Model predicted hospital death risk in older intensive care patients
The updated model scored 0.872 at distinguishing hospital deaths from survivors among 57,488 older intensive care patients across 206 hospitals.
Paper by Zhi Jin, Hua Sun, Yiting Zhou et al.
Our read
Biomarkers- Studied:
- Accepted, not yet published: cohort study of 57,488 older adults at their first intensive care admission (Independent replication, multicenter)
- What changed:
- The model scored 0.872 at distinguishing hospital deaths from survivors among older intensive care patients
- The catch:
- The evaluations measured predictions, not whether using them improved survival or nutrition care
Editor of record: Boris Djordjevic · AgingResearch.News
Adults aged 65 or older at their first intensive care admission were included in a multicohort study of hospital death prediction. Researchers developed the model using 33,671 patients and externally tested and updated it using 71,861 patients across 207 hospitals.
The model combined 83 predictors, including nutrition measures, blood tests, vital signs and illness severity. Its predictions closely matched observed death rates in the main test group. In a comparison involving 39,920 patients, it scored 0.871 versus 0.850 for an established severity-based model recalibrated for the comparison. The score measured how well predictions distinguished deaths from survivors. Performance remained similar in tests at hospitals left out of model development. Measures of nutritional reserves added modest predictive value alongside acute illness severity.
Why it matters
Nutrition risk overlaps with acute illness in older intensive care patients. The authors said the findings supported future electronic monitoring of nutrition risk.
Caveats
These evaluations tested risk predictions, not whether acting on them improved survival or nutrition care. The findings concern first intensive care admissions in adults aged 65 or older.
The paper
Zhi Jin, Hua Sun, Yiting Zhou et al.
Frontiers in Nutrition, Nov 12, 2026, CC BY, Accepted, not yet published
Comments
Be the first to comment.