Artificial intelligence (AI) is revolutionising the way chronic obstructive pulmonary disease (COPD) prognosis is assessed, according to a recent study evaluating the clinical utility of an AI-generated score known as CXR-Lung-Risk. This score, calculated from routine chest radiographs, estimates a patient’s risk of respiratory mortality expressed in “lung-risk years.” In a cohort of 4,226 patients with COPD, researchers found that higher CXR-Lung-Risk scores were strongly associated with increased risk of death from respiratory causes over a median follow-up of nearly seven years.
Superior Predictive Power of AI Risk Scores
The study demonstrated that for every five-year increase in CXR-Lung-Risk, the chance of respiratory mortality rose by 16%, independently of other clinical factors such as age, sex, smoking status, body mass index, and lung function test results. Notably, the CXR-Lung-Risk score outperformed the widely used Global Initiative for Chronic Obstructive Lung Disease (GOLD) grading system, achieving better discrimination with an area under the receiver operating characteristic curve of 0.76 compared to 0.61 for GOLD. Furthermore, pulmonary function tests declined in tandem with increasing lung-risk scores, confirming the biological relevance of the AI determinations.
This open-source AI tool offers a promising advance for clinicians managing COPD by providing a non-invasive, readily available means to stratify patients by risk and potentially guide tailored interventions aimed at reducing respiratory mortality. The ability to gain prognostic insights from a standard chest X-ray alone could facilitate earlier identification of high-risk individuals and optimise clinical decision-making in both primary and specialist care settings.
Overall, CXR-Lung-Risk embodies the potential of AI to augment traditional diagnostic methods, improve outcome predictions, and enhance personalised care for millions living with COPD worldwide. The study’s open-source AI approach further gives promise for widespread adoption and integration into routine clinical workflows, expanding personalised care options for the millions of individuals living with COPD worldwide.
Reference
Lee JH et al. Open-source AI model for predicting respiratory mortality in COPD from chest radiographs. Radiology: Cardiothoracic Imaging. 2025; https://doi.org/10.1148/ryct.250080.
