Benchmarking Early Deterioration Prediction Across Hospital-Rich and MCI-Like Emergency Triage Under Constrained Sensing

TL;DR

A leakage-aware benchmarking framework for early deterioration prediction that evaluates model performance under realistic, time-limited sensing conditions is presented and provides a clinically grounded benchmark to support the evaluation and design of deployable triage decision-support systems in resource-constrained settings.

Abstract

Emergency triage decisions are made under severe information constraints, yet most data-driven deterioration models are evaluated using signals unavailable during initial assessment. We present a leakage-aware benchmarking framework for early deterioration prediction that evaluates model performance under realistic, time-limited sensing conditions. Using a patient-deduplicated cohort derived from MIMIC-IV-ED, we compare hospital-rich triage with a vitals-only, MCI-like setting, restricting inputs to information available within the first hour of presentation. Across multiple modeling approaches, predictive performance declines only modestly when limited to vitals, indicating that early physiological measurements retain substantial clinical signal. Structured ablation and interpretability analyses identify respiratory and oxygenation measures as the most influential contributors to early risk stratification, with models exhibiting stable, graceful degradation as sensing is reduced. This work provides a clinically grounded benchmark to support the evaluation and design of deployable triage decision-support systems in resource-constrained settings.

Figures (4)

• Figure 1: Overview of the deterministic cohort construction and preprocessing pipeline. The workflow enforces first-hour temporal alignment, patient-level deduplication, and leakage-safe record linkage, and produces hospital-rich and MCI-like feature regimes for downstream benchmarking.
• Figure 2: Average precision across models under MCI-like (vitals-only, including bedside consciousness, oxygenation) and hospital-rich (all-features) regimes. Results shown for a representative split.
• Figure 3: Effect of cumulative feature availability on Average Precision for XGBoost (representative split).
• Figure 4: Global SHAP summaries for the XGBoost model under three feature regimes: (a) hospital-rich; (b) vitals and laboratory; (c) MCI-like setting using vitals only. Each point corresponds to a patient, colored by feature value (red = high, blue = low), with horizontal position indicating SHAP contribution to predicted deterioration risk.