Background
Pulmonary embolism (PE) is a common severe cardiovascular illness andcan be life-threatening if left untreated [
1‐
3]. However, clinical signs and symptoms have relatively low sensitivity and specificity in PE diagnosis [
4,
5]. Current diagnosis of PE still relies on computed tomographic pulmonary angiography (CTPA), which is costly, time-consuming and exposes patients to extra radiation [
6,
7]. Recent studies in North America showed that the detection rate of suspected PE is as low as 5% [
8]. Therefore, it is of great importance to avoid excessive CTPA by considering ruling out PE using clinical pretest probability scores.
Commonly used clinical pretest probability, such as Wells score and Geneva score, combines clinical symptoms with predisposing risk factors of venous thromboembolism and classifies patients with suspected PE into distinct risk categories [
9‐
11]. In addition to clinical pretest probability score, plasma level of D-Dimer, which is often elevated in the presence of acute thrombosis due to simultaneous activation of coagulation and fibrinolysis, is also used to rule out PE [
12,
13]. The combination of low pretest probability score and plasma D-Dimer level lower than 500 ug/L were used to rule out PE in an initial attempt, followed by age adjusted D-Dimer cutoff (500 ug/L for patients below 50 years, and age*10 ug/L for patients older than 50 years) and risk category specific D-Dimer cutoffs (1000 ug/L for patients with low pretest score, and 500 ug/L for patients with moderate pretest score) [
14‐
16]. Despite efforts in adjusting D-Dimer thresholds in different age groups and PE risk categories, specificities and negative predictive values of the PE rule out strategies are generally low, resulting in an overuse of CTPA among alternatively diagnosed patients with high plasma D-Dimer levels or moderate pretest probability score. What’s more, none of the current PE rule out strategies differentiate the failure rates across different PE severity grades, as major embolism and multiple embolisms require much more stringent diagnosis and treatment window than mild embolism [
17].
In this investigation, we aim to develop a comprehensive PE pretest score that incorporates clinical indications in Wells score and other important clinical predictors to guide clinicians in assessing PE risk with improved screening performances. The proposed PE Comprehensive Screening Score (PECSS) is designed to optimize the rule out performance among patients with severe PE and to achieve excellent operating characteristics among patients with moderate and mild PE. PE severity grades were evaluated based on CTPA and validation of PECSS was conducted in an independent held-out data set.
Methods
Data collection
We retrospectively collected data from patients presenting to the emergency department of Zhongshan Hospital (affiliated to Fudan University) from July 2018 to October 2022. Patients were included if they were 18 years and older, were suspected of PE or could not be ruled out of PE by the attending physicians. Patients were excluded if they did not have PE status or severity confirmed by CTPA, had missing D-Dimer values and were pregnant. Electronic medical records, CTPA, laboratory tests results and vital signs were collected and reviewed.
PE severity grade
The severity grade of PE was determined based on CTPA. PE was divided into 4 severity grade categories: 1) severe PE, which was defined as the pulmonary artery embolism or multiple pulmonary arteriole embolism; 2) moderate PE, defined as pulmonary artery branch embolism; 3) mild PE, defined as single pulmonary arteriole embolism; 4) no PE.
Study design
Patients meeting the inclusion and exclusion criteria were randomly divided into the derivation (80%) and validation (20%) set according to the stratified randomization scheme such that the distribution of PE severity grades were similar in the two data sets. Derivation set was used to construct the new screening approach and compare it to existing approaches, while the validation set was used to validate the findings. The primary endpoint was failure rate in patients with severe PE, which was defined as the proportion of PE cases failed to be ruled out in patients with severe PE. Secondary endpoints include failure rate in moderate, mild and total PE, defined as the proportion of PE cases failed to be ruled out in patients with moderate, mild and any PE, respectively. In addition, we also evaluated the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) associated with each screening approach.
Statistical analysis
Statistical analysis was performed using R statistical software version 4.0.3. Continuous characteristics were described using mean \(\pm\) standard deviation if normally distributed, or median (interquartile range) if normality was violated. Categorical characteristics were presented using frequency (percentage) per category. Statistical tests were performed using student t test or ANOVA for continuous variables (Wilcoxon rank sum test for skewed distribution), and Chi-square or Fisher’s exact test for categorical variables.
To construct the PE comprehensive screening score (PECSS), we first screened all 26 clinical predictors that were suspected to be associated with PE according to clinical consensus. Among all these predictors, Lactic Acid, Aspartate Transaminase and Lactic Dehydrogenase had missing proportions higher than 30% and were therefore dropped from the following analyses. Next, univariate association with PE status (PE vs no PE) of each predictor was assessed using univariate logistic regression model, and predictors with P values smaller than 0.05 were selected as candidate variables in models constructing PECSS. Candidate variables selected by univariate logistic regression models were then included in a searching algorithm, where a series of points were assigned to each predictor and the total points were added to the Wells score to construct PECSS. To facilitate clinical practice and in line with total Wells score, we considered 0, 0.5, 1.0 and 1.5 points for each candidate variable so that higher scores indicate higher probability of PE. Finally, optimal PECSS cutoffs and optimal D-Dimer cutoffs within each PECSS stratum were determined to minimize the failure rate in patients with severe PE.
Screening performances of PECSS, PECSS + D-Dimer and PECSS + PERC were evaluated and compared to Wells score, Wells score + D-Dimer and Wells score + PERC. Failure rates among patients with severe, moderate, mild and any PE were compared under these scoring systems, respectively. Sensitivity, specificity, PPV and NPV were also calculated for each screening approach.
Discussion
Due to the difficulties of PE diagnosis in early clinical stage and emergency medical setting [
18,
19], physicians often rely on excessive CTPA use to reduce the miss diagnosis rate of PE, especially for those with pulmonary artery embolism or multiple pulmonary arteriole embolism, which could lead to poor prognosis and high risk of death. To rule out PE among patients presenting to the emergency department, the PERC [
20] rule required 8 objective criteria, including age (less than 50 years old), medical history and several empirical indicator of clinical signs and symptoms [
21]. However, with the aging population and potential assessment bias introduced by attending physicians of various clinical experiences in PE diagnosis, the application of PERC rule in emergency medical setting is relatively limited [
22,
23].
In this investigation, we developed and validated an objective and easy-to-use scoring system to safely rule out PE among patients presenting to the emergency department. The newly proposed Pulmonary Embolism Comprehensive Screening Score (PECSS) combined important clinical predictors with indicators in the Wells score and achieved excellent performances across all PE severity grades. Particularly, we found that patients with age > 65 years old, anhelation, less pain, abnormal blood pressure, in critical condition when admitted, higher levels of pro-BNP, CRP, UA, WBC, NEUT, MONO and ALT, and lower levels of HGB, ALB and creatinine were at increased risk for PE. Through a comprehensive grid searching algorithm, PECSS identified 7 clinical predictors (anhelation, abnormal blood pressure, pro-BNP > 1000 pg/mL, CRP > 30 mg/L, UA > 300 umol/L, in critical condition when admitted and age > 65 years), in addition to indicators in Wells score (VTE symptoms, no alternative diagnosis, HR > 100 beats/min, immobilication / surgery and previous PE/VTE), and assigned corresponding scores to each of the predictor. All of the clinical predictors were easy to obtain from medical records and rountine blood tests in emergency department setting.
Identification of these candidate clinical predictors were supported by previous studies. Conditions such as pulmonary main embolism can lead to obstructive shock, which is a possible cause of hypotension. Elevated blood pressure can be explained by PE associated symptoms or stress [
24,
25]. Although a large number of studies have found pro-BNP as a biomarker for PE, optimal cut-off values are still inconsistent [
26‐
28]. In our study, we assigned 0.5 point to pro-BNP greater than 1000 ng/L in constructing PECSS and provided a reference for future research regarding the effect of pro-BNP on PE. Both CRP and UA are acute phase reactant, which are associated with increased risk of cardiovascular and cerebrovascular events [
29‐
33]. CRP has been shown to increase the tendency to thrombose through the mediation of interleukin 6 and monocyte chemoattractant protein 1 [
34,
35]. Plasma level of UA can be used as a risk marker for thrombosis and prognostic stratification among PE patients [
36,
37]. Therapies targeted at lowering UA are associated with reduced risk of major adverse cardiovascular events [
38‐
40].
To facilitate clinical application and further improve the screening performance, we incorporated plasma level of D-Dimer, which is often elevated in the presence of acute thrombosis, with PECSS. Optimal D-Dimer cutoff values were investigated in patients with low (PECSS \(\le\) 4), moderate (4 < PECSS \(\le\) 6) and high (PECSS > 6) PE risk. It turned out that D-Dimer only played a role in patients with moderate PE risk, where higher levels of plasma D-Dimer increased the pretest probability of PE. Therefore, patients with moderate risk (4 < PECSS \(\le\) 6) and low D-Dimer (D-Dimer \(\le\) 2.5 mg/L) could be safely ruled out of PE, whereas those with moderate risk (4 < PECSS \(\le\) 6) and high D-Dimer (D-Dimer > 2.5 mg/L) need to undergo CTPA before PE can be ruled out. D-Dimer were not expected to affect the decision of ruling out PE in patients with low and high pretest probability, since patients with low PECSS have very low risk of PE according to medical records and clinical symptoms where D-Dimer do not have enough clinical specificity, while those with high PECSS have high risk of PE such that they need to undergo CTPA regardless of their plasma D-Dimer level.
Screening performances of PECSS only, PECSS + D-Dimer, Wells only, Wells + D-Dimer were thoroughly compared in terms of failure rates across all PE severity grades, sensitivity, specificity, PPV and NPV. Compared to Wells only, PECSS only approach greatly reduced the failure rates in all PE severity grades (5.1% to 0, 16.5% to 2.2%, 16.0% to 0, and 9.6% to 0.5% in severe, moderate, mild and any PE, respectively). The huge improvement indicated that by including the additional seven clinical predictors chosen by variable selection and clinical meaningfulness, PECSS are able to identify patients with suspected PE risk more accurately and efficiently than Wells score. Although PECSS + D-Dimer and Wells + D-Dimer had similar performances in moderate and mild PE, PECSS + D-Dimer had three times lower failure rate in severe PE compared to Wells + D-Dimer (0.3% versus 0.9%). This is of particular clinical significance since servere PE consists of patients with pulmonary artery embolism or multiple pulmonary arteriole embolism and had the worst clinical prognosis [
41‐
44]. By improving the accuracy and effectiveness of pretest probability and targeting on the clinically beneficial patient population, the PECSS + D-Dimer would greatly reduce the mortality associated with PE. In terms of sensitivity and specificity, Wells score achieved the highest specificity, but at the cost of low sensitivity and NPV. PECSS had the highest sensitivity, while at the cost of low specificity and PPV. PECSS + D-Dimer had the best compromise between good sensitivity and specificity, and moderately high PPV and NPV. It should be noted that although PECSS had very high sensitivity, its specificity and positive predictive value were relatively low. This indicates that PECSS works well in identifying patients at high risk of PE, but relatively under-performs to rule out PE when patients are screened negative. PECSS + D-Dimer strategy improved upon PECSS only in terms of both specificity and positive predictive value.
Admittedly, our study has several limitations. Firstly, this is a single center study and generalization of the results require validation from multi-center set. Secondly, patients who were suspected of PE but did not undergo CTPA due to hemodynamic instability were not included in out study due to the uncertainty of their PE status. This would potentially lead to bias in applying PECSS and PECSS + D-Dimer approaches in the targeted patient population. Thirdly, radiological severity instead of clinical severity (such as hypotension, pulmonary hypertension, cardiac failure, and the need for oxygen therapy) were used when defining PE severity grades. However, they do not usually correlate perfectly with each other. Further, right ventricular dysfunction was not considered in defining the PE severity grades since echocardiography was missing for a large proportion of the study population. Although CTPA has been applied in right ventricular dysfunction, its reliability and accuracy cannot be guaranteed as echocardiography [
17,
18]. Finally, our study is retrospective in nature, and therefore, causal relationships could not be established between clinical predictors in constructing PECSS and PE diagnosis.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.