Automated Determination of Artifact and Real Alerts in Clinical Monitoring Data

Researchers at the University of Pittsburgh have developed a method for the automated determination of artifact and real alerts in clinical monitoring data. This innovative solution addresses the issue of alarm fatigue in acute care settings, where a large number of generated alerts are medically irrelevant. By combining machine learning with an active learning protocol, the technology can accurately isolate artifactual alerts in real time, significantly improving situational awareness for clinicians and reducing patient risks associated with attention saturation.

Description

The invention utilizes a machine learning approach to train models capable of real-time probabilistic determination of the nature of each bedside monitoring alert. It integrates numeric features across multiple monitoring parameters and systematically incorporates domain expertise from expert clinicians. The active learning protocol guides the robust adjudication of historical alerts by a committee of experts, yielding high-quality training data for model development. This method is independent of any particular monitoring device or parameter, making it universally applicable across different systems.

Applications

• Acute care and critical care monitoring
• Reducing alarm fatigue in healthcare settings
• Enhancing clinical decision-making and patient safety
• Improving the reliability of bedside monitoring systems

Advantages

This technology provides a principled, data-driven approach to artifact detection, operates in real time with incoming data from any monitoring device, and integrates multiple monitoring parameters simultaneously. The active learning protocol ensures high-quality training data while minimizing human effort. By accurately distinguishing between real and artifactual alerts, the technology enhances clinician confidence in monitoring systems and improves patient outcomes.

Invention Readiness

The technology has been empirically evaluated on a large collection of real-world clinical data, demonstrating its capability to isolate a large fraction of artifactual alerts accurately. Initial studies have shown that the method can reduce alarm fatigue and improve clinical outcomes.

IP Status

Software