DETER: A Clinical Deterioration Prediction Algorithm to Improve PatientCare with Devices-Based Telemetry and Generative AI

Traditional Early Warning Systems (EWS) for predicting clinical deterioration often rely on intermittent vital sign monitoring and aggregated scoring methods that may miss subtle physiological changes indicative of impending adverse events. Recent advances in continuous telemetry monitoring combined with artificial intelligence offer opportunities to enhance early detection capabilities beyond conventional approaches. We propose DETER, a novel deterioration prediction algorithm integrating continuous vital sign telemetry from device-based monitoring systems with Retrieval-Augmented Generation (RAG)-enhanced generative AI models. The system processes real-time physiological data including ECG, SpO2, blood pressure, temperature, respiratory rate, and heart rate variability alongside electronic medical records and clinical assessment data. The algorithm was developed using the MMIMC-IV dataset and validated using a dataset of 94 patients (59 males, 35 females; 74% aged >65 years) from the Cardio-thoracic Clinic of University Hospital of Patras. The generative AI architecture employs Transformer-based models for time-series prediction, achieving classification parameter rates of 96.5% accuracy with Area Under the Curve (AUC) values of 0.94 for high-risk deterioration events and 0.92 for moderate-risk events. DETER demonstrates superior predictive capability compared to traditional EWS, providing risk stratification with lead times of days to weeks rather than hours. The system generates personalized deterioration risk scores updated continuously, enabling proactive intervention strategies and optimized resource allocation. The integration of continuous device-based telemetry with RAG-enhanced generative AI represents a paradigm shift toward predictive, personalized patient monitoring. DETER's ability to forecast physiological trajectories with high accuracy and extended lead time has significant potential to reduce preventable deterioration events, optimize clinical workflows, and improve patient outcomes in acute and ambulatory care settings. Prospective validation studies are warranted to assess real-world clinical impact and implementation feasibility across diverse healthcare environments.