A Retrospective Cohort Study Characteristics of Chest X-Ray Findings in COVID-19 Patients that Presented to a Tertiary Care Facility in Dubai: Can a Severe Radiological Appearance Predict Clinical Outcome?

Multiple cases of pneumonia of unknown origin had been reported in Wuhan China to World Health Organization (WHO) on December 31, 2019. Following that, novel coronavirus 2019 was confirmed as the cause of those reports at the beginning of 2020. It was then renamed severe acute respiratory syndrome coronavirus (SARS-CoV-2) and on the 20th of February WHO declared the outbreak a global emergency [1]. Over 2.3 million confirmed cases and more than 160,000 deaths had been reported worldwide as of May 2020 [1].

The clinical presentations of COVID-19 infection range from mild respiratory symptoms to severe acute respiratory distress syndrome (ARDS). Chest X-ray (CXR) is a widely available diagnostic modality. As of March 2020, CXR was not considered a mandatory form of evaluation for all patients as per the American College of Radiology (ACR) [2] but is recommended for patients with clinical suspicion of moderate to severe COVID-19 while waiting for reverse transcriptase-polymerase chain reaction (RT-PCR) result [3]. A study performed by Cozzi et al. [4] in Italy demonstrated a CXR sensitivity of 67.1% for COVID-19 infection.

COVID-19 has been observed to mainly involve the respiratory system; many studies have described the typical features of the CXR. Globally, Cozzi et al. [4], Vancheri et al. [5], both in Italy, Weinstock et al. [6] in New York, Wong et al. [7] in Hong Kong have all demonstrated similar predominant patterns of peripheral, lower zone, involving both lung infiltrates which are mostly ground glass and consolidations.

The radiographic assessment of lung edema (RALE) score for CXR severity in ARDS had been developed by Warren et al. [8], with higher scores correlating well with worse outcomes. Cozzi et al. [4] and Wong et al. [7] have employed RALE on COVID-19 patient CXRs, concluding higher scores correlate with an increased risk of being admitted to the intensive care unit (ICU) [4, 8]. Likewise, there are various other established scoring systems like the lung severity score used in a study done by Toussie et al. [9]. These studies propose the use of severity scores for CXR as a predictor of severe outcomes like intubation and ICU admission [4, 9].

In this research, we will trace the characteristic features of CXR imaging obtained from COVID-19 patients who presented to the emergency department (ED) in a tertiary care facility. The findings will be classified by the RALE scoring system by two experienced radiologists ranging from normal, mild, moderate, and severe. Each category will be correlated with the risk of ICU admission, intubation as primary outcomes. Additionally, mortality will be studied in the severe category of patients.

The healthcare system in Dubai is divided into the private and government sector. There are tertiary care hospitals along with primary care health centers. The study will provide data for the regional health sector in assessing the use of CXR in risk stratification and triaging patients to efficiently use resources and divert the flow of patients.

Patients who presented to the ED between March 2020 and August 2020 suspected of COVID-19 were retrospectively collected and reviewed. The inclusion criteria addressed symptomatic patients, aged 18 years and above who met the COVID-19 case definition [10]. All patients who tested COVID-19 positive by nasopharyngeal RT-PCR swab (Cepheid Xpert Xpress SARS-CoV-2, GeneXpert^® Systems, Sunnyvale, CA, USA), had done a CXR (portable or departmental) at presentation, and were admitted were included in the study.

Patients below 18 years, pregnant, asymptomatic, presented with other clinical conditions and incidentally tested positive for COVID-19, transferred to another facility, or had missing results of RT-PCR or CXR in the file, were excluded from the study. 3,035 confirmed COVID-19 patients presented to the ED during the time period. From these, 404 patients fulfilled the inclusion and exclusion criteria (Fig. 1). We calculated 350 patients to sufficiently power the study at 80% to show a significant difference between the RALE categories with a p value of 0.05.

Image acquisition and analysis:

CXRs of patients who met the inclusion criteria were taken from the picture archiving and communication system and formatted to provide a digital radiograph. All patients had antero-posterior projection taken either in the department or by portable machine. All digital radiographs were compiled and divided into two separate USB flash drives and were blinded for the radiologists by deleting the hospital’s unique ID, name, demographics, and study comments.

Two consultant radiologists were given 1 USB each containing 202 radiographs. They reevaluated the CXRs retrospectively in order to describe the characteristics which included features of the type of infiltrates, peripheral or perihilar, zones involved, the dominance of lung involvement, and extra findings such as pneumothorax or pleural effusions.

Furthermore, each radiograph was divided into four quadrants, each quadrant was assessed to determine the extent of consolidation and density of the alveolar opacity to give a RALE score ranging from normal, mild, moderate, and severe (Fig. 2). The method of calculation was derived from the proposed validated method by Warren et al. [8] which was also used by Cozzi et al. [4] and the division to the category was left to the discretion of the radiologist. The RALE score was used as it is validated and has been used adequately for COVID-19 CXR’s as demonstrated by Cozzi et al. [4] and Ho Yuen et al. [4, 8]. The participating radiologists were familiar with the tool and it is used for scoring the CXR’s.

The electronic medical record system was utilized to collect patient information by the Emergency Medicine Resident investigators. A data tool was organized which included demographics and comorbidities, vitals which were initial respiratory rate (RR) and oxygen saturation (SPO2), ICU admission, intubation from the ED or during the hospital stay, and in-hospital mortality.

SPSS system was used for the analysis of the data. χ² test with a p value determined as <0.05 was used to establish whether there was a significant difference between the RALE categories and the rate of intubation and ICU admission from emergency or during the hospital stay.

The sample’s age ranged from 20 years to 88 years with a median of 48 years. 85% were males and 15% were females. Indians, Pakistanis, and UAE cumulatively accounted for 64% of the population.

Diabetes Mellitus was the most prevalent comorbidity, observed in 40% of the patients. Hypertension followed by 26%, obesity 9%, and asthma/COPD 3% (Table 1).

The initial RT-PCR swab resulted positive in 82% of the sample, while the remaining was positive by the second swab. The radiologists categorized the CXRs taken in the ED using the RALE score into normal, mild, moderate, and severe. From the 404 CXRs, the majority were mild accounting for 194 CXRs (48%), 79 were normal (20%), 95 were moderate (23%), and 36 severe CXRs (9%). To describe the most prominent features suggestive of COVID-19 infection in the sample population, features like the type of infiltrates (ground-glass opacity (GGO), consolidation, or both), perihilar versus peripheral, lung involvement, and zone dominance were studied (Table 2).

From the CXRs which had demonstrated signs of infection, 201 had GGO (50%), 50 had consolidations (12%), and 70 CXRs had both (17%). Characteristically, severe CXR’s predominantly demonstrated consolidations or both, opacities with consolidations, moderate largely revealed both, opacities with consolidations and mild mainly GGO.

Abnormal CXRs largely had peripheral infiltrates in 253 patients (63%). The infiltrates were seen bilaterally in 252 of CXRs (62%) mostly in the lower and mid-zone which were 183 (45%) and 120 (30%) CXRs, respectively.

Furthermore, we used the RALE score to observe clinically important endpoints like ICU admission and intubation. The mean time to intubation was 2 days, minimum of less than 1 day and a maximum of 27 days.

We studied the distribution of patients taken to ICU or intubated in ED versus during the hospital stay in each category. They were considered as separate entities to observe which category had more frequent and earlier intervention (Fig. 3, Table 3).

ICU admission and intubation from the ED were done for 29 patients equivalent to 7% of the sample, whereas intervention during hospital stay was noted in 148 patients accounting for 37% of the patients. From the 29 patients intubated with ICU care from the ED, 12 had severe (41%), 9 moderate (31%), 5 mild (17%), and 3 normal (10%) CXRs. On the other hand from the 148 patients taken to ICU after intubation during the hospital stay, 69 had mild (47%), 50 moderate (34%), 10 normal (7%), and 19 severe (13%) CXRs.

These differences in ICU admission and intubation from the ED between the RALE categories were found to be clinically significant using χ²X2 = 45.2 (df 3, n = 404) p value <0.001. Similarly, the intubation during hospital admission had a clinically significant difference between the categories as seen by the χ²X2 = 36.4 (df 3, n = 404) p value <0.001.

Sub-analysis of mortality revealed that the severe RALE score patients had a higher proportion of death as an outcome than any other category. 67% of patients in the severe category experienced mortality.

Initial vital parameters namely SPO2 and RR recorded in the ED were described for each category to observe whether clinical severity and CXR correspond. RR and SPO2 in the normal category had a range of 15–35/min with median 18/min and 80–100% with median 98%, mild 16–40/min with median 19/min and 43–100% with median 95%, moderate 17–40/min with median 24/min and 30–100% with median 90%, severe 16–42/min with median 26/min and 39–98% with median 83%.

During the global pandemic, the radiological approach should be implemented as a rapid method of triaging, diagnosing, and predictor of the patient outcome at initial contact. RT-PCR was initially seen to have sensitivities ranging from 37% to 71% during the Wuhan China outbreak [11, 12], later laboratory-based evaluations revealed better performances. However, as per the Centre for Disease Control and Prevention, there are still remains constraints to the tests that yield false-negative results due to inadequate sampling, viral load, and others [13].

RT-PCR while used for diagnostics with variable sensitivity has a debatable prognostic yield. Rajyalakshmi B et al. [14] suggested low cycle threshold values of RT-PCR as one of the prognostic variables, however, studies thereafter, Martínez et al. [15] questioned the correlation.

The Fleischner Society has published thoracic imaging consensus guidelines in COVID-19 [3]. If RT-PCR is unavailable, imaging can reveal an alternative diagnosis if present or, if findings suspicious for COVID-19 are seen, can guide further workup.

CXR has been proposed as an operationally simpler and during an infectious pandemic a relatively faster aid for decision-making with the additional benefit of portability and easy disinfection of surfaces which may help reduce cross infections [2]. The ACR has advised against the use of CT as the first line for the diagnosis of COVID-19 infection, rather portable and departmental CXR remain an important resource for imaging when needed [2].

The utility of CXR in suspecting and diagnosing COVID-19 can be done based on characteristic findings seen in the majority of images. Similar to the previously published studies [4‒7] our study revealed predominant features of bilateral GGOs seen mostly in the peripheral and lower to mid-zones (Table 2).

Additionally in our study, we looked at the distribution of GGO, consolidation, or both in the different RALE score categories. Most GGO were demonstrated in the mild category with a decreasing incidence in the moderate and severe. Both opacities and consolidations were seen predominantly in moderate CXRs. On the other extreme consolidations were found roughly equally in all categories, importantly all severe RALE CXRs had consolidations, with some having additional opacities as well.

While it can be argued that these findings are not specific to COVID-19 and initial radiographs may be normal as demonstrated in 20% of our sample, the use of CXR for diagnostics could be of value in patients suspected to have COVID-19 with initial false-negative RT-PCR which accounted for 18% of the 404 patients.

The RALE score has been published by Warren et al. [8] as a reliable indicator of worse outcomes in ARDS, with a higher RALE score correlating with lower patient PaO2/FiO2 and worse 28 days and 90 days mortality. Additionally, the inter-reliability of the score was found to be excellent [8].

The prognostic value of CXR in relation to ICU admission was explored by Diletta et al. [4] in Italy who showed a strong independent correlation of higher severity scores with a higher risk of intubation. Their data confirmed the RALE score as a valid standardized prognostic score with high interobserver reliability (ICC: 0.92 [95% CI: 0.88–0.95]).

This study reproduces similar statistically significant differences between the rates of intubation and ICU admission across the RALE severity categories. The frequency of intubation and ICU admission increased with worsening scores (Fig. 3; Table 3). Severe CXRs required earlier intervention with intubation and ICU care from the ED as compared to mild and moderate CXRs that later deteriorated requiring intubation and ICU care during the hospital stay.

Severe RALE CXR patients vitally had higher RRs and SPO2 observed in the ED at first contact and as compared to the other categories had the highest proportion of deaths to survivors. Interestingly, more than half of the patients in the moderate category were intubated at some point during hospital admission. Moderate CXRs as previously mentioned, mostly had both, consolidations and opacities, which can be an important consideration for admission to the ICU from ED for closer and more frequent evaluations for deteriorations.

This study provides data for the COVID-19-sinfected population of the studied city, from a tertiary care hospital. A major limitation to this study is its retrospective nature. From the initial 3,035 patients, the majority were excluded due to missing data or initial CXR, and lost to follow-up, as they were transferred to other facilities.

While the sample was multinational and included patients above 18 years with comorbidities, it was skewed to include more males than females. Pregnant and pediatric patients were excluded as most were transferred to obstetric/pediatric containing facilities.

Radiographs were blinded to patient demographics and clinical course to eliminate any potential recall bias during the assessment of CXR. As previously mentioned, Diletta et al. [4] had demonstrated high interobserver reliability for the RALE score. However, inter-reliability of the radiographers in this study could not be assessed as each radiographer received 202 CXRs for interpretation due to logistic reasons.

While new variant strains emerged over the course of the pandemic, RT-PCR testing had not evolved to differentiate between the various genotypes. The importance of screening and differentiating between the strains has not been established or considered in previous literature and was out of the scope for this study.

CXR was taken in equal distribution from a mobile unit or departmental, while mobile CXR provides poorer quality images. However, this distribution provides a realistic approach to the use of CXR as portable machines would be preferred for sicker patients and ease of disinfection.

Finally, clinical parameters were explored for each category. However, due to the scope of the study, we preferred to focus on the primary outcome and did not statistically correlate vital parameters with the CXR categories. Further study into the underlying risk factors producing the appearance of the CXR category could be undertaken in the future.

During the global pandemic, CXR is an inexpensive, simple tool to triage patients at initial contact. With characteristic bilateral lower zone GGOs suggestive of COVID-19 infection, the type of infiltrate seen can roughly predict the severity of the CXR.

Furthermore, its use can be extended to predicting outcomes. The use of the validated RALE score risk stratifies patients, with moderate to severe scores correlating well with a higher proportion of intubation, ICU admission from the ED, and mortality making it a useful piece of information when evaluating and triaging COVID-19 patients.

Ethical approval by Research Committee of Dubai health authority with Reference Number DRSEC/RRP/2020/43 and which was Dated on November 02, 2020. Consent was waived. Approval for Waiver of Consent for the study received by DSREC-GL07-202 21 MAR 2023. This research is compliant with the guidelines of human studies and was conducted ethically in accordance with the World Medical Association Declaration of Helsinki.

The authors declare that there is no conflict of interest regarding the publication of this article.

None.

Duaa Alhaddad: principle investigator, first author, data collection, and discussion. Ayeda Ahmed: author, data collection and data filter, and result analysis and discussion. Maryam Aldashtei: data collection and author. Ayoub Abedzadeh and Shaikh Sayeed Iqbal: CXR analysis. Feras Alnajjar: author and full review of paper. Yaser Armaghan: paper review.

Research data are not publicly available on legal or ethical. However, it is available from the correspondence author upon reasonable request.