Keywords pulmonary embolism - COVID-19 - D-dimer - computed tomography - diagnosis
Introduction
Prevalence of pulmonary embolism (PE) in patients with coronavirus disease 2019 (COVID-19)
is high, and PE diagnosis is challenging in this specific population.[1 ]
[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ] COVID-19-associated coagulopathy was described early on, as well as unusually high
D-dimer levels in a large majority of patients.[9 ]
[10 ]
[11 ] The D-dimer level seems to be proportional to the extent of lung damage, and was
also identified as an independent risk factor of in-hospital mortality.[10 ] PE is often suspected in COVID-19 patients with respiratory symptoms, but the use
of D-dimer to rule it out is a questionable strategy because of its increased levels
even in the absence of venous thromboembolism (VTE). Therefore, current validated
thresholds of D-dimer in non-COVID-19 outpatients with suspected PE may not help safely
reduce the number of fruitless computed tomography (CT) pulmonary angiography (CTPA)
in COVID-19 patients. Consequently, the International Society on Thrombosis and Haemostasis
guidelines suggested to directly confirm diagnosis using standard-of-care objective
testing, such as CTPA.[12 ] Most recent algorithms and guidelines aim to reduce imaging testing[13 ]
[14 ]
[15 ]
[16 ]; however, due to the lack of specificity of clinical signs of PE and the difficulty
to perform CTPA in the most severe COVID-19 patients, the application of such guidelines
in current clinical practice is challenging and potentially leads to misdiagnosing
PE or to excessively prescribe therapeutic anticoagulation without objective confirmation
of PE. Therefore, ruling out PE using a specific threshold for D-dimer could be useful
in reducing CTPA prescription, particularly in patients with severe COVID-19,[17 ] and in facilitating the management of imaging examinations for other patients.[18 ]
The purpose of this study was to derive a new diagnostic algorithm for PE with a D-dimer
cut-off value adjusted to the CT extent of lung damage, and to assess its safety for
the exclusion of PE. To determine the D-dimer thresholds, we used a retrospective
multicenter cohort study of COVID-19 patients with suspected PE (derivation set).
We then performed an external validation in an independent multicenter cohort of COVID-19
patients with suspected PE.
Methods
Design and Setting
This study is a retrospective, multicenter cohort study sponsored by the Groupe Hospitalier
Paris Saint Joseph (GHPSJ). The study design was approved by our institutional ethics
committee (IRB number: IRB 00012157) and registered on the National Institute of Health
data platform (INDS No. MR 4516150520). Patients' nonopposition to the use of their
data for research was also collected in accordance with the European regulation (General
Data Protection Regulation). Study reporting complied with the requirements of the
STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) statement
for observational studies in epidemiology (https://www.strobe-statement.org ).
Derivation Set
Between March 1 and May 8, 2020, all COVID-19 patients with respiratory symptoms who
had a CTPA for suspicion of PE were recorded in a database in two large academic hospitals
in Paris, France: GHPSJ and Hôpital Européen Georges-Pompidou (HEGP).[7 ] Patients from emergency rooms, general wards, or intensive care units (ICUs) were
included if they were over 18 years of age, were admitted for COVID-19 with respiratory
symptoms, and had a CTPA at baseline or during hospitalization for a suspected PE.
Diagnosis of SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) infection
was confirmed by a positive result of a reverse transcriptase polymerase chain reaction
assay or highly suggestive CT findings of COVID-19 pneumonia.[19 ] All the patients included were informed of the research protocol by letter, allowing
them to express their opposition to the use of their data, according to French legislation
and the institutional review board. The exclusion criterion was respiratory distress
syndrome explained by another disease than COVID-19. The decision to perform a CTPA
was not predefined and was left to the discretion of the clinicians in charge of the
patients.[20 ] In most cases, PE was suspected because of dyspnea, acute respiratory failure, tachycardia,
syncope, or respiratory worsening. To apply the European Society of Cardiology (ESC)'s
diagnostic algorithm,[13 ] the clinical probability of PE was calculated retrospectively using the simplified
revised Geneva[21 ] score on the same day of D-dimer collection. The clinical probability of PE was
classified as probable or unlikely, and the D-dimer threshold was adjusted for age.[15 ]
Validation Set
During the initial COVID-19 outbreak, five French academic medical centers associated
with the F-CRIN INNOVTE network (Saint Etienne University Hospital, Besançon University
Hospital, Brest University Hospital, Grenoble-Alpes University Hospital, and Amiens
University Hospital) recorded clinical, radiological, and biological data of all consecutive
COVID-19 patients with respiratory symptoms who had a CTPA for suspicion of PE. All
patients were included between February 12 and September 14, 2020. Diagnostic criteria
for COVID-19 were identical to those of the derivation set. Similarly to the derivation
set, the decision to perform a CTPA was not predefined and was left to the discretion
of the clinicians in charge of the patients. In most cases, PE was suspected because
of acute respiratory symptoms (chest pain, respiratory failure, or worsening respiratory
status).
Laboratory Data
For the derivation set, D-dimer assays were either Vidas D-dimers (Biomérieux, Marcy-l'Etoile,
France) or STA-Liatest D-Di (Diagnostica Stago, Asnières, France). For the validation
set, D-dimer assays were either Vidas D-dimers, STA-Liatest D-Di, or Innovance D-Dimers
(Siemens Healthcare Diagnostics, Marburg, Germany) according to local practice. D-dimer
levels were reported in ng/mL. Analyses were performed on both sets of patients using
the available D-dimer level within 24 hours prior to CTPA. As the D-dimer assays differed
from one center to another, we checked whether their performance differed to exclude
PE in patients with COVID-19. As the Vidas D-dimers is currently the assay with the
best sensitivity and negative predictive value (NPV) in the non-COVID-19 population,[21 ] we used it as the reference assay, like we would have in the general population.
CTPA Analysis
In both derivation and validation sets, all CTPAs were locally reviewed by two radiologists
to determine whether the findings were highly suggestive, indeterminate, or nonsuggestive
of COVID-19. The reading was performed according to the recommendations of the European
Society of Radiology and the European Society of Thoracic Imaging.[19 ] Lung damage extent was classified into two groups: <50% or ≥50%.[7 ] In each center, the same two radiologists locally confirmed or refuted the diagnosis
of PE, and, if present, whether or not it was sub-segmental. Discrepancies were resolved
through discussion with a third radiologist until consensus was reached.
Derivation of the Co-LEAD Algorithm
In the derivation set, we constructed receiver operating characteristic (ROC) curves
to define a D-dimer cut-off value that could safely exclude PE in COVID-19 patients.
We defined the threshold as the highest value associated with a maximum of one false
negative.[22 ]
[23 ] Then, as previously reported,[24 ] we found that D-dimer increased with lung damage extent, regardless of the presence
of PE. Consequently, we constructed two ROC curves, one for patients with lung damage
extent <50% and one for patients with lung damage extent ≥50%. Using the previous
definition allowing for one false negative, we conceived a diagnostic algorithm for
PE combining COVID-19-lung damage extent and adjusted D-dimer (Co-LEAD algorithm),
with a different threshold for each lung damage extent category. We then evaluated
the diagnostic performance of the proposed Co-LEAD algorithm in both derivation and
validation sets using sensitivity (Se), specificity (Sp), positive predictive value
(PPV), NPV, and negative likelihood ratio (NLR) defined as (1 − Se/Sp), the area under
the curve (AUC) of the strategy, number and proportion of patients with a false negative
diagnosis, and number of CTPAs that could have been avoided if the Co-LEAD algorithm
had been applied.
Statistical Analysis
Categorical variables are presented as number of patients (proportion) and quantitative
variables as median with interquartile range (IQR). Ninety-five percent confidence
intervals (95% CIs) of proportions were estimated using either the normal approximation
or the binomial method when proportions were close to 0 or 1. Comparison of proportions
was done, as appropriate, with the Chi-square or the Fisher test, and two means were
compared by a standard t -test or a Mann–Whitney test, as appropriate. The NCSS Statistical software was used
for calculations. A p -value <0.05 was deemed statistically significant. We calculated the corresponding
diagnostic indexes: sensitivity, specificity, PPV, and NPV with 95% CI using the usual
formula. The empirical (nonparametric) method of DeLong et al[25 ] was used to estimate AUC and compare the AUC of the derivation and the validation
sets.
Results
Derivation Set Characteristics
Among 337 COVID-19 patients with CTPA for suspected PE, 70 (20.8%) patients were diagnosed
with PE ([Table 1 ]). The D-dimer level at the time of CTPA was available for 267 (79.2%) patients.
The median D-dimer level was 1,340 ng/mL (IQR: 903–2,370) in patients without PE and
6,435 ng/mL (IQR: 2,935–3,750) in patients with PE (p < 0.001). In patients without PE, the median D-dimer level was 1,269 ng/mL (IQR:
833–2,092) in patients with lung extent damage <50%, and 2,070 ng/mL (IQR: 1,270–3,404)
in patients with lung extent damage ≥50% (p = 0.009). In patients with PE, the median D-dimer level was 6,167 ng/mL (IQR: 2,887–12,940)
in patients with lung extent damage <50% and 7,130 ng/mL (IQR: 3,376–14,500) in patients
with lung extent damage ≥50% (p < 0.001).
Table 1
Clinical characteristics of the COVID-19 patients in the derivation set and the validation
set
Derivation set
Validation set
p -Value[b ]
Total
(n = 337)
Patients with PE
(n = 70)
Patients without PE
(n = 267)
p -Value[a ]
Total
(n = 337)
Patients with PE
(n = 66)
Patients without PE
(n = 271)
Age, y (median, IQR)
66 (55–80)
66 (54–80)
66 (55–79)
0.79
65 (55–74)
65 (55–73)
65 (55–74)
0.052
Male sex, n (%)
207 (61.4)
39 (55.7)
168 (62.9)
0.27
214 (63.0)
49 (75.4%)
165 (61.8)
0.47
Time from illness onset to CTPA, days (median, IQR)
7 (3–10)
7 (4–13)
7 (3–10)
0.22
7 (4–12)
11 (6–15)
7 (3–10)
0.096
D-dimer level, ng/mL (median, IQR)
1,644 (981–3,750)
6,435 (2,935–3,750)
1,340 (903–2,370)
<10−6
1,600 (800–3,765)
5,340 (2,400–13,550)
1,285 (765–2,442)
0.71
BMI, kg/m2 (median, IQR)
26.3 (23.6–29.9)
29.7 (27.0–30.7)
26.1 (23.5–29.8)
0.057
27.5 (24.6–32.0)
27.5 (25.0–31.6)
27.4 (24.5–32.0)
0.003
Sub-segmental PE, n (%)
13 (18.6%)
10 (15.2%)
0.59
Risk factors for VTE
Previous VTE, n (%)
24 (7.2)
5 (7.1)
19 (7.3)
0.97
27 (8.0)
5 (7.7)
22 (8.1)
0.69
Active cancer, n (%)
32 (9.6)
5 (7.2)
27 (10.2)
0.45
53 (15.6)
9 (13,8)
44 (16.2)
0.017
Lung damage extent ≥ 50%[c ], n (%)
66 (26.1)
28 (40.0)
38 (22.5)
0.003
97 (29.0)
24 (36)
73 (27)
0.34
Abbreviations: BMI, body mass index; CTPA, computed tomography pulmonary angiography;
IQR, interquartile range; PE, pulmonary embolism; VTE: venous thromboembolism.
a Comparison between patients with PE and without PE in the derivation set.
b Comparison between all patients of the derivation set and the validation set.
c Evaluated at CT scan.
ESC Diagnostic Algorithm Performances in the Derivation Set
Fifty-eight (21.7%) Patients were classified as likely to have a PE, whereas 209 (78.3%)
patients were classified as unlikely and D-dimer testing was recommended before CTPA
to exclude PE ([Table 2 ]). Considering all the variables of the Geneva revised simplified score, only two
items differed between patients with and without PE: unilateral lower limb pain (10
[17.5%] vs. 8 [4.0%], p = 0.003) and pain on lower limb deep vein palpation and unilateral edema (11 [19.3%]
vs. 4 [2.0%], p < 0.0001). Applied to our derivation set, this algorithm had the following performances:
Se 100% (95% CI: 93.6–100.0), Sp 9.9% (95% CI: 6.3–14.8), PPV 22.8 (95% CI: 17.7–28.6),
and NPV 100.0 (95% CI: 83.8–100.0). Thus, if this algorithm was used, 246 CTPAs would
have been required. This diagnostic algorithm did not result in any false-negative
cases.
Table 2
Diagnostic performance of the different strategy
Derivation set
(N = 267)
ESC guidelines
Clinical probability[a ] then D-dimer <900 ng/mL
Clinical probability[a ] then Co-LEAD
D-dimer <900 ng/mL
Co-LEAD
Number of patients in whom PE can be excluded
21
48
55
53
61
Se
100%
98% (95–100)
98% (95–102)
98% (95–100)
98% (95–100)
Sp
10% (6–14)
22% (17–28)
26% (20–31)
25% (19–30)
28% (22–35)
PPV
23% (18–28)
25% (19–31)
26% (20–32)
26% (20–32)
27% (21–33)
NPV
100%
98% (94–100)
98% (95–100)
98% (94–100)
98% (95–100)
NLR
0.0
0.08
0.07
0.07
0.06
Efficiency
246
219
212
215
206
Efficacy
3.5
3.2
3.1
3.1
3.0
Number of false negative
0
1
1
1
1
Abbreviations: NLR, negative likelihood ratio; NPV, negative predictive value; PPV,
positive predictive value; Se, sensitivity; Sp, specificity.
Note: Efficiency: number of CTPA required; efficacy: number of CTPA scans needed to
diagnose one pulmonary embolism.
a Clinical probability has been retrospectively assessed by the simplified revised
Geneva Score.
Determination and Performances of D-dimer Threshold to Exclude PE in COVID-19 Patients
In the derivation set we obtained from the ROC curve a threshold of 900 ng/mL for
the exclusion of PE ([Supplementary Fig. S1 ] [available in the online version]). The strategy relying on the use of this specific
threshold in all the patients of the derivation set without previous estimation of
clinical probability had the following performance: Se 98.2% (95% CI: 94.7–100.0),
Sp 24.2% (95% CI: 18.4–29.9), PPV 25.6% (95% CI: 19.7–31.4), and NPV 98.1% (95% CI:
94.3–100.0). Among the 267 patients with a D-dimer value available at the time of
CTPA, 52 had D-dimer level <900 ng/mL. If this strategy was used, we would have performed
215 CTPAs. The strategy relying on using this specific threshold in the 209 patients
with PE considered as unlikely had the following performance: Se 98.2% (95% CI: 83.9–100.0),
Sp 22.3% (95% CI: 16.8–28.5), PPV 25.1% (95% CI: 19.5–31.4), and NPV 97.9% (95% CI:
88.9–99.9). If this strategy was used, we would have performed 219 CTPAs.
D-dimer Assays' Performance According to the Lung Damage Extent
We then made sure the diagnostic performance of the D-dimer assays was similar in
the two groups of lung damage extent (<50% vs. ≥50%) and that the differences in AUCs
were not statistically significant (0.80, 95% CI: 0.66–0.89 vs. 0.87, 95% CI: 0.77–0.92,
p = 0.34, [Supplementary Fig. S2 ] [available in the online version]). We obtained from the ROC curves a threshold
of D-dimer for each subgroup (≥50% and <50%): 1,700 ng/mL in the subgroup ≥50% and
900 ng/mL in the subgroup <50%. The strategy relying on the use of these specific
thresholds in the derivation set without previous estimation of clinical probability
([Supplementary Fig. S3 ] [available in the online version]) had the following performance: Se 98.2% (95%
CI: 94.7–100.0), Sp 28.4% (95% CI: 24.1–32.3), PPV 26.7% (95% CI: 20.7–32.7), and
NPV 98.4% (95% CI: 95.2–100.0). Among the 267 patients with a D-dimer value available
at the time of CTPA, 52 had D-dimer level <900 ng/mL. If this strategy was used, we
would have performed 206 CTPAs. The strategy relying on the use of these two specific
thresholds in the 209 patients with PE considered as unlikely had the following performance:
Se 98.2% (95% CI: 83.9–100.0), Sp 25.6% (95% CI: 19.8–32.0), PPV 25.9% (95% CI: 20.2–32.4),
and NPV 98.2% (95% CI: 90.3–99.9). If this strategy was used, we would have performed
212 CTPAs.
Derivation and Performance of the Co-LEAD Algorithm to Exclude PE in COVID-19 Patients
Using the previous definition allowing one false negative for the exclusion of PE,
we conceived a diagnostic algorithm for PE combining lung extent and D-dimer-adjusted
threshold (Co-LEAD, [Fig. 1 ]). The D-dimer threshold that safely excluded PE was 900 ng/mL in patients with lung
damage extent <50% and 1,700 ng/mL in patients with lung damage extent ≥50%. The Co-LEAD
algorithm entails that patients with a D-dimer level <900 ng/mL do not require CTPA,
whereas those with a D-dimer level ≥1,700 ng/mL require CTPA regardless of the disease
extent. When the D-dimer level is between 900 and 1,700 ng/mL, patients with lung
damage extent ≥50% on CT do not require contrast injection, whereas those with lung
damage extent <50% need contrast injection to confirm PE. This strategy had the following
performance: Se 98.0% (95% CI: 95.0–100.0), Sp 28.0% (95% CI: 22.0–35.0), PPV 27.0%
(95% CI: 21.0–33.0), and NPV 98.0% (95% CI: 95.0–100.0). With this strategy, we would
have performed 206 CTPAs and 88 patients would have had a two-step CT. The NLR of
this strategy was 0.06 (95% CI: 0.01–0.44). The AUC for the Co-LEAD algorithm was
0.63 (95% CI: 0.60–0.67). As predefined, there was one false negative of the diagnostic
algorithm (1.6%, 95% CI: 0.0–4.7). This patient was a 62-year-old woman with a single
antero-basal segmental PE and a D-dimer level of 650 ng/mL.
Fig. 1 Combined strategy for PE suspicion in COVID-19 patients: Co-LEAD algorithm. PE, pulmonary
embolism.
External Validation of the Co-LEAD Algorithm
In the validation set, 337 COVID-19 patients had a CTPA for suspected PE, which was
confirmed in 66 (19.6%) patients. The prevalence of PE did not differ between both
sets (p = 0.70). Clinical characteristics of patients were similar, except for body mass
index and active cancer ([Table 1 ]). D-dimer levels were available at the time of CTPA for 250 patients (74.2%) in
the validation set, a similar value to that of the derivation set (p = 0.12).
In the validation set, the Co-LEAD algorithm had a Se of 96.7% (95% CI: 88.7–99.6),
a Sp of 39.2% (95% CI: 32.2–46.1), a PPV of 33.9% (95% CI: 26.9–41.0), and a NPV of
97.4% (95% CI: 90.8–99.7) for the exclusion of PE ([Table 2 ]). The NLR was 0.08 (95% CI: 0.02–0.33). The AUC for the Co-LEAD algorithm in the
validation set was 0.68 (95% CI: 0.64–0.72) and did not differ from the AUC in the
derivation set (p = 0.097). There were two false-negative cases (2.6%, 95% CI: 0.0–6.2): a 58-year-old
man with lung damage extent <50%, a D-dimer level of 750 ng/mL, and a segmental PE
occurring 10 days after COVID-19 onset; and a 79-year-old man with active cancer,
a lung damage extent <50%, a D-dimer level of 510 ng/mL, and a segmental PE 14 days
after COVID-19 onset. No fatal PE occurred.
Using the Co-LEAD algorithm, in the 250 patients with a D-dimer level available at
the time of CTPA, 76 (30.4%) CTPAs would have been avoided, representing a 25.6% increase
of avoided CTPA compared with the aforementioned D-dimer threshold of 900 ng/mL.
Co-LEAD Algorithm Performance According to D-dimer Assays
In the pooled derivation and validation sets, the AUCs of the Vidas D-dimers, STA-Liatest
D-Di, and the Innovance D-Dimer did not differ significantly: 0.84 (95% CI: 0.75–0.90),
0.84 (95% CI: 0.74–0.90), and 0.85 (95% CI: 0.77–0.90) respectively, p = 0.94 and p = 0.93; [Fig. 2 ]). We also evaluated the performance of the Co-LEAD algorithm according to the D-dimer
assay used. As shown in [Table 3 ], the NLR always remained below 0.15 regardless of the assay used.
Fig. 2 Receiver-operating characteristic curves of D-dimer for PE diagnosis in COVID-19-suspected
PE patients according to the D-dimer assay in the whole population (validation and
derivation set). PE, pulmonary embolism.
Table 3
Diagnostic performance of the Co-LEAD strategy according to the D-dimer assay used
by centers
D-dimer assay
Centers
Patients (n )
Number of false negative with the Co-LEAD algorithm
Sensitivity
NPV
NLR
Vidas D-dimer
#1, #2
145
0
100.0
(97.9–100.0)
100.0
(97.9–100.0)
Not calculable
STA-Liatest D-Di
#0, #5, #2
178
1
97.2
(91.9–100.0)[a ]
97.8
(93.6–100.0)
0.09 (0.06–0.17)
Innovance D-Dimer
#3, #4
194
2
96.2
(90.9–100.0)[a ]
96.3
(91.3–100.0)
0.11
(0.02–0.19)
Abbreviations: NPV, negative predictive value; NLR, negative likelihood ratio.
Note: Centers: #0 GHPSJ, #1 HEGP, #2 CHU Brest, #3 CHU Besançon, #4 CHU Amiens, and
#5 CHU Grenoble-Alpes.
a One-sided 97.5% confidence interval.
Discussion
We propose a new diagnostic algorithm for PE in COVID-19 patients with suspected PE,
Co-LEAD, that combines D-dimer values using specific thresholds pending extent of
lung damage. This algorithm has a high sensitivity and a high NPV in both derivation
and validation cohorts, making this algorithm effective to exclude PE. To our knowledge,
this is the first study proposing an algorithm aiming at diagnosing PE in patients
with COVID-19.
PE is a frequent complication of COVID-19[1 ]
[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ] that concerns a large number of patients at different stages of the disease. A high
prevalence of PE was described in critically ill patients.[2 ]
[7 ]
[26 ] However, PE should not be considered as a late complication of COVID-19, but rather
as a specific and potentially early manifestation. Thus, in several reports, half
of PEs were diagnosed within 48 hours of admission.[3 ]
[7 ]
[27 ]
The diagnosis of PE is challenging in COVID-19 patients and the diagnostic algorithms
recommended by international guidelines for the outpatient population[13 ] might not be as useful because patients with COVID-19 have high D-dimer levels even
in the absence of VTE. To date, noncontrast CT has been considered the first-line
imaging tool and has quickly become a cornerstone in both the diagnostic workup and
follow-up of SARS-CoV-2 infection.[19 ] At admission or during follow-up, respiratory symptoms suggestive of PE are frequent
and may lead to contrast injection in almost all COVID-19 patients. It is thus necessary
to derive and validate a specific algorithm dedicated to this population, as it seems
inappropriate to perform CTPA to all COVID-19 patients. First, faced with a sudden
influx of patients, it is necessary to rationalize radiological examinations. Second,
COVID-19 patients seem more likely to present renal impairment,[17 ] and the benefit–risk balance of contrast injection should be carefully weighted.
The Co-LEAD strategy results in an important reduction (16% when compared with ESC
algorithm) in contrast dye injection, which seems relevant to reduce the risk of renal
impairment/failure, a poor prognosis criterion in critically ill COVID-19 patients.[17 ] Third, patient transportation is logistically difficult, particularly for the most
severe cases, and increases the risk of exposing other patients and medical staff
to the virus. The role of D-dimer in COVID-19 is still a matter of debate: some suggest
using D-dimer to suspect or to confirm VTE,[28 ]
[29 ]
[30 ] while others use D-dimer to initiate anticoagulant treatment outside of the usually
recommended practices.[31 ]
[32 ] Indeed, the so-called “super-high” COVID-19 D-dimer level led us to forget that
D-dimer testing is not specific but remains sensitive enough to safely exclude PE.
It was observed that D-dimer is correlated to severity and mortality in COVID-19 patients.[9 ]
[10 ]
[33 ] Moreover, the D-dimer level at hospital admission for COVID-19 is associated with
an increased in-hospital mortality, independent of VTE.[34 ]
The present Co-LEAD algorithm is the first algorithm designed to exclude PE in COVID-19
patients by using a D-dimer threshold adjusted to COVID-19 lung damage extent. Age-adjusted
D-dimer[15 ] or, more recently, clinical probability-adjusted D-dimer threshold[14 ]
[16 ]
[35 ] already showed excellent performances in non-COVID-19 patients. Since the damage
extent on CT seems to be associated with the intensity of COVID-19 coagulopathy,[24 ] and some reported that PE occurred more frequently in the affected lung areas, D-dimer
was adjusted here for lung extent damage on CT scan.[36 ] Given that CT is the routine examination performed in all hospitalized patients
with COVID-19 requiring oxygen therapy, information about the extent of lung damage
is almost always available. The Co-LEAD algorithm seems to have a better diagnostic
performance than the algorithm using age-adjusted D-dimer of the ESC guidelines. Unfortunately,
we were not able to compare the performance of Co-LEAD with the YEARS algorithm on
the derivation set because the post-hoc calculation of the clinical probability according
to YEARS was not feasible due to missing data.
In contrast to recommended algorithms,[13 ] Co-LEAD was constructed without using a pretest clinical probability score. This
choice was guided by our results showing comparable diagnostic performances when the
algorithm was applied to all patients in the derivation set rather than to those with
unlikely clinical probability according to the simplified revised Geneva score. Lastly,
as previously reported, a minority of COVID-19 patients had a high clinical pretest
probability and would have been directly eligible for CTPA. Not assessing clinical
probability simplifies the Co-LEAD algorithm and could improve its safety by reducing
the risk of misuse, as already shown for the PERC rule.[37 ]
[38 ]
[39 ] In fact, like the recent 4PEPS strategy, Co-LEAD offers a diagnostic strategy for
PE and results in a nonnegligible reduction in CTPA. Even if performances of the three
D-dimer assays show interesting results, performances can differ from one D-dimer
assay to another. Innovance D-Dimer and STA-Liatest D-Di have indeed a higher NLR,
and therefore a slightly lower performance, as already described in the non-COVID-19
population.[21 ]
[40 ] This could explain the two false-negative cases observed in the validation set,
for whom the Innovance D-Dimer assay was used. D-dimer assays may not have identical
results due to differences in antibody specificity because of variable affinity for
high- or low-molecular-weight fibrin degradation products and/or cross-linked and
non-cross-linked fibrin derivatives.
Our study has several limitations. First, regarding the applicability of this algorithm,
a major limit should be pointed out for patients with D-dimer between 900 and 1,700 ng/mL,
as it requires performing two scans: a first step including a noncontrast chest CT
to assess the extent of the disease, and then a CTPA according to the lung damage
extent. We are aware of the difficulties in assessing CT images in real time, and
this two-step protocol could be skipped by using the 900 ng/mL threshold in all COVID-19
patients. The Co-LEAD strategy moderately improves the efficacy and the efficiency
compared to a strategy based only on an adapted threshold of 900 ng/mL, by allowing
to avoid six iodine injections. It does nevertheless impose a two-step CT. However,
it has been shown that the extent of COVID-19 lesions on CT is stable over time beyond
the 10th day of illness[41 ]; the Co-LEAD thresholds could thus be used on the previous CT scan results, especially
in the most severe patients whose transport to the radiology department is neither
simple nor safe.[42 ]
Second, it is important to remember that nontransportable critically ill COVID-19
patients were not included in both derivation and validation sets because they were
not able to have CTPA. Therefore, the Co-LEAD diagnostic algorithm was not evaluated
in a population of unstable critically ill patients. On the other hand, it is important
to specify that the population included is heterogeneous, with both outpatients from
emergency departments and inpatients from general wards or ICUs. We can assume that
the characteristics of these patients could be different.
Third, our study is probably underpowered. It would have been desirable to get 100
events instead of 66 PEs in the validation cohort.[43 ] The upper limit of the 95% CI of the false-negative rate of the Co-LEAD algorithm
is above the recommended 3%.[23 ] To evaluate the performances of the current ESC guidelines in the derivation set,
the clinical probability for PE was calculated retrospectively. Lastly, due to the
retrospective nature of our study, we cannot exclude an indication bias of CTPA performance.
This strategy should therefore be validated in a management study including a 3-month
patients' follow-up, allowing for identification of thromboembolic complications following
the exclusion of PE in patients managed without CTPA.
Conclusion
The Co-LEAD algorithm using the D-dimer level with thresholds adapted to the lung
damage extent excludes PE in COVID-19 patients with a high sensitivity and high NPV
in both derivation and validation sets. This algorithm could reduce the number of
CTPA required to manage COVID-19 patients with a suspected PE. Further prospective
management studies are required to confirm this strategy in patients with COVID-19
and a suspected PE.
What is known about this topic?
What does this paper add?
