Escalation-Pathway Digital Twins for Predicting Failure-to-Rescue in General Hospital Wards

Abstract

Failure-to-rescue can occur even when physiological deterioration is correctly predicted because rescue depends on a time-critical chain of recognition, alerting, communication, clinical assessment, intervention, reassessment, and transfer. This paper develops an escalation-pathway digital twin whose prediction target is the probability that the response process will fail after deterioration has begun. The twin combines patient state, hospital-system state, and escalation state in a finite-state, time-dependent representation. Transition reliabilities and accumulated delays are integrated with a patient-specific rescue window to estimate conditional rescue probability, failure-to-rescue risk, and escalation-pathway reliability. A fixed-seed Monte Carlo simulation generated 50,000 synthetic episodes across normal operations, high workload, alert-system degradation, combined system stress, and automatic alert rerouting. Failure-to-rescue increased from 29.43% under normal operations to 86.70% under combined stress, while median escalation time increased from 73.0 to 145.0 minutes. Rerouting reduced alert failure from 36.96% to 13.58% under otherwise stressed conditions, but overall failure remained 82.34%, revealing downstream bottlenecks. Potential applications include overdue-action surveillance, backup routing, capacity monitoring, workflow-policy testing, and retrospective review. These findings demonstrate internal model behaviour rather than clinical effectiveness. Prospective validation, local calibration, human-factors evaluation, fairness auditing, and governance are required before clinical use.