Machine-Learning Validated Short Form of the Korean Version of the Sleep-Related Behaviors Questionnaire-10 Items: SRBQ-10.

Objective: This study aimed to develop a short form of the Korean version of the Sleep-Related Behaviors Questionnaire (SRBQ) and assess its validity and psychometric properties.

Method: We collected 300 responses from the EMBRAIN survey system and conducted exploratory and confirmatory factor analyses to group SRBQ items based on response similarity. The most representative item from each group was selected using eXtreme Gradient Boosting (XGBoost). The psychometric properties of the final 10 items were assessed using the Rasch model of item response theory (IRT).

Results: Based on the selected 10 key items, we developed the SRBQ-10-a data-driven shortened version of the SRBQ, which demonstrated excellent performance (0.96) in predicting the SRBQ score, despite having only 10 items, which is one-third of the items in the original SRBQ-32. In addition to its reliability, the photometric properties of the SRBQ-10 were in the theoretically expected order, with no overlap or reversal of scale order, confirming the validity of the item scale.

Conclusion: The SRBQ-10, a concise version of SRBQ, enables efficient screening of sleep-related behaviors in clinical settings. Our study framework combining classical test theory, XGBoost, and IRT can be applied to develop and validate shorter versions of other questionnaires.