UID preserving method improves clinical event timelines from discharge summaries
Anchoring Clinical Events in Time: UID-Preserving Multimodal Reconstruction and Source-Grounded Adjudication
Artificial Intelligence
Summary
Medical records often list events out of order or miss details, making it hard to understand a patient's treatment timeline. The authors created a system that links each event in a doctor's notes back to its source and uses both text and record data to build accurate timelines. They also built a tool that checks and compares these timelines against the original records. Their approach recovered many more events and matched expert doctors’ timeline assessments closely.
What this means in practice
- •For hospital data teams: Create accurate patient event timelines by linking narrative notes with structured records to improve treatment window analysis.
- •For medical software developers: Build clinical documentation tools that preserve event identity through text and data sources for better clinical timeline reconstruction.$Commercial implications: Enables advanced clinical timeline features that can be sold as enhancements to electronic health record systems.
Authors
Sayantan Kumar, Nicolas Grimaldi, Jack Cummins, Jeremy C. Weiss
Abstract
Clinical timelines support treatment-window analysis and leakage-free modeling, but discharge summaries often obscure chronology and structured EHR tables describe only part of the patient course. We present a UID-preserving framework that links each narrative event occurrence to its source span and retains that identity through text-only estimation, structured-evidence retrieval, timestamped source-row grounding, and joint revision. We also present GAVEL, an LLM judge that compares two UID-aligned timelines against the narrative and structured record, to augment prior matching and temporal assessments. Across six open-weight models and 40 mixed-critical-care summaries, the GLM 5.2 multimodal revision, as compared to its text-only variant, improved temporal agreement without reducing event recovery and performed competitively with clinician annotations, while other model revisions showed smaller gains and lower overall performance. Ablations showed that UIDs primarily preserve event retention, whereas source-row linkage supports temporal placement. Blinded human review upheld most GAVEL findings, and controlled adjudication favored multimodal over text-only GLM 5.2 but did not for DeepSeek V3.2. In developing the UID and judge pipeline, we are able to demonstrate 43\% increased event recovery, a framework competitive with clinician annotations, and a system with occurrence-level provenance for both reconstruction and evaluation.