Text Knows What, Tables Know When: Clinical Timeline Reconstruction via Retrieval-Augmented Multimodal Alignment
The research developed a method to accurately reconstruct clinical timelines by combining unstructured patient text with structured electronic health records (EHR) tables. The system uses a retrieval-augmented approach to use tabular data as an external temporal reference, correcting inaccuracies in the narrative timeline. This integration significantly improves the absolute timing of events without losing event content.