Key words COVID-19 - lung ultrasound - chest - LUS score - interoperator reliability
Introduction
Early diagnosis and follow-up of pneumonia are essential in the management of patients
with COVID-19 infection. To date, the current literature does not advocate lung ultrasound
(LUS) for the diagnosis of COVID-19. However, its role in monitoring lung conditions
is well known, especially in the critical care setting [1 ].
Chest radiography (CXR) is a less sensitive modality for the detection of COVID-19
lung disease compared to chest computed tomography (CT), with a reported baseline
CXR sensitivity of 69 % [2 ]. The most commonly reported CXR and CT findings of COVID-19 include lung consolidation
and ground glass opacities. Ground glass densities observed on CT may often have a
correlate that is extremely difficult to detect on CXR [3 ]. Furthermore, CT has been shown not only to be more sensitive than reverse transcriptase
polymerase chain reaction (RT-PCR) with a sensitivity of 98 % vs. 71 % in the diagnosis
of COVID-19, but also to correlate with disease progression and recovery [4 ]
[5 ]
[6 ].
Patients typically have bilateral multilobar ground-glass opacities (GGO) with a peripheral
or posterior distribution, and lesions detected on CT usually progress, with the greatest
severity of radiologic findings occurring around day 10 of symptom onset [7 ]. Despite its utility, CT is not readily available in many settings with limited
resources. In addition, the need for disinfection of the CT scanners after they are
used for suspected or confirmed COVID-19-positive patients results in delays in patient
care. Moreover, the risk related to radiological exposure has to be considered, especially
in view of these patients potentially needing repeated scans for clinical decision-making.
COVID-19 interstitial diffuse pneumonia typically involves the lung periphery, a feature
that makes it particularly suitable for ultrasound investigation [8 ]. LUS examination allows rapid and reliable diagnosis of lung consolidation, pleural
effusion, and interstitial–alveolar syndrome, and it is currently included in a consensus
statement on core competencies in point-of-critical-care ultrasound [9 ].
Among the advantages of LUS, the lack of radiation, simplicity, and execution speed
have to be highlighted. Compared to XCR, LUS also appears to be more sensitive for
an early detection of interstitial syndrome. It can be performed at bedside, which
is especially useful for sequential monitoring of patients in critical settings. Moreover,
in experimental models of acute respiratory distress syndrome (ARDS), LUS has proved
capable of detecting lung lesions before the development of hypoxemia [10 ]. Furthermore, in COVID infection settings, it is important to stress that LUS involves
and exposes a minimum number of health-care workers and medical devices to suspected
or confirmed cases, thus avoiding contamination of infrastructures and nosocomial
spreading of the virus. The main limitations of LUS lie in the areas of training,
operator variability, and reliability. Ultrasound qualitative description is operator-dependent
and based on artifacts due to the presence of air in the alveolar tissue, with only
images that reach the lung surface and the pleura being detected. Moreover, the acoustic
window may be limited by the rib cage. According to previous studies, LUS must be
performed according to a systematic examination protocol [11 ]
[12 ]
[13 ].
In the setting of COVID-19 pneumonia, some typical patterns have been described in
the literature [8 ]
[9 ]
[10 ]
[11 ]
[12 ]
[13 ]
[14 ] highlighting the features of pulmonary interstitial involvement:
B-lines
in COVID-19 pneumonia are visualized as clusters of B-lines, both in separate and
coalescent forms, sometimes giving the appearance of a shining white lung with a bilateral
specific patchy distribution that can be described as a “patchwork pattern”. Mixed
with a normal A pattern nearby, different from cardiac pulmonary edema, which shows
a homogeneous increase in B-lines with gravity
Pleural line
appears typically irregular or fragmented
Small peripheral consolidations
can be frequently visualized
Minimal data are available in the literature about interobserver reliability, need
for training, and level of expertise needed in order to perform a trustworthy examination.
Similarly, there is currently no agreement as to what constitutes an experienced LUS
operator.
The reliability of counting B-lines has been previously evaluated comparing expert
versus novice operator performance and expert performance versus the performance of
a software algorithm, and κ was 0.66–0.80 (95 % CI). Intra-rater reliability has also
been evaluated, ranging from 0.82 to 0.95 using the ICC (Intraclass Correlation Coefficient)
[15 ]
[16 ]
[17 ].
When considering the identification of pleural effusions, it has been shown that young
doctors in training can be easily taught to reliably perform thoracic ultrasound to
answer specific diagnostic questions and to guide safe intervention procedures [18 ].
Despite increasing interest in the technique, LUS training methods vary among centers
and are not standardized. Based on clinical experience, it has been hypothesized that
a short and easy-to-implement training program based on 25 LUS determinations supervised
by experts would be enough for trainees without expertise [19 ].
In order to provide early treatment for paucisymptomatic patients with COVID-19 infection,
we have implemented a “fast track” pathway in our emergency department in collaboration
with infectious diseases specialists. Together with the medical examination and oxygenation
parameters test, bedside LUS was performed to confirm or exclude the presence of viral
pneumonia.
The aim of this study was to evaluate the reproducibility of LUS in assessing the
features of COVID-19 pneumonia among operators with different levels of expertise
and to investigate any training effect.
Materials and methods
Study design, population, and setting
This study was a single-center prospective study. Patients with suspected COVID-19
who were referred to our “fast track” were enrolled to evaluate the ability of LUS
to discriminate the presence of interstitial pneumonia in association with nasopharyngeal
swab and clinical evaluation.
LUS was performed by two different operators, each performing repeat measurements
on the same subject on the same day and blinded to their colleague’s results.
The “expert” operator was a clinician with more than 5 years of LUS experience, while
the “novice” operator was a resident trainee in the emergency department with no expertise
in LUS but at least 1 year of abdominal ultrasound expertise. The “novice” operator
received a brief explanation and a dedicated training course of 60 minutes on how
to perform the measurements. All operators were blinded to the results obtained by
others.
Ultrasound imaging protocol
The US scans were all performed with Esaote My Lab 7. R
The convex probe (3–5 MHz) was used, with the widest acoustic window and maximum depth
of 10 cm, with the focus on the pleural line. The exam was performed in a sitting,
lateral, or supine position. As already reported in the literature, we have been using
a systematic protocol of scanning that is rapid and practical[20 ]. In each patient 12 areas were explored once by the operator and reported in the
LUS. The areas were registered as right and left lung areas (R and L) and divided
using the anterior and posterior axillary lines, resulting in three areas per hemithorax
(anterior, lateral, and posterior), with each of them split into superior and inferior.
Each area was examined in the sagittal and axial views ([Fig. 1 ]).
Fig. 1 Lung areas: both the right lung and the left lung are divided using the anterior
and posterior axillary lines, resulting in three areas per hemithorax (anterior, lateral,
and posterior), with each of them split into superior and inferior.
All lung areas were explored, and a score was defined for each one. The semiquantitative
assessment of pulmonary aeration loss can vary between 0 and 36:
0: normal lung (A-lines) ([Fig. 2 ]);
1: non-coalescent B-lines (B-lines occupying less than 50 % of the intercostal space
in the transverse plane) ([Fig. 3 ]);
2: coalescent B-lines (B-lines occupying more than 50 % of the intercostal space)
([Fig. 4 ]);
3: consolidation > 1 cm ([Fig. 5 ]).
Fig. 2 LUS score 0, normal A line pattern.
Fig. 3 LUS score 1, non-coalescent B-lines (B-lines occupy less than 50 % of the intercostal
space in the transverse plane).
Fig. 4 LUS score 2, coalescent B-lines (B-lines occupy more than 50 % of the intercostal
space); the arrows show the pleural irregularity.
Fig. 5 LUS score 3, consolidation > 1 cm.
The sum of the 12 different lung areas represents the LUS score (LUSs).
For each operator, the whole examination lasted approximately 15 minutes.
Main outcome measurements
Interoperator variability was assessed by comparing the LUSs obtained on the same
day by each operator. To establish if a training effect was present, the interoperator
agreement in the last 5 days was compared with the agreements found in the first 5
days.
Statistical analysis
Mean, median, standard deviation (SD), interquartile range (IQR), and frequencies
were used as descriptive statistics.
Agreement between the two operators was expressed using intraclass correlation coefficients
(ICCs) for single measurements and a 95 % confidence interval (CI) for the ICC. Systematic
differences were computed by means of the paired Student t-test. Bland-Altman plots
were constructed to visualize agreement and the limits of agreement were evaluated
together with their 95 % CI (18). The repeatability coefficients were also computed.
Statistical analysis was performed using SPSS (version 13.0 for Windows) and two-tailed
P-values less than 0.05 were considered significant.
Results
96 patients were enrolled during the COVID-19 outbreak between April 17 and April
27, 2020. The median age was 46 years old (IQR 15–88); 39 males (41 %) and 57 females
(59 %). Among the 96 patients, only 11 had known pulmonary disease: 1 case of COPD,
6 cases of asthma, and 4 other lung conditions.
No patients were excluded due to a suboptimal acoustic window.
The clinical characteristics of the patient population are described in [Table 1 ].
Table 1
Sample descriptive table of the population.
all patients
gender
male
39 (41 %)
female
57 (59 %)
age
RANGE
15–88
median
46
≤ 46 yrs
50 (52 %)
> 46 yrs
46 (48 %)
chronic pulmonary diseases
no
85
yes
11
The LUSs was 0 in 58 patients (60.4 %), ranged 1–3 in 24 patients (25.0 %), and ranged
4–14 in 13 patients (13.6 %) when performed by the expert operator. The LUSs for the
novice operator was 0 in 56 patients (58.3 %), ranged 1–3 in 27 patients (28.1 %),
and ranged 4–17 in 13 patients (13.6 %).
The ICC showed excellent agreement between the expert and the novice operator (ICC
0.975; 0.962–0.983) ([Fig. 6 ]). As shown by the Bland-Altman analysis, both operators achieved a similar LUSs
in the majority of patients, with only two cases showing a difference of 3 points.
In particular, the novice operator overestimated the LUSs compared to the expert operator
by a maximum of 3 points in only two cases, 17 instead of 14 and 8 instead of 5. A
maximum underestimation of 2 points was reported by the novice operator ([Fig. 7 ]).
Fig. 6 Linear correlation between expert and novice.
Fig. 7 Bland-Altman plot representing the level of agreement between the two operators.
In 5 cases the novice operator reported a pathologic ultrasound with an LUSs of 1
compared to a negative score of the expert operator. Furthermore, the novice operator
reported 3 cases as having a negative ultrasound, while the expert operator described
an LUSs of 1 in two cases and an LUSs of 2 in one case.
Age, gender, and chronic pulmonary diseases did not influence the reported LUSs. In
particular, the ICC was 0.973 (0.950–0.986) in males and 0.976 (0.959–0.986) in females.
The ICC in younger patients (≤ 46 yrs) was comparable with that seen in older patients
(> 46 yrs): 0.965 (0.940–0.980) vs. 0.973 (0.952–0.985). When considering the influence
of pulmonary disease, the ICC of affected patients was 0.967 (0.882–0.991), comparable
with the other patients with an ICC of 0.975 (0.962–0.984). In all analyses the difference
was not statistically significant, p < 0.001.
In order to test a potential improvement in the novice operator learning curve, we
divided the study population into two subpopulations: the first half (48 patients)
and the second half (48 patients). The ICC was 0.971 (0.949–0.984) for the first subpopulation
analysis and 0.981 (0.967–0.989) for the second subpopulation analysis, showing a
relative improvement in interobserver agreement ([Table 2 ]).
Table 2
Intra-class correlation between the two operators stratified according to different
factors.
ICC (95 % CI)
expert vs. novice
p-value
all patients
0.975 (0.962–0.983)
< 0.001
gender
male
0.973 (0.950–0.986)
< 0.001
female
0.976 (0.959–0.986)
< 0.001
age
≤ 46 yrs
0.965 (0.940–0.980)
< 0.001
> 46 yrs
0.973 (0.952–0.985)
< 0.001
chronic pulmonary diseases
no
0.967 (0.882–0.991)
< 0.001
yes
0.975 (0.962–0.984)
< 0.001
learning curve
first period
0.971 (0.949–0.984)
< 0.001
second period
0.981 (0.967–0.989)
< 0.001
Discussion
The main objective of this study was to assess the performance and reproducibility
of LUS in patients suspected of having COVID-19 when conducted by operators with different
levels of expertise.
Although the diagnosis of COVID-19 is made with a nasopharyngeal swab, detection of
pulmonary involvement is essential in the context of COVID-19 infection in order to
keep patients safe. So far, CT examination has been considered the gold standard in
view of the high sensitivity and positive predictive values. However, it must be taken
into consideration that this relates to the setting of a pandemic when there is a
very high “a priori” probability of disease in the presence of respiratory symptoms
[21 ].
Amidst the pandemic scenario characterized by limited technical resources, we consider
ultrasound to be a valid alternative to CT, in particular for the assessment of paucisymptomatic
patients. It is a safe and rapid method when used in conjunction with physical examination
and arterial-blood gas analysis parameters. In particular, a negative LUS examination
can allow the exclusion of pulmonary involvement with good accuracy [22 ].
Implementation of a standardized method is of utmost importance in order to use the
technique widely. Our results confirm that this is feasible and reproducible even
when performed by operators with different levels of expertise, including junior trainees
with a brief LUS course with a very rapid learning curve. Furthermore, US has the
advantage of being an easily portable system that could therefore also be used in
out-of-hospital settings.
Demographic features did not influence the reproducibility of the method.
Considering the importance of giving a positive or negative imaging result, we saw
a slight LUS discrepancy only in 8 patients. Greater discrepancies between the two
operators have been registered in LUS> 5, but these can be considered not clinically
relevant. Our study demonstrates that even a novice operator can easily exclude pulmonary
involvement of the virus.
Limitations of the study are the lack of extremely high values of the LUSs in relation
to the clinical features of our population and the lack of patients with possible
confounding factors such as chronic pulmonary disease.
Compliance with ethical standards
Ethical approval. The study protocol conforms to the ethical guidelines of the World
Medical Association Declaration of Helsinki. All patients gave their consent, and
the institutional review board of our hospital has been notified about the study.
