The main findings of this prospective, observational study assessing the ability LUS and CT to monitor pulmonary involvement in COVID-19 ICU patients are:

(1) LUSS has a moderate correlation with the CTSS for the quantification of pulmonary involvement, and a better correlation with the PaO2/FiO2 and Vd/Vt ratio than CTSS; (2) Serial LUSS is capable of detecting true changes in pulmonary involvement, whereas CTSS cannot; (3) Of the two scores, only a rise in LUSS after 2 weeks is significantly associated with mortality and detectable beyond measurement error. However, this association becomes insignificant again in the weeks afterward; (4) Addition of subpleural abnormalities did not lead to improvement of the existing LUSS.

Our results suggest LUS might be used as a substitute of chest CT for the monitoring of COVID-19 pneumonia severity in COVID-19 ICU patients. The correlation of the total LUSS and CTSS was moderate, which is comparable to previous studies [6, 7, 16]. Surprisingly, the CTSS did not correlate with PaO2/FiO2 ratio, and only correlated slightly with the Vd/Vt ratio. Even though the measurements were taken at time of CT. The LUSS correlated fairly with both. A worsening of the PaO2/FiO2 ratio over time would prompt a repeat CT in clinical practice to evaluate the progression of pulmonary involvement, but this thus seems redundant.

Although the degree of pulmonary involvement at presentation to the emergency department is associated with ICU admission and adverse events [6,7,8,9,10], previous studies failed to show an association between LUSS or CTSS at ICU admission and mortality (without it being incorporated in a composite end-point) [1, 7, 8, 10, 16, 27]. The only other prospective study assessing the prognostic value of serial LUS and CT, showed that the LUSS after 1 week did not differ significantly from the LUSS at admission, and did not show an association with mortality. However, our study showed that the LUSS might help differentiate between patients who survive or die if you follow them longer. Although the LUSS did not exceed the SDC (4.8) on a week-to-week basis, from week 2 onward the LUSS exceeded the SDC, and was significantly higher compared to admission in deceased patients. The mean difference between deceased patients and survivors was significant and also exceeded the SDC in week 2. This was not the case in weeks 3 and 4, possibly owing to the small sample size, as results did show a trend. In patients who survived, both LUSS and CTSS showed no decrease over time, stressing the discrepancy between clinical recovery and lack of image resolution [16].

Contrary to the LUSS, the CTSS seems not useful in monitoring ICU COVID-19 patients, which is at odds with the claims of the original CO-RADS protocol, which promoted the CTSS as a tool for follow-up [28]. Even though there was a statistically significant difference between patients who survived and those who died in week 1, it falls within the measurement error. Indeed, the CTSS never changed more than its SDC (3.9); neither between admission and following weeks, nor between surviving and deceased patients. This signifies that measurement errors based on interpretative differences between CTSS raters are greater than the true variation in score (based on actual pulmonary involvement changes). Thus, in clinical practice, the CTSS measurement error may suppress the true CTSS signal, i.e., a clinical CTSS change is more likely to be caused by measurement error than disease progression. This may be explained by the fact that the CTSS uses categories that are too broad, lacks responsiveness, and suffers from ceiling effects [25, 29]. In addition, the CTSS only includes ground glass opacities, omitting consolidations and fibrosis which are present later in the disease [13,14,15,16, 28]. Their incorporation might improve the CTSS. Other suggestions are increasing the number of categories of the CTSS, or using use of artificial intelligence (AI) to aid pulmonary involvement quantification [30].

Although there are other reasons to perform a CT—i.e., suspicion of a pulmonary embolism, or a super-infection—it could be argued that even for these indications CT does not have to be the first diagnostic step and the amount of CTs could at least be reduced. For instance, less invasive methods like routine screening of deep venous thrombosis by POCUS could obviate the need for further a CT pulmonary angiography [31,32,33]. Furthermore, although there is some evidence in immunocompromised patients that CT-guided broncho-alveolar lavage (BAL) results in a higher yield [34], BAL can also be performed without CT guidance [35]. In our cohort, a super-infection was suspected in 27.8% of CTs, which is comparable to what is reported in the most recent meta-analysis [36]. Since we found the posterior parts of the lungs were primarily involved when superinfection was suspected, we argue that a CT is not required to determine the optimal BAL location and these areas could be sampled empirically.

Some of the criticisms of the CTSS also apply to the LUSS, particularly the fact that its categories are too broad. However, we showed that including pleural abnormalities unfortunately did not improve the LUSS to a significant degree. First, this might be due to the fact that the LUSS already incorporates pleural abnormalities and consolidations to some degree, so by adding them again you might be counting them twice [20]. Second, we found there is more measurement error in pleural abnormality categories 1–3, than categories 0 and 4–5. This is no surprise as there is no clear definition of a ‘thickened’, ‘blurred’, or ‘irregular’ pleural line, nor is it clear when an ‘irregular’ pleura is actually caused by clear subpleural consolidation. As such clinicians will interpret these constructs in different ways [10, 17]. Moreover, it is unclear what the clinical relevance of pleural abnormalities of varying degrees is. As our results show they might not hold any additional value. In any case, universal standardization of these signs—as was done with other LUS signs—is paramount to facilitate comparison across studies [21]. All in all, we argue to keep using the existing LUSS—which has already been validated in multiple settings—and that restraint should be exercised in creating new, more complex and/or unvalidated measurement instruments [20, 25].

Maybe in the future AI can help improve LUS quantification of pulmonary involvement, and elucidate which findings have additional prognostic value. It has shown promise in a few studies, but further innovation and research is required before it will be ready for use on a larger scale [20, 37,38,39]. An important condition for AI-based quantification is that it should be directly applicable and useable at the bedside, so the benefits of the point-of-care nature of LUS are not lost.

In summary, we argue that if one would like to monitor ICU COVID-19 patients with an imaging modality, LUSS should be preferred over CTSS, at least as an initial step. Especially considering the cost, safety and time disadvantages associated with CT and the lack of its availability in many parts of the world.

First, this was a single center study with a relatively limited sample size. Still, this is the largest prospective study in consecutive COVID-19 ICU patients to investigate the ability of serial LUS and CT in monitoring and prognostication. Second, although we are a tertiary center, our case mix was reflective of the total ICU population, since ICU patients were divided across the country according to a fair-share principle. Accordingly, we believe selection bias was, therefore, minimized. Third, time between LUS and CT was always below 24 h, which limited imprecision of the correlations. Fourth, we only included patients who deteriorated or stagnated in their recovery, and did not perform a CT or LUS in patients that were recovering nor in patients in which the disease progressed to such a degree that further diagnostics and treatment were deemed no longer useful. Although, this does reflect daily practice, it would be interesting to include these groups in further research as well. Fourth, operators were blinded for CT results, reducing information bias to a minimum.