4081.0 - Evaluating an artificial intelligence alert system for early detection of patient deterioration: A secondary prevention approach in hospital care
Program: Health Informatics Information Technology
Session: Big Data HIIT Poster Session 1
Author: Debbie Humphries
See all authors and presenters →
Abstract
Background
Hospital-based secondary prevention reduces morbidity and mortality from patient deterioration. Traditional scoring systems, including Systemic Inflammatory Response Syndrome (SIRS), Modified Early Warning Score (MEWS), Quick Sequential Organ Failure Assessment (qSOFA), and Sequential Organ Failure Assessment (SOFA), have variable success. Artificial intelligence (AI) may improve these strategies by identifying subtle physiological trends, allowing early interventions preventing deterioration.
Objective
Evaluate an AI-based alert system’s performance in detecting patient deterioration compared to traditional scoring methods and explore how AI integration could improve secondary prevention.
Methods
A retrospective analysis was conducted on hospitalized patient data flagged by an AI system designed to predict clinical deterioration and intensive care unit (ICU) admission within 24 hours. The AI monitored vital signs (temperature, heart rate, blood pressure, respiratory rate) and laboratory parameters. To validate AI alerts, we retrospectively calculated SIRS, MEWS, qSOFA, and SOFA scores at the time alerts occurred. Predictive performance (accuracy, precision, recall, F1-score) for ICU admissions was compared. Pearson correlations assessed relationships between vital signs and ICU outcomes.
Results
Of AI-flagged patients, 42.9% required ICU admission, with alerts occurring on average 22.2 hours prior. SIRS and MEWS each demonstrated 78.6% accuracy, 71.4% precision, and 83.3% recall; qSOFA showed higher precision (100%) but lower recall (50%). Temperature negatively correlated (-0.532), while heart rate positively correlated (0.300) with ICU admission.
Discussion/Conclusions
The AI alert enabled earlier detection than traditional indices alone, potentially facilitating earlier interventions and improved resource allocation. Integrating AI into existing clinical scoring can strengthen secondary prevention efforts and enhance patient care quality.