|
|||
|
|
|
|
|
|
|
|
Abstract
Background: The applicability of a prediction model for suspected
deep vein thrombosis (DVT) that utilizes a clinical scoring
system in addition to a D-dimer measurement (CSD
model) has been evaluated in the emergency department
(ED) in only one study. This study also had outpatients
referred during the off hours of their vascular lab
and did not perform compression ultrasound (CUS) on
all patients.
Methods:In order to make our study applicable to clinical practice
in a real world ED setting the only inclusion criteria was
that the treating ED clinician felt the patient needed
evaluation for a DVT. All patients were assigned a clinical
score based on a previously published scoring system,
and a quantitative D-dimer and compression ultrasound
were obtained. Presence or absence of DVT was based
on ultrasound results as well as a three-month clinical
follow up.
Results: We prospectively studied 115 patients in our ED. We
found a prevalence of DVT of 34% (95% CI 25-43%). The
sensitivity of the D-dimer assay for DVT was 90% (compared
to 83-98% in other studies) and the sensitivity of
the clinical scoring system for DVT was 85% (compared
to 61-92% for other studies). However, the sensitivity of
the CSD model in our study was only 95% (compared to
97-100% for other studies) and two of the 18 ED patients
in the low clinical score and negative D-dimer subgroup
were found to have a DVT.
Conclusion:Application of the CSD model was not adequate to rule out DVT in our emergency patient population.
Introduction
Patients with suspected deep venous thrombosis (DVT)
present a challenge to emergency physicians. The availability
of compression ultrasound (CUS) testing during off hours is limited in many clinical settings. Common
practice is to consider a dose of low molecular weight
heparin in this situation. The bleeding risks from this
single dose of heparin are difficult to quantify, but are
likely very small.1,2
Recently there has been interest in a risk stratification
model for patients with suspected DVT based on a clinical
scoring system3 (Table 1) and the results of D-dimer
measurements (called the clinical score D-dimer or CSD
model throughout the paper). Although there have been
several studies published on this issue,3-11 the applicability
of this approach to emergency department (ED)
patients remains an open question. Of these studies only
Anderson, et al.4 recruited patients from the ED. However,
some of the patients in this study were referred to the ED
by their primary care clinicians after hours specifically for
evaluation of a possible DVT, and so this may have biased
their results.
We have evaluated patients in a real world emergency
department setting by prospectively recruiting patients
who presented to our ED and by having the only inclusion
criteria be that the treating ED clinician felt the patient
needed evaluation for DVT.
Methods
All patients who presented to the ED at our hospital
whose treating clinician felt needed evaluation for DVT
were eligible for the study. Exclusion criteria included
clinical suspicion of pulmonary emboli (PE), age less
than 18 years old, active therapeutic anticoagulation, or
refusal or inability to give informed consent. This was a
prospective observational study and was approved by our
hospital IRB.
Each patient had a clinical score assigned by their treating
clinician (see Table 1) and a D-dimer sent. D-dimer
assays were done in our central lab using the quantitative
enzyme immunoassay sandwich method with a final florescent
detection by VIDAS. Either an immediate CUS (if
during normal operating hours) or a delayed CUS within
12 hours was performed in our vascular lab. CUS was done using the Advanced Technology Laboratory (ATL
now owned by Phillips) HDI 5000 ultrasound machine
using L7-4 or C5-2 transducer or the Acuson Sequoia
ultrasound machine using either the 6LE or the 4VI transducer.
Our vascular lab protocol includes evaluation of
the common femoral, profunda femoral, superficial femoral,
popliteal, posterior tibial, peroneal, greater saphenous
and the lesser saphenous veins. Patients whose
CUS was delayed 12 hours were treated
at the discretion of the treating clinician,
typically with a subcutaneous dose of low
molecular weight heparin.
The prevalence of many diseases is different
in patients who present to the ED
versus those who present to their primary
care providers. To control for this possibility
we also recruited patients referred
directly to our vascular lab (VL group).
These patients had their clinical score
given to them by the vascular lab technician
before the ultrasound was done. A
D-dimer was run in our central lab. All the
vascular lab technicians were trained in
how to apply the clinical scoring model by
one of the authors of this study (JH). Ten patients had a clinical score assigned by the vascular lab
technician and by one of the authors of this study (JH)
independently, and in all 10 cases the assigned scores
were the same.
All patients gave informed consent and were followed
clinically for three months.
Results
Over the course of the study 122 patients were recruited
from our ED, and a convenience sample of 46 patients
was recruited in the vascular lab. From the ED population
only one patient refused consent, but two were felt to
be too demented by the treating clinician to give proper
informed consent. Two patients could not have their Ddimer
done due to technical reasons in the lab (and had
already left the ED so the samples could not be redrawn),
and so were dropped from the study. Two patients had
therapeutic levels on coumadin for other clinical indications,
and so were not eligible for the study. Of the 115
remaining patients recruited in the ED, two were pregnant
and were found not to have a DVT on evaluation (we did
not exclude pregnant patients). Patients whose treating
clinician felt needed evaluation for PE were not recruited
in the study and records on these patients were not kept.
Tables 2 and 3 summarize the results of our study. One
patient in the ED group with moderate clinical probability
and a positive D-dimer had an initial CUS that was
negative, but was found to have a DVT on follow up. One
patient in the VL group with a low clinical probability and
a positive D-dimer had an initial CUS that was negative
but was found to have a DVT on follow-up. There was
one death in the study group, which was in a patient with moderate clinical probability and negative D-dimer who
had a negative DVT on initial CUS and who died from
complications of their cancer not related to a thromboembolic
event.
For the ED patients the overall results showed that 42%
of patients had a low clinical probability, 32% had moderate
clinical probability and 26% had a high clinical
probability for DVT. Of patients with a negative D-dimer,
two of 18 patients with low clinical probability, one of
seven with moderate clinical probability and one of four
with high clinical probability had DVT. Of patients with
positive D-dimers four of 30 with low, 16 of 30 with moderate
and 15 of 25 with high clinical probability had DVT.
The ED patients had a DVT prevalence of 34% (95% CI
25-43%). The D-dimer sensitivity in these patients was
90% (95% CI 76-97%), and the sensitivity of a moderate or
high clinical score was 85% (95% CI 69-94%).
Analysis of the VL patients showed an overall prevalence
of 15%. Details for these patients are presented in Table
2. Historically our vascular lab had a 16% positive rate for
the CUS’s done to evaluate for DVT and so the present
study was consistent with this historical number, hence
comparison of disease prevalence between the ED and
outpatient populations was possible. The small number
of vascular lab patients limits further analysis of that data,
but for completeness we note that the sensitivity of Ddimer
for these patients was 88%, and the sensitivity of a
moderate or high clinical score for DVT was 91%.
Discussion
Several recent studies have looked at the combination
of a clinical scoring system and the results of a D-dimer
assay in patients being evaluated for DVT (CSD model)
and we found eight4-11 that have compiled data in such
a way as to allow detailed comparison to the present
study.
In four of the eight studies4,5,7,8 CUS was not performed
on patients in the low clinical score and negative D-dimer
subgroup (low-negative subgroup), and any positive
DVT reported in this subgroup was from the three month
clinical follow up. The retrospective study by Bates, et
al6 had CUS done on only some of the patients in the
low-negative subgroup, and again positive DVT in these
patients was from clinical follow up. All the studies
except Anderson, et al4 recruited outpatients for their
study population. Anderson’s study recruited patients
from the ED, but noted that some of these patients
were outpatients referred to the ED after hours. All the
studies except Johanning, et al10 and Lennox, et al11 used the Wells scoring system.3 Johanning, et al used a
clinical score limited to the presence of 2cm swelling in
the affected leg. Lennox, et al utilized a clinical scoring
system that had many aspects similar to the Wells’ scoring
system.
Table 4 summarizes results from these studies and compares
them to the present study. The data presented
includes total number of patients, overall prevalence of
DVT, sensitivity of the D-dimer assay used, sensitivity of
the clinical scoring system using moderate or high scores
as positive (even for studies whose authors noted only
high score should be considered positive we re-calculated
the sensitivity using moderate and high scores as
positive to make comparisons appropriate) and finally
the calculated sensitivity of the CSD model (again always
using moderate or high clinical score to be a positive
result). The last column in Table 4 uses the measured
sensitivities of the D-dimer assay and the clinical score
from each study, assumes these are independent, and
calculates the predicted sensitivity of the combined tests
(as discussed in the cascading example below).
As Baye’s theorem states, the predictive value of any test
is a function not only of the sensitivity and specificity of
the tests employed, but also of the prevalence of the disease
in the population being tested. It is therefore very
important to understand the overall disease prevalence
in the test population before generalizing the results of
one study to a different patient population.
We found a significantly higher prevalence of DVT in our
ED patients of 34% (95% CI 25-43%) than from our vascular
lab patients (15% in the present study and 16% historical
reference for our lab). This difference may reflect
self-selection by patients in presenting to the ED with a
higher likelihood of disease. It may also represent a higher
threshold for evaluation of DVT by the ED physician
than by a primary care clinician. This might be explained
by confounding factors encountered in the primary care
setting that are not seen in the ED, such as longevity of
complaints.
We had two of the total 161 patients (115 plus 46) have
DVT on follow up despite a negative initial CUS. This is
consistent with the published literature that shows about
1% of patients with an initial negative comprehensive
CUS have DVT on follow up.3,12
The specifics of the D-dimer test used also contribute
to results. Stein, et al13 reviewed the performance of
various D-dimer assays. Our lab utilizes a quantitative
D-dimer with a cut-off of 500ng/mL to define positive.
According to the manufacturer of this test (bioMerieux),
it is the only test in the United States approved by the FDA to exclude DVT in outpatients. Our study found the
D-dimer test to be 90% sensitive, within the 90-100%
sensitivity range noted for quantitative rapid ELISA Ddimer
test applied to DVT quoted in Stein, et al identical
to the 90% sensitivity found by Larsen14 for the same
VIDAS test and well within the range of 83-98% noted in
the comparison studies. The type of D-dimer assay done
must be taken into account when comparing sensitivities
and specificities. Further analysis of our data shows
that our D-dimer testing was 33% specific, consistent
with data published in Stein, et al (32-55%).
Five of the eight comparison studies4-9 did not perform
CUS on all patients in the low-negative subgroup, and so the
positives in this subgroup were from patients who had
positive results at the subsequent follow-up evaluation.
Some patients with DVT will have the condition
spontaneously resolve without treatment (see the discussion
below). Therefore, some patients with DVT may
have been missed in these studies, affecting the overall
results.
Kim and Kim15 followed 72 of 200 patients who developed
DVT after hip replacement. None of the patients
were treated, and all DVT resolved with no adverse
sequelae. Wang, et al16 followed 48 untreated patients
with DVT after total knee replacement and at follow up only one patient had residual thrombus, and no patients
suffered any complications from their untreated DVT. Both
these studies were in Asian patients, and the incidence in
Western patients is thought to be higher. Studies have
shown a prevalence of 70% of DVT in untreated patients
status post total hip replacement (THR), with a 1% incidence
of fatal pulmonary embolism (see discussion in
Kim, et al). Although this discussion centers on DVT after
orthopedic surgery, it illustrates the concept that many
patients with DVT may have resolution of this condition
even without treatment.
Patient outcome is certainly more important than the
presence or absence of DVT on CUS. The studies that did
not perform CUS on the low-negative subgroup3-9 did
follow all their patients for three months, and no adverse
clinical outcomes were noted. It may be that patients in
the low-negative subgroup have good clinical outcomes
even if they do have a DVT.
The study by Anderson4 utilized CUS to diagnose DVT
except in the low-negative subgroup, but intentionally
did not look for clots in the calf. As many as 25% of
untreated calf thrombi extend to proximal veins within a
week after presentation17.
All eight comparison studies4-11 concluded that patients in the low-negative subgroup did not need CUS and
could safely be followed clinically. The sensitivity of the
CSD model for these studies ranged from 97-100%, higher
than the 95% result from our study. Most of these studies
did not perform CUS in the low-negative subgroup, but
no bad outcomes occurred on three-month follow up.
With this limitation noted, the combined data from the
eight comparison studies show a total of seven of 868
patients in this subgroup (0.8%) had a DVT. Interestingly,
five of these seven were from the ED patients recruited by
Anderson4 (where 5 of 311 patients in the low-negative
subgroup had DVT). Their study had only a 97% sensitivity
for the CSD model, the lowest of the eight comparison
studies. They did not do CUS on this subgroup and positives
were only from clinical follow up.
Suppose we have a test with 90% sensitivity and we run
it on 100 patients with a given disease. By definition 10
of these patients would have a false negative result. Now
say we have yet another test with independent probability
from the first test but with 80% sensitivity and we
run it on the 10 patients with a negative result of the first
test. We would expect eight of these to have a positive
result, and so only two of the 100 initial patients with the
disease would be expected to be missed using this cascading
strategy, yielding an overall 98% sensitivity. If the
predictive values of the two tests were not independent,
the combined result would not be this high.
This very simplified example may not apply to the
present CSD model strategy. First, the two tests (clinical
score and D-dimer assay) may not be independent.
Furthermore, the sensitivity for the D-dimer result is for
all patients, weight averaging the D-dimer sensitivity for
each of the low, moderate and high clinical score subgroups.
It might be that the sensitivity of the D-dimer
assay is higher for the low clinical score group than it is on
average, and hence we might expect a result higher than
that for the simplified independent probability model
discussed above. Calculation of the predicted sensitivities
for each of the given studies assuming independent
probabilities of the clinical scoring system and the Ddimer
assay ranges from 98-99.9% (see Table 4).
Conclusion
A higher prevalence of certain diseases in patients who
present to the ED as compared to other outpatients is
not a new concept, and therefore before any evaluation
algorithm is applied to an ED population it is important
to test this algorithm specifically in ED patients. Our
study showed a prevalence of 34% in our ED patients,
higher than the 15% in the small control group recruited
in our vascular lab and higher than the historical 16%
prevalence from our vascular lab. Despite a D-dimer
sensitivity of 90% and an 85% sensitivity of the clinical
scoring model (using moderate or high scores as positive)
we found that two of 18 patients in the low-negative
subgroup had DVT. Based on the results of our study we
would caution against using the CSD model to obviate
the need for compression ultrasound in ED patients being
evaluated for DVT.
The tremendous variation in overall disease prevalence,
sensitivity of the clinical score and sensitivity of D-dimer
for DVT between the various studies analyzed is of concern.
As noted in the discussion above, if the group with
the lowest clinical score sensitivity had also had the lowest
D-dimer sensitivity, the sensitivity of the CSD model
would be expected to be only 93% or lower. The only
study to have looked at ED patients4 (although with some
percentage of these patients being outpatients referred
to the ED by their primary care physician) had a measured
sensitivity of the CSD model of only 97% (lowest of
the eight studies analyzed), missing five of 311 patients
with DVT. This study did not perform CUS on patients
in the low-negative subgroup so all the positives in this
subgroup were from the three-month follow up. Based
on these facts we would recommend a large multi-center
trial in ED patients before using the CSD model to obviate
the need for compression ultrasound to evaluate DVT in
the ED setting.
Based on the discussions above, at the present time we
would recommend that all patients whose treating ED
clinician feel need evaluation for DVT have CUS. Those
patients whose CUS must be delayed up to 12 hours can
be considered for treatment with a single dose of low
molecular weight heparin at the discretion of the evaluating
clinician.
The authors would like to thank the entire clinical staff of the St.
Vincent’s Hospital ED, Dr.’s Ukena, Josephs, Najjar, and the entire staff
of the vascular lab, without whose tireless efforts this unfunded study
could not have been accomplished.
Return to the Home Page