Sleepiness and Safety: Where Biology Needs Technology

Abstract

Maintaining human alertness and behavioral capability under conditions of sleep loss and circadian misalignment requires fatigue management technologies due to: (1) dynamic nonlinear modulation of performance capability by the interaction of sleep homeostatic drive and circadian regulation; (2) large differences among people in neurobehavioral vulnerability to sleep loss; (3) error in subjective estimates of fatigue on performance; and (4) to inform people of the need for recovery sleep. Two promising areas of technology have emerged for managing fatigue risk in safety-sensitive occupations. The first involves preventing fatigue by optimizing work schedules using biomathematical models of performance changes associated with sleep homeostatic and circadian dynamics. Increasingly these mathematical models account for individual differences to achieve a more accurate estimate of the timing and magnitude of fatigue effects on individuals. The second area involves technologies for detecting transient fatigue from drowsiness. The Psychomotor Vigilance Test (PVT), which has been extensively validated to be sensitive to deficits in attention from sleep loss and circadian misalignment, is an example in this category. Two shorter-duration versions of the PVT recently have been developed for evaluating whether operators have sufficient behavioral alertness prior to or during work. Another example is online tracking the percent of slow eyelid closures (PERCLOS), which has been shown to reflect momentary fluctuations of vigilance. Technologies for predicting and detecting sleepiness/fatigue have the potential to predict and prevent operator errors and accidents in safety-sensitive occupations, as well as physiological and mental diseases due to inadequate sleep and circadian misalignment.

Keywords: PERCLOS; Psychomotor Vigilance Test (PVT); Sleepiness; biomathematical models; drowsiness; fatigue; safety; vigilance.

Figure 1

Simulation using the Bayesian forecasting procedure to predict future performance of three individuals, measured with the 10-minute PVT, during total sleep deprivation. Performance is predicted starting from t = 44h of wakefulness, with mean (black line) and 95% confidence intervals (vertical lines). Individual predictions are based on traits identified from prior performance measurements up to 44h (black dots). The gray circles show the actual performance measurements during the 24h prediction period. Figure reprinted with permission from Van Dongen and colleagues.

Figure 2

Mean percent time of slow eyelid closures (PERCLOS) coherence for PVT lapse frequency across 42h of waking (triangles), as a function of the time base used to define an epoch. A distance-weighted least squares function was fit to the data. PERCLOS was measured by a human scoring videos of slow eyelid closures (Experiment 1) and by infrared retinal reflectance (Experiment 2, CMRL). In both experiments it had much higher coherence with PVT performance lapses of attention (i.e., high sensitivity to behavioral alertness) than any other technology evaluated in the experiments (i.e., two different EEG algorithms [EEG-1, EEG-2), two different eye blink technologies [Eye blink-1, Eye blink-2], and head movement sensor technology [Head sensor]). PERCLOS was also a better predictor of alertness than subjects’ self-reports of sleepiness by a visual analogue scale (i.e., VAS sleepiness). The accuracy of PERCLOS predictions of PVT performance increased as the time base for integrated assessments increased from 1 to 20 minutes. More recent work also supports the accuracy of PERCLOS for unobtrusive detection of sleepiness while performing a behavioral maintenance of wakefulness test and the PVT . Figure reprinted from Dinges and colleagues.^, .