Introduction

COVID-19 is associated with the development of acute respiratory distress syndrome (ARDS), intensive care unit admission, multi-organ failure, and death, especially in the elderly [Relationship between LUS-score and echocardiographic parameters

The relationships between LUS-score and conventional echocardiographic left and right ventricular parameters were assessed with univariable linear regression models. High LUS-score was not associated with elevated TR gradient (stand. β.coefficient: 0.16, P = 0.051), elevated E/E’ (stand. β.coefficient: 0.00, P = 0.99), reduced LVEF (stand. β.coefficient: − 0.03, P = 0.74) or reduced TAPSE (stand. β.coefficient: − 0.11, P = 0.14).

Relationship between LUS findings and VTE

Uni- and multivariable Cox and logistic regression models investigating the association between LUS findings (number of B-lines, B-line score, and LUS-score) and VTE are listed in Table 2. In univariable Cox (only VTE cases following the LUS examination) and logistic (all cases of VTE) regressions, an increasing number of B-lines, positive B-line score, and LUS-score were all significant markers of outcome.

Table 2 The relationship between VTE and LUS findings
Full size table

Number of B-lines was significantly associated with higher risk of VTE in multivariable Cox regression (HR = 1.14, 95% CI (1.03, 1.26), P = 0.016, per 1 B-line increase). This was also the case in multivariable logistic regression (OR = 1.09, 95% CI (1.01, 1.19), P = 0.038, per 1 B-line increase). B-line score also remained a significant marker of VTE during hospitalization in both multivariable Cox and logistic regression models (Cox regression: (HR = 9.79, 95% CI (1.87, 51.35), P = 0.007; logistic regression: OR = 6.10, 95% CI (1.69, 22.05), P = 0.006). Finally, LUS-score remained independently associated with VTE during hospitalization in the fully adjusted models (Cox regression: HR = 1.51, 95% CI (1.10, 2.07), P = 0.010 per 1-point increase; logistic regression: OR = 1.39, 95% CI (1.09, 1.79), P = 0.009, per 1-point increase). The associations between LUS findings and VTE are illustrated in Fig. 2.

Fig. 2
figure 2

Association between LUS findings and future VTE during hospitalization. Displaying the unadjusted incidence rates of VTE during hospitalization for COVID-19 (with 95% confidence intervals) for the population in relation to total number of B-lines (A) and LUS-score (B). LUS lung ultrasonography, VTE venous thromboembolic event

Full size image

Three sensitivity analyses were conducted using the same multivariable model. B-line score and LUS-score both remained significantly associated with the outcome in all sensitivity analyses. Meanwhile, the total number of B-lines only remained significant in logistic regression sensitivity analysis 2 and 3, (Table 2).

Optimal cutoff values for total number of B-lines and LUS-score were found to be, respectively, ≥ 15 B-lines and a LUS-score ≥ 5. The cutoff value for total number of B-lines demonstrated a sensitivity of 58.8% (IQR: 32.9; 81.6) and specificity of 76.1% (69.6; 81.9). Meanwhile, a positive B-line score had a sensitivity of 52.9% (IQR: 27.8; 77) and specificity of 82.2% (IQR: 76.2; 87.3). Finally, a LUS-score ≥ 5 had a sensitivity and specificity of, respectively, 58.8% (IQR: 32.9; 81.6) and specificity of 76.1% (69.6; 81.9).

Discussion

The present study was designed with the objective of assessing the association between lung ultrasound findings and development of VTE in hospitalized COVID-19 patients. We particularly aimed to assess the association between LUS findings and development of VTE, as several studies have coupled the severity of the SARS-CoV-2 infection with LUS findings [28, 29]. One retrospective cohort study investigated whether LUS could be used to detect COVID-19 in 42 patients with COVID-19 and 24 controls without COVID-19. In that study, the B-line score had a similar sensitivity to detect COVID-19 than chest X-ray and CT and improved specificity compared to chest radiograph [29]. However, the sample size was quite small, and they did not investigate how LUS findings were associated with the development of VTE. The results described in a systematic review[28] looking at the utility of LUS in COVID-19 suggest that LUS findings are very common in COVID-19 and that their mere presence in themselves may help detect COVID-19 but are not optimal for risk stratification. Rather, it seems the cumulative amount of LUS findings (B-lines, subpleural and lobar consolidations) may be more useful as a tool for risk stratification in hospitalized patients with COVID-19.

Using only the total number of B-lines may be inadequate as we observed many cases of confluent B-lines and subpleural and lobar consolidations, making it difficult to count the number of B-lines correctly. It seems the LUS findings in COVID-19 are numerous, and we, therefore, suggest a scoring system of the kind presented in this study to account for both the number of zones affected and the possibility of multiple LUS findings per zone. A scoring system like this requires external validation, but we hope that the present study can be hypothesis generating for future LUS studies in COVID-19 and help in elucidating the prognostic potential of LUS in COVID-19.

While we in the present study have found that LUS findings were associated with VTE, the study cannot answer the question of causality as there is a potential risk of reverse causation. That is, it may be that our LUS findings are present due to VTE instead of LUS findings occurring prior to later development of VTE. We used both Cox proportional hazard regression with the outcome being restricted to VTE diagnosed after the LUS examination and logistic regression with the outcome being development of VTE during the entire hospitalization as to observe possible differences that could be due to reverse causation. However, regardless of the method used to analyze the data, LUS findings were significantly associated with the outcome in both scenarios and remained so even after adjusting for possible confounding parameters.

In the present study, we observe that the patients who developed VTE had a higher number of pathological LUS findings. What the LUS detects is increased density of the lung parenchyma most likely due to inflammation related to pneumonitis rather than lung infarction due to emboli which is usually seen as triangular, hypoechoic, subpleural consolidations. Though, diffuse pulmonary intravascular coagulopathy also adds to the increased density of the parenchyma. Our findings suggest that patients who develop VTE have a more severe pneumonitis and probably also increased endotheliitis. Thus, the risk of develo** VTE is related to the clinical stage and severity of the disease and not to the extent of the pathological findings itself.

We used an 8-zone model as has previously been recommended [30] in our study. The lower lateral zones are, however, known to be difficult to obtain, in patients with prevalent respiratory failure in the semi-recumbent position. We, therefore, carried out sensitivity analyses to make sure that our results were not driven by inadequate lower lateral image clips in patients with the worst respiratory function. Still, the LUS findings remained significantly associated with the outcome in the majority of sensitivity analyses underscoring that our findings were not driven by differences in missing LUS zones. As the association between LUS findings and VTE remained significant in a 6-zone model, a lower number of zones may suffice in the clinical setting. However, further studies are needed to confirm this observation.

Strengths and limitations

This study is limited by a relatively small sample size. However, we used a multi-center design in which we included patients from all hospitals (8 hospitals) in Eastern Denmark. The multi-center design further strengthens our results as it makes our sample population much more representative of the general population. An important limitation to this study is that the observative design meant that not all participants had a CT angiography performed and thus some of the patients in the non-VTE group may actually have had an undetected VTE. As many of the patients had severe respiratory distress, the LUS examination was at times difficult to perform, especially in the lower lateral zones. Due to this, some of the patients had inadequate lower lateral zones for LUS analysis. In addition, many of the patients were too ill to collaborate to be in a seated position. All of this meant that we chose not to use a 12-zone LUS model which would otherwise have allowed us to assess the posterior zones and instead chose to only assess 8 zones (frontal and lateral zones) which limits our findings although the 8-zone model has been successfully used in several other studies [31, 32]. However, several sensitivity analyses were carried out which confirmed our findings in the primary analyses. A low-frequency probe was used for the study; however, a higher frequency probe would have improved resolution and may have aided in identifying even smaller subpleural consolidations which may have been missed. Due to the relatively small sample size, the number of events was low. However, our results remained stable in several different multivariable regression models and sensitivity models. Lastly, an important limitation is the variability in days from admission to LUS examination. A small team of investigators visiting two of the eight sites each day meant that each site was visited twice per week. Due to this approach, the median time from admission to ultrasound examination was 4 days, and as a result, a subset of our cohort probably represents a subset of COVID-19 patients requiring a longer hospital stay. This limits the generalizability of our results as they most likely reflect patients requiring longer hospital stays. The LUS investigators were experts in ultrasound and physicians without LUS experience may demonstrate poorer diagnostic performance. However, acquisition of LUS images is fairly simple and B-lines are easily recognized.

Conclusion

In hospitalized COVID-19 patients, pathological LUS findings were common, and the total number of B-lines, B-line score, and LUS-score were all associated with VTE. These findings indicate that the LUS examination may be useful in risk stratification and the clinical management of COVID-19. These findings should be considered hypothesis generating.