0006 A Prediction Model of Sleep Duration Based on Sleep Diary

Abstract Introduction Sleep diaries provide a subjective record of individual sleep and are widely used for evaluating daily sleep. However, they remain limited in capturing long-term and continuous aspects of sleep behavior. The two-process model has played a role in analyzing the interaction between sleep homeostasis (S(t)) and circadian rhythms (C(t)), thereby advancing the understanding of the sleep–wake rhythm. Recent studies have applied the model to actigraphy data, although practical issues–such as the requirement for continuous device use–still pose challenges. This study proposes implementing the two-process model based on sleep diary to predict sleep duration, with the aim of exploring its potential application as a decision aid. Methods Sleep diaries and actigraphy data from 25 individuals without sleep disorders were analyzed (measurement period: 10.9±1.1 days, mean±SD). Binary data were generated from the sleep diaries by setting sleep periods as 0 and wake periods as 1, and C(t) was estimated using cosinor analysis. S(t) was designed to decrease during sleep periods and increase during wake periods, with sleep debt incorporated into S(t) based on each participant’s habitual and recovery sleep duration. For n days of prior data, the values of n and k were determined by applying C(t) and S(t)+k to minimize the error in estimating sleep duration. Using these optimized parameters, the two-process model was then applied to predict sleep duration for the subsequent day. Results Sleep data from 178 nights across 25 participants were evaluated by comparing predicted sleep duration with actigraphy-derived reference values. The root mean squared error (RMSE) of prediction was 1.31±0.60 h, significantly smaller than the RMSE obtained without incorporating sleep debt (p< 0.01, Wilcoxon signed-rank test). The correlation coefficient between individual mean predicted sleep duration and reference sleep duration was 0.81 (p< 0.001). Conclusion The proposed sleep diary–based model demonstrated reliable agreement with actigraphy-derived values. This approach can offer an accessible tool for evaluating sleep–wake patterns, with potential applications in decision-making and personalized sleep management. Support (if any) This study was supported by funds donated to Mind the SHIM (Seoul national university hospital Health In Mind) Center, Seoul National University Hospital, Republic of Korea.