Donghyeon Kim
KO EN
KO EN
GitHub LinkedIn
Back to Research
Research Detail

A Context-Adaptive Gated Embedding Framework for Advanced Clinical Decision-Making

This study proposes a hierarchical clinical decision support framework that estimates diagnostic context via partial-label automated ICD coding and reinjects it into irregular ICU time-series forecasting through context-adaptive gating for mechanical ventilation transition prediction. By conditioning temporal interpretation on diagnostic context, the framework substantially improves rare-event detection.

Venue Mathematics (submitted)
Year 2026
Type Journal Paper
Clinical Decision Support SystemAutomated ICD CodingICU Time-seriesMechanical Ventilation PredictionPartial-Label LearningExtreme Multi-Class ClassificationTCNGatingRare Event Detection
Abstract

This study proposes a hierarchical clinical decision support framework that estimates diagnostic context via partial-label automated ICD coding and reinjects it into irregular ICU time-series forecasting through context-adaptive gating for mechanical ventilation transition prediction. By conditioning temporal interpretation on diagnostic context, the framework substantially improves rare-event detection.

Connection to Portfolio Roles
Research Contribution

Proposed a context-adaptive gated embedding framework that reinjects diagnostic context from automated ICD coding into ICU time-series prediction.

Portfolio Relevance

Shows modeling judgment that connects diagnostic context and time-series signals hierarchically instead of treating data surfaces as isolated inputs.

Method / Judgment Signal

Combines partial-label diagnostic context, TCN temporal representations, and gating to strengthen rare transition-event prediction.

Modeling / Evaluation

This research is read as a basis for NLP/LLM modeling and evaluation design.

System Judgment

The research result is connected to data, evaluation, and operating decisions in project work.

Application Context

Domains such as medical, finance, and recommendation are treated as application contexts, not the top-level identity.

Summary

This study proposes an integrated ICU CDSS framework in which diagnostic information and time-series signals are not treated as separate tasks, but are linked hierarchically so that higher-level diagnostic context conditions lower-level temporal interpretation. A key idea is to redefine automated ICD coding not as a terminal prediction task, but as a representation-learning stage for downstream intervention prediction.

Why It Matters

ICU EHR data are difficult to exploit directly because measurements are irregularly sampled, missingness is structural, and diagnostic information is often incomplete or weakly coded. Automated ICD coding is an extreme multi-class problem with a long-tailed label space, while mechanical ventilation prediction suffers from severe imbalance because clinically important transitions such as ONSET and WEAN are rare. Prior work has usually focused either on temporal patterns alone or on ICD coding accuracy itself, leaving the linkage between diagnostic context and intervention prediction underexplored.

Contribution

The proposed CAGE framework first encodes irregular ICU time-series using a three-channel VAL/MSK/DELTA representation and estimates diagnostic context through a partial-label automated ICD coding module. The resulting probability-weighted diagnostic embedding is then reinjected into TCN-based temporal features via context-adaptive gating before four-class prediction of ONSET, WEAN, STAY ON, and STAY OFF transitions. The framework achieved hit@1 0.4863, hit@3 0.7302, hit@5 0.8063, and hit@10 0.8801 in automated ICD coding, and reached Macro-AUC 98.2, Macro-AUPRC 77.4, and F1-score 79.4 for intervention prediction, showing substantial gains in rare-event detection.