Machine learning models for classifying non-specific neck pain using craniocervical posture and movement.

Area of Science:

• Biomechanics and Physical Therapy
• Machine Learning in Healthcare
• Occupational Health

Background:

• Non-specific neck pain (NSNP) is prevalent among office workers.
• Craniocervical posture (CCP) and cervical kinematics are commonly used screening methods.
• The predictive performance of machine learning (ML) models for NSNP classification needs further investigation.

Purpose of the Study:

• To compare the predictive performance of ML models for classifying individuals with and without NSNP.
• To evaluate datasets including CCP and cervical kinematics during protraction and retraction (CKdPR).

Main Methods:

• An exploratory, cross-sectional study involving 773 public service office workers (PSOWs).
• Five datasets were created: CCP, cervical kinematics during protraction, cervical kinematics during retraction, CKdPR, and a combination of CCP and CKdPR.
• Four ML algorithms (random forest, logistic regression, Extreme Gradient boosting, support vector machine) were trained and evaluated using AUC, accuracy, precision, recall, and F1-score.

Main Results:

• The random forest model using the CKdPR dataset achieved the highest AUC (0.892) and F1-score (0.832) for classifying NSNP.
• The random forest model using the CCP dataset showed the lowest performance (AUC, 0.738; F1, 0.715).

Conclusions:

• ML algorithms demonstrate superior performance in classifying NSNP compared to classical statistical methods.
• Cervical kinematics during protraction and retraction (CKdPR) datasets yield better ML model performance for NSNP prediction than CCP datasets.