In late December 2019, a novel coronavirus, SARS-CoV-2, was isolated from patients with bilateral pneumonia. Soon after, the clinical syndrome caused by SARS-CoV-2 was labeled COVID-19 by the World Health Organization. This highly transmissible and virulent disease has had a devastating effect, overwhelming hospitals worldwide with critically ill patients. Although knowledge of the wide range of clinical features of COVID-19 is growing fast, the pathophysiology underlying the most common complications has not yet been fully elucidated [1, 2]. COVID-19 is a systemic disease associated with vascular inflammation and endothelial injury. The role of hypercoagulability is certainly significant in the diverse clinical manifestations of COVID-19 [3]. The emergence of thromboembolic complications is common, with thrombotic complexity and coagulation disorders emerging as a critical issue in COVID-19 patients who consequently sustain an increased risk of pulmonary thromboembolism (PE) [4]. While coagulation disorders often occur in severe cases with poor prognosis, the nature of this abnormality is not yet clear. Moreover, there is a lack of explicit indications regarding the best algorithm for diagnosing PE in COVID-19 patients. In particular, it is unclear whether the latest guidelines issued in 2019 by the European Respiratory Society and the European Society of Cardiology for the diagnosis and therapy of acute PE can be successfully applied to COVID-19 patients with clinical characteristics of PE [5, 6]. Prior to the COVID-19 pandemic, the Geneva and Wells scores were the most commonly employed to predict PE in the general population, either alone or in combination with D-dimer [7]. The combination of Wells and Geneva Scores, along with D-dimers, allows doctors to efficiently screen patients and minimize needless radiological imaging [8]. PE Rule-out criteria (PERC) and the YEARS clinical decision rule are used to identify individuals who are unlikely to have a PE [9, 10]. Another excellent pretest probability measure for ruling out PE and reducing imaging tests is the 4-Level Pulmonary Embolism Clinical Probability Score (4PEPS) [11]. However, because COVID-19 patients have a distinct thrombotic environment, the utility of these scores in predicting PE has not been thoroughly studied. Herein, we aimed to analyze new clinical and laboratory parameters and compare existing scoring systems (Modified Wells score, Revised Geneva Score, Simplified Geneva Score, PERC score, 4PEPS score, YEARS Score) used in assessing the clinical likelihood of PE in patients with SARS-CoV-2 infection.

From February 2020 to August 2021, 420 patients with dyspnea and positive rapid antigen test underwent CTPA due to suspected PE. 140 patients could not be found in the hospital information system: 120 were discharged immediately after CTPA was performed, 10 were sent back for scanning to a separate institution from which they were sent, and 10 had no records or were lost to follow-up. These patients, for whom none of the data relevant for this study was available, were excluded from the study. An additional eight patients were excluded due to negative RT-PCR test for SARS-CoV-2, while two others were excluded due to DVT and immediate admission to ICU (Fig. 1).

Fig. 1.

Flowchart of the included patients who underwent CTPA due to clinical suspicion of PE. CTPA, computed tomography pulmonary angiography; DVT, deep vein thrombosis; ICU, intensive care unit; PE, pulmonary embolism; RT-PCR, real-time reverse transcriptase-polymerase chain reaction.

The characteristics of included patients and the differences in biometrics, comorbidity, biochemical parameters, and treatment outcomes according to PE are shown in Table 1. Overall, 76 out of 270 included patients had PE registered on CTPA. The incidence of PE in our cohort was 28.1% (95% CI: 21.8–34.5).

For those admitted due to respiratory difficulties, the group of patients with PE had a significantly longer period of the disease duration (10 days) compared to the patients with no PE (8 days) (p = 0.006). The D-dimer and hs-Troponin levels were significantly higher in the PE group (respective 10.2 $\mu$g/L vs 5.3 $\mu$g/L, p 0.001; 22.2 ng/L vs 13.1 ng/L, p = 0.005). In the PE group, higher counts of platelets (289 vs 239 [$\times$10${}^9$/L], p = 0.016) and leukocytes (9.4 vs 8.2 [$\times$10${}^9$/L], p = 0.030) were observed. Also, the PE group contained more patients with CKD and heart failure (Table 1).

The consistency of averages of WELLS and 4PEPS scores calculations were both excellent: ICC of 91.2% (95% CI: 88.7–93.1) and 90.1% (95% CI: 93.3–95.9), respectively. The inter-rater agreement on YEARS score was very good: Cohen’s kappa = 82.9% (95% CI: 73.2–92.7).

3.1 The CDD-10 Score

The univariate logistic analysis associated eight possible predictors with the possibility of PE, including two existing scores: 4PEPS and modified Wells (Supplementary Table 1). Three of the independent predictors gave a unique statistically significant contribution to the multiple logistic regression model, namely, D-dimers, days of illness before admission, and the pre-existence of chronic kidney disease ($\chi$${}^2$ = 21.04, DF = 3, p 0.001). Based on this, a new proposed score consisting of three variables was defined: chronic kidney disease, D-dimers, and 10 days of illness before admission (CDD-10 score). This score accurately classifies 75.4% of PE cases, with cutoff value set at 0.5 resulting in AuROC 0.672, p 0.001 (Tables 2,3, Fig. 2A).

Table 2.Multiple logistic regression model coefficients and developed scoring system.

Variable	Coefficient	Std. error	p	Odds ratio	Variable value	CDD-10 score points
D-dimer	0.0308	0.0123	0.013	1.03	0.5 $\mu$g/L	–1.5
					$>$10.0 $\mu$g/L	2
					$>$20.0 $\mu$g/L	3
Illness duration prior to admission	0.0724	0.0284	0.011	1.08	$>$8 d	1
					$>$11 d	2
Existing chronic kidney disease	1.023	0.467	0.029	2.78	Yes	0.5
Constant	–2.193	0.367	0.001

Table 3.Comparison of different instruments predicting pulmonary embolism.

		PE (N = 76)	No PE (N = 194)	p-value*
Scores/Criteria for PE
	4PEPS [pts]	5 (3–8)	4 (2–6)	0.001
	Modified Wells [pts]	3.0 (0.8–3.3)	1.5 (0.0–3.0)	0.004
	Revised Geneva score [pts]	4 (4–6)	4 (3–6)	0.557
	Simplified Geneva score [pts]	2 (2–3)	2 (1–3)	0.354
	PERC (PE excluded) [n/N (%)]	1.3% (–1.3–3.9)	6.3% (2.7–9.9)	0.099
	YEARS (PE excluded) [n/N (%)]	7.9% (3.0–16.4)	14.4% (9.8–20.2)	0.160
Scores/models		AUC (95% CI)	Difference (95% CI) †	p-value
	CDD-10 model	0.69 (0.61–0.73)	0.00 (–0.04–0.05)	0.890
	CDD-10 score	0.67 (0.61–0.73)	/	/
	4PEPS	0.63 (0.56–0.69)	0.05 (–0.05–0.14)	0.329
	Modified Wells	0.59 (0.52–0.65)	0.084 (–0.01–0.18)	0.088
*For continuous variables Mann-Whitney U test, and for incidence rate comparison Chi-square test; † - compared to CDD-10 score.

Fig. 2.

Receiver operating curve analysis according to PE. (A) Original CDD-10 regression model. (B) CDD-s10 scoring system.

The new scoring system proposes a low clinical probability (CP) of PE (less than 2%) at a CDD-10 score 0. A CDD-10 score of 0 to 2.5 corresponds to moderate CP (20–40%) and a CDD-10 score over 3 corresponds to a high CP (more than 50%). PE prevalence in the low category was 0%; moderate category, 23.9% (95% CI: 17.5%–31.7%); and in the high category, 42.9% (95% CI: 28.2%–62.4%), (Table 4, Fig. 2B).

Table 4.Applicability and differences of the existing scores and CDD-10 scores by validation in different cohorts.

	N	PE [n (%)]	Clinical Probability Score
			Very low	Low	Moderate	High
4PEPS score			0	0–5	6–12	$>$12
Original validation cohort #1 [11]	1548	332 (21.4%)	3 (2.5%)	76 (10.0%)	207 (34.1%)	46 (75.4%)
Original validation cohort #2 [11]	1669	196 (11.7%)	5 (1.4%)	79 (8.2%)	95 (28.4%)	17 (68.0%)
Our COVID-19 cohort	267	76 (28.5%)	1 (7.1%)	39 (24.3%)	33 (41.2%)	0 (0.0%)
Our cohort and D-dimer 10 µg/L	267	76 (28.5%)	0 (0%)	1 (0.6%)	2 (2.27%)	0 (0.0%)
	N	PE [n (%)]	Clinical Probability Score
			Low	Moderate	High
Modified Wells score			0–1.5	2–6	6.5–12.5
Original cohort [14]	1239	217 (17.5%)	25 (3.4%)	112 (27.8%)	80 (78.4%)
Our COVID-19 cohort	267	76 (28.5%)	36 (22.8%)	34 (36.2%)	6 (40.0%)
Our cohort and D-dimer 0.01 $\times$ Age	267	76 (28.5%)	2 (1.3%)	0 (0%)	0 (0%)
	N	PE [n (%)]	Clinical Probability Score
			Low	Moderate	High
CDD-10 score			0	0–2.5	$>$3
Our COVID-19 cohort	270	74 (27.4%)	0 (0.0%)	47 (23.9%)	27 (42.9%)
Our cohort and D-dimer 0.01 $\times$ Age	270	74 (27.4%)	0 (0.0%)	2 (1.0%)	0 (0%)
4PEPS, 4-Level Pulmonary Embolism Clinical Probability Score; PE, pulmonary embolism.

While COVID causes a tendency to thromboembolic incidents, the pathophysiological mechanism that causes disorders of the coagulation system, as well as the diagnostic procedure that would enable efficient selection of patients with a high risk of PE, are still not completely clarified [18]. The main finding of our study is the newly proposed CDD-10 score, based only on three clinical-laboratory criteria (D-dimers, days of illness prior to admission, and the presence of chronic kidney disease) which demonstrated the highest accuracy in predicting PE in patients with SARS-CoV-2 infection. According to our study, pre-test probability scores that were frequently used for PE prediction in the general population showed low sensitivity and specificity for PE prediction in the SARS-CoV-2 cohort of patients. We found no difference regarding age, gender, comorbidities, other biochemical parameters, outcomes, or other pre-test probability scores between the two groups, except for chronic renal and heart failure, platelet and leukocyte levels, troponin levels, D-dimer levels, and days of illness before admission which were significantly higher in PE patients than in non-PE patients.

In our cohort, for whom CTPA was performed for clinical suspicion of thromboembolic complications, the incidence of PE was 28.1%, significantly higher than in the usual population of non-COVID patients who visit the emergency department due to dyspnea [18]. We believe there are two main explanations for this: (1) COVID-19 as a disease in which the chance of developing PE is increased, and (2) the selection of patients in whom the clinician performed CTPA because PE was the most likely differential diagnosis [19]. According to these findings, the real incidence of PE in all patients with confirmed SARS-CoV-2 infection is still unclear and probably underestimated.

The problem or the advantage of all pre-COVID emergency PE-related scoring systems is that they exclude the possibility of PE and thus avoid excessive use of CTPA. Any such scoring system is based on the exclusion of PE or reducing the number of false-negative predictions. Therefore, no system aims for a good sensitivity (or the resulting high positive predictive value). Secondly, the negative predictive values of these systems must be precise. In our case, where a high incidence reached over 28%, as opposed to validation cohorts of observed scoring systems [11, 14] of 12, 18 or a maximum of 21%, this value must be even more reliable.

Precisely due to the low sensitivity, no score can have symmetrical receiver operating curves with the surface under the curve trending to 100%. A statistically significant reduction in area under ROC for existing scores, in contrast to their validation cohorts, supports our thesis that in the COVID-19 population, the scores for patients who present for dyspnea (with a high incidence of PE) are not sufficient. Their specificity is lower, which can be seen from the increase in the incidence of PE (exceeding the agreed intervals) in low and moderate clinical prediction score groups. In our study, PE was more often observed in patients with CKD. Several processes can account for these observations. Because of elevated levels of procoagulant factors, reduced endogenous anticoagulants, and fibrinolytic activity in CKD patients, they are at risk of clot formation and thrombosis [20].

In contrast to our findings, an Italian multi-center retrospective investigation of 689 COVID-19 patients found that CKD was not predictive of PE incidence [21]. However, in a study by Inge H.Y Luu et al. [22] COVID-19 patients with positive CTPA more often had CKD than patients in whom CTPA was negative for PE (p 0.03), confirming our results. In addition to the aforementioned molecular pathways, patients with CKD are more likely to have other documented risk factors for PE, such as congestive heart failure and immobility. In our cohort, more prevalent PE in patients with a history of CHF can be explained by decreased left ventricular systolic function, increased venous stasis, and chronic inflammation in the cardiovascular system [23]. Patients with PE experienced a longer delay from beginning of symptoms to hospitalization in the Fauvel et al. [24] research, in line with our findings.

TnI levels were observed to be considerably higher in the PE group. As a result of pulmonary vascular obstruction, right ventricular pressure may contribute to right ventricular dilatation and myocardial ischemia, resulting in a rise in Troponin I (TnI) levels [25]. Furthermore, PE induces a rise in tension in the right ventricle and pericardium, which can constrict the coronary arteries and cause partial myocardial ischemia and necrosis of myocardial cells, resulting in the release of TnI [26]. On the other hand, increased levels of troponin can also be seen in the setting of actual SARS-CoV-2 infection causing myocardial damage by non-ischemic myocardial processes, such as acute respiratory infection, sepsis, systemic inflammation, pulmonary thrombosis, cardiac adrenergic hyperstimulation during cytokine storm syndrome, and perhaps myocarditis. In a systematic review of four studies including 374 patients, cardiac TnI levels were considerably higher in those with severe COVID-19 infection compared to those with non-severe disease (OR 25.6, 95% CI: 6.8–44.5) [27]. Furthermore, TnI elevation in many patients may be exacerbated by concurrent renal failure, which was found to be more frequent in the PE group in our research. It can be concluded that the elevated levels of troponin in our PE patients are multifactorially caused [28].

Many patients with SARS-CoV-2 infection and respiratory failure appeared to have hypoxemia out of proportion to the impairment in lung compliance, which might be explained by pulmonary thrombosis, in some cases subclinical or radiologically unconfirmed [29]. In a study by Mirsadraee et al. [30], D-dimer levels did not discriminate between patients with and without PE in which screening CTPA was performed for patients with COVID-19 on admission to the ICU. This contradicts the results of our study but can be explained by the fact that these were patients in the ICU where the inflammatory and hypercoagulable component is particularly emphasized independently of PE existence [30]. Although a significant number of studies have confirmed the association of elevated D-dimer levels with PE in COVID patients, increased D-dimer levels alone cannot be used to confirm PE diagnosis [31, 32, 33, 34]. This is all the more the case because D-dimer values are increased even in COVID-19 individuals who do not have PE due to thromboinflammation or COVID-19-associated coagulopathy [35, 36].

Considering the above, the results of our research are not surprising. Namely, the levels of D-dimers were significantly higher in our PE group compared to the non-PE group. Due to the high levels of D-dimers in COVID-19 patients, even in the absence of PE, some authors recommended a higher D-dimer threshold to select patients appropriate for CTPA, based on the Youden index [29, 35, 36, 37, 38]. Clinical suspicion of PE in patients with COVID-19 pneumonia is often diminished because the signs and symptoms of COVID-19 pneumonia mimic those of PE which sometimes remain unrecognized: the clinical presentation of PE may overlap with that of COVID-19 pneumonia which may hinder the recognition of PE symptoms in patients who are already complaining of dyspnea. As a result, current estimates may significantly underestimate the real PE incidence in COVID-19, as revealed by autopsy investigations [39].

Kirsch et al. [40] verified the utility of the Wells score in predicting PE in a retrospective cohort of 64 hospitalized COVID-19 patients (HCP). In this study, a Wells score of 4 or above was strongly linked with PE development (p = 0.04). The AUC-ROC curve for the prediction of PE in HCPs, calculated for an optimal value of Wells score between 1 and 2, was 0.54, lower than in our cohort of patients with PE [40]. The study by Scardapane et al. [41] found no significant correlation between Wells score and PE in a cohort of 43 HCPs (median age 65 years, 51.16% males), as opposed to the Revised Geneva Score, which was significantly higher in PE patients than in non-PE patients (mean 4 + 2 vs 2 + 2, p = 0.01). In our study, the Revised Geneva score did not reach statistical significance in the difference between the groups with or without PE. Polo Fritz et al. [42] conducted a similar research, based on 41 HCPs (median age 71.7 years, 73% females) undergoing CTPA. The Wells score was found not clinically useful for predicting PE.

Although Wells score did not prove to be reliable in predicting PE in COVID patients, it has been widely used to predict PE in the general population, stratifying patients into three groups with low (1.3% prevalence), moderate (16.2% prevalence), and high risk (37.5% prevalence), according to their pre-test chance of developing PE [43]. The score had an AUC of the ROC calculated for predicting PE in the general population of 0.632 (95% CI: 0.574–0.691) [44] which was similar to our cohort of patients with PE and COVID-19 (AUC 0.61 (95% CI: 0.54–0.67)).

The fundamental disadvantage of the Wells score is the inclusion of a physician’s subjective judgment among factors, i.e., “PE is the most likely diagnosis”, as previously stated by Klok and colleagues [45]. Especially in COVID-19 management, physicians will most often suspect PE if patients present with hypoxemia and tachycardia, thus limiting the utility of this score in predicting PE. It has been observed that using the Wells score and D-dimer together improves the test’s sensitivity and specificity [46, 47]. Kampouri et al. [47] discovered that a Wells score $>$2 paired with a D-dimer value $>$3000 ng/L offered a highly specific prediction rule with a sensitivity of 57.1% and a specificity of 91.6% in a retrospective investigation of 443 HCPs (median age 68.68 years, 57.7% males). To date, no studies have been conducted to assess the accuracy of the 4PEPS score in predicting PE in the SARS-CoV-2 cohort of patients. The 4PEPS score has an AUC of 0.63 (95% CI: 0.57–0.69) in our cohort of patients with SARS-CoV-2 infection, indicating that it is neither specific nor sensitive for predicting PE in patients with SARS-CoV-2 infection.

Compared to the existing scores, our newly proposed CDD-10 score, based only on three clinical-laboratory criteria (D-dimers, days of illness before admission, and existing CKD), showed the highest accuracy in predicting PE, with its AUC-ROC of 0.669 and 0.672, respectively. None of the other four scores (Revised and Simplified Geneva score, PERC, YEARS) reached statistical significance in our cohort of patients; according to our study, they should not be routinely used for PE prediction in patients with SARS-CoV-2 infection. YEARS algorithm and PERC are used to rule out PE, reducing unnecessary CTPA. In a retrospective study of 93 COVID-19 patients with acute respiratory failure, Porfidia et al. [48] found no difference in terms of age, gender, and PERC between COVID-19 patients with positive CTPA and those with negative CTPA for PE, with the exception of D-dimer $>$1000 ng/mL and the indication to undergo CTPA based on YEARS algorithm. However, there were patients without confirmed PE who had D-dimer $>$1000 ng/mL and an indication to undergo CTPA based on the YEARS algorithm, indicating that the YEARS algorithm is not sufficiently specific.

In our study, there was no difference between the PE group and the non-PE group regarding the YEARS algorithm and PERC. Accordingly, these algorithms should not be routinely used in the stratification of patients who need to undergo CTPA for suspected PE. According to our findings, one factor that may have the greatest influence on inter-rater reliability during the calculation of pre-test probability scores is “PE as a first or equally likely diagnosis”, based primarily on the physician’s impression during admission after considering the patient’s medical history and clinical status. “PE as a first or equally likely diagnosis” was mostly considered if patients had unexplained dyspnea, higher than expected D-dimer levels, or low or rapid drop in oxygen saturation not explained by concurrent pneumonia or other possible factors.

According to our study, previously used pre-test probability scores for PE showed low sensitivity and specificity for PE prediction in the SARS-CoV-2 cohort of patients. Only the 4PEPS score and Modified Wells score reached statistical significance in hospitalized COVID-19 patients (HCP). Our newly proposed CDD-10 score, based only on three clinical and laboratory parameters, demonstrated the highest accuracy in PE prediction among HCP. Further prospective studies are necessary to determine other risk factors and to develop algorithms for better risk stratification of SARS-CoV-2 patients who require CTPA for suspected PE.

AUC, Area under the ROC curve; CHF, Chronic heart failure; CDD-10, Chronic kidney disease, D-dimers, 10 days of illness before admission; CKD, Chronic kidney disease; CI, Confidence interval; COPD, Chronic Obstructive Pulmonary Disease; COVID-19, Coronavirus disease 2019; CPTP, Clinical pre-test probability; CTPA, Computerized tomography pulmonary angiography; DVT, Deep venous thrombosis; HFNC, High-flow nasal cannula; HCP, Hospitalized COVID-19 patients; ICU, Intensive Care Unit; IQR, Interquartile range; LOS, length of hospital stay; LVEF, Left ventricular ejection fraction; MV, Mechanical ventilation; NIV, Non-invasive ventilation; NSAID, Non-steroid inflammatory drug; OR, Odds ratio; PCR, Polymerase chain reaction; PE, Pulmonary embolism; 4PEPS, 4-Level Pulmonary Embolism Clinical Probability Score; PERC, Pulmonary embolism rule-out criteria; RBBB, Right bundle branch block; ROC, Receiver operating curve; RT-PCR, Reverse-transcriptase polymerase chain reaction; TnI, Troponin I; TPA, Tissue plasminogen activator; VTE, Venous thromboembolism; YEARS, YEARS study algorithm; WHO, World Health Organization.

The data sets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

MM, OB and MPI designed the research study. MM, OB, MPI, IJ, IG, IS, MZP and KD performed the research. MM, OB, MPI, IJ, IG, IS, MZP and KD provided help and advice on writing and reviewing the article. MM, OB and MPI analyzed the data. MM, OB, MPI, IJ, IG, IS, MZP and KD wrote the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript.

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Ethics Committee of the University Hospital of Split, Croatia (500-03/21-01/140, Ur.br. 2181-147/01/06/M.S.-21-02). Written informed consent was not required for participants with emerging infectious diseases and this study had no interventions. Due to the retrospective analysis of data, the need for informed consent was waived by our institution.

Not applicable.

This research received no external funding.

The authors declare no conflict of interest.