Relating process and outcome metrics for meaningful and interpretable cannulation skill assessment: A machine...

Area of Science:

• Medical Simulation
• Machine Learning in Healthcare
• Clinical Skills Assessment

Background:

• Healthcare quality relies on clinician skills, with cannulation errors in hemodialysis posing significant risks.
• Objective assessment and effective training are crucial for improving patient safety during hemodialysis procedures.

Purpose of the Study:

• To develop and evaluate a machine learning approach for objective assessment of cannulation skills.
• To utilize a highly-sensorized simulator and objective metrics for skill evaluation.

Main Methods:

• 52 clinicians performed cannulation tasks on a sensorized simulator.
• Machine learning models (SVM, SVR, EN) were trained using force, motion, and infrared sensor data.
• Skill was assessed using both discrete classification and a continuous skill representation.

Main Results:

• Support Vector Machine (SVM) model achieved high accuracy in skill prediction.
• Support Vector Regression (SVR) model provided a continuous assessment of skill and outcomes.
• Elastic Net (EN) model identified critical process metrics influencing cannulation success.

Conclusions:

• The sensorized simulator and machine learning assessment offer significant advantages over traditional training methods.
• This approach can substantially improve the effectiveness of skill assessment and training for hemodialysis cannulation.
• Enhanced training has the potential to improve clinical outcomes for hemodialysis patients.