AgingResearch.NewsJoin free
The Lancet Digital Health

Chest X-ray model flags chronic obstructive pulmonary disease risk

Adding chest images to clinical scoring better distinguished future cases from non-cases, with scores of 0.73 versus 0.66 in ever-smokers and 0.70 versus 0.60 in never-smokers.

Paper by Saman Doroodgar Jorshery, Jay Chandra, Anika S. Walia, +17 more authorsVineet K. Raghu · Broad Institute · Brigham and Women's Hospital
Figure 1. Schematic diagram of the study.
Open the figure at full size
Figure 1. Schematic diagram of the study.Schematic diagram of the study.Jorshery et al.

Our read

Biomarkers
Studied:
Cohort study of 27,848 outpatients without diagnosed COPD or lung cancer, 6 years (Dose-response, independent replication, registered NCT03154346)
What changed:
COPD prediction improved by 0·08 on the AUC scale when chest X-ray analysis was added to the clinical risk score
The catch:
The model could flag existing undiagnosed COPD rather than future disease

Editor of record: Boris Djordjevic · AgingResearch.News

Paper
Follow @agingbrief

Among 27,848 outpatients aged 50–80, a chest X-ray model helped identify people at higher risk of chronic obstructive pulmonary disease (COPD), a disease that restricts airflow. The women and men had no known COPD, emphysema or lung cancer. This retrospective model validation study used routine X-rays taken at two Boston hospitals in 2013–14.

Researchers combined a deep-learning model with an existing clinical risk score. The model had originally been developed to predict deaths from lung disease. The combination better separated future COPD cases from non-cases. The comparison used the area under the receiver operating characteristic curve, a measure of this separation. Higher values mean better separation. Within six years, COPD developed in 12.4% of 12,550 ever-smokers and 3.8% of 15,298 never-smokers.

Why it matters

Earlier detection of chronic obstructive pulmonary disease is a challenge for care in midlife and older age. The authors suggest that high-risk patients might benefit from spirometry, a breathing test, and subsequent preventive care.

Caveats

The authors note that whether the model improves early detection needs testing in a prospective trial. The findings came from adults aged 50–80 at one Boston hospital system, most of whom were non-Hispanic White.

From the paper

Participants

… we tested whether CXR-Lung-Risk was associated with pulmonary function and plasma protein abundance using data from 2097 individuals with a CXR image in PBHS (mean age 51·4 years [SD 17·0], 919 [43·8%] male, 1178 [56·2%] female; 1292 [61·6%] White; p 15).

Results

Result

… a graded association with incident COPD was observed across moderate (adjusted hazard ratio [aHR] 1·7 [95% CI 1·5–1·9]) and high (aHR 3·4 [2·9–3·9]) CXR-Lung-Risk categories compared with the low-risk category, after adjusting for risk factors and radiologist findings on the CXR report.

Results

Limitation

First, the primary analysis used patients having routine CXR at a single hospital system in Boston, and most were non-Hispanic White.

Discussion

Quoted word for word from the full text on PubMed Central.

Comments

Be the first to comment.

More on Deep learning