TRACE models daily clinical trajectories for inpatient risk prediction
Original title: TRACE: Tracking Multimodal Clinical Trajectories for Inpatient Risk Prediction
On nearly 180,000 stays from MIMIC-IV, eICU, and MIMIC-III, authors report the highest AUROC in all nine settings.
TRACE organizes EHR data, chest x-rays, radiology reports, and discharge summaries into day-level trajectories. Temporal attention encodes each trajectory and lets missing modalities emit no tokens. A phenotype-guided contrastive objective handles heterogeneous disease distributions. Authors evaluate mortality, length of stay, and readmission and report the top AUROC across nine dataset-task settings.
Unconfirmed: AUROC and accuracy rankings are author-reported and not independently confirmed here.
- Tasks cover in-hospital mortality, prolonged length of stay, and readmission.
- Cohorts total nearly 180,000 stays from MIMIC-IV, eICU, and MIMIC-III.
- TRACE had the highest accuracy in eight of nine settings, authors report.
- Ablations support the attention masks and contrastive objective.
From the original article
1 min read at huggingface.co
Accurate inpatient risk prediction requires modeling heterogeneous clinical evidence that accumulates irregularly over time. This challenge spans in-hospital mortality, prolonged length of stay, and readmission, each with task-specific observation windows.
By Minjun Kim, Jong Hak Moon · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.