Patent No. 6553252 Method and system for predicting human cognitive performance

 

Patent No. 6553252

Method and system for predicting human cognitive performance (Balkin, et al., Apr 22, 2003)

Abstract

An apparatus and method for predicting cognitive performance of an individual based on factors including sleep history and the time of day. The method facilitates the creation of predicted cognitive performance curves that allow an individual to set his/her sleep times to produce higher levels of cognitive performance. The method also facilitates the reconstruction of past cognitive performance levels based on sleep history.

Notes:

FIELD OF THE INVENTION

This invention relates to a method for predicting cognitive performance of an individual based on that individual's prior sleep/wake history and the time of day.

BACKGROUND OF THE INVENTION

Maintenance of productivity in any workplace setting depends upon effective cognitive performance at all levels from command/control or management down to the individual soldier or worker. Effective cognitive performance in turn depends upon complex mental operations. Many factors have been shown to affect cognitive performance (e.g., drugs or age). However, of the numerous factors causing day to day variations in cognitive performance, two have been shown to have the greatest impact. These two factors are an individual's prior sleep/wake history and the time of day.

Adequate sleep sustains cognitive performance. With less than adequate sleep, cognitive performance degrades over time. An article by Thorne et al. entitled "Plumbing Human Performance Limits During 72 hours of High Task Load" in Proceedings of the 24.sup.th DRG Seminar on the Human as a Limiting Element in Military Systems, Defense and Civil Institute of Environmental Medicine, pp. 17-40 (1983), an article by Newhouse et al. entitled "The Effects of d-Amphetamine on Arousal, Cognition, and Mood After Prolonged Total Sleep Deprivation" published in Neuropsychopharmacology, vol. 2, pp. 153-164 (1989), and another article by Newhouse et al. entitled "Stimulant Drug Effects on Performance and Behavior After Prolonged Sleep Deprivation: A Comparison of Amphetamine, Nicotine, and Deprenyl" published in Military Psychology, vol. 4, pp. 207-233 (1992) all describe studies of normal volunteers in which it is revealed that robust, cumulative decrements in cognitive performance occur during continuous total sleep deprivation as measured by computer-based testing and complex operational simulation. In the Dinges et al. article entitled "Cumulative Sleepiness, Mood Disturbance, and Psychomotor Vigilance Performance Decrements During a Week of Sleep Restricted to 4-5 Hours Per Night" published in Sleep, vol. 20, pp. 267-277 (1997) it is revealed that on fixed, restricted daily sleep amounts, cumulative reduced sleep also leads to a cognitive performance decline. Thus, in operational settings, both civilian and military, sleep deprivation reduces productivity (output of useful work per unit of time) on cognitive tasks.

Thus, using computer-based cognitive performance tests, it has been shown that total sleep deprivation degrades human cognitive performance by approximately 25% for each successive period of 24 hours awake. However, it also has been shown that even small amounts of sleep reduce the rate of sleep loss-induced cognitive performance degradation: Belenky et al. in their article entitled "Sustaining Performance During Continuous Operations: The U.S. Army's Sleep Management System," published in 20.sup.th Army Science Conference Proceedings, vol. 2, pp. 657-661 (1996), disclose that a single 30-minute nap every 24 hours reduces the rate of cognitive performance degradation to 17% per day over 85 hours of sleep deprivation. This suggests that recuperation of cognitive performance during sleep accrues most rapidly early in the sleep period. No other factor besides the amount of sleep contributes so substantially and consistently to the normal, daily variations in cognitive performance.

In addition to sleep/wake history, an individual's cognitive performance at a given point in time is determined by the time of day. In the early 1950s, Franz Halberg and associates observed a 24-hour periodicity in a host of human physiologic (including body temperature and activity), hematologic, and hormonal functions, and coined the term `circadian` (Latin for `about a day`) to describe this cyclic rhythm. Halberg showed that most noise in experimental data came from comparisons of data sampled at different times of day.

When humans follow a nocturnal sleep/diurnal wake schedule (for example, an 8-hour sleep/16-hour wake cycle, with nightly sleep commencing at approximately midnight), body temperature reaches a minimum (trough) usually between 2:00 AM and 6:00 AM. Body temperature then begins rising to a maximum (peak) usually between 8:00 PM and 10:00 PM. Likewise, systematic studies of daily human cognitive performance rhythms show that speed of responding slowly improves across the day to reach a maximum in the evening (usually between 8:00 PM and 10:00 PM) then dropping more rapidly to a minimum occurring in the early morning hours (usually between 2:00 AM and 6:00 AM). Similar but somewhat less consistent rhythms have been shown from testing based on various cognitive performance tasks. Thus, superimposed on the effect of total sleep deprivation on cognitive performance noted above was an approximately .+-.10% percent variation in cognitive performance over each 24-hour period.

Various measures have been shown to correlate, to some extent, with cognitive performance. These include objective and subjective measures of sleepiness (or its converse, alertness). Some individuals familiar with the art use "sleepiness" to indicate the opposite of "alertness" (as is the case in the present document). "Drowsiness" often is used interchangeably with "sleepiness" although some familiar with the art would argue that "sleepiness" pertains specifically to the physiological need for sleep whereas "drowsiness" refers more to the propensity or ability to fall asleep (independent of physiological sleep need) or the subjective feeling of lack of alertness. The term "fatigue" has been used as a synonym for "sleepiness" by the lay population, but those familiar with the art do not consider "fatigue" to be interchangeable with "sleepiness"--rather, "fatigue" is a broad term that encompasses more than just the effects of sleep loss per se on performance. Likewise, "cognitive performance" has been defined as performance on a wide variety of tasks, the most commonly used being vigilance tasks (tasks requiring sustained attention). From vigilance and other tasks, some researchers use accuracy as their measure of cognitive performance, while others use reaction time (or its inverse, speed). Still others use a measure that is calculated as speed multiplied by accuracy, that is the amount of useful work performed per unit of time (also known as throughput). Those familiar with the art generally agree that vigilance tasks are appropriate measures of cognitive performance under conditions of sleep deprivation, and that either reaction time (speed) or some measure that takes reaction time into account (e.g., throughput) is a valid and reliable way of measuring cognitive performance.

The Multiple Sleep Latency Test (MSLT) is a widely accepted objective measure of sleepiness/alertness. In the MSLT, individuals try to fall asleep while lying in a darkened, quiet bedroom. Various physiological measures used to determine sleep or wakefulness are recorded (eye movements, brain activity, muscle tone), and time taken to reach the first 30 seconds of stage 1 (light) sleep is determined. Shorter latencies to stage 1 are considered to indicate greater sleepiness (lower alertness). Sleep latencies under 5 minutes are considered to be pathological (i.e., indicative of a sleep disorder or sleep deprivation). During both total and partial sleep deprivation, latency to sleep on the MSLT (alertness) and performance decline (i.e., sleepiness as measured by MSLT increases). However, although there is a correlation between MSLT-determined sleepiness/alertness and cognitive performance (greater sleepiness as indexed by MSLT corresponding to poorer cognitive performance), this correlation has never been shown to be perfect and for the most part is not strong. As a result, the MSLT is a poor (i.e., unreliable) predictor of cognitive performance.

Subjective measures of sleepiness/alertness also have been shown to correlate (albeit weakly) with cognitive performance. Hoddes et al., in their article entitled "Quantification of Sleepiness: A New Approach" published in Psychophysiology, vol. 10, pp. 431-436 (1973) describe the Stanford Sleepiness Scale (SSS), a subjective questionnaire used widely to measure sleepiness/alertness. In the SSS, individuals rate their current level of sleepiness/alertness on a scale from 1 to 7, with 1 corresponding to the statement, "feeling active and vital; alert; wide awake" and 7 corresponding to the statement "almost in reverie; sleep onset soon; losing struggle to remain awake." Higher SSS scores indicate greater sleepiness. As with the MSLT, during both total and partial sleep deprivation, scores on the SSS increase. However, as with MSLT, the correspondence between SSS-determined sleepiness/alertness and cognitive performance decrements is weak and inconsistent. As a result, the SSS also is a poor predictor of cognitive performance. Some other examples of subjective measures of sleepiness/alertness include the Epworth Sleepiness Scale described by Johns in his article entitled "Daytime Sleepiness, Snoring, and Obstructive Sleep Apnea" published in Chest, vol. 103, pp. 30-36 (1993), and the Karolinska Sleepiness scale described by Akerstedt and Gillberg in their article entitled "Subjective and Objective Sleepiness in the Active Individual" published in International Journal of Neuroscience, vol. 52, pp. 29-37 (1990). The correspondence between these subjective measures and cognitive performance also is weak and inconsistent.

In addition, factors modifying cognitive performance may not correspondingly affect objective or subjective measures of sleepiness/alertness, and vice versa. For example, the Penetar et al. article entitled "Amphetamine Effects on Recovery Sleep Following Total Sleep Deprivation" published in Human Psychopharmacology, vol. 6, pp. 319-323 (1991), disclose that during sleep deprivation, the stimulant drug d-amphetamine improved cognitive performance but not sleepiness/alertness (as measured by the MSLT). In a similar study, caffeine given as a sleep deprivation countermeasure maintained elevated cognitive performance for over 12 hours while the effects on subjective sleepiness, vigor and fatigue transiently improved but then decayed. Thorne et al. in their article entitled "Plumbing Human Performance Limits During 72 hours of High Task Load" in Proceedings of the 24.sup.th DRG Seminar on the Human as a Limiting Element in Military Systems, Defense and Civil Institute of Environmental Medicine, pp. 17-40 (1983), describe how cognitive performance continues to decline over 72 hours of sleep deprivation whereas subjective sleepiness/alertness declined over the first 24 hours but subsequently leveled off. The findings that cognitive performance and measures of sleepiness/alertness are not always affected in the same way indicate that they are not interchangeable. That is, measures of sleepiness/alertness cannot be used to predict cognitive performance, and vice versa.

Methods and apparatuses related to alertness detection fall into five basic categories: a method/apparatus for unobtrusively monitoring current alertness level; a method/apparatus for unobtrusively monitoring current alertness level and providing a warning/alarm to the user of decreased alertness and/or to increase user's alertness level; a method/apparatus for monitoring current alertness level based on the user's responses to some secondary task possibly with an alarm device to warn the user of decreased alertness and/or to increase user's alertness level; methods to increase alertness; and a method/apparatus for predicting past, current, or future alertness.

These methods and apparatuses that unobtrusively monitor the current alertness level are based on an "embedded measures" approach. That is, such methods infer alertness/drowsiness from the current level of some factor (e.g., eye position or closure) assumed to correlate with alertness/drowsiness. Some recently issued patents of this type include U.S. Pat. No. 5,689,241 to J. Clarke, Sr., et al. disclosing an apparatus to detect eye closure and ambient temperature around the nose and mouth; U.S. Pat. No. 5,682,144 to K. Mannik disclosing an apparatus to detect eye closure; and U.S. Pat. No. 5,570,698 to C. Liang et al. disclosing an apparatus to monitor eye localization and motion to detect sleepiness. An obvious disadvantage of these types of methods and apparatuses is that the measures are likely detecting sleep onset itself rather than small decreases in alertness.

In some patents, methods for embedded monitoring of alertness/drowsiness are combined with additional methods for signaling the user of decreased alertness and/or increasing alertness. Recently issued patents of this type include U.S. Pat. No. 5,691,693 to P. Kithil describing a device that senses a vehicle operator's head position and motion to compare current data to profiles of "normal" head motion and "impaired" head motion. Warning devices are activated when head motion deviates from the "normal" in some predetermined way. U.S. Pat. No. 5,585,785 to R. Gwin et al. describes an apparatus and a method for measuring total handgrip pressure on a steering wheel such that an alarm is sounded when the grip pressure falls below a predetermined "lower limit" indicating drowsiness. U.S. Pat. No. 5,568,127 to H. Bang describes a device for detecting drowsiness as indicated by the user's chin contacting an alarm device, which then produces a tactile and auditory warning. U.S. Pat. No. 5,566,067 to J. Hobson et al. describes a method and an apparatus to detect eyelid movements. A change in detected eyelid movements from a predetermined threshold causes an output signal/alarm (preferably auditory). As with the first category of methods and apparatuses, a disadvantage here is that the measures are likely detecting sleep onset itself rather than small decreases in alertness.

Other alertness/drowsiness monitoring devices have been developed based on a "primary/secondary task" approach. For example, U.S. Pat. No. 5,595,488 to E. Gozlan et al. describes an apparatus and a method for presenting auditory, visual, or tactile stimuli to an individual to which the individual must respond (secondary task) while performing the primary task of interest (e.g., driving). Responses on the secondary task are compared to baseline "alert" levels for responding. U.S. Pat. No. 5,259,390 to A. MacLean describes a device in which the user responds to a relatively innocuous vibrating stimulus. The speed to respond to the stimulus is used as a measure of the alertness level. A disadvantage here is that the apparatus requires responses to a secondary task to infer alertness, thereby altering and possibly interfering with the primary task.

Other methods exist solely for increasing alertness, depending upon the user to self-evaluate alertness level and activate the device when the user feels drowsy. An example of the latter is U.S. Pat. No. 5,647,633 and related patents to M. Fukuoka in which a method/apparatus is described for causing the user's seat to vibrate when the user detects drowsiness. Obvious disadvantages of such devices are that the user must be able to accurately self-assess his/her current level of alertness, and that the user must be able to correctly act upon this assessment.

Methods also exist to predict alertness level based on user inputs known empirically to modify alertness. U.S. Pat. No. 5,433,223 to M. Moore-Ede et al. describes a method for predicting the likely alertness level of an individual at a specific point in time (past, current or future) based upon a mathematical computation of a variety of factors (referred to as "real-world" factors) that bear some relationship to alterations in alertness. The individual's Baseline Alertness Curve (BAC) is first determined based on five inputs and represents the optimal alertness curve displayed in a stable environment. Next, the BAC is modified by alertness modifying stimuli to arrive at a Modified Baseline Alertness Curve. Thus, the method is a means for predicting an individual's alertness level, not cognitive performance.

Another method has been designed to predict "work-related fatigue" as a function of number of hours on duty. Fletcher and Dawson describe their method in an article entitled "A Predictive Model of Work-Related Fatigue Based on Hours of Work" published in Journal of Occupational Health and Safety, vol. 13, 471-485 (1997). In this model a simplifying assumption is made--it is assumed that length of on-duty time correlates positively with time awake. To implement the method, the user inputs a real or hypothetical on-duty/off-duty (work/rest) schedule. Output from the model is a score that indicates "work-related fatigue." Although this "work-related fatigue" score has been shown to correlate with some performance measures, it is not a direct measure of cognitive performance per se. It can be appreciated that the fatigue score will be less accurate under circumstances when the presumed relationship between on-duty time and time awake breaks down--for example when a person works a short shift but then spends time working on projects at home rather than sleeping or when a person works long shifts but conscientiously sleeps all the available time at home. Also, this method is obtrusive in that the user must input on-duty/off-duty information rather than such information being automatically extracted from an unobtrusive recording device. In addition, the model is limited to predictions of "fatigue" based on work hours. Overall, this model is limited to work-related situations in which shift length consistently correlates (inversely) with sleep length.

Given the importance of the amount of sleep and the time of day for determining cognitive performance (and hence estimating productivity or effectiveness), and given the ever-increasing requirements of most occupations on cognitive performance, it is desirable to design a reliable and accurate method of predicting cognitive performance. It can be appreciated that increasing the number of relevant inputs increases cognitive performance prediction accuracy. However, the relative benefits gained from such inputs must be weighed against the additional burdens/costs associated with their collection and input. For example, although certain fragrances have been shown to have alertness-enhancing properties, these effects are inconsistent and negligible compared to the robust effects of the individual's sleep/wake history and the time of day. More important, the effect of fragrances on cognitive performance is unknown. Requiring an individual to keep a log of exposure to fragrances would be time consuming to the individual and only result in negligible gains in cognitive performance prediction accuracy. In addition, while the effects of the sleep/wake history and the time of day on cognitive performance are well known, the effects of other putative alertness-altering factors (e.g., job stress and workload), how to measure them (their operational definition), and their direction of action (cognitive performance enhancing or degrading) are virtually unknown.

An important and critical distinction between the present invention and the prior art is that the present invention is a model to predict performance on tasks with a cognitive component. In contrast, previous models involving sleep and/or circadian rhythms (approximately 24-hour) focused on the prediction of "alertness" or "sleepiness." The latter are concepts that specifically relate to the propensity to initiate sleep, not the ability to perform a cognitive task.

Although sleepiness (or its converse, alertness) could be viewed as an intervening variable that can mediate cognitive performance, the scientific literature clearly shows that cognitive performance and alertness are conceptually distinct, as reviewed by Johns in the article entitled, "Rethinking the Assessment of Sleepiness" published in Sleep Medicine Reviews, vol. 2, pp. 3-15 (1998), and as reviewed by Mitler et al. in the article entitled, "Methods of Testing for Sleepiness" published in Behavioral Medicine, vol. 21, pp. 171-183 (1996). Thomas et al. in the article entitled "Regional Cerebral Metabolic Effects of Prolonged Sleep Deprivation" published in Neurolmage, vol. 7, p. S130 (1998) reveal that 1-3 days of sleep loss result in reductions in global brain activation of approximately 6%, as measured by regional cerebral glucose uptake. However, those regions (heteromodal association cortices) that mediate the highest order cognitive functions (including but not limited to attention, vigilance, situational awareness, planning, judgment, and decision making) are selectively deactivated by sleep loss to a much greater extent--up to 50%--after three days of sleep loss. Thus, decreases in neurobiological functioning during sleep restriction/deprivation are directly reflected in cognitive performance degradation. These findings are consistent with studies demonstrating that tasks requiring higher-order cognitive functions, especially those tasks requiring attention, planning, etc. (abilities mediated by heteromodal association areas) are especially sensitive to sleep loss. On the other hand, brain regions such as primary sensory regions, are deactivated to a lesser degree. Concomitantly, performance (e.g., vision, hearing, strength and endurance tasks) that is dependent on these regions is virtually unaffected by sleep loss.

Consequently, devices or inventions that predict "alertness" per se (e.g., Moore-Ede et al.) putatively quantify the brain's underlying propensity to initiate sleep at any given point in time. That is, devices or inventions that predict "alertness" (or its converse "sleepiness") predict the extent to which sleep onset is likely. The present invention differs from such approaches in that the nature of the task is accounted for--i.e., it is not the propensity to initiate sleep that is predicted. Rather, the present invention predicts the extent to which performance of a particular task will be impaired by virtue of its reliance upon brain areas most affected by sleep deprivation (heteromodal association areas of the brain). The most desirable method will produce a highly reliable and accurate cognitive performance estimate based on the sleep/wake history of an individual and the time of day.