Impaired decision making following 49 h of sleep deprivation

Authors


William D. S. Killgore, Department of Behavioral Biology, Walter Reed Army Institute of Research (WRAIR), 503 Robert Grant Avenue, Silver Spring, MD 20910, USA. Tel.: (301) 319-9391; fax: (301) 319-9979; e-mail: william.d.killgore@us.army.mil

Summary

Sleep deprivation reduces regional cerebral metabolism within the prefrontal cortex, the brain region most responsible for higher-order cognitive processes, including judgment and decision making. Accordingly, we hypothesized that two nights of sleep loss would impair decision making quality and lead to increased risk-taking behavior on the Iowa Gambling Task (IGT), which mimics real-world decision making under conditions of uncertainty. Thirty-four healthy participants completed the IGT at rested baseline and again following 49.5 h of sleep deprivation. At baseline, volunteers performed in a manner similar to that seen in most samples of healthy normal individuals, rapidly learning to avoid high-risk decks and selecting more frequently from advantageous low-risk decks as the game progressed. After sleep loss, however, volunteers showed a strikingly different pattern of performance. Relative to rested baseline, sleep-deprived individuals tended to choose more frequently from risky decks as the game progressed, a pattern similar to, though less severe than, previously published reports of patients with lesions to the ventromedial prefrontal cortex. Although risky decision making was not related to participant age when tested at rested baseline, age was negatively correlated with advantageous decision making on the IGT, when tested following sleep deprivation (i.e. older subjects made more risky choices). These findings suggest that cognitive functions known to be mediated by the ventromedial prefrontal cortex, including decision making under conditions of uncertainty, may be particularly vulnerable to sleep loss and that this vulnerability may become more pronounced with increased age.

Introduction

Sleep loss is associated with decrements in basic cognitive functions including alertness, reaction time, attention, and vigilance (Dinges et al., 1997; Glenville et al., 1978; Horne et al., 1983; Wesensten et al., 2004). While there is considerable evidence to suggest that sleep deprivation impairs performance on simple tasks involving alertness and vigilance, the effects of sleep loss on higher-order cognitive processing are less clear (Harrison and Horne, 2000; Jones and Harrison, 2001). These complex cognitive processes, often collected under the umbrella term ‘executive functions,’ include a wide variety of abilities such as planning, sequencing, organizing, and coordinating willful and directed action, concept formation, mental abstraction, flexible thought, self-monitoring, selective attention, inhibition, and conflict resolution (Jones and Harrison, 2001; Jonides et al., 2002; Lezak, 1995). A substantial body of neuropsychological evidence suggests that these diverse executive functions are largely the product of neural activity within the prefrontal cortex (Robbins, 1998). Results from positron emission tomography studies suggest that as little as 24 h of sleep deprivation (i.e. one night of sleep loss) significantly reduces prefrontal metabolic activity (Thomas et al., 2000). Moreover, reduced activity within the prefrontal cortex is associated with corresponding decrements in performance on tasks requiring complex cognitive processing (Thomas et al., 2003), suggesting a neurobiological basis for executive function deficits observed during sleep deprivation.

Several studies have reported that 24–36 h of sleep loss impairs certain types of executive functions such as supervisory control (Nilsson et al., 2005), problem solving, divergent thinking capacity (Horne, 1988; Linde and Bergstrom, 1992), verbal creativity, flexibility, inhibition (Harrison and Horne, 1998), and cognitive set shifting (Wimmer et al., 1992), whereas in other studies, no specific effects on tasks of executive function were found after similar amounts of sleep loss (Binks et al., 1999; Drummond et al., 2004; Sagaspe et al., 2003). One aspect of executive function that is relevant to real-world problem-solving, but which has not been adequately explored in studies of sleep deprivation, is the ability to make advantageous decisions under conditions of uncertainty. A prominent theory of decision making, known as the somatic marker hypothesis (Damasio, 1996), suggests that the ability to decide advantageously requires intact functioning of the ventromedial prefrontal cortex, a sector of the frontal lobes that shows evidence of decreased metabolic activity during sleep deprivation (Thomas et al., 2003). The ventromedial prefrontal cortex is believed to be important for forming and maintaining links between certain classes of stimuli or experience and the associated emotional or ‘feeling’ state with which it was previously paired. This hypothesized linkage between feelings and complex situations allows emotions to bias or guide decision making. Damage to the ventromedial prefrontal cortex appears to disrupt the ability to utilize these emotional ‘markers’ to guide decision making, thus leading to deficits in the ability to learn from past experience and make advantageous decisions (Bechara et al., 2000a). Because sleep deprivation significantly reduces glucose metabolism within the prefrontal cortex (Thomas et al., 2000, 2003), including the ventromedial regions involved in making decisions based on learning of reward and punishment contingencies (Bechara et al., 1994, 1997, 2000b), we predicted that sleep-deprived volunteers would show a deficit in decision making that would mimic the performance of patients with damage to the ventromedial prefrontal cortex.

Bechara et al. (1994, 1997, 2000b) at the University of Iowa have developed a paradigm known as the Iowa Gambling Task (IGT), which mimics real-world decision making and has been shown to differentiate patients with damage to the ventromedial prefrontal cortex from healthy controls and other groups of brain-injured patients. Results from recent functional neuroimaging studies also confirm that performance on this task correlates with brain activity in the ventromedial prefrontal and orbitofrontal cortices (London et al., 2000). The IGT requires participants to choose among four decks of cards that differ in their inherent payoff schedules. Two of the decks yield high immediate payoffs but relatively larger long-term penalties, leading to a net loss overall; the other two decks yield low immediate payoffs but even smaller long-term losses, resulting in an overall net gain by the end of the game (Bechara et al., 1999). When performing the Gambling Task, normal volunteers learn to avoid selecting from the riskier ‘bad’ decks and show a preference for selecting from the less risky ‘good’ decks. In contrast, patients with lesions to the ventromedial prefrontal cortex show a preference for choosing from the disadvantageous decks and seldom learn that their decisions consistently lead them toward large long-term losses The latter phenomenon has been described as ‘myopia for the future’ (Bechara et al., 2000b, p. 2189) and is a hallmark of ventromedial prefrontal lobe damage.

For the present study, we predicted that sleep-deprived volunteers would show a deficit in decision making on the IGT similar to, though less severe than, patients with ventromedial prefrontal cortex lesions. Volunteers were tested on the IGT while fully rested and again following 49.5 h of total sleep deprivation. Our goal was to evaluate the effect of sleep deprivation on the ability to learn from implicit reward and punishment contingencies and to apply such information toward an advantageous decision making strategy.

Methods

Subjects

Study participants initially included 48 healthy volunteers (38 men; 10 women), ranging in age from 19 to 39 years (mean age = 25.2, SD = 5.7). Due to a technical malfunction in the computer administration, data for the IGT were incomplete for 14 participants. Therefore, the present results are based on 34 (24 men; 10 women) volunteers (mean age = 24.9, SD = 5.5). Participants were recruited through advertisements placed at local universities in the Washington, DC metropolitan area. Study requirements and procedures were explained and written informed consent was obtained from all participants. Prior to participation, volunteers were screened (medical history and medical exam) for past and present physical, neurological, psychiatric, and sleep problems. All participants were free of any history of psychiatric or neurological problems or other medical issues that would pose a significant health risk or affect cognitive performance. Volunteers who reported daily caffeine consumption in excess of 400 mg/day and/or use of tobacco products within the last 36 months were also excluded from the study. Participants were required to abstain from alcohol, stimulants (including caffeine-containing foods and beverages such as chocolate and coffee), and other psychoactive drugs for 48 h prior to the arrival at the laboratory. Compliance with these requirements was verified by a urine drug screen upon arrival and at regular intervals throughout the study. The protocol and procedures of this study were approved by the Walter Reed Army Institute of Research Human Use Review Committee and the US Army Human Subjects Research Review Board.

Participant motivation

Participants were paid a base fee of $1500 for successful completion of all study procedures, plus a ‘performance bonus’ of ‘up to $500’ to ensure high motivation to perform well on all the tasks. Participants were informed that the bonus would be dependent upon their level of performance on the IGT and other cognitive tasks and that the money they won on the game would contribute partly toward their total bonus earnings. However, the exact criterion used to determine bonus payment was left unspecified to the participants. In actuality, the bonus criterion was set low enough that all participants were guaranteed of receiving the maximum $500 bonus as long as they completed the tasks at a chance or better level. Thus, participants were motivated to perform at their best in order to receive the maximum bonus, but were never at risk of not receiving the maximum $500 payment as long as they attempted to complete the task.

Iowa Gambling Task

The IGT assesses the ability to modify a decision making strategy based on implicit learning of punishment and reward contingencies (Bechara et al., 1994, 2000b). The task is presented as a card game on a computer screen. Volunteers were presented with four identically appearing decks of cards (labeled A′, B′, C′, and D′ at baseline and K′, L′, M′, N′ following sleep deprivation). Volunteers were instructed to try to win as much money as possible by selecting cards, one at a time, from among the four decks. In order to ensure a high level of motivation, participants were informed that their performance on the Gambling Task would contribute partially to a ‘performance bonus’ at the end of the study. They were informed that their performance on the Gambling Task would contribute toward a maximum bonus of ‘up to $500.’ Each time they selected from a deck, volunteers won or lost a predetermined amount of money which was displayed on the computer screen and which they believed would contribute to their final bonus earnings. For two of the decks in the IGT, the amount of money won is generally high, but occasionally an extremely high-penalty loss is incurred. Consequently, these two decks are considered to be ‘disadvantageous’ or ‘bad decks’ because playing them consistently will lead to a net loss. For the other two decks, the amount of money won is typically less than that for the first two decks, but the penalties are much lower. Consequently, these latter two decks are considered to be ‘advantageous’ or ‘good decks’ because playing them consistently will lead to a net gain. Volunteers were informed that ‘some decks are better than others’ but were not given any other clues as to which decks to choose. Each Gambling Task session consisted of 100 card selections.

The following instructions represent a slight modification of the original instructions that were provided by the test developer (A. Bechara, University of Iowa) with the IGT and were read aloud individually to every participant just prior to each administration of the Gambling Task:

In front of you on the screen, there are 4 decks of cards. When we begin the game, I want you to select one card at a time by clicking on a card from any deck you choose. Each time you select a card, the computer will tell you that you won some money. I don't know how much money you will win. You will find this out as you go along. Every time you win, the green bar at the top of the screen gets bigger. Every so often, when you click on a card, the computer will tell you that you won some money as usual, but then it will say that you lost some money as well. I don't know when you will lose or how much. You will find out as you go along. Every time you lose, the green bar at the top of the screen gets smaller. You are absolutely free to switch from one deck to the other at any time, and as often as you wish. The goal of the game is to win as much money as possible and avoid losing as much money as possible. Remember, your performance on this task will contribute to the total performance bonus at the end of the study, worth up to a maximum of $500. You won't know when the game will end. Simply keep on playing until the computer stops. You will have $2000 of credit, shown by the green bar, to start the game. You have borrowed this money, so the red bar is a reminder that you borrowed $2000 to play the game. The red bar keeps track of how much money you have to pay back before we see whether you won or lost. The only hint I can give you, and the most important thing to note, is this: Out of these four decks of cards, some are worse than others. To win, you should try to stay away from bad decks. No matter how much you find yourself losing, you can still win the game if you avoid the worst decks. Also, the computer does not change the position of the decks once the game begins. It does not make you lose at random, or make you lose money based on the last card you picked. Once you pick a card, you have to wait until the computer says, ‘PICK A CARD’ before you can select your next card. When you are ready, you may pick the first card by clicking on it with the mouse. Remember, try to win as much as you can and try to avoid losing as much as you can. Put your headphones on and you may begin.

Study design and procedures

The present data were collected as part of a larger study in which the effects of stimulant medications on performance and alertness during sleep deprivation were evaluated (Wesensten et al., in press). Volunteers participated in groups of four at a time. They reported to the laboratory at 18:00 hours (Day 1) and were fitted with electrodes for polysomnographic recording. During the first night, subjects were allowed to acclimate to the laboratory and the basic study procedures were explained. Participants were introduced to the computer console where testing would occur and were given practice on several other computerized tasks. Subjects were not, however, introduced to the IGT until the following day. Bedtime was at 23:00 hours, and participants remained in bed with lights out for eight continuous hours. Following awakening at 7:00 hours on Day 2, volunteers began regular psychomotor vigilance testing at bihourly intervals throughout the sleep deprivation period. At 8:35 hours Day 2, volunteers completed a rested baseline session of the Gambling Task (A′, B′, C′, D′), where C′ and D′ were the advantageous decks. At 8:35 hours Day 4 (following 49.5 h of continuous wakefulness), volunteers completed an alternate version of the gambling task (K′, L′, M′, N′), where K′ and M′ were the advantageous decks. Performance was scored by first dividing the 100 Gambling Task trials (selections) into five blocks of 20 contiguous selections each. Within each Block, the number of selections from the advantageous decks (e.g. C′ and D′) and disadvantageous decks (e.g. A′ and B′) were summed separately. A net score was calculated for each Block by subtracting the number of cards chosen from the disadvantageous decks from the number of cards chosen from the advantageous decks (e.g. net score = [C′ + D′] – [A′ + B′]). Positive net scores indicated advantageous responding whereas negative net scores indicated disadvantageous responding (Bechara and Damasio, 2002).

Results

Net scores (indicating the relative number of card selections for good versus bad decks) were entered into a 2 × 5, Condition (rested baseline versus sleep deprived) × Block (selections 1–20, 21–40, etc.), repeated measures analysis of variance with Greenhouse–Geisser corrections. Fig. 1 shows mean net scores on the Gambling Task at rested baseline versus sleep deprived (i.e. following 49.5 h of continuous wakefulness). For comparison purposes, a reference line is provided which was derived by averaging reported Gambling Task performances from five previously published samples of patients with documented ventromedial prefrontal lesions (Bechara and Damasio, 2002; Bechara et al., 1999, 2000a,b).

Figure 1.

When tested at a well-rested baseline (bsl00043), participants demonstrated a gradual linear shift away from disadvantageous decks (A′ and B′) and toward advantageous decks (C′ and D′). In contrast, following 49.5 h of sleep deprivation (bsl00001), the same subjects showed a significantly different pattern of performance that was best characterized by a quadratic function, demonstrating a shift in behavior toward advantageous decks in the first half of the game followed by a reversal of behavior toward progressively more choices from disadvantageous decks in the latter half of the game. For comparison, the figure also shows a composite line calculated by extrapolating and summarizing Gambling Task data reported from five separate studies (total n = 48) of patients with damage to the ventromedial prefrontal cortex (bsl00066) (Bechara et al., 1999, Bechara and Damasio, 2002; Bechara et al., 2000a,b).

As seen in Fig. 1, a significant interaction between Condition and Block was found for net scores, F[4, 128] = 5.80, P = 0.001. This interaction was deconstructed by conducting a repeated measures trend analysis on net scores across Blocks separately for the rested baseline versus sleep-deprived conditions. For the rested baseline condition, a significant effect of Block was found, F[4, 128] = 10.19, P < 0.001; trend analyses revealed that volunteers showed a linear shift in selection behavior away from bad decks (A′ and B′) and toward good decks (C′ and D′) across blocks as the game progressed, FLinear [1, 32] = 21.35, P < 0.001. Following 49.5 h of sleep deprivation, a significant Block effect was also observed, F[4, 132] = 2.81, P < 0.05; however, in contrast to the rested baseline condition, under sleep-deprived conditions volunteers failed to show the expected linear shift in selection away from bad decks and toward good decks, FLinear [1, 33] = 0.52, NS. Instead, when sleep deprived, volunteers showed a trend of initially shifting toward more selections from the good decks in the first half of the game, followed by a reversal of this trend, progressively sampling more frequently from the bad decks during the latter half of the game, FQuadratic [1, 33] = 6.12, P = 0.019 (see Fig. 1). Pairwise comparisons (with Bonferroni adjustment of α = 0.01) between rested baseline and sleep-deprived conditions at each block revealed that under sleep-deprived conditions, volunteers selected significantly fewer cards from the good decks relative to bad decks compared to rested baseline, but only during the final 20-card block, t[32] = 3.40, P = 0.002. No other comparisons were significant (all P’s > 0.01).

We were also interested in determining the extent to which age mediates the detrimental effects of sleep deprivation on decision making performance. Therefore, in separate analyses at baseline and following sleep deprivation, we correlated net scores obtained during the final 20-card block of the game (Block 5) with volunteer age. Block 5 was chosen for comparison because this block was statistically significant between the baseline and sleep-deprived conditions in the previous analysis (see Table 1) and this was where the greatest learning effect would be expected. As shown in Fig. 2, at rested baseline age was not related to Gambling Task performance during the last block, r = 0.07, NS. However, when the same participants were tested again at 49.5 h of sleep deprivation, age was significantly negatively correlated with net score, r = −0.40, P = 0.019. That is, the older the volunteer, the greater the adverse effect of sleep deprivation on decision making; older volunteers selected more frequently from the ‘bad’ or risky decks than did younger volunteers.

Table 1.  Mean net scores and standard deviations for the Iowa Gambling Task during rested baseline and following 49.5 h of continuous wakefulness
Block of cardsCondition
BaselineSleep deprivedP-value
  1. The values are expressed in mean (SD).

  2. *Significant after Bonferroni correction at α = 0.01.

1–20−2.79 (8.80)−0.67 (5.12)0.14
21–402.97 (7.40)2.06 (6.09)0.54
41–602.79 (10.59)3.45 (6.84)0.77
61–805.15 (11.03)2.30 (8.60)0.20
81–1009.27 (11.07)0.42 (9.11)0.002*
Figure 2.

The relationship between age and decision making during the final 20-card Block of the Gambling Task was examined at baseline and again following 49.5 h of sleep deprivation. (a) At baseline, age was not significantly correlated with decision making performance during the final Block of the game. (b) However, following sleep deprivation, the same subjects showed a significant negative correlation between age and advantageous decision making. When deprived of sleep for 2 days, older subjects tended to make riskier choices than younger subjects during the final 20-card selections of the game.

Discussion

Relative to resting baseline, performance on the IGT was significantly impaired following 49.5 h of sleep deprivation. Whereas fully rested volunteers performed normally on the IGT (i.e. showing a gradual shift in behavior away from choosing ‘bad’ decks and toward selecting from ‘good’ decks), when tested again following 49.5 h of total sleep deprivation these same volunteers showed a pattern of responding that became significantly more disadvantageous as the game progressed. When compared to their resting performance, sleep-deprived volunteers in our study appeared less able to weigh the immediate benefits of short-term rewards against the greater costs of long-term penalties, a cognitive ability that appears to rely heavily on the integrity of the prefrontal cortex (Bechara, 2001). The IGT has been shown to be clinically sensitive to brain lesions within the ventromedial prefrontal and orbitofrontal cortices; thus, the present findings are consistent with the assertions of Horne and colleagues that the prefrontal cortex and the complex cognitive processes it mediates may be particularly vulnerable to continuous wakefulness (Harrison and Horne, 2000).

Because sleep loss is associated with decreased blood flow in the ventral regions of the prefrontal cortex (Thomas et al., 2000), we hypothesized that sleep-deprived participants would show deficits analogous to those seen in patients with lesions to the ventromedial prefrontal lobes. The present data support this hypothesis. As is evident in Fig. 1, sleep-deprived volunteers showed an initial shift in decision making behavior toward more frequent selections from low-risk decks during the first half of the game, but midway through the task shifted their strategy and began sampling with progressively greater frequency from the high-risk decks. While not as severe as seen in brain-injured patients, this shift in behavior bears a striking resemblance to the general pattern of IGT performances reported in the literature for patients with lesions to the ventromedial prefrontal cortex (dashed line, Fig. 1). Although our participants were less impaired globally than brain-injured patients seen in a clinical setting, the similarity between the pattern of performances exhibited by sleep-deprived volunteers and patients with ventromedial prefrontal cortex lesions is notable and suggests that sleep loss may adversely affect the functioning of these same regions of the prefrontal cortex.

Given the evidence that the ventromedial prefrontal cortex shows significant reductions in metabolic activity as a result of prolonged wakefulness (Thomas et al., 2000), it is not surprising that there are concomitant changes in behavior associated with alterations in the functioning of that region. The ventromedial prefrontal cortex is an important convergence zone whereby environmental stimuli are believed to be linked with emotional body states (Bechara et al., 2000a; Damasio, 1994, 1996). These stimulus–emotion links have been termed ‘somatic markers’ because they mark a particular choice alternative as feeling ‘good’ or ‘bad.’ Somatic markers then have the potential to guide or bias decision making by quickly flagging a possible option with a somatic state that alerts the decision maker as to the ‘goodness’ or ‘badness’ of a particular decision or course of action (Bechara et al., 2000a). These processes often operate covertly below the level of conscious awareness to bias decision making (Bechara et al., 1997). When considered in the context of other data suggesting that sleep deprivation affects emotional processes including mood (Dinges et al., 1997; James and Gregg, 2004; Lingenfelser et al., 1994) and the evaluation of emotional events (Zohar et al., 2005), the present study raises the possibility that prolonged wakefulness has a particularly detrimental effect on those types of executive functions that are heavily dependent on the ability to integrate emotion with other cognitive processes.

Recent evidence suggests that the prefrontal cortex may be especially vulnerable to the effects of sleep loss with increasing age (Munch et al., 2004). We therefore examined the extent to which the decision making deficits observed in the present study might be affected by age. At rested baseline, there was no significant linear relationship between age and net score performance during the final Block of the game. When the same volunteers were tested again following 49.5 h of sleep deprivation, however, there was a significant correlation between age and the tendency to select from the riskier decks. The significant negative correlation between age and net score in the final Block of the game suggests that, relative to their younger counterparts, older participants were more adversely affected by sleep loss – resulting in less advantageous selections. These findings are consistent with recent data that show that frontal regions of the brain are more vulnerable to the effects of prolonged wakefulness in older individuals (Munch et al., 2004) and suggests that executive capacities may become more vulnerable to disruption by sleep loss with increasing age. Other studies also suggest that sleep deprivation may have different effects on individuals as a function of age. For instance, Webb (Webb, 1985; Webb and Levy, 1982) found that, relative to younger volunteers, older participants displayed greater performance decrements on simple measures of attention and cognitive processing as a result of sleep deprivation. More recent data suggest that while older individuals start out with slower reaction time performances at baseline, they do not show declines in performance following sleep deprivation to the same degree as younger participants (Philip et al., 2004). Our results raise the possibility that while older individuals may be more resistant to sleep deprivation induced decrements in simple reaction time, these same individuals may be more adversely affected than younger ones on tasks that rely on higher-order cognitive abilities under the control of the prefrontal cortex.

While we believe that the present study provides important new data concerning the effects of sleep loss on higher-order cognitive processes, several caveats must be kept in mind when interpreting our results. First, it could be argued that the present results were due to efforts to increase stimulation and excitement during a long sleep-deprivation period, rather than to the effects of sleep loss on executive control. Specifically, the sleep-deprived participants in the present study may have selected the riskier decks merely to increase the level of stimulation and counteract their fatigue by creating some excitement. While this explanation is feasible, it is probably unlikely, because the subjects all believed that the bonus was directly related to their performance on the task. Given that most participants, if not all, chose to participate in the study primarily for the possibility of monetary compensation, it seems highly unlikely that they would choose to risk a substantial portion of the money for a slight increase in stimulation while playing the game. Second, this study was limited by the fact that each participant served as their own control. Because we did not include a separate non-sleep-deprived control group, it is possible that the changes in performance that we observed could be explained by practice effects and familiarity with the task upon repeat administration. This explanation is unlikely, however, as recent data from the laboratory of Bechara and colleagues suggest that the two versions of the task used in the present study (i.e. A′, B′, C′, D′ and K′, L′, M′, N′) produce virtually no practice effect when administered in the same order used in the present study (A. Bechara, personal communication, 22 July 2005). Those data suggest that healthy non-sleep-deprived participants show no significant difference in performance across three alternate forms of the IGT administered in serial order. Finally, since we did not include a postrecovery sleep condition, it is impossible to know whether the observed deficits in performance return to normal following adequate sleep. We are currently undertaking a study that includes a postrecovery condition in order to answer these questions. With the aforementioned limitations in mind, we believe that the present data provide important information regarding the detrimental effects of continuous wakefulness on the higher-order executive functions involved in decision making. Overall, these findings are consistent with neuroimaging data showing relative reductions in prefrontal metabolic activity during sleep deprivation and raise the possibility that the specific types of executive functions that are mediated by the ventromedial prefrontal cortex may be particularly vulnerable to sleep loss and/or less amenable to compensatory effort than those mediated by other regions of the prefrontal cortex.

Acknowledgements

The views expressed in this article are those of the authors and do not reflect the official policy or position of the Department of the Army, the Department of Defense, the US Government, or any of the institutions with which the authors are affiliated.

Ancillary