P 113 The dynamics of human REM sleep investigated by the analysis of time dependent transition probabilities

Abstract

Background Sleep can be conceptualized as a sequence of discernable vigilance states (sleep stages). When the polysomnogram is decomposed into bouts of subsequent epochs of the same sleep stage it can be shown from sufficiently large data bases that bout lengths follow a stochastic process characterized e.g. by exponential of power law distributions. A complementary method of analysis calculates the probability of transition to another stage dependent on the time spent in the initial stage (time dependent transition probabilities). Variations of these transition probabilities over time may reveal details of the dynamics of sleep stage change not evident from the analysis of bout length distributions alone. Detailed analyses of transition probabilities have been conducted e.g. for rat data (Bassi et al., 2009), but rarely for human polysomnographic data. Methods We used polysomnograms from a large database ( n = 373) to calculate time dependent transition probabilities with a focus on the dynamics of the interval between two REM episodes (NREM interval). Transition probabilities were estimated from observed transitions during time intervals Δ t and normalized to unit time (1 min). Results Transition probabilities are high during the first 10 min into an NREM interval probably reflecting REM inertia, i.e. the persistence of the preceding (biological) REM period even if a certain amount of NREM intervenes. This REM inertia is interrupted by > 60 s of intervening wake. In the further course of the interval, transition probabilites NREM to REM are reduced to very low values (0.005) for about 50 min. indicating a relatively REM refractory period. After this period, transition probabilites rise again. Although some segments of the curves are compatible with exponential or power-law distributions, the pronounced variations of transition probabilities over time cannot be explained by a static stochastical distribution. Discussion/Conclusions Inertia effects and relatively REM refractory periods arguably reflect the biological mechanisms underlying ultradian NREM-REM cycling. While inertia effects have direct counterparts in the rat data, the rat equivalent of REM suppression is less clear. The calculation of time dependent transition frequencies is a promising method that can give additional information complementing the analysis of bout durations.