TRACE Organizes Daily Clinical Events into Trajectories for Inpatient Risk Prediction
Minjun Kim · hf · 2026-10-21
The paper proposes TRACE (TRAjectory modelling of multimodal Clinical Events) for inpatient risk prediction across mortality, length of stay, and readmission tasks.
Motivation: patient status evolves over time and is documented through heterogeneous, irregularly recorded events — labs, medications, imaging, notes — that do not form a complete or uniformly sampled multimodal record. Existing work focuses on modality fusion, admission-level relational structure, or missing-modality compensation, with limited attention to how evidence evolves temporally during admission.
TRACE organizes recorded clinical events into daily-scale global-local temporal trajectories to model evidence evolution, improving prediction on all three tasks.
More from Research
- Stanford research shows switching agent harnesses mid-run cuts cost and boosts performance — StanfordAILab · 2026-10-21
- Ropedia releases open W0 world model trained on 10,000 hours of egocentric human video — liuziwei7 · 2026-10-21
- ROTATE, an open-ended training method for generalizable teamwork agents, wins NeurIPS 2026 Spotlight — PeterStone_TX · 2026-10-21
- EdgeKernelBench: LLM-Generated Kernels Hit 64.5% Correct but Only 25.5% Fast on Edge Backends — encodelab · 2026-10-21
- MMEnterprise-Bench from KAIST Tests Enterprise Agents on Conflicting, Evolving Multimodal Records — Seohyun Lee · 2026-10-21
- NTU and Alibaba's FocusGUI: a lightweight adapter filters multimodal evidence to stop GUI agent error compounding — NanyangTechnologicalUniversity · 2026-10-21