TRACE improves hospital risk prediction by tracking daily clinical data
TRACE: Tracking Multimodal Clinical Trajectories for Inpatient Risk Prediction
Artificial Intelligence
Summary
Predicting risks like death or longer hospital stays is hard because patient data comes from many sources and changes every day. The authors created TRACE, a method that organizes daily patient information from health records, X-rays, and reports into a timeline to better understand how a patient's health evolves in the hospital. TRACE uses a smart attention system to handle missing data and groups similar patient cases by disease traits. Tested on large hospital datasets, TRACE predicted outcomes more accurately than previous methods.
What this means in practice
- •For hospital data teams: Integrate daily clinical event timelines from multiple data types to improve predictions of patient mortality, readmission, and length of stay in hospital systems.
- •For medical imaging software developers: Develop imaging tools that align chest X-rays and radiology reports with patient clinical trajectories to enhance automated risk assessment capabilities.
Authors
Minjun Kim, Jong Hak Moon
Abstract
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. Existing multimodal approaches combine structured health record data, chest x-rays, and clinical text, but primarily focus on modality fusion or missing-modality compensation, with limited attention to temporal trajectories over an admission. We propose TRACE (TRAjectory modeling of multimodal Clinical Events), a daily-scale framework organizing clinical events into day-level trajectories, separating day-invariant global context from day-specific local events drawn from EHR data, chest x-rays, radiology reports, and discharge summaries. The trajectory is encoded using temporal attention, allowing unavailable modalities to emit no tokens. A phenotype-guided contrastive objective further organizes admissions under heterogeneous disease distributions. Across cohorts from MIMIC-IV, eICU, and MIMIC-III comprising nearly 180,000 total stays, we evaluate TRACE on mortality, length of stay, and readmission prediction. TRACE achieves the highest AUROC across all nine dataset-task settings and the highest ACC in eight, outperforming baselines. Ablation studies validate our attention masks and contrastive learning, while trajectory-level attribution localizes risk-associated hospital days and modalities.