Harmonizing patient-reported outcome measures for nasal complaints using traditional and machine learning methods

Area of Science:

• Otolaryngology
• Medical Informatics
• Psychometrics

Background:

• Nasal obstruction assessment relies on diverse measurement instruments with varying scales.
• Data harmonization is crucial for comparing scores across different instruments but is underutilized.
• Standardizing nasal obstruction data enhances research and clinical analysis.

Purpose of the Study:

• To identify the optimal method for harmonizing nasal obstruction scores from multiple instruments.
• To develop a transformation method for equivalence between the NOSE, Utrecht-Q, and SCHNOS scales and the ENFAS scale.
• To enhance the precision of score prediction between different nasal complaint instruments.

Main Methods:

• Compared linear equating, Item Response Theory (IRT), and machine learning (ML) methods.
• Utilized linear regression, random forest, support vector machine, and neural networks for harmonization.
• Evaluated harmonization quality using root-mean-square error (RMSE) in 5-fold cross-validation with 1324 patients.

Main Results:

• Machine learning methods yielded superior harmonization results with the lowest RMSEs.
• ML approaches outperformed IRT, a common psychometric harmonization technique.
• All tested ML models demonstrated effectiveness in score prediction and harmonization.

Conclusions:

• Machine learning methods offer high-quality data harmonization for nasal obstruction assessment.
• ML methods present a powerful alternative to linear equating and IRT for harmonizing nasal complaint scores.
• Implementing ML can significantly improve data harmonization quality in specific nasal surgery research scenarios.