An approach to automatic process deviation detection in a time-critical clinical process.

Area of Science:

• Medical Informatics
• Patient Safety
• Clinical Process Analysis

Background:

• Minor errors in complex medical processes can escalate, leading to adverse patient outcomes.
• Real-time detection of process deviations is crucial for preventing or mitigating medical errors.

Purpose of the Study:

• To develop an automated system for detecting errors and deviations in trauma resuscitation workflows.
• To improve the accuracy of identifying deviations in real-time medical processes.

Main Methods:

• Video review and activity coding of 95 pediatric trauma resuscitations.
• Comparison of activity traces against a knowledge-driven workflow model using conformance checking.
• Iterative system repair based on analysis of false alarms and model discrepancies.

Main Results:

• Initial system achieved 66.6% accuracy with a 0.42 F1-score, detecting 73% false alarms.
• Repaired system demonstrated high accuracy: 95.2% (0.85 F1-score) in validation and 98.5% (0.96 F1-score) in testing.
• Analysis of 95 resuscitations identified 1060 deviations, including errors of omission, commission, and scheduling.

Conclusions:

• The developed approach effectively detects deviations in complex medical processes like trauma resuscitation.
• Assessing detected deviations is vital for refining knowledge-driven models to accurately reflect actual clinical practice ('work as done').
• This method can inform the design of decision support systems to enhance patient safety.