arXiv

Model improves heart risk prediction in lung screening

An observational evaluation involving 53,452 lung-screening participants at 33 US hospitals found that the model outperformed cardiovascular risk calculators.

Paper by Yuxiang Lai, Jiashuo Zhang, Liang Zhao, Xiaofeng Yang
Lungs surround a heart with narrow and wider vessel openings, while small upright, faceless people stand scattered around the chest section.

Our read

Biomarkers
Studied:
Preprint: study of 53,452 people
What changed:
Heart-death prediction was more accurate with CE-Agent than clinical calculators, with a concordance index of 0.731
The catch:
Comparator calculators used reference blood pressure, lipids, and kidney function instead of participants’ measurements

Editor of record: Boris Djordjevic · AgingResearch.News

Follow @agingbrief

Among 53,452 lung-screening participants from 33 US hospitals, researchers tested a risk-prediction framework using observational data. They also evaluated it in 6,814 multi-ethnic adults. The Causal Envelope Agent used published evidence to constrain how risk factors influenced predictions before examining outcomes. It then used cohort data to estimate local baseline risks and contributions from features on chest computed tomography scans.

In lung screening, the model outperformed cardiovascular risk calculators. It also matched established lung cancer screening models, with predicted risks closely matching observed outcomes. When clinical or scan information was missing, it produced ranges expressing uncertainty, intended to guide record review or selective imaging for patients near decision thresholds.

Why it matters

Lung cancer and cardiovascular disease are the leading causes of death in lung-screening populations. Assessing both risks together could help guide follow-up from a single screening visit.

Caveats

The evaluation tested risk predictions in these adult populations, not whether using the model improved health. The work is a preprint and has not been peer-reviewed.

Comments

Be the first to comment.