Retrospective analysis of the diagnostic accuracy of lung ultrasound for pulmonary embolism in patients with and without pleuritic chest pain

We combined individual patient data from one prospective monocentric study, Reissig 2001 [16], and two prospective multicentric studies, Nazerian 2014 [12] and Nazerian 2017 [17], enrolling consecutive patients with suspected PE.

After analysis of the existing literature on the topic, we selected for convenience the only 3 studies that reported complete information about the presence or absence of pleuritic pain on presentation. The studies had the following characteristics: (a) original publication; (b) prospective cohort study of patients with an objectively confirmed diagnosis of symptomatic PE; (c) record of presence or absence of pleuritic chest pain at presentation; (d) LUS performed in all enrolled patients.

One investigator, who was not a co-author of the three original studies included in the analysis, used the Quality Assessment of Studies of Diagnostic Accuracy included in Systematic Reviews-2 (QUADAS-2) tool to assess the methodological quality [18]. This tool is composed of two parts: risks of bias and concerns regarding applicability. The former was assessed in four domains patient selection, index test, reference standard and flow and timing, and the latter was assessed in three domains patient selection, index test and reference standard.

A core group of investigators (PN, GV and AR) developed the process for obtaining patient level data and the planned analyses, and all the co-authors approved them before the data collection phase.

After the investigators agreed to share their data, the databases were anonymously transferred to a central location under the auspices of PN. Data were checked, explanations for coding and uncertain data were clarified and a single pooled database was developed.

Reissig 2001 and Nazerian 2017 enrolled patients suspected of PE without differentiating the risk score, whereas Nazerian 2014 enrolled patients with Wells score > 4 (likely) or a positive d-dimer that underwent MCTPA.

The Wells score included the following items: clinical signs and symptoms of deep vein thrombosis (DVT) (+ 3), PE is most likely diagnosis or equally likely (+ 3), heart rate > 100 bpm (+ 1.5), immobilization at least 3 days or surgery in the previous 4 weeks (+ 1.5), previous objectively diagnosed PE or DVT (+ 1.5), hemoptysis (+ 1), malignancy with treatment within 6 months or palliative care (+ 1) [19]. Patients were categorized as PE likely if Wells score was > 4 and PE unlikely if ≤ 4. Cut-off values for d-dimer was < 500 ng/ml and not age-adjusted. All studies reported whether patients had pleuritic chest pain, that was defined as acute onset sharp pain exacerbated by breathing or coughing. Additional file 1: Table S1 reports LUS criteria for PE diagnosis and the reference test used in each study to formulate a final diagnosis of PE.

In all studies LUS was performed by scanning the whole chest in 2 anterior, 2 lateral and 2 posterior chest areas per side; in each area, all the intercostal spaces were scanned searching for pulmonary infarctions. Details about the ultrasound machines and transducers used are reported in the method section of each study. Investigators performing LUS were blinded to diagnostic tests results and to all the clinical information except for symptoms of presentation and visible physical signs. The pattern considered positive for lung infarction was visualization of a pleural-based anechoic consolidation, wedge or round shaped, with sharp margins, without air bronchograms, of a minimum size measured at the pleural level of 0.5 cm with or without an associated small pleural effusion (Fig. 1). Two studies, Nazerian 2014 and Nazerian 2017, also reported the performance of a limited LUS examination based on a single ultrasonographic scan in the most painful chest area indicated by the patient.

Data points are expressed as mean ± standard deviation. The diagnostic performance of LUS in all patients, and in patients with and without pleuritic chest pain was assessed by calculating accuracy (ROC curves), sensitivity, specificity, positive and negative predictive values, and likelihood ratios. The extended McNemar and the McNemar tests were used to compare sensitivities and specificities of LUS in patients with and without pleuritic chest pain, of global chest LUS examination approach versus a single LUS scan performed in the most painful area [20]. Using the same tests, we also compared two pre-test strategies for the prediction of PE: the combination between the clinical Wells score with the d-dimer test (Wells + d-dimer) versus the combination of the Wells score with the result of the LUS exam (Wells + LUS). The unpaired Student’s t-test was used to compare normally distributed data. Chi-square test was used for the comparison of variables expressed as proportions. To evaluate the most efficient strategy to rule-out PE, we compared the conventional approach recommended by international guidelines [2], i.e., Wells score unlikely (≤ 4) combined with negative d-dimer, to a LUS-based approach, i.e., Wells score unlikely combined with negative LUS. Efficiency is a statistical parameter well suited when diagnostic strategies based on a combination of different tests, are compared; it is the result of the number of true positive and negative test results of all positive and negative test results observed [21]. Failure rate (false negative proportion) of the Wells + d-dimer approach was calculated as the number of patients with a final diagnosis of PE in the group with Wells score ≤ 4 and negative d-dimer divided by all patients in the same group, whereas failure rate of the Wells + LUS approach was calculated as the number of patients with a final diagnosis of PE in the group with Wells score ≤ 4 and negative LUS divided by all patients in the same group. Efficiency of the Wells + d-dimer approach was calculated as the number of patients with Wells score ≤ 4 and negative d-dimer divided by all included patients, whereas efficiency of the Wells + LUS approach was calculated as the number of patients with Wells score ≤ 4 and negative LUS divided by all included patients. Efficiency of the Wells + LUS + dimer approach was calculated as the number of patients with Wells score ≤ 4, negative LUS and negative dimer divided by all included patients. Finally, we calculated the diagnostic accuracy of a third strategy: Wells + LUS + dimer (wells score unlikely, negative d-dimer, and negative LUS). A p-value < 0.05 indicates statistical significance. All p-values are two sided. Calculations were performed using SPSS and STATA statistical package (version 25.0, SPSS Inc., Chicago, Illinois, and version 13.0, STATA Corp, College Station, Texas).

Among the 872 patients suspected of PE enrolled in the studies, 279 (32%) were diagnosed with PE. Additional file 1: Table S1 shows the derivation of the source studies used in this population-based analysis. The main characteristics of patients according to presence or absence of PE are shown in Table 1. Table 2 shows the leading diagnosis in all patients and in patients with and without pleuritic chest pain. D-dimer, measured in 808 patients, was positive in 340 out of 549 patients without PE (61.9%) and in 244 out of 259 with PE (94%, p < 0.001).

Five patients with wells score ≤ 4 (unlikely) and negative d-dimer had a final diagnosis of PE; 4 out of these 5 patients presented with pleuritic chest pain. Additional file 1: Table S2 shows the quality assessment of each study.

Table 3 reports true positive, false positive, true negative and false negative results of LUS in the overall population and in patients with and without pleuritic pain. Table 4 reports the derived diagnostic performance of LUS in the same groups of patients. Sensitivity of LUS for the diagnosis of PE in patients with pleuritic chest pain (81.5%, 95% CI 70–90.1) was superior to patients without pleuritic chest pain (49.5%, 95% CI 38.8–77.6, p < 0.001), while specificity was similar (95.4% for patients with pleuritic chest pain and 94.8% for patients without, p = 0.86).

Considering the databases of the two studies reporting the findings of a simplified LUS exam focused on the painful chest area, this latter technique was applied in 156 patients. This simplified LUS showed similar sensitivity and specificity when compared to the whole chest LUS examination, respectively, 78.7%, 95% CI 64.3–89.3 vs 83%, 95% CI 69.2–92.3, p = 0.48 for sensitivity and 95.4%, 95% CI 89.6–98.5 vs 94.5%, 95% CI 88.4–98, p = 1 for specificity.

To evaluate the most efficient strategy to rule-out PE, we compared the conventional approach recommended by international guidelines [2], i.e., Wells score unlikely combined with negative d-dimer measurement, and a LUS-based approach, i.e., Wells score unlikely combined with negative LUS [17]. To this aim, data are derived from 451 patients enrolled in Reissig 2001 and Nazerian 2017 with available d-dimer, while Nazerian 2014 was not included because the study enrolled only patients with high pre-test probability of PE (Wells score likely) or a positive d-dimer. The overall analysis of these 451 patients, without differentiation for pleuritic pain, showed that the failure rate of the conventional approach was significantly lower when compared to the LUS-based approach (4.1% vs 12.4% p = 0.01) (Table 5). However, considering the subgroup of 141 patients complaining of pleuritic chest pain at presentation (Table 6), the LUS-based approach had a non-significant lower failure rate and higher sensitivity than the conventional approach (respectively, 3.7% vs 6.7%, p = 0.42 and 93% vs 90.7%, p = 1), but was significantly more efficient (56.7% vs 42.5%; p = 0.02) and more specific (78.6% vs 57.1%; p < 0.001).

Considering a strategy combining Wells score with a simplified LUS examination on a single focused scan in the most painful area, the above-listed results were confirmed, with lower failure rate and higher sensitivity that did not reach statistical significance (respectively, 4.9% and 90.7%, p = 0.65and p = 1), and significantly higher efficiency and specificity (respectively, 57.4% and 78.6%, p = 0.01and p < 0.001) compared to Wells score + d-dimer.