Machine Learning Based Linking of Patient Reported Outcome Measures to WHO International Classification of...

Area of Science:

• Rehabilitation Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Patient-reported outcome measures (PROMs) are crucial for assessing functional health in healthcare.
• The World Health Organization (WHO) recommends using the International Classification of Functioning, Disability and Health (ICF) framework for classifying patient functioning.
• Integrating PROMs with the ICF framework is essential for comprehensive health assessment across all sectors.

Purpose of the Study:

• To optimize machine learning (ML) methods for automatically linking PROM data to ICF categories for chronic low back pain (cLBP).
• To identify the minimum set of PROMs required for accurate linking without performance degradation.
• To enhance the classification of activity limitations and participation restrictions using ML.

Main Methods:

• Utilized data from 806 cLBP patients who completed comprehensive PROMs and ICF assessments.
• Developed and refined random forest (RF) methods to classify ICF activity and participation categories based on PROM data.
• Identified key PROM items and validated linking performance using minimal item subsets.

Main Results:

• Novel RF methods demonstrated improved performance over existing ML linking techniques (ROC-AUC 0.73-0.81).
• Variable importance analysis indicated that a minimal set of 15-24 PROMs could maintain linking performance.
• The optimized ML models accurately predicted limitations and restrictions within ICF core categories for cLBP.

Conclusions:

• Optimized ML-based methods provide accurate prediction of functional limitations and participation restrictions in cLBP patients.
• Reducing the number of PROMs to a minimal set of 15 items does not significantly impact the accuracy of ICF linking.
• These findings support the efficient and accurate use of PROMs integrated with the ICF framework in clinical practice.