Chest X-Rays Predict Adverse Outcomes in COVID-19 Pneumonia Patients
- In this multicenter study, 1,367 chest X-rays (CXRs) of 405 patients with COVID-19 were assigned a Radiographic Assessment of Lung Edema (RALE) score and were also processed with a computer algorithm that is commercially available in Europe
- Among the 323 patients who had serial CXRs, those who died or received mechanical ventilation had greater RALE and AI score changes than those who recovered/did not have mechanical ventilation (P<0.001 to 0.013)
- The addition of demographic, clinical and laboratory information to regression models significantly improved the predictive value of CXRs
Studies have shown that subjective Radiographic Assessment of Lung Edema (RALE) scores of chest X-rays (CXRs) capture the extent of pulmonary involvement in patients with COVID-19 pneumonia. RALE scores also predict adverse outcomes such as hospitalization and ICU admission.
Subscribe to the latest updates from Pulmonary & Critical Care Advances in Motion
Likewise, several objective artificial intelligence (AI)–based algorithms have proved accurate in assessing CXRs of COVID-19 patients for the severity of lung involvement and distinguishing between moderate and severe pneumonia.
Shadi Ebrahimian, MD, research fellow, and Mannudeep K. Kalra, MD, DNB, director of the Webster Center for Quality and Safety in the Department of Radiology at Massachusetts General Hospital, and their colleagues from the Mass General & BWH Center for Clinical Data Science, have demonstrated additional benefits of CXRs for evaluating hospitalized patients with COVID-19 pneumonia. In Scientific Reports, they say AI is as robust as RALE for predicting whether patients will recover and/or need mechanical ventilation.
The study included 405 adults with COVID-19: 226 from Mass General (site A) and 179 from four hospitals of a tertiary care system in Daegu, South Korea (site B). Of these, 323 patients had serial CXRs, for a total of 1,367.
Two thoracic radiologists assigned each CXR a RALE score, which reflected the extent and severity of lung involvement from COVID-19 pneumonia. Two other team members processed each CXR with an AI algorithm that is commercially available in Europe (qXR v2.1 c2, Qure.ai Technologies, Mumbai, India). The algorithm reports the percentage of lung area projected to have COVID-19–related findings (AI score). Each CXR was processed in less than five seconds.
RALE and AI Scores
The RALE and AI scores were strongly positively correlated:
- Site A: r2, 0.79; P<0.0001
- Site B: r2, 0.86; P<0.0001
- Entire dataset: r2, 0.83; P<0.0001
There was strong agreement between sites in the changes over serial CXRs for RALE and AI scores (Site A, 75%; Site B, 77%).
Among patients with serial CXRs, those who died or received mechanical ventilation had greater RALE and AI score changes than those who recovered/did not have mechanical ventilation (P<0.001 to 0.013).
Addition of Other Data
Combining baseline RALE and/or AI scores with other information (e.g., patient age, gender, smoking history, body mass index, white blood cell count and peripheral oxygen saturation) improved the ability to predict outcomes.
Simpler Is Better
Most imaging and AI literature about COVID-19 focus on CT scans, and it's heartening that CXRs are capable of predicting outcomes. They are faster and administer a substantially lower radiation dose than most chest CT protocols, and CXR units are more portable, easier to sterilize and more accessible, even at the bedside when necessary.
CXRs also help confirm the placement of the central lines and other life support catheters and devices frequently needed for critically ill patients with COVID-19.
Unlike the RALE score, AI scoring is quantitative and automated. The rapid information it provides about the likelihood of mechanical ventilation and other adverse outcomes should be useful for both management of individual patients and the planning of resources needed for patient care during peaks of the pandemic.
view original journal article Subscription may be required
View all COVID-19 updates
Learn more about research in the Department of Radiology