Introduction

The novel coronavirus Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) and the coronavirus disease (COVID-19) was first reported as a cluster of cases of pneumonia in Wuhan, Hubei Province on 31 December 2019.1 COVID-19 subsequently spread at alarming rates and was declared a global pandemic by the World Health Organisation (WHO) on 11 March 2020.1 COVID-19 primarily affects the respiratory system. Chest X-ray (CXR) changes have been reported in almost 50% of cases with a sensitivity reaching 69% for abnormal findings.2, 3 Typical changes are classically bilateral mid-to-lower zone predominant patchy airspace consolidation or ground glass opacities, with pleural effusions and adenopathy as rarer findings. Severe cases often demonstrate acute respiratory distress syndrome (ARDS)-type patterns with bilateral diffuse patchy airspace changes developing over a short period of time. A CXR is an accessible, straight-forward radiological test that can help assess progression or improvement in lung parenchymal changes in patients with COVID-19, including those in an intensive care unit (ICU) or high dependency unit (HDU) setting. Despite these advantages, there are few existing severity scoring systems in literature, and no formally widely accepted CXR template for reporting COVID-19 infection.

The Brixia scoring system is an 18-point CXR scoring system developed to facilitate the clinical grading of CXR reports into five different severity categories in hospitalised patients with COVID-19 infection.4 Higher Brixia Scores are associated with higher inpatient mortality.4 This was also true in the prediction of mortality in a cohort of patients presenting to the emergency department.5 While infiltrates alone may be identified on CXR in patients suffering from COVID-19, ground glass opacities are more commonly observed (in 62.8% of cases). By definition, ground glass opacities can also reflect alveolar disease, interstitial disease or a combination of both. Consolidation was observed as the next most common lung parenchymal change in 57.7% of cases.3 The original Brixia scoring system demonstrated a correlation between higher CXR scores and mortality, however, did not examine any association between scores and likelihood of ICU admission or requirement for non-invasive ventilatory support (NIV) or intubation.

Our study aimed to devise a novel CXR severity scoring system modified from the Brixia score which considers infiltrates, ground glass opacities and consolidation. This study also examines the correlation of CXR scores with clinical outcomes including ICU admission, requirement for non-invasive ventilatory support (NIV), mortality and to assess if this system may help predict the need for intubation and ICU admission in particular patients.

Methods Patient selection

Between March 2020 and May 2020, 325 polymerase chain reaction (PCR) positive COVID-19 patients were admitted in our institution. All of these patients were enrolled in this retrospective cohort study. Age, sex and clinical outcomes including the need for NIV, ICU admission, intubation and mortality were collected for each patient. Patient demographics for those included in this study are displayed in Table 1.

Table 1. Patient demographics from those included in study Patient demographics Number of Patients 325 Age (years); mean (range) 65 (26–99) Gender, number of patients (%) Male 203 (63) Female 122 (38) Clinical outcome, number of patients (%) Non-invasive ventilation requirement 29 (9) Intubation and ICU admission 44 (14) Death 59 (18) Not requiring intubation and survived 225 (69) The chest X-ray scoring system

The most widely referenced CXR scoring system to date is the Brixia scoring system. This system divides both lungs into three discrete upper, middle and lower zones. Each lung zone is ascribed a score of 0–3, based on disease burden. A score of 0 represents no lung abnormalities, while a score of 1 indicates interstitial infiltrates, a score of 2 indicates interstitial and alveolar infiltrates with interstitial predominance and a score of 3 indicates interstitial and alveolar infiltrates with alveolar predominance.4

Our modified Brixia scoring system aimed to place more emphasis on consolidative changes in cases of COVID-19 than the original model and create a simple, easily applicable tool for non-radiologists in the clinical setting. This system was devised by four consultant radiologists, each with a minimum of 10 years’ experience working in tertiary referral centres, including two expert chest fellowship-trained radiologists. The modified Brixia scoring system utilises similar anatomical landmarks to divide lungs into three discrete zones. Table 2 details the landmarks used, with the lungs divided into six discrete zones, labelled ‘A’ to ‘F’. The modified Brixia scoring system uses different terminology to score disease severity in each zone with a score of 0 representing normal lung parenchyma, a score of 1 indicating interstitial changes only, a score of 2 indicating non-confluent airspace consolidation +/− interstitial changes and a score of 3 indicating confluent consolidation +/− interstitial changes). Confluent consolidation was defined as consolidation occupying 50% or more of any given lung zone. The breakdown of these scores is described in Table 3; Table 4 provides the original Brixia scoring system, which is provided for comparison. Figures 1 and 2 show CXRs for two patients to demonstrate this system in use.

Table 2. Details of anatomical landmarks used to divide lungs into six discrete zones for the modified Brixia scoring system Right and left upper zones: Above the inferior border of the aortic arch, that is, upper lungs. Zone A on right. Zone D on left Right and left mid zones: Below the inferior border of the aortic arch to above the inferior margin of the right inferior pulmonary vein, that is, mid-lung zone. Zone B on right. Zone E on left Right and left lower zones: Below the inferior wall of the right inferior pulmonary vein, that is, lung bases. Zone C on right. Zone F on left Table 3. Criteria used to score each lung zone for determining overall modified Brixia score Score Lung parenchymal findings 0 Normal 1 Interstitial only changes 2 Non-confluent airspace opacification +/− interstitial changes 3 Confluent consolidation +/− interstitial changes Table 4. Scoring criteria from the original Brixia scoring system Score Lung parenchymal findings 0 No lung abnormalities 1 Interstitial infiltrates 2 Interstitial and alveolar infiltrates with interstitial predominance 3 Interstitial and alveolar infiltrates with alveolar predominance Modified Brixia score of 4 in a patient with COVID 19 pneumonia. Horizontal lines are drawn following guidelines provided in Table 2 to divide each lung into three discrete zones for scoring. Modified Brixiana score of 12 in a patient with COVID-19 pneumonia. Horizontal lines are drawn following guidelines provided in Table 2 to divide each lung into three discrete zones for scoring.

As with the original system, scores from all six lung zones are added together to obtain an overall score ranging between 0 and 18.

Chest X-ray scoring

For each patient admitted with COVID-19 to our institution, the initial hospital admission CXR or the first CXR where COVID-19 infection was clinically suspected was reviewed and scored independently by two staff radiologists. In the case of multiple CXRs during admission, the CXR associated with the first clinical suspicion for COVID-19 in the CXR request was interpreted as the ‘index’ CXR. Discrepancies in scoring were resolved through a third independent reader. When available, prior imaging was reviewed to screen for pre-existing chronic background CXR changes. In the setting of chronic background CXR changes, these were not taken into account for the purpose of marking disease severity. The scoring system was also applied to all appropriate patient’s CXRs nearest to pre-intubation and the last chest X-ray prior to death where ICU and death occurred respectively.

Statistical analysis

The median scores of the ICU admission/pre-intubation CXRs was calculated and compared to their median initial hospital admission CXR scores. The same was calculated for the death outcome subgroup. For non-normal data, we used a Mann–Whitney U-test to compare groups. The interobserver variability of the scoring system was calculated based on weighted kappa, and a 95% confidence interval was determined (CI). All calculations were performed using SPSS software (P-values of <0.01 were considered statistically significant).

Results

There were a total of 325 hospitalised patients who tested positive for COVID-19 infection. Of this cohort, 203 (63%) were male and 122 (38%) female. The mean age was 65 years (range 26–99 years). In this study, 29/325 (9%) were treated with non-invasive ventilation, 44/325 (14%) were intubated and admitted to ICU and 59/325 (18%) passed away during admission. And, 225/325 (69%) did not require intubation and survived. A total of 428 CXRs were reviewed. In addition, 325 of these were admission CXRs, with a further 59 CXRs performed prior to intubation and ICU admission, and 44 CXRs taken prior to patient’s passing away. A third reviewer was required 72/425 (17%) times. The independent third reviewer resolved discrepancies for 35 admission CXRs, 13 CXRs performed prior to intubation and ICU admission and 24 CXRs prior to a patient passing away. The inter-rater reliability, Cohen’s Kappa Weighted score, was 0.7286 (substantial agreement).

Chest X-ray admission scores

The median score of all admission CXRs was 3 (IQR 0–6.5). The median score of admission CXRs of those who did not require ICU admission and survived was 1.5 (IQR 0–5). The median admission score of those intubated was 9 (IQR 4.75–12). The admission median score of those who were treated with NIV was 7.5 (IQR 1–11). For those patients that passed away, the median admission score was 4.5 (IQR 4.5–13.5). A breakdown of the distribution of modified Brixia CXR scores are provided in Figure 3 and Table 5.

Distribution of modified Brixia scores throughout three groups in study. Blue represents CXR scores for all hospitalised patients testing positive for COVID-19 on admission, red represents CXR scores for patients requiring intubation and green represents CXR scores for patients who passed away from COVID-19.

Table 5. Breakdown of median CXR modified Brixia score for all patients and subgroups Median chest X-rays score (IQR) All admissions 3 (0–6.5) Subgroups NIV admission 7.5* (1–11) ICU admission, pre-intubation 9 (4.75–12), 11.5* (9–14) RIP, last X-ray prior to death 4.5 (IQR 4.5–13.5), 7.5* (4.5–13.5) Not intubated and survived admission 1.5 (0–5) ICU, intensive care unit; NIV, non-invasive ventilation; RIP, loss of life. * P < 0.001 versus all admission median score. Change in chest X-ray scores during admission based on outcomes

On further analysis of CXR scores for the ICU group, the median scores of the pre-intubation ICU CXRs was 11.5 (IQR 9–14.125), this increased from a median admission CXR for this group of 9 (P-value 0.0088). In the group that died, the median score of the last CXR taken prior to death was 7.5 (IQR 4.5–13). This also increased from a median admission CXR score of 4.5 (P-value 0.00068) The change in Brixia score for these patient cohorts is demonstrated in Figure 4.

Change in modified Brixia scores for patients with COVID-19 requiring ICU admission and intubation from 9 to 11.5 (blue) and those who passed away from 4.5 to 7.5 (RIP, red).

Sensitivity analysis

A sensitivity/specificity analysis was performed using a receiver operating curve as a measure of the accuracy of the severity scoring system to predict intubation from all admission CXRs. With a cut-off of 6, the scoring system had a sensitivity of 77% and specificity of 73% in predicting the need for intubation (Fig. 5). This value was found to provide optimal sensitivity and specificity. A range of sensitivities and specificities for predicting intubation is given in Table 6.

Receiver operating curve for ICU patients to predict intubation.

Table 6. Range of sensitivities and specificities for scores predicting intubation and ICU admission Threshold Specificity Sensitivity 3.25 0.6 0.8 3.75 0.6 0.8 4.25 0.689 0.77 4.75 0.72 0.755 5.25 0.74 0.73 5.75 0.77 0.733 6.25 0.8 0.71 6.75 0.84 0.64 7.25 0.86 0.6 7.75 0.88 0.5 8.25 0.9 0.5 Discussion

COVID- 19 remains a serious viral respiratory illness with over 196 million confirmed cases and over 4 million deaths worldwide.6 While PCR testing is the diagnostic test for COVID-19 infection, CXRs have a sensitivity of 89.0%, specificity of 60.6% and a positive predictive value of 87.9%.7 A recent study found 69% of patients having an abnormal CXR at the time of hospital presentation and 80% of all patients had radiographic abnormalities sometime during their hospitalisation.8 With high rates of abnormal CXRs in hospitalised patients and the rapidly evolving parenchymal abnormalities, CXRs represent a useful tool in monitoring disease severity and progression. The ease of access, particularly in the setting of ICU portable rounds, makes the CXR a useful, accessible, cheap option for assessing COVID-19 patients.

There is no formal standard template or widely accepted severity system for reporting CXRs of those with COVID-19. The most referenced severity scoring tool to date is the Brixia score. Our scoring system has similar anatomical zones (A–F) but differs in the severity scoring, with scores of 2 and 3 reflecting lung parenchymal non-confluent airspace opacification +/− interstitial changes and confluent consolidation +/− interstitial changes respectively. This compares to Brixia score of 2 for interstitial and alveolar infiltrates with an interstitial predominance and a score of 3 for interstitial and alveolar infiltrates with an alveolar predominance.4 This modified scoring system was devised to address the more common ground glass changes and consolidation changes seen in 62.8% and 57.7% of CXRs infected with COVID-19 respectively.3 These changes and resulting new severity scoring system was also based on the authors own experience of hospitalised COVID-19 patients with consolidation being the predominant pattern in the CXR seen as opposed to simply interstitial abnormalities. The authors also felt that consolidation would be a more user friendly marker for non-radiologists applying a severity score to this population.

Prior papers have shown that higher Brixia scores have been linked to mortality.4, 5 Our scoring system also illustrates this. In addition, our scoring system demonstrates higher scores on admission CXRs in those with poorer outcomes, that is, patients requiring NIV, intubation and ICU admission and death. While sicker patient cohorts may already be announced by laboratory findings and clinical parameters, there may be a disconnect between a patient’s clinical picture and CXR findings at a given point in time. Therefore, CXRs may highlight patients that might require higher levels of respiratory support early during their admission and further aid clinical decision making.

Patients that were intubated and admitted to ICU had an increase in their baseline admission scores, from 9 to 11.5 (P < 0.01) at the time of intubation and this was also seen in the subgroup who passed away during their hospitalisation, with a median of 4.5 on admission rising to 7.5 prior to death (P < 0.01). This increase in CXR scores from baseline correlated with poorer clinical outcomes. Lower scores in the patient group who passed away compared to those who were admitted to ICU are attributed to higher levels of palliation and, thus, were less likely to have serial radiographs noting the extent of progressive changes compared to that of the ICU group.

In this patient population, and particularly for the use of allocating resources, a tool to predict ICU admission is invaluable. In fact, Balbi et al.5 showed that the percentage of lung involvement and PaO2/FiO2 ratio were significant predictors of the need for ventilatory support. In this retrospective study, with a simple CXR scoring system, we were able to predict the need of intubation with a sensitivity of 77% and specificity of 73%. Previous studies have demonstrated Brixia scores to correlate with patient mortality but failed to establish a relationship between scores and likelihood for need for ICU admission and intubation.

In terms of limitations of this study, the readers were not blinded to outcome when reading the subgroup scoring on the pre-intubation and last CXR prior to death which could introduce a source of bias. We do not have the time point when patients went on NIV for respiratory support and, thus, the results of the NIV group may lag behind when they actually when on NIV during their admission. Co-morbidities and background chronic CXR changes were noted but not included as a positive score for severity, only new changes were included. The quality of CXRs varied with AP and portable CXRs often being of poorer quality. It should also be known that the cause of death was not reported and many of these patients may have died due to underlying conditions with COVID-19 infection being a contributor to death.

In conclusion, we have developed a modified Brixia CXR scoring system for use in the assessment of patients presenting with COVID-19 pneumonia. Higher scores are associated with intubation, need for non-invasive ventilation and death. This scoring system is a simple tool that may help predict patients requiring intubation.

Ethical approval

Ethical approval was granted from the local internal ethics committee in accordance with ethical standards as laid down by the latest amendment of the 1964 Declaration of Helsinki. For this type of study, formal consent is not required.