Intensity–complexity post-triage prioritization to shorten waiting times in emergency departments

Emergency department triage reliably identifies patients at immediate risk, although it compresses heterogeneous clinical profiles into a small number of urgency categories, leaving discretion in how patients are prioritized within the same category. We propose a second-stage prioritization framework that makes two complementary clinical dimensions explicit: intensity (acute physiological severity) and complexity (longer-term vulnerability and management burden). Using ED triage data from a medium-large hospital, we derive patient-level intensity and complexity measures for patients assigned to a high-volume intermediate-acuity class through a domain-adapted large language model pipeline that integrates structured variables with free-text triage documentation; we validate these measures against independent physicians assessment on a representative 10% subsample. We then evaluate the operational impact using a discrete-event simulation of ED flow under congestion, comparing first-come–first-served sequencing with an intensity–complexity–aware prioritization rule. The proposed approach reduces average waiting time in the targeted class by 23–39 minutes per patient and yields spillover reductions in lower-acuity classes of approximately 24–44 minutes per patient. Robustness analyses indicate that these gains persist under alternative modeling assumptions, including a context-specific empirical non-FCFS comparator, and validation-calibrated LLM label noise. These improvements arise by limiting delay for patient profiles whose service requirements are most sensitive to waiting time, thereby mitigating downstream congestion effects.