Latent Factor Decomposition Model: Applications for Questionnaire Data.

Area of Science:

• Statistical modeling
• Data science
• Clinical informatics

Background:

• Clinical questionnaire data presents challenges like missing fields and diverse data types.
• Existing methods for analyzing such data are often not robust, statistically sound, or interpretable.
• Principal Component Analysis (PCA) is a common dimensionality reduction technique but has limitations with mixed data types and missing values.

Purpose of the Study:

• To propose a novel latent factor modeling framework for analyzing clinical questionnaire data.
• To extend principal component analysis to handle both categorical and quantitative data with missing elements simultaneously.
• To provide a statistically sound and interpretable method for clinical data analysis.

Main Methods:

• Developed a latent factor model framework extending PCA.
• The model handles categorical and quantitative data with missing values.
• Simultaneously computes principal components (basis) and patient projections in a latent space.

Main Results:

• The proposed model effectively handles mixed data types and missing data.
• Demonstrated application in Irritable Bowel Syndrome (IBS) symptom analysis.
• Found significant correlations between patient projections and standardized symptom scales.

Conclusions:

• The latent factor model offers a robust and interpretable approach for clinical questionnaire data analysis.
• The framework facilitates clustering and inference on complex patient datasets.
• This method has broad applicability across various clinical research areas.