Rapid Diagnostic Model for Critical Illness Polyneuropathy Based on Electrophysiological Data

Area of Science:

• Neurology
• Intensive Care Medicine
• Biomedical Engineering

Background:

• Critical illness polyneuropathy (CIP) is a frequent cause of muscle weakness in critically ill patients.
• Current diagnostic methods are time-consuming (60-90 min) and invasive, limiting their use in intensive care settings.
• There is a need for rapid, efficient, and minimally invasive diagnostic approaches for CIP.

Purpose of the Study:

• To develop and validate a rapid, efficient, and minimally invasive diagnostic model for CIP.
• To compare the performance of various machine learning algorithms for CIP diagnosis.
• To identify key electrophysiological features for accurate CIP detection.

Main Methods:

• Machine learning models including SVM-RBF, random forest, and XGB were evaluated using electrophysiological data from 134 CIP patients and 135 controls.
• Feature selection was performed to identify the most predictive electrophysiological parameters.
• Model performance was assessed using cross-validated areas under the receiver operating characteristic curve (AUC).

Main Results:

• The Support Vector Machine with Radial Basis Function (SVM-RBF) model, utilizing seven key electrophysiological features from the peroneal and ulnar nerves, achieved an AUC of 0.93.
• This rapid model demonstrated strong performance in an independent validation set (AUC = 0.88).
• Distal compound muscle action potential (CMAP) of the peroneal nerve was identified as a critical diagnostic feature.

Conclusions:

• A rapid diagnostic model for CIP was successfully developed using SVM-RBF and seven electrophysiological features.
• This model facilitates timely, efficient, and minimally invasive diagnosis of CIP in critically ill patients.
• The developed model and associated code are publicly available for further research and application.