Mining statistically significant associations for exploratory analysis of human sleep data

Area of Science:

• Sleep Science
• Data Mining
• Biostatistics

Background:

• Complex sleep data, including polysomnography, clinical summaries, and questionnaires, presents challenges for pattern discovery.
• Existing methods may not effectively integrate diverse data types for comprehensive sleep analysis.

Purpose of the Study:

• To introduce a specialized association rule mining technique for extracting patterns from multifaceted sleep data.
• To identify clinically relevant associations between polysomnographic events, clinical summaries, and questionnaire responses.

Main Methods:

• Developed a specialized association rule mining technique.
• Applied chi-squared analysis to ensure statistical significance (P < 0.05).
• Utilized data from 242 human subjects, integrating polysomnographic recordings, clinical summaries, and sleep questionnaire responses.

Main Results:

• Successfully extracted clinically interesting associations among polysomnographic, summary, and questionnaire variables.
• Demonstrated the technique's ability to link specific sleep events (e.g., REM sleep) with factors like caffeine intake.
• Identified potential utility of association mining for variable selection in logistic regression.

Conclusions:

• The specialized association rule mining technique effectively uncovers significant patterns in complex sleep data.
• This approach reveals clinically meaningful insights into sleep patterns and associated factors.
• Association mining shows promise as a preliminary step for predictive modeling in sleep research.