A Two-Level Clinical Novelty Prioritization Framework for Disease Trajectories Discovered from Longitudinal Electronic Health Records
A two-level framework ranks 99,527 MIMIC-IV disease trajectories by literature novelty and clinical priority.
The paper proposes a two-level method for prioritizing 99,527 branching disease-trajectory motifs from a frozen, patient-disjoint MIMIC-IV v3.1 pipeline. Motifs were embedded with BAAI/bge-small-en-v1.5 and compared with 166,972 pre-2023 PubMed references; positive recombination affected 8.63% of motifs but barely changed global ranking. In the high-semantic subgroup, branch-enhanced forecasting reduced conditional event-time MAE by 20.1 days and sMAPE by 2.37 percentage points. The authors present this as literature-relative prioritization, not causal discovery or clinician-adjudicated novelty.
- Ranks 99,527 frozen MIMIC-IV v3.1 branching disease-trajectory motifs.
- Uses BAAI/bge-small-en-v1.5 against 166,972 pre-2023 PubMed references.
- Positive recombination in 8.63% of motifs barely changed rankings.
- High-semantic forecasting reduced event-time MAE by 20.1 days.
- Authors do not claim causal discovery or clinician-adjudicated novelty.
- Organizations
- BAAI
- AI models
- BAAI/bge-small-en-v1.5
From the original article
1 min read at arxiv.orgGraph-based analysis of electronic health records (EHRs) yields disease-trajectory catalogs, but structural support does not establish semantic novelty. We present a two-level framework to prioritize 99,527 branching motifs previously discovered by an upstream patient-disjoint, train-only MIMIC-IV v3. 1 pipeline and frozen after validation-based catalog selection.
By Jin-Jia Zheng · arxiv.org
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.