Abstract
Accurate prediction of clinical outcome requires models that can jointly reason over longitudinal diagnostic narratives and structured electronic health records (EHRs). Current multimodel approaches often treat patient data as static or rely on early/late feature concatenation, overlooking the temporal progression of clinical events and the semantic relationships between diagnostic findings. We propose TS-GNN, a lightweight Temporal-semantic Graph Neural Network that represents each hospital admission as a multimodal graph composed of sequential chest X-ray reports enriched with structured clinical features. Each node fuses hierarchical ClinicalBioBERT embeddings with a six-phase feature schema capturing demographics, acuity, surgical status, care intensity, interventions, and biological state. Temporal edges encode the chart-time progression of diagnostic events, while semantic edges connect clinically similar findings, enabling long-range evidence propagation. A multitask GNN architecture jointly predicts in-hospital survival, length of stay, and 30-day readmission. We evaluate on MIMIC-IV; TS-GNN demonstrates consistent improvements across hierarchical feature phases and temporal windows, outperforming strong tabular and sequential baseline models. Despite integrating rich multimodal context, the framework remains computationally lightweight with sparse graph construction and moderate parameterization. These results highlight the value of temporal-semantic graph reasoning for robust and scalable clinical decision support.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMERGE_031.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=BpB0rbAvYy
BibTex
@InProceedings{PalRah_Relational_MICCAISAT2026,
author = { Pal, Rahul AND Singh, Azad AND Kumar, Sanoj AND Mishra, Deepak},
title = { { Relational Learning of Temporal and Semantic Clinical Structured EHRs for Enhanced Patient Outcome Forecasting } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17266},
month = {pending},
page = {pending}
}