Predicting sleep latency from the three-process model of alertness regulation

This paper presents a modification of the quantitative ''three-process model of alertness regulation'' to predict sleep latency in connection with irregular sleep/wake patterns. This model uses a circadian and a homeostatic component (sleep loss) that are summed to yield predicted alertness (on a scale of 1 to 20) across a specified time span. The timing of sleep from two studies of irregular sleep were used as input to the model. The predicted alertness at bedtime was regressed on empirical sleep latency from two studies. The maximum R(2) (0.88) was reached for an exponential function, with the model acrophase set to 2048. The predictions were cross validated on another set of sleep latency data from an irregular sleep study and a maximum R(2) of 0.65 was obtained. In both studies, the prediction from the model explained more variance that did self-rated alertness at bedtime. Cross validation was also carried out successfully with published data from two studies of shift work. It was concluded that sleep latency on irregular schedules may be predicted with accuracy from knowledge of the prior sleep/wake pattern. This may have practical consequences for rest/actvity management.