Diagnostic performance of D-dimer in predicting venous thromboembolism and acute aortic dissection

Abstract

Background

D-dimer is elevated in a variety of conditions. The purpose of this study was to assess the positive predictive value of D-dimer to rule in patients with confirmed pulmonary embolism, deep vein thrombosis, acute aortic dissection or thrombosis of the upper extremity in comparison to patients with elevated D-dimer for other reasons.

Methods and results

We studied 1334 patients presenting to the emergency department with pulmonary embolism (n=193), deep vein thrombosis (n=73), acute aortic dissection (n=22), thrombosis of the upper extremity (n=8) and 1038 controls. The positive predictive value was increased with higher D-dimer concentrations improving the ability to identify diseases with high thrombus burden. Patients with venous thromboembolism, acute aortic dissection and thrombosis of the upper extremity showed a maximum positive predictive value of 85.2% at a D-dimer level of 7.8 mg/L (95% confidence interval (CI) 78.1 to 90.4). The maximum positive predictive value was lower in cancer patients with venous thromboembolism, acute aortic dissection and thrombosis of the upper extremity, reaching 68.9% at a D-dimer level of 7.5 mg/L (95% CI 57.4 to 78.4). The positive likelihood ratio was very consistent with the positive predictive value. Using a cut-off level of 0.5 mg/L, D-dimer showed a high sensitivity of at least 93%, but a very low specificity of nearly 0%. Conversely, an optimised cut-off value of 4.6 mg/L increased specificity to 95% for the detection of life-threatening venous thromboembolism, acute aortic dissection or thrombosis of the upper extremity at the costs of moderate sensitivities (58% for pulmonary embolism, 41% for deep vein thrombosis, 65% for pulmonary embolism with co-existent deep vein thrombosis, 50% for acute aortic dissection and 13% for thrombosis of the upper extremity). Using the same cut-off in cancer patients, higher values were observed for sensitivity at a specificity level of more than 95%. The area under the curve for the discrimination of venous thromboembolism/acute aortic dissection/thrombosis of the upper extremity from controls was significantly higher in cancer versus non-cancer patients (area under the curve 0.905 in cancer patients, 95% CI 0.89 to 0.92, vs. area under the curve 0.857 in non-cancer patients, 95% CI 0.84 to 0.88; P=0.0349).

Conclusion

D-dimers are useful not only to rule out but also to rule in venous thromboembolism and acute aortic dissection with an at least moderate discriminatory ability, both in patients with and without cancer.

Introduction

D-dimer testing is well established for the exclusion of venous thromboembolism (VTE) among patients with a low or moderate clinical pre-test probability.^1–14 A sequential diagnostic strategy has been recommended in the clinical routine to reduce unnecessary advanced imaging and costs.^{3,4,8,15–17}

D-dimer values below a prespecified cut-off signal a negligible or absent thrombus formation. In contrast, the presence of D-dimer in circulating plasma above a particular threshold can well herald the possibility of an undetected thrombus, with values generally thought to be proportional to clot burden.^{1,4,5,18–22} However, D-dimer can be raised in other conditions including malignancy, infection, pregnancy, post-surgery, inflammation/trauma, disseminated intravascular coagulopathy and renal impairment.^1,2,23,24 Therefore, measurement of D-dimer is not recommended for rule-in of VTE or acute aortic syndrome due to low specificity and positive predictive value (PPV). Unfortunately, little is known about the specificity and PPV of very high D-dimers among symptomatic patients presenting to an emergency department (ED).^1,9

Thus, we aimed to explore the PPV of D-dimer in order to identify patients with confirmed VTE, acute aortic dissection (AAD) or thrombosis of the upper extremity (TUL) among patients having elevated D-dimer for various reasons.

Materials and methods

Patient selection and data collection

We retrospectively analysed data from 1334 symptomatic patients who were admitted between February 2012 and November 2015 to the ED of the University Hospital of Heidelberg, and in whom D-dimer values above the diagnostic cut-off were measured. D-dimer levels were routinely determined in patients with suspected VTE and a low to moderate probability based on the Wells score. In the present cohort, D-dimers were either measured before hospital admission or during ED stay at the discretion of the attending physician. An elevated D-dimer was defined by a value exceeding 0.5 mg/L. Pre-test probabilities were systematically calculated retrospectively, and final diagnoses were adjudicated by two independent emergency physicians and a third cardiologist in case of discordance. Standard data at admission contained information on patient demographics, history, physical findings, imaging results including compression ultrasound (CUS) with Doppler, computed tomography (CT) angiography, transthoracic/transesophageal echocardiography and details of medical treatment.

According to our local protocol and current guidelines,^1,9,15,16 all patients with elevated D-dimer above the cut-off (0.5 mg/L) and clinical suspicion for deep vein thrombosis (DVT) and/or pulmonary embolism (PE) were subjected to either CUS or CT.

Laboratory measurements

D-dimers were measured in the central laboratory using the Innovance D-dimer assay with a diagnostic cut-off point of 0.5 mg/L (Siemens Healthcare Diagnostics Products GmbH, Marburg, Germany). The assay principle is a particle-enhanced, immune-turbidimetric assay for the quantitative determination of cross-linked fibrin degradation products (D-dimers) in human plasma for use on coagulation analysers (CS-5100; Siemens Healthcare Diagnostics Products GmbH, Marburg, Germany). Other laboratory parameters included high-sensitivity troponin T (hsTnT; Roche Diagnostics, Rotkreuz, Switzerland), C-reactive protein (CRP) and N-terminal pro brain natriuretic peptide (NT-proBNP). Regarding hsTnT, serial measurements were also analysed obtained at the discretion of the attending physician as at least two consecutive hsTnT samples are required to differentiate an acute from a chronic myocardial injury. hsTnT is an important diagnostic tool not only for patients with suspected acute coronary syndrome (ACS), but also for several life-threatening differential diagnoses such as acute PE, decompensated heart failure, Takotsubo cardiomyopathy or acute aortic syndrome. In addition, hsTnT provides valuable information on the presence of acute myocardial injury in myocarditis. The prevalence of mild to moderate hsTnT elevation requires additional measurements to detect a significant concentration change in a considerable proportion of these patients. The vast majority of patients were admitted to hospital for a median of 5 days. In these patients, additional hsTnT measurements between day 1 and day 5 (commonly one additional measurement) were indicated for the assessment of prognosis.

In patients with PE, DVT, AAD and TUL additional biomarkers were available in the majority of cases. In particular, moderate or severe PE is systematically associated with functional impairment of the right ventricle and myocardial injury. Therefore, a subanalysis on the benefit to add NT-proBNP, hsTnT or both to D-dimer was performed in PE versus DVT/AAD/TUL.

Cases and controls

Patients with unequivocally confirmed PE, DVT, AAD or TUL formed the cases and patients with miscellaneous reasons for D-dimer elevation after the exclusion of VTE or aortic syndrome formed the controls. These cases comprised patients with ACS, heart failure, arrhythmias, pneumonia or other infections, hypertensive crisis, lung disease, bleeding, cancer and non-cardiac chest pain.

Definitions

The diagnosis of PE or DVT required either thrombus detection on CT angiography or a positive CUS. Likewise, the diagnosis of AAD or intramural haematoma required visualisation by CT angiography, magnetic resonance angiography (MRA) or echocardiography. A non-ST-segment elevation acute coronary syndrome (NSTE-ACS) was suspected in the presence of chest discomfort and a moderate to high clinical probability.²⁵ The diagnosis of a myocardial infarction was made according to the criteria of the third version of the universal myocardial infarction definition²⁶ and required a cardiac troponin (cTn) elevation above the 99th percentile with a rise and/or fall of consecutive measurements, together with signs or symptoms of myocardial ischaemia. The study was approved by the ethics committee of the University of Heidelberg and performed in accordance with the Declaration of Helsinki. Informed consent of the individual patient was not required.

Statistical analysis

Continuous variables were tested for normal distribution and were presented either as means with 95% confidence intervals (CIs), or as mean ± standard deviation (SD). Percentages were rounded. The normality of data distribution was assessed by the Kolmogorov–Smirnov test. Groups were compared using the χ² test for categorical variables and the Mann–Whitney U test for continuous variables. Tests of significance were also conducted using Student’s t test, when appropriate. To evaluate the PPVs and specificities of D-dimer we analysed three different strategies to derive cut-offs: (a) the recommended rule-out cut-off at 0.5 mg/L; (b) C-statistics using the receiver-operating characteristic (ROC) optimised cut-off balancing sensitivities and specificities; and (c) a cut-off that provided specificities greater than 95% yielding different sensitivities. ROC curves were plotted and areas under the curves (AUCs) were calculated with 95% CIs. Optimal cut-off values were determined by ROC analysis according to the methodology of DeLong. Sensitivity and specificity were calculated to assess the diagnostic performance of D-dimer. PPVs and negative predictive values (NPVs) were also reported.

All hypothesis testing was two-tailed and P values less than 0.05 were considered statistically significant. Statistical analyses were performed using MedCalc (Ostend, Belgium) and GraphPad Prism (La Jolla, California, USA) software packages.

Results

Elevated D-dimer concentrations were measured in 1334 patients (age 64 ± 18 years, 701 women and 633 men). Of these, 296 patients (22%) had confirmed PE (n=193), DVT (n=73), AAD (n=22) or TUL (n=8), 69% of all 296 patients as first event. Among the remainder, elevated D-dimer concentrations were observed in 1038 patients who received a final diagnosis of ACS (n=155), pneumonia or other infection (n=190), hypertensive crisis (n=61), heart failure (n=110), arrhythmia (n=41), major bleeding (n=7), lung disease (n=46) and non-cardiac chest pain (n=428). These groups served as the control group (age 65 ± 18 years, 555 women and 483 men). The clinical characteristics of the study population are presented in Table 1. Information about diagnostics and outcome is provided in >Table S1, available in the Supplemental Material.

Table 1.

Clinical characteristics of patients presenting with PE, DVT, AAD or TUL.

Variables	Overall	PE	DVT	AAD	TUL	P value
	(n=296; 100%)	(n=193; 65%)	(n=73; 25%)	(n=22; 7%)	(n=8; 3%)
Demographics
Age, years	63 ± 17	64 ± 16	60 ± 19	63 ± 14	44 ± 17	0.0151
Age >70 years	113 (38%)	80 (41%)	26 (36%)	6 (27%)	1 (13%)
Male sex	150 (51%)	102 (53%)	34 (47%)	14 (64%)	0 (0%)
Female sex	146 (49%)	91 (47%)	39 (53%)	8 (36%)	8 (100%)
Clinical presentation
First event	205 (69%)	138 (72%)	48 (66%)	12 (55%)	7 (88%)
Pain at presentation	201 (68%)	123 (64%)	51 (70%)	20 (91%)	7 (88%)
Duration of pain (hours)	186 ± 583	233 ± 712	108 ± 190	90 ± 225	92 ± 170	0.0151
Clinical prediction rule
Wells score (original version)	2.6 ± 2.0	2.9 ± 2.2	1.8 ± 1.0	–	1.6 ± 1.2	0.0004
Symptoms and signs
Dyspnoea	166 (56%)	148 (77%)	14 (19%)	4 (18%)	0 (0%)
Chest pain	101 (34%)	72 (37%)	6 (8%)	17 (77%)	6 (75%)
Limb swelling	84 (28%)	34 (18%)	43 (59%)	0 (0%)	7 (88%)
Leg pain	71 (24%)	22 (11%)	47 (64%)	2 (9%)	0 (0%)
Coughing	36 (12%)	35 (18%)	0 (0%)	1 (5%)	0 (0%)
Vertigo	28 (10%)	27 (14%)	0 (0%)	1 (5%)	0 (0%)
Back pain	22 (7%)	13 (7%)	1 (1%)	7 (32%)	1 (13%)
Tachycardia	16 (5%)	15 (8%)	0 (0%)	1 (5%)	0 (0%)
Syncope	14 (5%)	12 (6%)	1 (1%)	1 (5%)	0 (0%)
Abdominal pain	7 (2%)	6 (3%)	0 (0%)	1 (5%)	0 (0%)
Haemoptysis	7 (2%)	7 (4%)	0 (0%)	0 (0%)	0 (0%)
Risk factors
Diabetes mellitus	39 (13%)	30 (16%)	6 (8%)	3 (14%)	0 (0%)
Smoke	83 (28%)	52 (27%)	18 (25%)	10 (46%)	3 (38%)
Immobilisation >3 days	83 (28%)	52 (27%)	27 (37%)	3 (14%)	1 (13%)
Family history	39 (13%)	19 (10%)	17 (23%)	3 (14%)	0 (0%)
Pregnancy	2 (1%)	1 (1%)	1 (1%)	0 (0%)	0 (0%)
Cancer	84 (28%)	65 (34%)	14 (19%)	3 (14%)	2 (25%)
Stroke	20 (7%)	15 (8%)	4 (6%)	0 (0%)	1 (13%)
History of VTE/AAD	90 (30%)	55 (29%)	25 (34%)	9 (41%)	1 (13%)
Coagulation disorder	39 (13%)	19 (10%)	17 (23%)	3 (14%)	0 (0%)
LV dysfunction/heart failure	41 (14%)	30 (16%)	7 (10%)	2 (9%)	2 (25%)

PE: pulmonary embolism; DVT: deep vein thrombosis; AAD: acute aortic dissection; TUL: thrombosis of the upper limb; VTE: venous thromboembolism; LV: left ventricle.

The mean age of all patients with VTE, AAD and TUL was 63 ± 17 years (146 women and 150 men). Within patients with VTE or AAD the age did not differ, while patients with TUL were younger (44 ± 17 vs. 63 ± 17 years; P=0.0151). The Wells score showed lower values in patients with DVT as compared to patients with PE (1.8 ± 1.0 vs. 2.9 ± 2.2; P<0.0001) (Table 1).

At initial presentation, the most prevalent leading symptom was dyspnoea, occurring in 56% of all patients with PE, DVT, AAD and TUL. Comparing subgroups, the prevalence of dyspnoea was higher in patients with PE, showing rates up to 77%. Symptoms differed significantly among the groups, with dyspnoea, chest pain and coughing being documented in the majority of patients with PE and AAD, whereas swelling and pain of the lower or upper limb were more prevalent in patients with DVT or TUL. Patients with PE had the longest duration of pain after onset in comparison to AAD patients with the shortest periods (233 ± 712 vs. 90 ± 225 hours; P=0.3651).

Regarding the risk factors for thromboembolic diseases, smoking, immobilisation, history of VTE/AAD and co-existent cancer disease were most prevalent. Cancer was most frequently encountered in PE patients (n=65; 34%), with gastrointestinal (n=17; 26%), urological (n=15; 23%) and haematological (n=10; 15%) cancer as the most common types (Table S2).

D-dimer

Mean serum D-dimer levels in the entire study population (n=1334) were 3.1 ± 5.2 mg/L, whereby patients with PE, DVT, AAD and TUL showed a mean value of 7.9 ± 8.0 mg/L (P<0.0001). D-dimer values were significantly higher in the AAD group than in the VTE group (11.7 ± 11.5 vs. 7.7 ± 7.5 mg/L; P=0.0224). Likewise, patients with PE (8.1 ± 7.3 mg/L; P=0.1717) and DVT (6.5 ± 8.0 mg/L; P=0.5168) showed higher albeit not significant D-dimer levels when compared to patients with manifested TUL (4.5 ± 9.2 mg/L). Massive PE led to higher D-dimer concentrations than submassive PE (9.7 ± 6.5 vs. 6.6 ± 7.8 mg/L; P=0.0038). Patients with PE and co-existent DVT exhibited mean D-dimer values of 8.7 ± 6.8 mg/L. The control group (n=1038) showed a mean D-dimer level of 1.8 ± 3.0 mg/L and differed significantly from every other subgroup (P < 0.0001) (Figure 1). Controls consisted of patients with ACS (n=155, D-dimer 1.5 ± 1.7 mg/L), pneumonia or other infection (n=190, D-dimer 1.8 ± 2.4 mg/L), hypertensive crisis (n=61, D-dimer 1.8 ± 4.5 mg/L), heart failure (n=110, D-dimer 1.9 ± 2.3 mg/L), arrhythmia (n=41, D-dimer 1.5 ± 1.2 mg/L), major bleeding (n=7, D-dimer 2.0 ± 1.2 mg/L), lung disease (n=46, D-dimer 1.4 ± 1.3 mg/L) and non-cardiac chest pain (n=428, D-dimer 1.8 ± 3.7 mg/L).

Figure 1.

Admission D-dimer levels in patients with PE (8.1 ± 7.3 mg/L), DVT (6.5 ± 8.0 mg/L), PE with coexistent DVT (8.7 ± 6.8 mg/L), AAD (11.7 ± 11.5 mg/L) and TUL (4.5 ± 9.2 mg/L). *P<0.05 vs. control group; ^#P<0.05 vs. AAD. VTE: venous thromboembolism; PE: pulmonary embolism; DVT: deep vein thrombosis; AAD: acute aortic dissection, TUL: thrombosis of the upper limb.

Patients with cancer demonstrated higher D-dimer values (8.9 ± 7.5 mg/L, n=84) and more often had a history of concomitant cardiovascular risk factors when compared to non-cancer patients with VTE, AAD or TUL (P=0.1579). Among the different entities of cancer, patients with cancer of the oropharynx, the hypopharynx or the breast showed the lowest (6.5 ± 6.0 mg/L, n=15) and patients with cancer of the gastrointestinal tract, central nervous system (CNS) and cancer of unknown primary origin (CUP) the highest values of D-dimer (11.7 ± 9.1 mg/L, n=25).

Concerning symptoms, patients with haemoptysis, syncope, abdominal or back pain revealed the highest D-dimer values (max. 16.5 ± 13.6 mg/L in patients with haemoptysis, n=7). Characteristics of D-dimer are presented in Table S2 and basic laboratory parameters in Table S3.

Overall diagnostic performance

Overall cohort

The performance of different D-dimers to discriminate VTE or AAD from controls is displayed in detail in Table S4. We found an AUC of 0.857 (95% CI 0.84 to 0.88; P < 0.0001) for the entire cohort of patients with PE, DVT, AAD and TUL. The AUC was 0.906 (95% CI 0.89 to 0.92; P < 0.0001) for all 193 PE patients, 0.762 (95% CI 0.74 to 0.79; P < 0.0001) for DVT patients, 0.852 (95% CI 0.83 to 0.87; P < 0.0001) for AAD patients and 0.554 (95% CI 0.52 to 0.58; P=0.6469) for TUL patients versus control subjects (n=1038). Patients with PE and co-existent DVT revealed the highest AUC of 0.938 (95% CI 0.92 to 0.95; P < 0.0001) (Figure S1). Comparison of the AUCs showed statistical significance for PE, DVT, ‘PE with co-existent DVT’ and TUL when compared to the overall group (Table S5).

Cancer patients

The prevalence of cancer among all patients with PE, DVT, AAD or TUL was 28% (84 patients with cancer vs. 212 without cancer). The AUC for cancer patients with PE, DVT, AAD and TUL as compared against controls was 0.905 (95% CI 0.89 to 0.92; P < 0.0001) (Table S4; Figure S2). Cancer patients with PE and co-existent DVT presented an AUC value of 0.935 (95% CI 0.92 to 0.95; P < 0.0001), whereas AAD cancer patients showed an AUC of 0.962 (95% CI 0.95 to 0.97; P < 0.0001). No statistical significance could be detected in any of the cancer subgroups in comparison to the overall group (Table S5). The AUC for discrimination of VTE, AAD and TUL from controls was significantly higher in cancer versus non-cancer patients (AUC 0.905 in cancer patients, 95% CI 0.89 to 0.92, vs. AUC 0.857 in non-cancer patients, 95% CI 0.84 to 0.88; P=0.0349).

Other biomarkers

A subanalysis on the ability of additional biomarkers reflecting myocardial injury and/or haemodynamic compromise to discriminate PE from DVT, AAD and TUL was performed. Adding hsTnT to D-dimer demonstrated a non-significant trend to increase the AUC by 0.097 (AUC for D-dimer alone: 0.607 (SE 0.04) vs. AUC for D-dimer+hsTnT: 0.705 (SE 0.03); P=0.0565). Neither the addition of NT-proBNP (ΔAUC 0.022; P=0.8447) nor hsTnT plus NT-proBNP (ΔAUC 0.121; P=0.2225) increased the AUC of D-dimer alone significantly (Table S6).

PPV and likelihood ratios

The PPVs to detect PE, DVT, AAD and TUL according to the three different strategies to derive the optimal D-dimer cut-off are listed in Table S4 and depicted in Figures S3 and S4.

Using C-statistics to calculate cut-offs, the maximum PPV was 85.2% (95% CI 78.1 to 90.4) at a D-dimer level of 7.8 mg/L for the entire cohort of patients with VTE, AAD and TUL. Maximum PPVs were 80.6% (95% CI 72.3 to 86.8) at a D-dimer cut-off of 7.2 mg/L for PE, 55.6% (95% CI 33.7 to 75.4) at a D-dimer cut-off of 15.5 mg/L for DVT, 50.0% (95% CI 31.7 to 68.3) at a D-dimer cut-off of 13.3 mg/L for AAD and 16.7% (95% CI 2.6 to 60.4) at a D-dimer cut-off of 24.6 mg/L for TUL, respectively. Patients with PE and co-existent DVT demonstrated a maximum PPV of 73.9% (95% CI 63.2 to 82.4) at a D-dimer cut-off of 7.8 mg/L. Among cancer patients with VTE, AAD and TUL, a maximum PPV of 68.9% (95% CI 57.4 to 78.4) was achieved at a D-dimer level of 7.5 mg/L (Table S4).

Using the rule-out cut-off level of 0.5 mg/L, D-dimer showed a sensitivity of at least 93% for the detection of VTE, AAD or TUL, but a specificity of nearly 0%. In order to obtain a satisfying specificity greater than 95% for the detection of any of the life-threatening conditions (VTE, AAD and TUL), the cut-off value has to be set at 4.6 mg/L, yielding a sensitivity of 58% for PE, 41% for DVT, 65% for ‘PE with co-existent DVT’, 50% for AAD and 13% for TUL. Using the same cut-off value of 4.6 mg/L in cancer patients, consistently higher values for sensitivity were obtained at a specificity level of greater than 95% (Table S4).

In contrast to predictive values that depend on disease prevalence, likelihood ratios are not affected by disease prevalence.²⁷ Therefore, we also analysed positive likelihood ratios (PLRs) and negative likelihood ratios (NLRs). It is generally accepted that a PLR greater than 10 is suggestive of a good rule-in tool and a NLR of less than 0.1 of a good rule-out tool.²⁸ At a cut-off value of 0.5 mg/L, the PLR showed low values among the disease entities PE, DVT, AAD and TUL. In contrast, NLR was only less than 0.1 in the AAD, PE as well as the ‘PE with co-existent DVT’ group, which strongly points towards a good rule-out ability at this cut-off value (Table S4). In order to obtain a PLR greater than 10 and NLR of 0.14 for VTE, AAD and TUL, it is necessary to raise the D-dimer cut-off level to 4.5 mg/L and 0.6 mg/L, respectively. Cancer patients of the same groups showed a PLR greater than 10 at D-dimer levels of 4.1 mg/L and a NLR less than 0.1 at D-dimer levels of less than 0.9 mg/L. Therefore, especially among patients with cancer (Figure S2) D-dimer levels less than 0.9 mg/L revealed a high diagnostic power to rule out PE, DVT, AAD and TUL (at the cut-off level of 0.9 mg/L: sensitivity 96%, specificity 43%, PLR 1.7, NLR <0.1). The lowest NLR in non-cancer patients was 0.14 at a D-dimer cut-off level of 0.6 mg/L (sensitivity 98%, specificity 15%, PLR 1.2).

Discussion

The established role of D-dimer testing is almost exclusively restricted to rule out VTE or AAD in patients with low or moderate pre-test probability.^3,18,29–31 In the presence of elevated D-dimers that were measured at the discretion of the attending physician, there is no conclusive guidance on the diagnostic strategy, and physicians select the appropriate imaging method depending on the clinical context. Given the lack of specificity, the measurement of D-dimer is at the moment discouraged by current guidelines in the presence of a high clinical probability.^{9,16,17,32–35} Nevertheless, D-dimer concentrations contain diagnostic and prognostic information – and if available for any reason – could be useful to select the appropriate diagnostic method.²¹ To evaluate PPVs and specificities of D-dimer we analysed three different strategies to derive cut-offs: (a) the recommended rule-out cut-off at 0.5 mg/L; (b) C-statistics using the ROC optimised cut-off balancing sensitivities and specificities; and (c) a cut-off that provided specificities greater than 95% yielding different sensitivities.

This present retrospective analysis of a large cohort presenting to an ED demonstrates several important findings. First, very high D-dimer concentrations improve the ability to discriminate between diseases with very high thrombus burden such as VTE or AAD versus other diseases that are associated with less thrombus burden such as NSTE-ACS, pneumonia and other infections, hypertensive crisis or heart failure. Three different strategies were tested to evaluate the importance of D-dimer on the prediction of specific pathologies versus controls. Using the threshold at the rule-out cut-off in predicting and discriminating VTE or AAD from controls was ineffective as specificities and PPVs were extremely low. In contrast with our findings, others report sensitivities ranging between 91% and 97% and specificities between 40% and 70%^14,36 including low PPV between 9% and 20% among patients with DVT.³⁷ C-statistics resulted in a cut-off that balanced sensitivities and specificities. In our cohort, cut-offs at approximately 10-times rule-out cut-off provided excellent (AUC >0.9) or at least moderate (AUC >0.75 to 0.89) discriminatory performance to differentiate VTE or AAD from controls. Likewise, D-dimer cut-offs that yield specificities greater than 95% proved useful for the prediction of VTE or AAD, regardless of the presence or absence of cancer. Notably, the PPV of D-dimer was found to increase with the use of higher cut-offs.^36,38 In accordance with previous studies,^{1,4,5,18–22} D-dimers were found to correlate positively with thrombus burden and to increase gradually according to the severity of the disease. Substantially increased levels were observed in patients with massive PE (9.7 ± 6.5 mg/L) or AAD (11.7 ± 11.5 mg/L) in contrast to patients with submassive PE (6.6 ± 7.8 mg/L) or TUL (4.5 ± 9.2 mg/L).

Second, the prevalence of PE or DVT increased significantly with higher values of D-dimer. PE prevalence was 1.4% in the 0.5–1.0 mg/L group and already 80.4% in the group with D-dimer levels above 5 mg/L. The values are comparable with the data from Tick et al.³⁶ on 1515 consecutive patients who presented in the ED with clinically suspected PE and revealed a prevalence of 15% in the 0.5–1.0 mg/L group and 61% in the group with D-dimer levels above 4.0 mg/L. The present study revealed an almost hundredfold increased risk of PE in patients with D-dimer levels higher than 4.6 mg/L when compared to patients with D-dimer levels less than 1.0 mg/L. Regarding patients with PE and DVT, a cut-off level of 4.6 mg/L showed a sensitivity of 58% and 41%, respectively, at a specificity of greater than 95%. Analysing the areas under the ROC curves of PE, DVT and TUL, D-dimer testing of patients with PE and co-existent DVT revealed the best diagnostic performance of 0.938 (95% CI 0.92 to 0.95; P < 0.0001) in predicting VTE, whereas ROC curves of DVT or TUL demonstrated the lowest values (0.762 and 0.554, respectively). However, it should be noted that the AUC is affected by both sensitivity and specificity at each possible threshold.³⁹ For this reason, we calculated likelihood ratios that are largely independent of disease prevalence. PLRs were very consistent with PPVs (Table S4).

Third, D-dimer testing for rule-out of VTE is strongly discouraged in patients with cancer due to a very low specificity. As compared to patients free from malignancy, cancer patients are more likely to present with a more extensive thrombus burden and the development of thrombosis at atypical sites such as upper extremity thrombosis.^40,41 Previously published data showed that about 45% of patients (n=60) with a bilateral DVT had an existing malignancy and iliaco-caval/proximal locations of DVT were accompanied by a higher incidence of new cancer at 2 years follow-up when compared to distal lesions.^40,42,43 In our study, a cut-off level of 0.5 mg/L in cancer patients with PE revealed a sensitivity of 100% but a specificity of 0%, whereas the sensitivity and specificity rose to 69% and 95% at a cut-off level of 4.6 mg/L. Similar to patients without cancer, cancer patients with PE and co-existent DVT had the highest AUC of 0.935 (95% CI 0.92 to 0.95; P < 0.0001). Lower maximum PPV at ROC optimal cut-offs in cancer patients confirm the inferior diagnostic performance when compared to non-cancer patients. Nevertheless, very high D-dimer approximately 10-times higher than the rule-out cut-off still suggest the presence of a high thrombus burden with a PPV of approximately 54%.

Fourth, D-dimer testing has been recommended only for the rule-out of VTE and AAD. Regarding AAD, previous studies had investigated D-dimer with heterogeneous findings concerning sensitivities and specificities.^19,44,45 A small study on 24 patients with AAD reported a sensitivity of 100% and a specificity of 68.6% at a cut-off value of 0.5 µg/mL.⁴⁶ Similar rates for sensitivity and specificity were described in a meta-analysis of available studies on using D-dimer in order to rule out AAD, reporting a sensitivity of 97% and a specificity of 56% with a NLR of 0.06 and a PLR of 2.43.⁴⁵ These results were discordant with our findings for patients with AAD, as our study reports a specificity of 0% (95% CI 0.0 to 0.4) and a PLR of 1.0 (95% CI 1.0 to 1.0). Conversely, sensitivities (95% CI 84.6 to 100.0) and NPVs were 100% and NLR of 0 demonstrated an excellent ability for the exclusion of AAD. In patients with cancer, similar values were observed with a higher AUC of 0.962 (95% CI 0.95 to 0.97; P < 0.0001) confirming the excellent performance of D-dimer to discriminate AAD from controls. In order to obtain a prespecified specificity greater than 95%, the cut-off was determined to be 4.6 mg/dl in cancer and non-cancer patients. When using this cut-off among cancer patients, our data revealed a higher sensitivity of 67% (95% CI 9.4 to 99.2) as compared to 50% (95% CI 28.2 to 71.8) in non-cancer patients. Our findings suggest that the strategy derived for VTE based on very high D-dimer values could be extended to patients with suspected AAD for the prediction of tissue trauma and inflammation.

Study limitations and future directions

There are several limitations that have to be addressed. First, D-dimers were not measured routinely in all symptomatic patients in the ED. Therefore, a selection bias resulting in overoptimistic results and overestimated PPVs cannot be fully excluded. On the other hand, the uncritical overuse of imaging methods for screening of VTE or AAD must be avoided. We regard this study as a hypothesis generating first evidence to stimulate forthcoming systematic evaluation. Second, our observations were made in an internal medicine ED and not in an interdisciplinary ED, and thus may have a different spectrum and prevalence of diseases prohibiting the uncritical extrapolation of findings. Third, the study findings, i.e. D-dimer cut-offs, have not been prospectively validated yet. Fourth, the retrospective design of the study and the small number of patients with AAD and TUL must be emphasised bearing the risk of a sample size error. A prospective, ideally study, with physicians randomly blinded to routinely collected D-dimers, would be required to test the validity of our findings at prespecified cut-offs, and in order to provide a better estimate on the clinical consequences. Fifth, the indication for CT scan was left at the discretion of the attending physician. For example, 46 patients with elevated D-dimer above 10-times cut-off for other reasons than PE or AAD did not receive a CT scan, leaving some uncertainties on the rates of true and false positives. Finally, our used D-dimer cut-off level of 0.5 mg/L was not age-adjusted as this was not part of guideline recommendations until 2019.

Conclusions

Our findings support the hypothesis that D-dimer approximately 10-times higher than the general rule-out cut-off may help to rule in VTE and AAD with an at least moderate discriminatory ability as compared to other less life-threatening diseases, and to discriminate life-threatening conditions such as VTE or AAD from miscellaneous conditions that may be associated with unspecific mild D-dimer elevations. Unfortunately, many patients might still escape a correct diagnosis of VTE or AAD due to unspecific clinical presentation. The finding of very high D-dimer values – either measured with previous assessment of pretest probability or not – might result in a reconsideration on the presence of an otherwise missed condition. This would decrease the numbers of false negatives, however, at the cost of more unnecessary imaging methods as PPV is 75–85%. The study findings also support the concept that D-dimer concentrations are directly related to clot burden in VTE or tissue trauma in AAD. In addition, we could demonstrate that D-dimer measurement might also improve the detection of VTE and AAD in cancer patients, albeit with lower predictive power. Whether D-dimers might be recommended for their ability to rule in rather than rule out VTE or AAD and at which cut-offs requires confirmatory evidence from prospective validation in independent external cohorts.

Supplemental Material

Supplementary material is available at European Journal of Preventive Cardiology online.

Conflict of interest

The authors declare that there is no conflict of interest.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.