Comparison of lung ultrasound scoring systems for the prognosis of COVID‐19 in the emergency department: An international prospective cohort study

The purpose of this study was to evaluate whether the lung ultrasound (LUS) scores applied to an international cohort of patients presenting to the emergency department (ED) with suspected COVID‐19, and subsequently admitted with proven disease, could prognosticate clinical outcomes.


Introduction
The SARS-CoV-2 (COVID- 19) global pandemic has had a profound impact, resulting in an estimated 18 million deaths worldwide by the end of 2021. 1 Point-of-care ultrasound in the emergency department (ED) has been valuable in many parts of the world during the pandemic as it is readily available, easy to sanitise and portable, and can immediately supply clinical information. 2,3In particular, the utility of lung ultrasound (LUS) for the diagnosis and prognosis of COVID-19 pneumonia for patients presenting to the ED has been explored. 4,5owever, while the LUS scores have been developed for assessing the severity of COVID-19, [6][7][8][9] there have not been any comparative studies for their prognostic value on an international cohort of patients.
Lung ultrasound is known to be more sensitive for the detection of lung pathology than plain X-ray and is comparable to computed tomography (CT), particularly for peripheral or pleural lesions. 10Therefore, LUS is thought to be ideal for the evaluation of COVID-19 pneumonia, which mainly affects the peripheral lung parenchyma, interstitium and pleura. 11It has grown in favour during the pandemic due to its relative availability, portability, rapidity and affordability, in comparison with other imaging modalities, such as CT. 12 Although chest CT gives arguably more comprehensive images, the infection control requirements when scanning COVID-19-positive patients could lead to decreased efficiency of the CT scanner throughput and increased risk of nosocomial infection.The main utility of LUS for COVID-19 has been proposed for the diagnosis and prognosis of the disease, with the LUS scores devised to risk stratify the severity of lung disease. 4,5ung ultrasound has been useful for the evaluation of COVID-19 pneumonia, given that most of the findings are peripheral based. 13,14Unfortunately, the diagnostic utility of LUS for COVID-19 has been limited by non-specific findings, which overlap with other viral respiratory infections. 5,157][8][9] These LUS scores have been used to stratify the severity of disease, to assist with disposition and prognostication.
Although LUS is unlikely to replace nucleic acid [reverse transcriptase polymerase chain reaction (RT-PCR)] testing for definitive diagnosis, 15,16 it may have utility in predicting clinical deterioration and indicating the need for ventilatory support.If proven to accurately predict clinical deterioration, LUS findings could be integrated with clinical findings at the time of ED presentation, to triage patients to either outpatient management (low-risk features), ward admission (moderate-risk features) or intensive care unit (ICU) admission (high-risk features). 17This could potentially impact patient outcomes, resource allocation and departmental flow, particularly in times of crisis and overcrowding.
The objective of this study was to evaluate whether the LUS scores of an international cohort of patients presenting to the ED with suspected COVID-19 could prognosticate invasive ventilation, ICU admission or death, when admitted to hospital with proven disease.

Study design
This was an international, multicentre, prospective, observational cohort study of patients presenting to ED who were admitted to hospital with a positive RT-PCR test for COVID-19, received LUS and were followed up for the composite outcome of intubation, ICU or high-dependency unit (HDU) admission or death.The study was approved by the Monash Health Human Research Ethics Committee (RES-20-0000-284A) with each international site obtaining local approval.The study was prospectively registered with the Australian and New Zealand Clinical Trials Registry (ACTRN12620000734965) and has been reported according to the Strengthening the Reporting of Observational Studies in Epidemiology statement. 18

Study setting and population
The study was conducted across participating international sites, which were tertiary referral centres with an annual census of adult patients ranging between 70,000 and 100,000 located in the United States, Canada and Australia.The sites included the Buffalo General Hospital (Buffalo, NY, USA), Ottawa Hospital (Ottawa, Ontario, Canada), Health Sciences Centre (Winnipeg, Manitoba, Canada), London Health Sciences Centre (London, Ontario, Canada), Alfred Hospital (Melbourne, Victoria, Australia) and Gold Coast University Hospital (Gold Coast, Queensland, Australia).The study population included patients aged 18 years or older who presented to the ED and were admitted with COVID-19 confirmed on RT-PCR testing.Patients were excluded if they were deemed to be for palliation, had pre-existing lung disease (pneumonectomy, malignancy, pleurodesis and interstitial lung disease) or tested negative for COVID-19.

Study procedures
Sites asynchronously commenced in the study, with patients then recruited as a convenience sample at each site.The clinicians enrolling patients across all sites included both emergency physicians (n = 42) and senior trainees, that is trainees in their final years of training (n = 16), who had been credentialed within their own institute to perform LUS, which included at least 50 prior LUS scans.These clinicians screened potentially eligible patients when working on a clinical shift.Patients were approached and provided with written information regarding the study with the opportunity to ask questions before electronic consent was obtained, as required by the local ethics committee.Machines were disinfected, and LUS was performed in full personal protective equipment as per each participating site's local policy.Clinicians performing the LUS were blinded to any prior imaging, including X-ray or CT.Findings were documented in the medical records, and data were later entered into the REDCapâ database (Vanderbilt University, Nashville, TN, USA).Images were quality-assured at each site by the ultrasound site lead.LUS findings were entered as per findings in each zone, with the LUS scores later calculated by the lead author who was blinded to their outcomes.

Intervention
Utilising a point-of-care ultrasound machine, a 12-zone LUS protocol was performed 19,20 (Figure 1), unless the patient could only be scanned anteriorly, using a curvilinear or phased array probe with a 'lung' preset (artefact reduction features, such as tissue harmonic imaging or cross-beam enhancement, disabled), a depth set to at least 8-10 cm and the focus set to the pleural line.A 3-to 4-second cine loop was obtained in each region.Additional images could be recorded, including the use of a high-frequency linear probe, at the discretion of the ED clinician.Clinicians performing the scan documented pleural findings, including absence of lung sliding, pleural line abnormalities and pleural effusions, and lung findings, including Blines (number/distribution) and consolidation (subpleural < 1 cm, >1 cm or hepatisation).The LUS scores were calculated according to the previously published systems, selected by the authors to represent the main variety of protocols, including two 12-zone protocols giving 0-36 total scores (referred to as the 'de Alencar score' 9 and the 'CLUE score' 6 ), 12-zone protocol with lung and pleural findings giving a 0-60 total score ('Ji score' 8 ) and 11-zone protocol giving a 0-33 total score ('Tung-Chen score' 7 ).In all cases, higher scores denote more significant changes.

Outcomes
The primary composite outcome was the need for endotracheal intubation and ventilation, ICU or HDU (hereafter combined as ICU) admission from ED, or in-hospital mortality.Secondary outcomes included the components of the primary composite outcome as well as the composite end point of intubation, ICU admission or in-hospital mortality within 72 h of enrolment.Other secondary outcome measures included the correlation of the LUS scores with level of respiratory support and fraction of inspired oxygen at enrolment and hospital length of stay.Clinical information was obtained when LUS was performed, including days since symptom onset, comorbidities, COVID-19 vaccination status, heart rate, respiratory rate, blood pressure, oxygen saturation and any oxygen delivery (mode/ fraction of inspired oxygen).Additionally, findings of any chest X-ray or CT scan performed on ED presentation were recorded, as reported by a radiologist.

Sample size
Sample size calculations assumed that the primary composite outcome of endotracheal intubation, ICU admission or death would occur in around a third of admissions based on data Lung ultrasound scores for COVID-19 prognosis from a prior study, 21 with the assumption that disease patterns would remain constant (e.g.virulence of the virus).A sample size of 1000 COVID-19 admissions was planned, to allow the true prevalence of the composite end point to be estimated to within three percentage points either side of the estimated prevalence, at a 95% confidence interval (CI) level.Under these assumptions, the study would have greater than 80% power if a participant with a high LUS score [defined as mean + one standard deviation (SD)] had a true probability of 0.374 or greater of meeting the primary end point (Table S1).

Data analysis
The primary analysis comprised logistic regression modelling of intubation, ICU admission and death, with the LUS scores analysed individually as predictor variables.An association between the LUS score and the composite end point was determined if the 95% CI for the associated regression coefficient did not cross the null.Explained variability was described using McFadden's pseudo-R 2 .Continuous data were reported as mean (SD) or median [interquartile range (IQR)].Categorical data were presented as frequency and percentage.
Associations between the primary composite end point and patient characteristics, chest X-ray and CT findings were described as odds ratios (OR), with statistical significance testing performed using the chi-squared test for categorical predictors and logistic regression for continuous predictors.Agreement between the LUS scores was enumerated using pairwise Pearson coefficients, with agreement by lung zone analysed with linear weighted kappa.Level of respiratory support, fraction of inspired oxygen and hospital length of stay were analysed using ordered logistic regression, linear regression and negative binomial regression, respectively.
Missing data were assumed to be missing at random.For regression analysis modelling, missing data were multiplyimputed using ordered logistic regression, with LUS values for other lung zones for that participant as predictors.Complete case analysis and single imputation using the mean of nonmissing lung zone scores for that participant were also reported.Formal adjustment of CIs for multiplicity was not performed.Analyses were performed using Stata V17.0 (StataCorp, College Station, TX, USA).

Primary outcome
Complete data were available for all lung zones for 118 participants, including 19 who met the primary outcome and 99 who did not.Summary statistics for the de Alencar, CLUE, Ji and Tung-Chen scores are shown in Table 2 and Figure 2, including complete case and single imputation analyses.Of the 11 participants, eight were with missing data only from the posterior zones (R5, R6, L5 or L6) and the rest with missing data from the posterior and lateral zones, allowing robust multiple imputation for these participants.
Logistic regression analysis findings of the primary outcome by the LUS score are shown in Table 3.Following multiple imputation of missing data, no association was seen between increased the LUS score and the composite primary end point for the de Alencar score (OR = 1.04; 95% CI: 0.97-1.11;P = 0.29), the CLUE score (OR = 1.03; 95% CI: 0.96-1.10;P = 0.40), the Ji score (OR = 1.02; 95% CI: 0.97-1.07;P = 0.40) and the Tung-Chen score (OR = 1.02; 95% CI: 0.97-1.08; Figure 3).Complete case and single imputation analyses showed similar results.Pseudo-R 2 for the complete case and single imputation analyses varied from a minimum of 0.004 (Ji score, single imputation analysis) to a maximum of 0.013 (de Alencar score, complete case analysis).

Lung ultrasound scores
Pairwise Pearson's correlation coefficients for the four LUS scores are shown in Table 6.Pearson's correlation coefficient ranged between 0.82 and 0.97 for all pairs of the LUS scores, denoting a strong, positive correlation between scores.
Agreement by lung zone using linear weighted Kappa is displayed in Table 7. Substantial zone-wise agreement was seen between the de Alencar and CLUE scores, with linear weighted Kappa of 0.90.The linear weighted Kappa statistic for other pairs of the LUS scores varied between 0.39 and 0.65, consistent with moderate agreement.
Chest X-ray and CT Chest X-ray and CT findings are summarised in Table 8.Chest X-ray imaging was performed on a total of 127 of the recruited participants, whereas chest CT imaging was performed on 48 recruited participants.One or more abnormalities were found on chest X-ray in 92% of participants, with the most common abnormalities being ground-glass opacities (GGO; 57%) and consolidation (20%).No statistically significant associations were found between chest X-ray findings and the primary outcome.One or more abnormalities were found on chest CT imaging in 98% of participants who underwent this modality, most commonly GGO (81%), consolidation (19%) and mediastinal lymphadenopathy (17%); an association was seen between bullous changes (0% vs. 40%; P < 0.001) and the primary outcome of intubation, ICU admission or death.

Discussion
In an international cohort of patients admitted from ED with COVID-19, a representation of published LUS scores did not predict invasive ventilation, ICU admission or death.However, the different LUS scoring systems correlated reasonably well with each other, and scores were in keeping with other outcomes, such as the level of respiratory support and duration of hospital admission.
Invasive ventilation and death due to COVID-19 is predominately related to the development of acute respiratory distress syndrome (ARDS), which typically occurs within a week of contracting the virus. 22,23ARDS is characterised by diffuse alveolar damage, which manifests as hypoxia and bilateral infiltrates on chest imaging. 24,258][9]26 Therefore, the timing of LUS being performed during the patient's hospital journey is critically important for the purposes of prognostication.
Two studies performed LUS on patients admitted to hospital with COVID-19 at around 1-week postadmission that predicted invasive ventilation, ICU admission and death. 4,8These were both single-centre, prospective cohort studies, with Ji et al.March 2020 with the LUS scores > 12/60 (12-region protocol; lung and pleura findings) and Trias-Sabria et al. 4 recruiting 36 participants in Barcelona, Spain, in April 2020 with the LUS scores ≥ 24/36 (12-region protocol), that predicted ICU admission, invasive ventilation and death.However, both of these studies may have been biased by being a selected cohort of admitted patients with known COVID-19 with either wellestablished pneumonia on non-invasive respiratory support or admitted for other reasons with incidental mild disease.It is also unclear whether the addition of pleura findings adds value to the LUS severity score. 13n contrast, our cohort consisted of undifferentiated patients who presented to the ED with respiratory symptoms suggestive of potential COVID-19 and were then admitted to hospital.By excluding patients who were discharged from the ED, we removed the potential bias from having an over-representation of patients with mild disease and low scores.The multicentre, prospective cohort study by Sun et al. 27 in February to March 2020 recruited 402 patients in Wuhan, China, who presented to ED with COVID-19, had the LUS scores > 15/36 (12-region protocol) that predicted mortality, but may have been biased by including patients who were discharged.Conversely, two studies were similar to our eligibility criteria, the single-centre, prospective cohort study by de Alencar et al. 9 from March to May 2020 demonstrated that 180 patients in Brazil, with the LUS scores > 18/36 (12-region protocol), and the multicentre, prospective cohort study by Tung-Chen et al. 7 from March to September 2020 in Spain, with the LUS scores > 10, in combination with other markers, were both predictive of mortality.
Compared with these earlier studies performed at the start of the pandemic, the negative outcome of our study could reflect the changing scenario of the COVID-19 pandemic, with the majority of patients recruited in early 2021.9][30] This was further supported in our data in that other imaging modalities, such as X-ray and CT scans, were also not able to prognosticate outcomes. 31However, the LUS scoring systems did correlate with disease severity in terms of level of respiratory support, estimated fraction of inspired oxygen and hospital length of stay.These findings suggest that, although the LUS scores may accurately describe pathological lung changes, their prognostic value appears to be influenced by other patient, disease and system factors that contribute to these outcomes, which have evolved over time.
The initial study design intended a much larger sample size, to allow for an accurate estimate of the true prevalence of the composite end point.Unfortunately, the intended sample size was not achieved due to a lack of funding and increased clinical pressures associated with the COVID-19 pandemic, raising the concern that the study may be underpowered to detect a true association.Exploratory post hoc power analysis was used to estimate the achieved power for a range of possible effect sizes, based upon the achieved sample size and the observed    with the severity of COVID-19 disease, including current respiratory support requirements and hospital length of stay, suggesting that the change in disease patterns and treatments may have cofounded the primary outcome.Further work is required to establish the use of the LUS scores in COVID-19 disease prognostication, including the determination and prospective validation of a standardised scoring system, as well as the integration of these scores into management pathways for patients with pandemic respiratory viral disease.

Figure 3 :
Figure 3: Odds ratio for logistic regression of ICU/HDU admission, intubation or death by lung ultrasound score.HDU, high-dependency unit; ICU, intensive care unit.

Table 1 :
Patient characteristics by composite outcome of death, ICU or HDU admission, or intubation.
ª 2023 The Authors.Australasian Journal of Ultrasound in Medicine published by John Wiley & Sons Australia, Ltd on behalf of Australasian Society for Ultrasound in Medicine.AJUM May 2024 27 (2) 79 Lung ultrasound scores for COVID-19 prognosis

Table 1 .
Continued a Early: January to June, Late: July to December of the specified year.b COVID-19 vaccination data missing for 24 patients in the 'No death, ICU or HDU admission or intubation' group and for 12 patients in the 'Death, ICU or HDU admission or intubation' group.

Table 2 :
Summary statistics by COVID-19 lung ultrasound score; complete cases and single imputation by mean score.
80 AJUM May 2024 27 (2) ª 2023 The Authors.Australasian Journal of Ultrasound in Medicine published by John Wiley & Sons Australia, Ltd on behalf of Australasian Society for Ultrasound in Medicine.
8recruiting 280 participants in Wuhan, China, in January to Lung ultrasound score by composite outcome of death, intubation or ICU/HDU admission; complete cases and single imputation by mean score.HDU, high-dependency unit; ICU, intensive care unit.ª 2023 The Authors.Australasian Journal of Ultrasound in Medicine published by John Wiley & Sons Australia, Ltd on behalf of Australasian Society for Ultrasound in Medicine.

Table 3 :
Logistic regression of composite outcome of death, intubation or ICU/HDU admission by COVID-19 lung ultrasound score; complete cases, simple imputation by mean score and multiple imputation by ordered logistic regression.

Table 4 :
Secondary outcomes and subgroup analyses by COVID-19 lung ultrasound score.