Early Risk Prediction with Temporally and Contextually Grounded Clinical Language Processing

2025-11-26

AI Summary

This study addresses key challenges in early type-2 diabetes risk prediction from EHR clinical notes—namely, long document length, irregular temporal spacing, complex time dependencies, and constraints on privacy and computational resources. To this end, we propose HiTGNN, a Hierarchical Temporal Graph Neural Network that jointly models fine-grained temporal structures and domain-specific medical knowledge graphs, and ReVeAL, a lightweight verification framework integrating large language model distillation with test-time inference optimization to enable efficient, privacy-preserving, and interpretable predictions. Evaluated on real-world multicenter EHR data, our approach significantly improves short-term risk prediction: AUC increases by 5.2% and sensitivity by 12.7%, while maintaining strong fairness across diverse demographic subgroups. Our core contribution is the first integration of temporal graph modeling, clinical knowledge embedding, and lightweight large-model inference for risk prediction from clinical text.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Clinical notes in Electronic Health Records (EHRs) capture rich temporal information on events, clinician reasoning, and lifestyle factors often missing from structured data. Leveraging them for predictive modeling can be impactful for timely identification of chronic diseases. However, they present core natural language processing (NLP) challenges: long text, irregular event distribution, complex temporal dependencies, privacy constraints, and resource limitations. We present two complementary methods for temporally and contextually grounded risk prediction from longitudinal notes. First, we introduce HiTGNN, a hierarchical temporal graph neural network that integrates intra-note temporal event structures, inter-visit dynamics, and medical knowledge to model patient trajectories with fine-grained temporal granularity. Second, we propose ReVeAL, a lightweight, test-time framework that distills the reasoning of large language models into smaller verifier models. Applied to opportunistic screening for Type 2 Diabetes (T2D) using temporally realistic cohorts curated from private and public hospital corpora, HiTGNN achieves the highest predictive accuracy, especially for near-term risk, while preserving privacy and limiting reliance on large proprietary models. ReVeAL enhances sensitivity to true T2D cases and retains explanatory reasoning. Our ablations confirm the value of temporal structure and knowledge augmentation, and fairness analysis shows HiTGNN performs more equitably across subgroups.

Problem

• Predicts chronic disease risk from EHR clinical notes
• Models temporal dependencies in longitudinal patient data
• Balances predictive accuracy with privacy and resource constraints

Innovation

• Hierarchical temporal graph neural network for patient trajectories
• Lightweight verifier models distill large language model reasoning
• Temporal and knowledge augmentation for equitable risk prediction