Score from chest scans predicted patients’ risk of death
In 7,028 patients tested separately, the score distinguished those who died from survivors with a prediction-performance measure of 0.78.
medRxiv
Researchers analysed 28,509 patients undergoing scans to assess blood flow to the heart in an observational study at 12 centres across four countries. They developed an artificial intelligence-based score in 15,037 patients and tested it in separate groups of 6,444 and 7,028 patients.
The score combined measures of bone, muscle and fat in four locations from routine low-dose chest scans. It also included age, sex, body mass index—a measure of weight relative to height—and calcium deposits in the heart’s arteries. During a median follow-up of 3.5 years, 4,697 patients died. In the external testing group, the score’s area under the curve was 0.78, a measure of how well it distinguished deaths from survival. It outperformed a model using demographic information and artery calcium alone.
Why it matters
The study addresses whether body composition can help assess survival prospects, a question relevant to lifespan. The authors suggested that combining complex body measurements into one score could help guide individualised prevention.
Caveats
This observational study tested prediction, not whether changing body composition prevents deaths; its simulations were not interventions in patients. The paper is a preprint, not yet peer-reviewed, and studied patients undergoing heart scans.
- Machine learning
- Body mass index
- Coronary artery disease
- Body composition index
- Coronary artery calcium score
- Humans
The paper
Show 37 more authors
Anna M Marcinkiewicz, Assiata Kamagate, Aakash Shanbhag, Waseem Hijazi, Mark Lemley, Jianhang Zhou, Joanna X Liang, Giselle Ramirez, Shiva Mostafavi, Meghana Urs, Clemens P Spielvogel, Leandro Slipczuk, Mark Travin, Erick Alexanderson, Isabel Caraval-Juarez, René R S Packard, Mouaz Al-Mallah, Terrence D Ruddy, Andrew J Einstein, Attila Feher, Edward J Miller, Wanda Acampa, Stacey Knight, Viet T Le, Steve Mason, Vinicius F Calsavara, Panithaya Chareonthaitawee, Samuel Wopperer, Alan C Kwan, Lixia Wang, Debiao Li, Elliot K Fishman, Felipe Lopez-Ramirez, Daniel S Berman, Jacek Kwiecinski, Damini Dey, Marcelo Di Carli,medRxiv · 10 Aug 2026 · Preprint, not peer-reviewed
