Optimizing Winter-Type Seasonality Criteria in East Asian Populations: A Machine Learning Approach Using the Seasonal Pattern Assessment Questionnaire.

backgroundThe Seasonal Pattern Assessment Questionnaire (SPAQ) evaluates seasonal variations in mood and behavior. According to Kasper's criteria, individuals meeting diagnostic standards for seasonal affective disorder (SAD) or subsyndromal SAD (S-SAD) are categorized as winter or summer type based on the month they "feel worst." However, in East Asian countries with hot, humid summers, relying solely on the "feel worst" item may misclassify seasonality. This study aimed to refine Kasper's criteria using machine learning to improve identification of winter-type seasonality.

methodsAmong 495 participants from a mood disorder cohort, SPAQ data from SAD and S-SAD cases were clustered using the K-Modes algorithm into winter type and other types. A decision tree algorithm identified winter seasonality with minimal SPAQ items.

resultsClustering aligned with additional SPAQ items beyond Kasper's criteria. Respondents selecting a winter month or "no particular month" as "feel worst" were classified as winter type if they also chose a winter month for "gain most weight," "sleep most," or "socialize least." Those selecting a summer month as "feel worst" were considered winter type if they marked a winter month for "gain most weight" or "sleep most."

conclusionThis study evaluated seasonality in a South Korean early-onset mood disorder cohort using SPAQ and Kasper's criteria. Incorporating atypical vegetative symptoms and reduced social activity improved winter seasonality classification accuracy. The revised criteria may facilitate more precise identification and management of seasonal symptoms in East Asia.