Use of a Latent Topic Model for Characteristic Extraction from Health Checkup Questionnaire Data.

Area of Science:

• Medical Informatics
• Data Science in Healthcare
• Patient-Reported Outcomes

Background:

• Subjective responses in health checkup questionnaires pose challenges for accurate topic extraction.
• Developing models to interpret this subjective data is crucial for comprehensive patient assessment.

Purpose of the Study:

• To develop and evaluate a model for extracting meaningful topics from subjective patient questionnaire data.
• To assess the utility of this model in identifying distinct patient groups based on their responses.

Main Methods:

• Utilized latent Dirichlet allocation (LDA), a latent topic model, to extract 30 topics from questionnaire data.
• Grouped 4381 participants based on these extracted topics and compared laboratory test results (e.g., glucose, eGFR) between groups.
• Compared the LDA model's grouping with hierarchical clustering.

Main Results:

• The latent topic model successfully identified a distinct group of participants reporting subjective urinary disorder symptoms.
• Significant differences (p < 0.05) in laboratory measurements were observed between groups, correlating with questionnaire response patterns.
• The latent topic model proved effective in isolating smaller, specific participant cohorts.

Conclusions:

• Latent topic modeling is a valuable tool for extracting characteristics from large, item-rich questionnaires, especially for identifying small, distinct patient groups.
• Questionnaire data, analyzed via this model, provides useful criteria for patient condition assessment alongside chief complaints and medical history.