, or female
hormone intake; deficiency of antithrombin, protein C, or protein S;
presence of the lupus anticoagulant; or cancer. All patients initially
received unfractionated or low-molecular-weight heparin at therapeutic doses. During high-risk situations, including surgery, trauma,
or immobilization, patients received thromboprophylaxis with a
low-molecular-weight heparin according to local standard practice
based on national or international guidelines.
Diagnosis of deep vein thrombosis was established by a positive
finding on venography or color duplex sonography (in case of
proximal thrombosis). The diagnosis of pulmonary embolism was
confirmed either by ventilation-perfusion scanning or by spiral
computed tomography. Patients with symptomatic pulmonary embolism and deep vein thrombosis were classified as having a
pulmonary embolism.
Patients entered the study at the time of discontinuation of oral
anticoagulation. The end point of the study was recurrent symptomatic deep vein thrombosis confirmed by venography or color duplex
sonography (in case of proximal thrombosis of the contralateral leg)
or recurrent symptomatic pulmonary embolism confirmed by
ventilation-perfusion scanning and/or spiral computed tomography.
An adjudication committee that consisted of independent radiologists established the diagnosis. The ethics committee of the Vienna
University Hospital approved the study, and all patients provided
written informed consent to participate.
Selection of Risk Factors for Development of
the Model
Clinical variables (age at venous thrombosis, sex, location of VTE,
and body mass index [determined by dividing weight in kilograms by
the square of height in meters]) and laboratory variables (factor V
Leiden mutation, prothrombin G20210A mutation, D-dimer, and in
vitro thrombin generation) were preselected by the principal investigators (SE, PAK) as relevant risk factors on the basis of the
following criteria: Independent confirmation of the impact on the
recurrence risk, simplicity of assessment, and reproducibility. Smoking status was not selected because even data on the relevance of
smoking with regard to the risk of a first VTE are conflicting,9 –11 and
the impact of smoking on recurrence risk has never been investigated
systematically. In addition, all patients were advised to refrain from
smoking for general medical reasons, so smoking behavior may have
changed over time.
Blood Sampling and Laboratory Analyses
At study entry, after patients had fasted, blood was collected into
1/10 volume of trisodium citrate 0.11 mmol/L and immediately
centrifuged for 20 minutes at 2000g. The plasma was stored at
⫺80°C. Determination of antithrombin, protein C, and protein S;
diagnosis of a lupus anticoagulant; and screening for factor V Leiden
and for prothrombin G20210A were performed as described previously.12 D-dimer was determined by ELISA (Asserachrom D-dimer,
Boehringer Mannheim, Germany). In vitro thrombin generation was
determined in platelet-poor plasma by use of a commercially
available assay (Technothrombin TGA, Technoclone, Vienna, Austria) that monitors the fluorescence generated by thrombin cleavage
of a fluorogenic substrate over time on activation of the coagulation
cascade by recombinant human tissue factor (final concentration
7.16 pmol/L) and negatively charged phospholipids (3.2 ␮mol/L).
The level of peak thrombin was used as a read-out variable. The
technicians were unaware of patient characteristics at all times.
1631
Statistical Analysis
Continuous variables are described by median and quartiles and
categorical variables by frequency and percentage. Median and
quartiles of the follow-up distribution were estimated by the KaplanMeier method with reverse meaning of the status indicator.13 Annual
recurrence rates and cumulative recurrence rates were estimated with
the actuarial (life table) and Kaplan-Meier methods, respectively.14
We started developing a risk model by fitting a Cox proportional
hazards model using all clinical variables and performing forward
selection with P⫽0.5 on laboratory variables. Because forward
selection usually leads to violation of nominal type I errors, we
assessed significance of the included variables by bootstrap zerocorrected 95% confidence intervals (CIs).15 These CIs were computed by the percentile method from 1000 resamples drawn with
replacement from the original data set, with refitting of the model on
each of the resamples and assignment of a regression coefficient of
zero to a variable not selected in a particular sample. Only variables
for which the 95% CIs excluded a regression coefficient of zero were
further considered. Next, we developed a model based on the clinical
variables and peak thrombin, because only this laboratory variable
was significant in the first step. We eliminated clinical variables if
they were not significant in this model and if their exclusion from the
model did not alter the results on the remaining variables. The model
that included sex, location of VTE, and peak thrombin could still be
overoptimistic in the sense that in future samples, the effect of these
variables on recurrence could be less than expected from the model
(regression to the mean effect). The amount of overoptimism can be
expressed by a shrinkage factor ⬍1, which was again computed by
use of bootstrap resampling and used to multiply the coefficients of
the final model to obtain a model free of optimism and suitable for
predictive purposes.16 We repeated the computation of the risk
model replacing peak thrombin with D-dimer and compared results.
Because it showed a very skewed distribution, D-dimer entered these
computations using its logarithm. For both models, we computed
goodness-of-fit tests.17 By multiplying the estimated regression
coefficients of the final risk model by the shrinkage factor, a risk
score from a patient’s particular value of sex, location of VTE, and
D-dimer can be obtained, which directly translates into estimates of
cumulative recurrence rates at various time points during follow-up.
This translation is depicted in a nomogram that includes D-dimer that
can be used for routine purposes. We expressed the predictive ability
of the model using a 0.632 bootstrap estimate of the discrimination
index.18,19 Time-dependent receiver operating characteristic curves
at 12 and 60 months were computed by the method of Heagerty and
colleagues.20 To assess the predictive ability of the models, we
computed bootstrap cross-validated risk scores for each patient; in
each bootstrap resample, we estimated model coefficients, and using
these numbers, we computed shrunken risk scores for those patients
not included in that particular resample. Finally, each patient’s risk
score was the overall average of such cross-validated risk scores for
that patient. This process resembles validation in an independent
sample. The risk scores were stratified into quartiles and compared
with the observed time to recurrence by Kaplan-Meier analysis,
accompanied by a log-rank trend test. All models were estimated
with all available complete cases with regard to covariate information. Statistical analysis was performed with the SAS system version
9.2 (2008; SAS Institute Inc, Cary, NC). For graphs and nomograms,
the statistical software R was used (www.r-project.org, mainly using
F. Harrell’s “Design” package).
Results
Study Population
The study population comprised 929 patients (Table 1). The
median follow-up was 43.3 months (25th percentile 14.7
months, 75th percentile 78.5 months). Symptomatic recurrent
VTE was seen in 176 (18.9%) of 929 patients (deep-vein
thrombosis in 100 patients and pulmonary embolism in 76
patients; 3 embolisms were fatal). In 160 patients, VTE
recurred spontaneously; in 16 patients, recurrence was pro-
1632
Circulation
April 13, 2010
Table 1.
Patient Characteristics
Variable
No. of
Subjects
Age, y
929
Female sex
929
Location of first VTE
929
Distal deep vein thrombosis
Value
54 (43, 63)
367 (40.0)
164 (17.7)
Proximal deep vein thrombosis
327 (35.2)
Pulmonary embolism
438 (47.1)
Body mass index, kg/m2
909
Body mass index ⬎30 kg/m2
27.1 (24.4, 30.1)
236 (26.0)
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
D-Dimer, ␮g/L
832
355 (236, 558)
Peak thrombin, nmol/L
629
358 (283, 439)
Factor V Leiden mutation
916
224 (24.4)
Factor II G20210A mutation
915
49 (5.4)
Duration of anticoagulation, mo
929
6.6 (6.1, 8.0)
Observation time, mo
929
43.3 (14.7, 78.5)
Values are median (25th, 75th percentiles) or frequency (%).
voked by surgery or trauma. According to Kaplan-Meier
analysis, the cumulative probabilities of recurrence (95% CI)
after 2, 5, and 10 years were 13.8% (11.6% to 16.5%), 24.6%
(21.6% to 28.9%), and 31.8% (27.6% to 37.4%), respectively
(Figure 1). The annual recurrence rate (SE) was 8.9% (1.0%)
in the first year of follow-up and was lower in subsequent
years (5.4% [0.9%], 3.6% [0.8%], 4.0% [1.0%], 5.4% [1.3%],
0.9% [0.6%], 5.6% [1.7%], 1.5% [1.1%], 2.0% [1.4%], 0%
[0%], and 6.1% [4.2%]).
Development of the Risk Model
Table 2 shows the univariate and bootstrap zero-corrected
multivariable hazard ratios associated with the risk factors. In
this first assessment of laboratory variables, only peak thrombin was a significant risk factor, whereas the presence of
factor V Leiden or prothrombin G20210A failed significance.
Also in this evaluation, D-dimer did not reach significance,
such that whenever a model included peak thrombin, D-dimer
would not add important information (hazard ratio 1.21, 95%
CI 0.87 to 1.53, P⫽0.662). When we evaluated D-dimer
without considering peak thrombin at the same time, the
association with recurrence of thrombosis reached significance, with higher values being associated with a higher risk
of recurrence (Table 2). With regard to clinical factors, only
male sex, proximal deep vein thrombosis, and pulmonary
embolism were related to a higher recurrence risk, and these
variables entered the final risk models (Table 3).
The ability of these final models to discriminate patients at
low or high recurrence risk was comparable whether peak
thrombin or D-dimer levels were used, and for both models,
there was no indication for lack of fit (model-based hazard
ratio using peak thrombin per 100 nmol/L⫽1.38, 95% CI
1.17 to 1.63, c-index 0.664, test for lack of fit P⫽0.228;
model-based hazard ratio using D-dimer per doubling⫽1.27,
95% CI 1.08 to 1.51, c-index 0.651, test for lack of fit
P⫽0.539). Because D-dimer is a well-standardized and
widely established parameter, we further developed our
model based on D-dimer levels and clinical variables.
Validation of the Risk Model
To validate the risk model, we used an internal validation
procedure based on bootstrap cross-validation in the following way: A bootstrap sample of 929 patients was drawn with
replacement from the original sample. Owing to drawing
with replacement, some patients appeared several times in
this resample, whereas others were missing. The resample
constituted a training sample in our validation process, and
the risk model was reestimated from the training sample data,
which yielded slightly different coefficients associated with
sex, location of VTE, and D-dimer than seen in the original
total sample. The patients not included in the training sample
were used as an independent validation set, for which
shrunken risk scores were computed on the basis of the
estimates from the training sample multiplied by the overall
shrinkage factor, which was estimated as 0.88. This process
was repeated 1000 times. Finally, each patient was used in
validation samples several times, and his or her risk scores
obtained in each of those validation samples were averaged to
obtain 1 risk score for each patient of the original sample.
These risk scores were divided into quartiles that defined
patients with expected low risk, intermediate low risk, intermediate high risk, and high risk. The cumulative probabilities
(95% CI) of recurrence within 5 years of patients within risk
quartiles (lowest to highest risk, respectively) were 9.2%
(5.4% to 15.5%), 21.0% (14.9% to 29.2%), 29.7% (21.5% to
40.0%), and 33.1% (25.5% to 42.1%) in the D-dimer– based
model (P⬍0.001 for trend). Figure 2 shows the receiver
operating characteristic curves with the cross-validated and
shrunken risk scores, assessing the ability to predict recurrence up to 12 months and up to 60 months.
Nomogram for Risk of Recurrent VTE
Figure 1. Overall cumulative recurrence rate in 929 patients with
a first unprovoked VTE estimated by Kaplan-Meier analysis, with
95% CIs (dotted lines).
We developed a nomogram that can be used to calculate risk
scores and expected cumulative recurrence rates from a
patient’s sex, location of initial VTE, and D-dimer levels
(Figure 3 and the online-only Data Supplement) in individual
patients. The precision of the estimated cumulative recurrence rates is expressed in 95% CIs. A straight line must be
Eichinger et al
Table 2.
Risk Assessment Model to Predict Recurrent VTE
1633
Univariate and Bootstrap Zero-Corrected Multivariable Cox Regression Models With All Risk Factors
Variable
Crude HR
95% CI
P
Bootstrap Zero-Corrected
Multivariable HR*
95% CI
P†
Sex (male vs female)
1.91
1.37–2.67
⬍0.001
2.01
1.30 –3.22
0.002
Age (per 10 y)
0.95
0.86–1.06
0.395
0.96
0.80–1.16
0.730
BMI (per 5 kg/m2)
1.19
1.02–1.38
0.026
1.03
0.78–1.36
0.806
BMI ⬎30 vs BMI ⱕ30 kg/m2
1.18
0.85–1.64
0.338
1.05
0.65–1.70
0.806
Location: proximal vs distal thrombosis
2.76
1.57–4.84
⬍0.001
1.69
0.85–3.79
0.138
Pulmonary embolism vs distal thrombosis
3.15
1.83–5.44
⬍0.001
2.77
1.44–6.24
⬍0.001
Peak thrombin (per 100 nmol/L)
1.31
1.11–1.54
0.001
1.36
1.15–1.65
0.002
D-dimer (per doubling)
1.24
1.05–1.45
0.010
1.34‡
1.06–1.70‡
0.016‡
Factor V Leiden mutation
1.54
1.12–2.11
0.007
1.68
1.00–2.71
0.106
Factor II G20210A mutation
1.24
0.70–2.18
0.458
1.00
0.38–2.19
1.000
HR indicates hazard ratio.
*Median from 1000 resamples.
†Computed by P⫽2 min 关f (HRⱕ1), f (HRⱖ1)兴, where f(HRⱕ1) denotes the relative frequency of an HRⱕ1 among the 1000 resamples.
‡From the bootstrap zero-corrected multivariable model excluding peak thrombin.
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
drawn upward to the points to determine how many points a
patient will receive for sex, location of VTE, and level of
D-dimer. The sum of the points for each predictor must be
located on the total point axis. By drawing a straight line
downward, the patient’s cumulative recurrence rate after 1
and 5 years can be found.
Discussion
In the present analysis of patients with a first unprovoked
VTE, the overall recurrence risk was high. VTE recurred in
approximately one fourth of the patients within 5 years after
discontinuation of anticoagulation, and these data are in good
agreement with findings from other cohorts.21–26 Although
recurrence rates were slightly higher when patients with
isolated calf vein thrombosis were excluded,27 the recurrence
risk of patients with unprovoked calf vein thrombosis has
been reported to be as high as 25% 6 years after discontinuation of anticoagulation.28 According to current guidelines,
long-term anticoagulant treatment is recommended for all
patients with unprovoked VTE except those with isolated calf
vein thrombosis, in whom treatment for 3 months is sug-
gested to be sufficient.2 The level of evidence for the
recommendation relative to the latter group, however, is low.
Because at least some of these patients could have a higher
risk of recurrence and might therefore benefit from extended
anticoagulation, we decided not to restrict the present analysis
to patients with proximal deep vein thrombosis or pulmonary
embolism only; however, we did not include patients with
VTE related to a transient, reversible risk factor such as
surgery, trauma, or female hormone intake, because these
patients have a well-established low risk of recurrence and
should receive anticoagulation for only 3 months.2
One must keep in mind that despite an overall high
recurrence risk, the majority of patients with a first unprovoked VTE will stay recurrence free for many years and will
be exposed to a considerable risk of bleeding when the
aforementioned guidelines are applied. Therefore, identifica-
Table 3. Multivariable Cox Regression Models Including
D-Dimer (A) and Peak Thrombin Levels (B)
Variable
Hazard
Ratio
95% CI
P
1.90
1.31–2.75
⬍0.001
Model A
Male vs female sex
Pulmonary embolism vs distal thrombosis
2.60
1.49–4.53
⬍0.001
Proximal vs distal thrombosis
2.08
1.16–3.74
0.01
D-dimer (per doubling)
1.27
1.08–1.51
0.005
Model B
Variable
Male vs female sex
2.05
1.36–3.09
⬍0.001
Pulmonary embolism vs distal thrombosis
2.32
1.32–4.09
0.004
Proximal vs distal thrombosis
1.88
1.03–3.44
0.04
Peak thrombin (per 100 nmol/L)
1.38
1.17–1.63
⬍0.001
Figure 2. Receiver operating characteristic curves based on
cross-validated risk scores from the model that included sex,
localization, and D-dimer for recurrence up to 12 months (solid
line) and 60 months (dashed line). AUC indicates area under the
curve.
1634
Circulation
Points
0
April 13, 2010
10
20
30
40
50
60
70
80
90
100
male
Sex
female
proximal DVT
Location
distal DVT
pulmonary embolism
D-Dimer (µg/l)
100
150
200
250
400
500
750
1000
1500
2000
Total Points
0
50
100
150
200
250
300
12 months cumulative recurrence rate
0.02
0.04
0.06
0.08
0.1 0.12 0.15
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
60 months cumulative recurrence rate
0.1
0.2
0.3
0.4
95% confidence intervals (CI) for cumulative recurrence rates
Time point
Recurrence Rate
12 months
2%
1.1 - 3.7
4%
2.6 - 6.2
6%
4.0 - 9.0
60 months
95% CI
8%
5.7 - 11.0
10%
7.3 - 14.0
12%
8.4 - 17.0
15%
9.7 - 23.0
10%
5.8 - 17
20%
14 - 29
30%
24 - 37
40%
28 - 55
50%
35 - 68
tion of patients with unprovoked VTE in whom the risk of
recurrence is low enough to consider short-term anticoagulation is a major goal in the management of these patients.
It was therefore our aim to develop a simple risk assessment model in patients with unprovoked deep vein thrombosis or pulmonary embolism. In our model, the combination of
location of the initial VTE event and the patient’s sex
together with D-dimer or peak thrombin levels was well
suited to discriminate between low- and high-risk patients.
The predictive ability of the model was comparable whether
D-dimer or peak thrombin levels were used. Because D-dimer
is a well-standardized, widely established parameter that can
be easily applied in daily practice, we further developed our
risk model on the basis of this parameter.
We computed a nomogram to calculate risk scores and to
estimate the cumulative recurrence rate in an individual
patient. Examples of how to apply the nomogram are given in
0.5
350
Figure 3. Nomogram to compute shrunken
risk scores and estimate cumulative recurrence rates of recurrent VTE by use of sex,
location of VTE, and D-dimer. How to use
the nomogram: For each value of sex, location of VTE, and D-dimer, read the “points”
at the very top of the nomogram. Add up
the points assigned to each variable to
obtain total points. Finally, map total points
to expected cumulative proportion recurrence-free at 12 and 60 months. As an
example, a female patient with distal deep
vein thrombosis and a D-dimer level of 237
␮g/L would be given 0 points for female
sex, 0 points for distal thrombosis, and 28
points for the D-dimer level. The total number of points adds up to 28 and translates
into cumulative recurrence rates after 12
and 60 months of approximately 1.5% and
7.5%, respectively. The patient was free of
recurrence after 12 years. Another male
patient had an initial pulmonary embolism and
a D-dimer level of 743 ␮g/L. His score yields
217 points, which corresponds to an
expected cumulative recurrence rate of 7.9%
after 1 year and 28.5% after 5 years. The
patient had a recurrence after 14 months.
the legend to Figure 3. D-dimer was used as a continuous
variable, which, in contrast with the use of a dichotomized
variable, allows for more variability with regard to different
levels. Points for the risk model were based on the regression
coefficients obtained from our final model and reflect the
impact of the respective variable on the recurrence risk.
Scoring systems are widely applied for risk prediction in many
diseases and most often use single-digit numbers to simplify
calculation. In contrast, a larger scale (1 to 100 points) is used in
our model to avoid loss of precision and to better relate the
continuous D-dimer levels to the risk estimates.
Our model has undergone an extensive cross-validation
process. We divided the present study cohort into test and
validation samples, thereby mimicking independent validation. This process was repeated 1000 times, and the results
were averaged to avoid dependence of the validation results
on a particular partition of the study cohort. Patients were
Eichinger et al
Risk Assessment Model to Predict Recurrent VTE
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
assigned to different risk categories according to their risk
score. When we calculated cumulative recurrence rates for
patients within quartiles of the risk score, these risk categories
corresponded well with the recurrence rate, because patients
with lower scores had lower recurrence rates. Our risk model
could now be applied to independent cohorts to evaluate its
validity in patients with different thrombotic risk profiles.
The model was particularly well suited to identify patients
at low risk of recurrence. Women, patients with isolated calf
deep vein thrombosis, and patients with low levels of
D-dimer all gained lower scores, which indicates a lower
probability of recurrence. In patients within the lowest
quartile of the risk score, for instance, the annual recurrence
rate was as low as 1.9%. The model does not predict,
however, whether a single patient will have recurrence or not,
because this is influenced by a large variety of genetic and
acquired factors, some of which are still unknown.
Predicting the risk of recurrence by a combination of risk
factors has thus far been attempted in only 1 other study.27
Compared with the present study, the number of patients and
recurrences were lower in that study, and the mean observation time was only 18 months. No combination among the 69
potential predictors of recurrence satisfied the authors’ criteria of a low-risk group (⬍3% per year) in male patients with
unprovoked VTE. In women, however, a low-risk group with
an annual recurrence risk of 1.6% (95% CI 0.3 to 4.6) could
be identified when none or only 1 feature (signs of postthrombotic syndrome, D-dimer level ⱖ250 ␮g/L, body mass index
ⱖ30 kg/m2, or age ⱖ65 years) was present.
Some limitations of the present study need to be addressed.
Patients with strong thrombophilic factors, including a natural
inhibitor deficiency, a lupus anticoagulant, or double heterozygous or homozygous carriers of factor V Leiden or the
prothrombin mutation, were not included because they received long-term anticoagulation after the first venous thrombosis. Our model, therefore, cannot be applied to these
patients. Clinical and laboratory variables that were used to
develop the risk model have been preselected on the basis of
their established relevance for recurrence risk and solid
methodology for assessment. It would stand to reason that if
other determinants of the risk of recurrence, such as high
clotting factor levels or residual vein thrombosis, were
included in the model, risk prediction could improve. It was
our aim, however, to develop a simple model that could be
easily applied in routine care. Although we developed the
model within one of the largest data sets of patients with
venous thrombosis, the prediction of recurrence could be
improved by increasing the number of patients. As pointed
out above, variables to estimate recurrence rates in the present
study were selected on the basis of practical considerations
with regard to methodology and general availability of a
specific risk factor, as well as statistical considerations, ie,
only variables were selected for which we could statistically
prove an association with recurrence rates. With a larger data
set, the set of variables could be selected with less uncertainty, thus providing more precise estimates of recurrence
probabilities for individuals. D-dimer was measured only
once shortly after discontinuation of anticoagulation. It remains to be investigated whether prediction of recurrence can
1635
be improved by serial D-dimer measurements during followup. D-dimer values were missing in 97 of 929 patients;
however, clinical and laboratory characteristics did not differ
between patients with and without D-dimer values (data not
shown).
Prediction of the risk of recurrence is important to determine the optimal duration of anticoagulation. By use of a
simple, easy to apply scoring system, the assessment of the
recurrence risk in patients with a first unprovoked VTE
without strong thrombophilic defects can be improved. Anticoagulants are very effective in preventing recurrent VTE,
but their duration is limited by the risk of bleeding. By use of
our risk assessment model, identification of those patients
with unprovoked VTE in whom the recurrence risk is low
enough to consider a limited duration of anticoagulation can
be achieved. The model should undergo external validation
before it is applied in routine clinical practice. Our prediction
model could be used to stratify patients according to their
recurrence risk in randomized clinical trials that investigate
the optimal duration of anticoagulation in patients with
unprovoked deep vein thrombosis or pulmonary embolism.
Sources of Funding
This study was supported by the Oesterreichische Nationalbank
(Jubiläumsfonds), the Medizinisch-Wissenschaftlicher Fonds des
Bürgermeisters der Bundeshauptstadt Wien, and the Wiener
Städtische Versicherung. The sponsors had no role in the design and
conduct of the study; in the collection, management, analysis, or
interpretation of the data; or in the preparation, review, or approval
of the manuscript.
Disclosures
None.
References
1. Douketis JD, Gu CS, Schuman S, Ghirarduzzi A, Pengo V, Prandoni P.
The risk for fatal pulmonary embolism after discontinuing anticoagulant
therapy for venous thromboembolism. Ann Intern Med. 2007;147:
766 –774.
2. Kearon C, Kahn SR, Agnelli G, Goldhaber S, Raskob GE, Comerota AJ;
American College of Chest Physicians. Antithrombotic therapy for
venous thromboembolic disease: American College of Chest Physicians
Evidence-Based Clinical Practice Guidelines (8th Edition). Chest. 2008;
133:454S–545S.
3. Zhu T, Martinez I, Emmerich J. Venous thromboembolism: risk factors
for recurrence. Arterioscler Thromb Vasc Biol. 2009;29:298 –310.
4. Verhovsek M, Douketis JD, Yi Q, Shrivastava S, Tait RC, Baglin T, Polli
D, Lim W. Systematic review: D-dimer to predict recurrent disease after
stopping anticoagulant therapy for unprovoked venous thromboembolism. Ann Intern Med. 2008;149:481– 490.
5. Hron G, Kollars M, Binder BR, Eichinger S, Kyrle PA. Identification of
patients at low risk for recurrent venous thromboembolism by measuring
thrombin generation. JAMA. 2006;296:397– 402.
6. Eichinger S, Hron G, Kollars M, Kyrle PA. Prediction of recurrent venous
thromboembolism by endogenous thrombin potential and D-dimer. Clin
Chem. 2008;54:2042–2048.
7. Tripodi A, Legnani C, Chantarangkul V, Cosmi B, Palareti G, Mannucci
PM. High thrombin generation measured in the presence of thrombomodulin is associated with an increased risk of recurrent venous thromboembolism. J Thromb Haemost. 2008;6:1327–1333.
8. Besser M, Baglin C, Luddington R, van Hylckama Vlieg A, Baglin T.
High rate of unprovoked recurrent venous thrombosis is associated with
high thrombin-generating potential in a prospective cohort study.
J Thromb Haemost. 2008;6:1720 –1725.
9. Tsai AW, Cushman M, Rosamond WD, Heckbert SR, Polak JF, Folsom
AR. Cardiovascular risk factors and venous thromboembolism incidence:
1636
10.
11.
12.
13.
14.
15.
16.
17.
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
18.
19.
20.
Circulation
April 13, 2010
the longitudinal investigation of thromboembolism etiology. Arch Intern
Med. 2002;162:1182–1189.
Pomp ER, Rosendaal FR, Doggen CJ. Smoking increases the risk of
venous thrombosis and acts synergistically with oral contraceptive use.
Am J Hematol. 2008;83:97–102.
Samama MM. An epidemiologic study of risk factors for deep vein
thrombosis in medical outpatients: the Sirius study. Arch Intern Med.
2000;160:3415–3420.
Kyrle PA, Minar E, Bialonczyk C, Hirschl M, Weltermann A, Eichinger
S. The risk of recurrent venous thromboembolism in men and women.
N Engl J Med. 2004;350:2558 –2563.
Schemper M, Smith TL. A note on quantifying follow-up in studies of
failure time. Control Clin Trials. 1996;17:343–346.
Kaplan EL, Meier P. Non-parametric estimation from incomplete observations. J Am Stat Assoc. 1958;53:457– 481.
Austin PC. Using the bootstrap to improve estimation and confidence
intervals for regression coefficients selected using backwards variable
elimination. Stat Med. 2008;27:3286 –3300.
Harrel FE. Regression Modeling Strategies: With Applications to Linear
Models, Logistic Regression and Survival Analysis. New York, NY:
Springer Verlag; 2001.
Moreau T, O’Quigley J, Lellouch J. On Schoenfeld’s approach for testing
the proportional hazards assumption. Biometrika. 1986;73:513–515.
Efron B, Tibshirani RJ. Improvements on cross-validation: the
632⫹bootstrap method. J Am Stat Assoc. 1997;92:548 –560.
Harrel FE, Lee KL, Mark DB. Multivariable prognostic models: issues in
developing models, evaluating assumptions and adequacy, and measuring
and reducing errors. Stat Med. 1996;15:261–387.
Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for
censored survival data and a diagnostic marker. Biometrics. 2000;56:
337–344.
21. Beyth RJ, Cohen AM, Landefeld CS. Long-term outcomes of deep-vein
thrombosis. Arch Intern Med. 1995;155:1031–1037.
22. White RH, Zhou H, Romano PS. Length of hospital stay for treatment of
deep venous thrombosis and the incidence of recurrent thromboembolism.
Arch Intern Med. 1998;158:1005–1010.
23. Heit JA, Silverstein MD, Mohr DN, Petterson TM, Lohse CM, O’Fallon
WM, Melton LJ III. Predictors of recurrence after deep vein thrombosis
and pulmonary embolism: a population-based cohort study. Arch Intern
Med. 2000;160:761–768.
24. Hansson PO, Sörbo J, Eriksson H. Recurrent venous thromboembolism
after deep vein thrombosis: incidence and risk factors. Arch Intern Med.
2000;160:769 –774.
25. Schulman S, Lindmarker P, Holmström M, Lärfars G, Carlsson A, Nicol
P, Svensson E, Ljungberg B, Viering S, Nordlander S, Leijd B, Jahed K,
Hjorth M, Linder O, Beckman M. Post-thrombotic syndrome, recurrence
and death 10 years after the first episode of venous thromboembolism
treated with warfarin for 6 weeks or 6 months. J Thromb Haemost.
2006;4:734 –742.
26. Prandoni P, Noventa F, Ghirarduzzi A, Pengo V, Bernardi E, Pesavento
R, Lotti M, Tormene D, Simioni P, Pagnan A. The risk of recurrent
venous thromboembolism after discontinuing anticoagulation in patients
with acute proximal deep vein thrombosis or pulmonary embolism: a
prospective cohort study in 1626 patients. Haematologica. 2007;92:
199 –205.
27. Rodger MA, Kahn SR, Wells PS, Anderson DA, Chagnon I, Le Gal G,
Solymoss S, Crowther M, Perrier A, White R, Vickars L, Ramsay T,
Betancourt MT, Kovacs MJ. Identifying unprovoked thromboembolism
patients at low risk for recurrence who can discontinue anticoagulant
therapy. CMAJ. 2008;179:417– 426.
28. Schulman S. The effect of the duration of anticoagulation and other risk
factors on the recurrence of venous thromboembolisms: Duration of
Anticoagulation Study Group. Wien Med Wochenschr. 1999;149:66 – 69.
CLINICAL PERSPECTIVE
Venous thromboembolism (VTE) is a common disease with a high recurrence risk that is particularly high among patients
in whom VTE occurred in the absence of a triggering factor. These patients are candidates for indefinite anticoagulation.
Despite the high recurrence risk, the majority of patients will remain recurrence free for many years and will unnecessarily
be exposed to a bleeding risk with treatment. Therefore, identification of patients in whom the risk of recurrence is low
enough to consider short-term anticoagulation is a major goal in patient management. It was our aim to develop a simple
risk assessment model in patients with unprovoked VTE. In a prospective cohort study, 929 patients were followed up for
a median of 43.3 months after discontinuation of anticoagulation. A total of 176 patients (18.9%) had recurrent VTE.
Preselected clinical and laboratory variables (age, sex, location of VTE, body mass index, factor V Leiden, prothrombin
G20210A mutation, D-dimer, and in vitro thrombin generation) were analyzed, and those that were significantly associated
with recurrence were used to compute risk scores. The combination of location of the initial VTE event and the patient’s
sex together with D-dimer levels was well suited to discriminate between low- and high-risk patients. With these variables,
we developed a nomogram that can be used to calculate risk scores and to estimate the cumulative probability of recurrence
in an individual patient. In conclusion, by use of a simple scoring system, the assessment of the recurrence risk in patients
with a first unprovoked VTE can be improved.
Go to http://cme.ahajournals.org to take the CME quiz for this article.
Risk Assessment of Recurrence in Patients With Unprovoked Deep Vein Thrombosis or
Pulmonary Embolism: The Vienna Prediction Model
Sabine Eichinger, Georg Heinze, Lisanne M. Jandeck and Paul A. Kyrle
Downloaded from http://circ.ahajournals.org/ by guest on June 15, 2017
Circulation. 2010;121:1630-1636; originally published online March 29, 2010;
doi: 10.1161/CIRCULATIONAHA.109.925214
Circulation is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231
Copyright © 2010 American Heart Association, Inc. All rights reserved.
Print ISSN: 0009-7322. Online ISSN: 1524-4539
The online version of this article, along with updated information and services, is located on the
World Wide Web at:
http://circ.ahajournals.org/content/121/14/1630
Data Supplement (unedited) at:
http://circ.ahajournals.org/content/suppl/2010/03/29/CIRCULATIONAHA.109.925214.DC1
Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally published
in Circulation can be obtained via RightsLink, a service of the Copyright Clearance Center, not the Editorial
Office. Once the online version of the published article for which permission is being requested is located,
click Request Permissions in the middle column of the Web page under Services. Further information about
this process is available in the Permissions and Rights Question and Answer document.
Reprints: Information about reprints can be found online at:
http://www.lww.com/reprints
Subscriptions: Information about subscribing to Circulation is online at:
http://circ.ahajournals.org//subscriptions/