Predictors of interindividual differences in vulnerability to neurobehavioral consequences of chronic partial sleep restriction

Abstract

Interindividual differences in the neurobehavioral response to sleep loss are largely unexplained and phenotypic in nature. Numerous factors have been examined as predictors of differential response to sleep loss, but none have yielded a comprehensive view of the phenomenon. The present study examines the impact of baseline factors, habitual sleep–wake patterns, and homeostatic response to sleep loss on accrued deficits in psychomotor vigilance during chronic partial sleep restriction (SR), in a total of 306 healthy adults that participated in one of three independent laboratory studies. Findings indicate no significant impact of personality, academic intelligence, subjective reports of chronotype, sleepiness and fatigue, performance on working memory, and demographic factors such as sex, ethnicity, and body mass index, on neurobehavioral vulnerability to the negative effects of sleep loss. Only superior baseline performance on the psychomotor vigilance test and ability to sustain wakefulness on the maintenance of wakefulness test were associated with relative resilience to decrements in vigilant attention during SR. Interindividual differences in vulnerability to the effects of sleep loss were not accounted for by prior sleep history, habitual sleep patterns outside of the laboratory, baseline sleep architecture, or homeostatic sleep response during chronic partial SR. A recent theoretical model proposed that sleep–wake modulation may be influenced by competing internal and external demands which may promote wakefulness despite homeostatic and circadian signals for sleep under the right circumstances. Further research is warranted to examine the possibility of interindividual differences in the ability to prioritize external demands for wakefulness in the face of mounting pressure to sleep.

Statement of Significance.

Interindividual differences in response to chronic sleep restriction (SR) are well documented yet remain unexplained. The ability to identify in advance those most susceptible to the negative effects of sleep loss would have significant implications for many sensitive occupations. This is the largest and most comprehensive analysis to date to examine baseline predictors of phenotypic response to sleep loss. Only baseline psychomotor vigilance performance and ability to sustain wakefulness were associated with relative resilience to decrements in vigilant attention during chronic SR. Other variables, including personality, demographic factors, neurobehavioral performance, prior sleep history, and homeostatic sleep response, did not account for interindividual differences in vigilant attention decrements. Further research is warranted to investigate neurobiological mechanisms underpinning the phenotypic response to sleep loss.

Introduction

Despite the established health benefits of adequate sleep duration and quality, approximately 15% of US adults report sleeping less than the recommended minimum of 7 h per night [1].

Insufficient sleep is associated with greater morbidity, mortality, and financial costs [2]. Reports of medical incidents and performance failures have consistently cited insufficient sleep or human error as a result of sleep loss as a contributing factor. Sleep pathology, voluntary curtailment of sleep, and circadian misalignment have been linked with motor vehicle accidents [3–6], occupational accidents [7], aviation accidents [8], failures in combat operations [9], and medical errors [10–14], as well as major disasters such as the grounding of Exxon Valdez oil tanker and nuclear plant accidents at Three Mile Island and Chernobyl [15].

The adverse effects of chronic sleep loss are wide ranging with pronounced effects on neurobehavioral functions, particularly vigilant attention. However, the magnitude of the deficit in vigilant attention resulting from sleep loss varies substantially between individuals. Large interindividual differences in vulnerability to vigilance decrements induced by sleep loss have been robustly reported: some individuals show marked decrements in performance under conditions of sleep loss, while others show only moderate decrements, and still others show few to no decrements—even when sleep loss is severe. These interindividual differences are stable over time, highly replicable, and observed across different conditions of sleep loss, suggesting a trait-like phenotypic response [16–18]. For a review of the literature on this phenotypic effect and potential underlying biological mechanisms, see Tkachenko et al. [18].

Several efforts to predict phenotypic vulnerability to sleep loss have been undertaken to identify which individuals are at greater risk for the deleterious effects of sleep loss on neurobehavioral performance. To date, the single best predictor of the individual differences in neurobehavioral performance deficits resulting from in sleep loss is baseline performance [16, 19]. Several studies have shown that aspects of vigilant attention performance, indexed by the psychomotor vigilance test (PVT), are predictive, to varying extents, of vigilance decrements induced by sleep loss [20–22]. Simply put, the best predictor of impairment on the PVT following sleep loss is performance on the task while rested. Neuroimaging studies have also revealed that baseline measures of functional connectivity and dynamic connectivity states can be used to predict vulnerability to decrements in vigilant attention induced by sleep loss [23, 24]. However, studies investigating other potential predictors of vulnerability have yielded inconsistent findings. As vulnerability to sleep loss is likely influenced by a number of factors, there is a need to examine potential predictors of vulnerability together to evaluate the differential contributions of baseline measures. Furthermore, there is presently no single point of reference to document the relative utility, or lack thereof, of various baseline measures in predicting vulnerability to the effects of sleep loss on vigilant attention.

While rested performance on the PVT reliably predicts performance under conditions of sleep loss, it is unclear whether rested performance on other measures of neurobehavioral function can be used to improve predictions of PVT outcomes under such conditions. Age may confer resilience to performance decline under conditions of sleep loss, but this has only been observed when comparing samples of a wide age range with limited effects among more homogenous, younger populations [25–27]. While other demographic factors, including sex and race, have been implicated with interindividual differences in physiological response to sleep loss, they have not been reliably associated with neurobehavioral vulnerability to sleep loss across single or repeated exposures to sleep loss [16, 27–31]. Self-reported chronotype, as measured by the Morningness–Eveningness Questionnaire (MEQ), has been shown to discriminate between average individuals and those resilient to the effects of sleep loss on PVT performance during total sleep deprivation [32], but MEQ scores could not discriminate the most vulnerable individuals from those who were resilient, demonstrating its limited predictive power. Despite the relatively large literature on the influence of personality on cognitive performance during sleep loss, variability in study design, level of laboratory controls, and the metrics used to index personality, as well as the lack of studies examining the influence of personality on performance across repeated exposures to sleep loss, make it difficult to support any single personality factor as a reliable predictor of individual response. Finally, the potential role of intelligence or academic achievement in resilience to sleep loss has not been explicitly evaluated, despite claims in several publications that interindividual differences are not accounted for by standardized measures of intelligence (IQ) [33, 34]. In summary, while many studies declare that interindividual differences in vigilance decrements induced by sleep loss cannot be explained by baseline measures, the relative utility of these baseline measures in predicting vulnerability has not been directly assessed in a single, large-scale study.

Furthermore, it is unclear whether basal differences in sleep architecture, sleep physiology, and habitual sleep–wake dynamics contribute to phenotypic vulnerability to sleep loss. While interindividual differences in sleep architecture (i.e. the distribution of different sleep stages across a sleep period) are relatively stable across multiple sleep EEG recordings in rested conditions [35–38], it is unclear whether vulnerability groups differ in sleep architecture parameters at baseline. The accumulation and dissipation of homeostatic sleep pressure across extended wakefulness has not been reliably associated with changes in neurobehavioral performance [39, 40], though a recent study suggests that larger increases in slow-wave activity, a marker of homeostatic sleep pressure, induced by sleep loss are associated with greater performance decrements on the PVT [41]. There is also evidence that individuals with habitually higher homeostatic sleep pressure may be predisposed to greater vulnerability to accumulating neurobehavioral decrements with sleep loss [40, 42]. A rested baseline comparison of power spectral data between individuals classified as vulnerable or resilient to total sleep deprivation on the basis of PVT performance decline found that vulnerable subjects had more spectral power in the theta band of waking EEG at baseline, suggesting higher homeostatic sleep pressure at rested wakefulness [40]. Habitual sleep behaviors, including the duration and timing of sleep, can contribute to sleep debt buildup and habitual homeostatic sleep pressure. While most studies have reported no difference in habitual bedtimes between individuals classified as vulnerable or resilient to neurobehavioral performance decrements [29, 30, 40, 43, 44], there is some evidence that individuals classified as resilient have more variability in self-reported sleep–wake schedules as compared with vulnerable individuals [40], and that individuals who are rigid in their sleep–wake patterns exhibit greater performance decrements on the PVT during total sleep deprivation relative to individuals with flexible sleep–wake patterns [45]. Given the inconsistent findings, further research is needed to elucidate the relationships between baseline sleep parameters, habitual sleep duration and timing, homeostatic response to sleep loss, and interindividual vulnerability to the neurobehavioral effects of sleep loss.

To evaluate the contribution of factors proposed to influence the interindividual vulnerability to sleep loss, the present study leveraged a sample of more than three hundred healthy adults who participated in one of three independent, highly controlled, in-laboratory chronic sleep restriction (SR) studies, to examine (1) the relative utility of various baseline measures in predicting vigilant attention decrements induced by sleep loss and (2) whether individuals classified by vulnerability group (i.e. vulnerable, intermediate, or resilient) exhibited differences in baseline sleep architecture, habitual sleep patterns, or sleep homeostatic response.

Methods

Study participants

Three-hundred and six healthy individuals (46% female) participated in one of three laboratory studies. Eligibility for the laboratory studies included the following: typical sleep–wake patterns, including habitual nocturnal sleep duration between 6.5 and 8.5 h and habitual wake times between 06:00 and 09:30 with no evidence of habitual napping or sleep disturbances; no engagement in shift work or transmeridian travel in the 2 months prior to the start of the laboratory study; no present medical or psychological conditions, or a history of brain injury; no current substance use, evaluated via urine toxicology. Eligibility was assessed using clinical interviews, questionnaires, physical exams, and blood and urine tests. Participants abstained from caffeine, alcohol, tobacco, and medications (excluding birth control), during the week prior to the laboratory study. In the week prior to the laboratory study, sleep–wake patterns were monitored via actigraphy, sleep diaries, and time-stamped phone records for time in and out of bed. A total of N = 331 participants initiated the study, however n = 25 (8.9%) did not complete the protocol (n = 19 withdrew due to personal reasons, n = 3 were withdrawn for noncompliance or inability to perform the neurobehavioral tasks, and n = 3 were withdrawn due to eligibility issues).

Experimental study design

Data were derived from three independent chronic SR experiments that employed an identical chronic partial SR paradigm. Three-hundred and six participants completed the laboratory protocol in groups of three to five participants at a time. To ensure that the effects of SR were not attributable to the laboratory environment, a subset of participants were randomized to a nonsleep-restricted control condition (n = 28); the remaining n = 278 participants were randomized to the SR condition. Given the focus of the study, the final analytic sample included only the n = 278 participants randomized to the SR condition. All participants received two baseline nights (B1 and B2) of 10 h time in bed (TIB; 22:00–08:00) sleep opportunity per night. Participants in the control condition continued to receive 10 h TIB sleep opportunity on each night for the remainder of the laboratory protocol, while the participants in the SR condition underwent five consecutive nights (SR1–SR5) of 4 h TIB (04:00–08:00) sleep opportunity per night. Participants were provided at least one recovery night of 10 h TIB sleep opportunity prior to departure from the laboratory. Participant sleep–wake patterns continued to be monitored for 1 week after leaving the laboratory via actigraphy and sleep–wake diaries.

All participants underwent the same study procedures, with the exception of neurobehavioral testing during times when the control condition participants were asleep and the sleep-restricted subjects remained awake (i.e. 22:00–04:00). Participants were continuously monitored by trained research staff to ensure participant safety and adherence to the study protocol. Regular meals were provided throughout the protocol (08:30–10:00, 12:30–14:00, and 18:30–20:00). Laboratory lighting was kept constant at <50 lux during periods of scheduled wakefulness and <1 lux during scheduled sleep times; temperature was kept between 22 and 24°C. Participants were not permitted to nap or engage in any vigorous physical activity while in the laboratory. Nurses performed daily clinical checks on all participants and a physician was informed of any clinical issue.

Beginning at 08:00 on each laboratory day, participants completed a 30-min computerized neurobehavioral assessment every 2 h during scheduled wakefulness. For the purposes of this study, neurobehavioral performance was determined by the daily average of each outcome from test bouts administered between 10:00 and 19:00 each day to ensure that all participants contributed an equal number of assessments and that the influence of different circadian phases and sleep inertia was minimized [46]. The studies were approved by the Institutional Review Board of the University of Pennsylvania. Subjects provided written informed consent prior to participation.

Measures

Demographic information

Demographic data were collected via self-report and included age, sex, and race. Body mass index (BMI) data were also collected at study intake. Data were available for age, race, and sex from N = 278 participants, and N = 277 participants for BMI. Number of years of education was also collected via self-report and used as a proxy for socioeconomic status; data were available from N = 275 participants.

Neurobehavioral assessments

Psychomotor Vigilance Test

The PVT is an established reaction time (RT) task of vigilant attention that is highly sensitive to the attentional deficits resulting from sleep loss [47–49]. Unlike other neurocognitive tests, the PVT has negligible aptitude or practice effects [50]. Participants are asked to depress the “space” bar on the computer keyboard in response to a visual stimulus on a computer screen; visual stimuli are presented on the screen at a random interstimulus interval of 2–10 s for a total of 10 min. The primary outcomes for the PVT were lapses of attention (i.e. RT >500 ms), errors of commissions (i.e. false starts, RT <130 ms), and the mean of the fastest 10% of RTs. Data were available from N = 278 participants. Participants were categorized by vulnerability class using a tertile split of the change in PVT lapses from the baseline (B2) to the fifth day of SR (SR5). This yielded n = 93 resilient participants with minimal change in PVT lapses, n = 93 vulnerable participants who exhibited the greatest increase PVT lapses, and n = 92 intermediate participants. Performance on the PVT measured by lapses across the SR protocol for each vulnerability group is shown in Figure 1.

Figure 1.

Change in PVT lapses across days in the study protocol (B2—baseline; SR1–SR5—first through fifth nights of SR). Resilient (Type 1) subjects are shown by blue circles, Intermediate (Type 2) subjects are graphed in green triangles, and Vulnerable (Type 3) subjects are represented by red stars. Error bars denote 95% confidence interval.

Digit Symbol Substitution Test

The Digit Symbol Substitution Test (DSST) is a 90-s computerized test adapted from the Wechsler Adult Intelligence Scale [51]. The DSST displays nine line symbols that are associated with numbers (i.e. 0–9) at the top of the screen; participants are presented with one of the symbols and prompted to respond by entering the corresponding number using the keyboard as quickly and accurately as possible. The pattern of symbol and number pairs is selected at random from a set of 24 possible patterns to reduce the possibility of practice effects. The primary outcome for this task was the total number of correct responses. Data were available from N = 278 participants.

Digit Span Task

The Digit Span Task (DS) is a computerized version of the DS adapted from the Wechsler Adult Intelligence Scale that measures working memory [51]. Participants are presented with a series of numerical digits, one at a time, after which they are asked to recall the series by entering the correct order of numbers using the keyboard. When a participant correctly responds to the DS prompt, the length of the series is increased by one number for the following prompt to determine the maximum length that can be achieved. Both the forward and reverse versions of the DS were administered. The four outcome measures for the task included the maximum length and total number correct in the forward and backward versions. Data were available from N = 273 participants; the task was added to the study protocol after n = 4 participants completed the study and data were missing for one participant.

Maintenance of Wakefulness Test

The Maintenance of Wakefulness Test (MWT) is an objective measure of an individual’s ability to maintain wakefulness [52–54]. Participants were asked to remain awake as long as possible under soporific conditions, with a maximum time limit of 30 min. Wakefulness was monitored with EEG, EMG, and EOG. The primary outcome was the time to the first occurrence of a microsleep (i.e. 10 s of unambiguous sleep evidenced by EEG theta activity). The MWT was administered at B2 and SR5 between 14:00 and 17:00. The MWT was not collected in one study (n = 63) and therefore, data were available from a total of N = 215 participants.

Hayling Sentence Completion Test

The Hayling test is a measure of response initiation and suppression and is often used as a proxy for executive functioning [55]. The Hayling test consists of two sets of 15 sentences, each missing the last word. Each sentence is read aloud by a trained examiner and participants were asked to complete the sentence. In the first set of questions, participants can complete the sentence with the first word that comes to mind. In the second set, participants were asked to complete the sentence with a word that does not fit. The duration of time to respond (i.e. response latency) are recorded in seconds and converted to scaled scores. Data were available from N = 275 participants, data from n = 3 participants were missing.

North American Adult Reading Test

The North American Adult Reading Test (NAART) consists of 61 irregular words which are read aloud by the participant and scored for pronunciation accuracy according to American and Canadian standards. It is used as a measure of premorbid IQ [56, 57]. Primary outcome measures include Verbal IQ (VIQ), Performance IQ (PIQ), and Full Scale IQ (FSIQ). Due to high collinearity among the three outcome variables, only FSIQ was used in the present analyses. Complete data were available from N = 276 participants, data from n = 2 participants were missing.

Subjective ratings of fatigue and sleepiness

Profile of Mood States

The Profile of Mood States (POMS) is a 65-item self-report measure of mood that asks participants to indicate how they feel at the moment using a 5-point Likert-type scale ranging from “1” (Not at all) to “5” (Extremely). The POMS Fatigue-Inertia subscale was used in this study [58]. Data were available from N = 278 participants.

Karolinska Sleepiness Scale

The Karolinska Sleepiness Scale (KSS) is an established assessment of subjective sleepiness that prompts participants to rate their level of sleepiness on a 9-point Likert-type scale, which ranges from “1” (very alert) to “9” (very sleepy) [59, 60]. Data were available from N = 278 participants.

Visual Analog Scale—Fatigue

The Visual Analog Scale—Fatigue (VAS-F) scale prompted participants to rate their feeling of fatigue at that moment [61, 62]. To rate their fatigue, participants used a cursor to continuously move a marker along an unindexed reference line from an initial central location to indicate how they felt in that moment relative to the anchors (“fresh as a daisy” and “tired to death”) at each end of the line. Data were available from N = 278 participants.

Other self-report measures

Morningness–Eveningness Questionnaire

The MEQ is a 19-item self-assessment measure chronotype, evaluating the degree to which individuals are active and alert at different times of day. MEQ scores range from 16 to 86, with higher scores indicative of morning type [32]. Complete MEQ data were available from N = 264 participants. Of the remaining n = 14 MEQ surveys, n = 3 were missing, n = 3 were incomplete (i.e. not all questions were answered), and n = 8 had one survey question with more than one recorded response.

Eysenck Personality Inventory

The Eysenck Personality Inventory (EPI) is a 57-item yes/no questionnaire designed to measure two dimensions of personality: extraversion and neuroticism. The instrument yields total scores for Neuroticism, Extraversion, Psychoticism, and a Lie/Desirability score that serves as a measure of validity [63, 64]. Data were available from N = 273 participants for the Neuroticism, Psychoticism, and Lie/Desirability subscales. Of the n = 5 participants without EPI data, n = 3 were missing and n = 2 had incomplete survey responses. Data were available for n = 272 participants for the Extraversion subscale, due to n = 1 incomplete responses.

Beck Depression Inventory-II

The Beck Depression Inventory-II (BDI-II) is a 21-item questionnaire that measures depression symptomatology. BDI scores range from 0 to 63, with higher values indicative of more severe symptomatology [65]. BDI data were available from N = 278 participants.

Objective sleep measures

Polysomnography

Polysomnography (PSG) recordings were collected using the EEG Sandman Suzanne portable digital recording system (128-Hz sampling) on both baseline nights (i.e. B1 and B2) and the first and fifth nights of SR (i.e. SR1 and SR5). Sleep was scored from the C3–A2 derivation by trained technicians using the criteria of Rechtschaffen and Kales [66]. Sleep onset was determined as the observation of ≥3 consecutive 30 s epochs of stage 2–4 or rapid eye movement (REM) sleep. The primary sleep outcomes were the total sleep time (TST), sleep onset latency (SOL), number of awakenings (NWAK), the proportion of TST spent in slow-wave sleep (SWS; the sum of stage 3 and stage 4 sleep), and the proportion of TST spent in REM sleep. The second baseline night (B2) was used as a reference to reduce effects of the novelty of the laboratory environment on sleep. Data were available from n = 263 participants for B2, n = 256 participants for SR1, and n = 253 participants for SR5, which yielded a total of N = 772 nights of valid PSG data. Missing data were either due to hardware malfunction or excessive artifact in the EEG for sleep scoring.

Actigraphy

Study participants were monitored with wrist actigraphy during the week prior to (PRE) and following (POST) participation in the laboratory study. Data were collected using the Phillips Respironics Actiwatch Spectrum with epoch length set to 1 min. The “Actiware 5” software developed by Phillips Respironics was used to download the data and extract the variables of interest. Outcome measures included the average bedtime and wake time, and TST, as derived from the software, as well as the standard deviation in the timing of sleep onset and awakening, and average TSTs across the seven nights of actigraphy monitoring. Data from the week prior to study participation were available from N = 276 participants. Following laboratory SR, actigraphy data were available from N = 276 participants. Missing actigraphy data were due to hardware malfunction or participants not wearing the actiwatch.

Statistical analyses

The two primary objectives of the manuscript were evaluated independently. The first objective was to evaluate baseline predictors of individual differences in the change in PVT performance across SR (objective 1) and the second was to evaluate mean differences in baseline sleep architecture, habitual sleep patterns, and homeostatic response to SR by vulnerability classification (objective 2).

Objective 1

For the analysis of baseline predictors of vigilant attention, the primary outcome was the change in PVT lapses from the second baseline day to the fifth day of SR (i.e. Δ PVT lapses = SR5 − B2). Independent variables included: (1) baseline performance on the PVT, (2) performance on the DSST and DS, (3) performance on the MWT, (4) subjective reports of sleepiness and fatigue, (5) demographic variables, (6) personality factors, and (7) academic intelligence. Data were visually examined to check for outliers and Q–Q plots were used to evaluate normality of each variable independently. Due to the low frequency of several ethnic categories on the demographic questionnaire, the variable was dichotomized into Caucasian and Non-White groups. The Non-White group included African American, Asian, Hispanic, Native Hawaiian/Pacific Islander, and more than one ethnicity. The four outcomes on the DS were collinear. To reduce the observed collinearity, a Principal Components Analysis was performed to create a single composite index to reflect overall performance. Four study outcomes exhibited nonnormal distributions and these data were transformed as follows: POMS Fatigue-Inertia, PVT Lapses, and PVT False Starts were subjected to a reciprocal transformation (1/(x + 1)); VAS-F responses showed severe rightward skew with a possible floor effect and these data were transformed with a log 10(x + 1) function; MWT data showed significant leftward skew with a clear ceiling effect (35% of participants achieved the maximum score), so data were dichotomized to indicate whether or not each individual was able to remain awake during the test; the slope of PVT lapses from baseline to the fifth day of SR (i.e. Δ PVT lapses = SR5 − B2) was severely right skewed and was transformed using a cubed root function (x^(1/3)) to retain information about directionality of change over time. The remaining variables were not different from a normal distribution with no outliers.

Patterns of missing data among all independent variables are shown in Supplementary Figure S1. All missing data were assumed to be missing at random. Independence of residuals was confirmed by a Durbin–Watson statistic of 1.07. Homoscedasticity and linearity were assessed by visual inspection of a plot of studentized residuals versus unstandardized predicted values. There was no evidence of multicollinearity, as assessed by partial correlations below 0.55 among the independent variables and tolerance values greater than 0.45. Casewise diagnostics noted three outliers based on studentized deleted residuals greater than ±3 standard deviations. These cases were nevertheless included in the model as they were deemed to be within the expected range of values on each measure. There were no leverage values greater than 0.3 or values for Cook’s distance above 1. Normality of residuals was confirmed with visual inspection of the histogram of residuals as well as the Normal P–P plot of regression standardized residuals.

Multiple imputation was used to address missing data as it performs well in small sample sizes and when using large multiple regression models with as much as 50% missing data (although there were at most 22.7% of missing data in the present sample) [67]. Multiple imputation was performed in SPSS (Version 27.0; IBM Corp., 2016) using predictive mean matching with 50 maximum iterations (random seed = 950). All analysis variables were used as predictors under the assumption that data were missing at random. Consistent with recommendations by von Hippel et al., any variables with severely skewed distributions were transformed prior to multiple imputation [68]. Furthermore, 40 imputed datasets were generated to ensure that data imputation did not introduce variance based on recommendations by Graham et al. [69]. The resulting dataset containing the original data with missing cases along with the 40 imputed complete datasets was imported into R version 3.5.1 using the “foreign” package [70, 71].

Seven sequential multiple linear regression models on each imputed dataset were conducted using the “mice” package in R, with independent variables introduced in predetermined blocks as indicated below (models 1–7) [72]. The sequential order of the predetermined blocks was determined by expected predictive utility of each category of variables based on review of the literature.

• Model 1: Δ PVT Lapses ~ PVT Performance [Baseline Performance]

• Model 2: * ~ * + DSST + Digit Span [Neurobehavioral Performance]

• Model 3: * ~ * + MWT [Physiological Wakefulness]

• Model 4: * ~ * + KSS + POMS + VAS [Subjective Reports]

• Model 5: * ~ * + MEQ + Sex + Age + Education + Ethnicity + BMI [Demographics]

• Model 6: * ~ * + EPI [Personality]

• Model 7: * ~ * + NAART + Hayling [Academic Intelligence]

Analyses run on each dataset were pooled according to Rubin’s rules [73]. Pooled F and R² values were obtained using the R code provided by van Ginkel et al. and “pool.r.squared” command in the “mice” package, respectively [72, 74]. Analyses investigating the impact of baseline variables on change in PVT lapses across SR were repeated with listwise deletion and full information maximum likelihood estimation as alternative means of handling missing data (see Supplement).

Objective 2

Participants were classified into vulnerability categories by a tertile split using the difference in PVT lapses from the baseline to the fifth day of SR (i.e. Δ PVT lapses = SR5 − B2), consistent with prior research [16]. For vulnerability classification, participants in the first tertile were classified as “resilient.” Participants in the second tertile were classified as “intermediate” and participants in the third tertile were classified as “vulnerable.”

Objective 2a

To examine whether baseline sleep architecture differed among vulnerability groups (i.e. /resilient, /intermediate, /vulnerable), a one-way multivariate analysis of variance (MANOVA) was used. Several outliers were observed in the data, and TST, SOL, and NWAK had nonnormal distributions. Analyses were run on nontransformed data and did not exclude any outliers as the MANOVA is fairly robust to violations of these assumptions [75]. Six multivariate outliers were observed, as assessed by Mahalanobis distances greater than 27.88 given 9 dependent variables. Three of these outliers were within the resilient group, while the remaining three outliers were in the intermediate group. The multivariate outliers were also included in the analysis to retain maximum information. There was no multicollinearity, as assessed by Pearson correlations (all r < 0.70) [76].

Objective 2b

To examine whether habitual sleep–wake patterns during the week prior to and the week following the laboratory experiment differed by vulnerability groups, two one-way MANOVAs were conducted. For these analyses, average bedtimes and wake times were transformed to numeric variables to represent distance from midnight in hours, with negative values indicating times prior to midnight (i.e. −0.5 indicates 30 min prior to midnight or 23:30). Only one outlier was observed for a Type 2 individual in standard deviation in TST. The outlier was not removed for the analysis, as it contains valuable information regarding variability in sleep schedules. Average TST, sleep, and wake times were normally distributed for each group and no multicollinearity among dependent variables was observed. Measures of standard deviation in these variables were slightly skewed to the right but were not concerning for the present analyses after review of the Q–Q normality plots. One multivariate outlier was observed in the PRE data, and 7 multivariate outliers were observed in the POST data, as assessed by a Mahalanobis distance greater than 22.46 given 6 dependent variables. All data were included in the analyses given the large sample size and to retain maximum information. Homogeneity of variance–covariance matrices was not met, as assessed by Box’s test of equality of covariance matrices (p < 0.001). However, given the roughly equal sample sizes, this was not of significant concern. There was homogeneity of variances, as assessed by Levene’s test of homogeneity of variance (p > 0.05).

Due to normative age-related changes in sleep architecture, age was included as a covariate in the analyses for objectives 2a and 2b. If any of the omnibus tests of group differences were statistically significant, post hoc analyses were conducted to examine pairwise differences between vulnerability groups on each variable of interest.

Objective 2c

Mixed-effects models were used to examine whether change in sleep architecture variables differed among vulnerability groups (i.e. resilient, intermediate, or vulnerable) over several days of SR (B2, SR1, and SR5). Analyses were repeated with age added as a covariate in each model to account for natural changes in sleep architecture with age and prior findings suggesting older individuals are less impacted by sleep loss as compared with younger participants [25–27, 77–80]. Maximum likelihood estimation was used to address missing data in each of the models. Bonferroni adjusted p values were used to account for multiple comparisons in any post hoc analyses following a significant omnibus test [81]. All statistical analyses were conducted in SPSS Version 27.0 [82].

Results

Two hundred and seventy-eight healthy individuals completed the SR protocol. Demographic information is presented in Table 1. Participants were on average 31.0 ± 7.8 (SD) years of age (range: 21–50 years) with 14.6 ± 2.0 years of education, and had an average BMI of 24.8 ± 3.6. Forty-six percent were female, and approximately 38% identified as Caucasian, 53% identified as African American, and approximately 9% identified as either Asian, Hispanic, or more than one race, yielding 63.1% of participants identified as Non-White. On average, participants had a BDI score of 1.4 (range 0–12) and an MEQ score of 40.1 (range 26–55). Baseline performance on all measured variables averaged across vulnerability groups is displayed in Table 2.

Table 1.

Demographic characteristics of all study participants and by vulnerability group

	All participants		Resilient		Intermediate		Vulnerable
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Age (yr)	31.02	7.81	31.95	7.92	31.90	8.14	29.22	7.11
Education (yr)	14.63	1.96	14.82	2.02	14.80	1.96	14.27	1.88
BMIa	24.83	3.60	24.40	3.45	25.24	3.56	24.84	3.76
MEQb	40.12	5.94	40.28	6.10	40.25	6.35	39.82	5.38
BDI-IIc	1.37	2.11	1.10	1.79	1.28	2.01	1.74	2.43
	N	%	N	%	N	%	N	%
Sex
Male	150	53.96	46	49.46	53	57.61	51	54.84
Female	128	46.04	47	50.54	39	42.39	42	45.16
Race
Caucasian	105	37.77	34	36.56	36	39.13	35	37.63
African American	149	53.60	54	58.06	47	51.09	48	51.61
Otherd	24	8.63	5	5.38	9	9.78	10	10.75

Means and standard deviations for age, body mass index, morningness–eveningness scores, and depression ratings. Counts and percentage of sex and race.

^aBody mass index.

^bMorningness–Eveningness Questionnaire.

^cBeck Depression Inventory.

^dEncompasses individuals self-identified as Asian, Hispanic, Native Hawaiian/Pacific Islander, and more than one ethnicity.

Table 2.

Baseline performance on all independent measures by vulnerability group

	Vulnerability group
	Resilient		Intermediate		Vulnerable		All
	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
PVT Raw Lapsesa	1.26	2.58	1.80	2.16	3.41	4.31	2.15	3.28
PVT Fastest 10% RT (ms)a	198.84	21.14	200.83	18.55	205.21	20.69	201.63	20.27
PVT False Startsa	1.26	1.72	1.53	1.82	1.58	1.58	1.46	1.71
DSST # Correctb	58.57	8.09	55.62	9.78	54.74	8.93	56.31	9.07
DS-Forward # Correctc	5.76	1.28	5.63	1.28	5.38	1.20	5.59	1.26
DS-Forward Maxc	6.87	1.17	6.74	1.18	6.58	1.03	6.73	1.13
DS-Back # Correctc	5.05	1.78	4.77	1.76	4.25	1.51	4.69	1.71
DS-Back Maxc	6.20	1.65	5.90	1.64	5.55	1.51	5.88	1.62
MWT (min)d	22.92	9.42	21.77	9.28	16.82	10.29	20.68	9.96
KSSe	2.97	1.37	2.97	1.25	3.22	1.30	3.05	1.31
POMS-Fatiguef	0.93	1.40	0.73	1.20	0.62	4.90	0.76	3.02
VAS Exhausted/Energeticg	33.64	34.97	26.47	31.79	12.58	20.76	24.22	30.96
EPI Extraversionh	14.93	3.60	15.37	4.16	15.74	3.65	15.35	3.81
EPI Lie/Desirabilityh	12.70	3.96	12.61	4.11	11.45	4.27	12.26	4.14
EPI Neuroticismh	5.32	2.34	5.18	2.26	5.52	2.25	5.34	2.28
EPI Psychoticismh	2.71	1.70	2.96	1.75	2.92	1.80	2.86	1.75
NAART FSIQi	105.67	8.90	103.76	8.33	104.55	9.12	104.66	8.80
Hayling Scaled Totalj	5.51	1.43	5.26	1.47	5.30	1.40	5.36	1.43

^aPsychomotor Vigilance Test.

^bDigit Symbol Substitution Test.

^cDigit Span Test.

^dMaintenance of Wakefulness Test.

^eKarolinska Sleepiness Scale.

^fProfile of Mood States—Fatigue.

^gVisual Analog Scale.

^hEysenck Personality Inventory.

ⁱNorth American Adult Reading Test.

^jHayling Sentence Completion Test.

Association between baseline variables and the change in PVT lapses across SR

Pooled results for multiple linear regression models run on each of the 40 imputed datasets are presented in Table 3. The full model including all a priori defined baseline measures was significant for the pooled analyses, R² = 0.181, F(21,992153) = 2.58, p = 0.0001; adjusted R² = 0.11. Only the addition of MWT performance (model 3) significantly improved model fit, ΔR² = 0.021, ΔF(1,424.5) = 4.42, p = 0.036. None of the other models resulted in significant model improvement (p > 0.15). To develop a parsimonious model, a final multiple regression model was run that limited the analysis to baseline PVT performance and MWT. This parsimonious model was statistically significant, R² = 0.121, F(4,15130) = 8.30, p = <0.001, adjusted R² = 0.108, indicating that increased PVT lapses across the five nights of SR were significantly associated with more PVT lapses at baseline (t = −4.3, p < 0.001) and shorter time to fall asleep at baseline, assessed by the MWT (t = −2.22, p = 0.028). A post hoc analysis examined whether PVT lapses and MWT performance at baseline were predictors of vulnerability group classification. This model was statistically significant, R² = 0.214, F(2,5199) = 32.52, p < 0.001, adjusted R² = 0.208, indicating that vulnerability to SR was principally associated with more PVT lapses at baseline (t = −8.0, p < 0.001), but not with time to fall asleep at baseline on the MWT (t = −1.90, p = 0.06).

Table 3.

Model parameters and fit using sequential linear regression with multiple imputation

	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6		Model 7
Fixed effects	Estimate	Std. error	Estimate	Std. error	Estimate	Std. error	Estimate	Std. error	Estimate	Std. error	Estimate	Std. error	Estimate	Std. error
Intercept	2.569**	0.668	3.069**	0.745	3.052**	0.739	3.034**	0.765	3.418*	1.136	3.049*	1.234	3.921*	1.597
PVT Lapses	−1.059**	0.233	−0.990**	0.246	−0.962**	0.248	−0.814*	0.264	−0.741*	0.269	−0705*	0.270	−0.695*	0.271
PVT False Start	−0.213	0.254	−0.201	0.255	−0.188	0.253	−0.214	0.255	−0.213	0.258	−0.159	0.260	−0.122	0.262
PVT Fastest 10% RT	−0.002	0.003	−0.002	0.003	−0.002	0.003	−0.001	0.003	−0.0003	0.003	−0.0001	0.003	−0.0001	0.003
DSST # Correct			−0.010	0.006	−0.009	0.006	−0.010	0.006	−0.011	0.007	−0.010	0.007	−0.009	0.007
Digit Span			−0.004	0.034	−0.012	0.034	−0.022	0.035	−0.014	0.036	−0.009	0.036	−0.018	0.039
MWT					−0.283*	0.135	−0.279*	0.133	−0.278*	0.132	−0.268*	0.133	−0.266*	0.133
KSS							0.027	0.050	0.030	0.052	0.020	0.052	0.021	0.052
POMS							0.078	0.218	0.064	0.220	−0.013	0.222	−0.006	0.222
VAS-Fatigue							−0.210	0.147	−0.181	0.152	−0.158	0.154	−0.203	0.160
Sex									−0.011	0.126	−0.002	0.126	−0.004	0.126
Age									−0.011	0.008	−0.008	0.008	−0.008	0.008
Education									−0.036	0.032	−0.047	0.032	−0.045	0.032
Ethnicity									−0.131	0.128	−0.056	0.131	−0.090	0.135
MEQ									−0.0001	0.011	0.001	0.011	0.001	0.011
BMI									0.019	0.017	0.019	0.017	0.018	0.017
EPI Extraversion											0.032	0.015*	0.032*	0.015
EPI Lie/Desirability											−0.024	0.015	−0.029	0.016
EPI Neuroticism											0.004	0.025	0.001	0.026
EPI Psychoticism											−0.024	0.033	−0.027	0.034
Hayling Scaled Score													−0.046	0.042
NAART													−0.005	0.009
Model fit
R 2	0.098		0.106		0.127		0.136		0.154		0.176		0.181
F	9.880		6.393		5.966		4.371		3.105		2.759		2.581
ΔR2	0.098		0.008		0.021		0.009		0.018		0.022		0.005
ΔF	9.880		1.210		4.415		0.859		0.873		1.666		0.857
Sig. ΔF (p value)	<0.001		0.300		0.036		0.462		0.514		0.155		0.424

*p < 0.05.

**p < 0.001.

Baseline sleep architecture and vulnerability group

Descriptive statistics for all polysomnographic measures on the second baseline night in the laboratory (B2) for each vulnerability group are presented in Table 4. All participants were provided a 10 h sleep opportunity on the baseline night. Participants averaged 513 ± 56 min (mean ± SD) of TST, with an average SOL of 19 ± 20 min. Participants had 18 ± 12 measured awakenings throughout the night and 171 ± 47 stage shifts. The percentages of sleep stages relative to TST were as follows (mean ± SD): 8% ± 5% in stage 1 sleep, 53% ± 8% in stage 2 sleep, 14% ± 7% in SWS, and 25% ± 5% in REM sleep, with 71 ± 35 min until onset of the first REM episode. Age had a significant effect on baseline PSG measures (F(9,251) = 9.16, p < 0.001). Older participants had shorter TST, more awakenings, spent a smaller proportion of TST in SWS, and spent a larger proportion of TST in REM. Differences among vulnerability groups on baseline sleep architecture were not statistically significant after adjusting for age, F(18,504) = 0.84, p = 0.66; Pillai’s V = 0.06; partial η² = .03.

Table 4.

Polysomnographic measures during baseline sleep (B2) by vulnerability group

	Group	Mean	Std. Dev.	N
TST (min)a	Resilient	509.68	50.31	88
	Intermediate	512.11	59.08	87
	Vulnerable	517.55	57.67	88
SOL (min)b	Resilient	17.59	19.09	88
	Intermediate	22.21	23.63	87
	Vulnerable	16.34	14.81	88
NWAKc	Resilient	18.59	11.27	88
	Intermediate	17.09	11.18	87
	Vulnerable	18.08	12.47	88
% SWSd	Resilient	14.38	7.50	88
	Intermediate	14.56	7.12	87
	Vulnerable	13.70	7.58	88
% REMe	Resilient	24.67	3.82	88
	Intermediate	24.61	5.16	87
	Vulnerable	24.34	6.00	88
Stage shiftsf	Resilient	169.19	43.37	88
	Intermediate	170.55	44.94	87
	Vulnerable	172.70	51.45	88
REM latency (min)g	Resilient	69.75	31.32	88
	Intermediate	71.65	34.77	87
	Vulnerable	71.69	35.43	88
% stage 1h	Resilient	8.15	5.33	88
	Intermediate	8.35	5.09	87
	Vulnerable	8.69	4.64	88
% stage 2i	Resilient	52.62	7.88	88
	Intermediate	52.58	7.55	87
	Vulnerable	53.21	7.92	88

^aTotal sleep time in minutes.

^bSleep onset latency in minutes.

^cNumber of awakenings after sleep onset.

^dPercent of TST spent in slow-wave sleep.

^ePercent of TST spent in REM sleep.

^fNumber of stage shifts throughout the night.

^gREM SOL in minutes.

^hPercent of TST spent in stage 1 sleep.

ⁱPercent of TST spent in stage 2 sleep.

Habitual sleep patterns outside the laboratory and vulnerability group

Descriptive statistics for all actigraphy measures for each vulnerability group in the weeks prior to and following the laboratory study are presented in Table 5. During the week prior to the study, participants averaged 8 ± 0.6 h of TST per night, with an average bedtime of 23:50 ± 54 min and wake time of 07:52 ± 53 min. Additionally, participants had a standard deviation of 38 min in their bedtimes, 43 min in their wake times, and 49 min in TST. Age had a significant effect on habitual sleep–wake patterns (F(6,264) = 5.6, p < 0.001; Pillai’s V = 0.11; partial η² = .11), specifically on the average bedtime and wake time (both p < 0.001), with older adults going to bed and waking up earlier. There were no differences in actigraphy measures among vulnerability groups in the week prior to the laboratory study (F(12,530) = 0.76, p = 0.70; Pillai’s V = 0.03; partial η² = .017).

Table 5.

Actigraphy measures prior to (PRE) and following (POST) laboratory participation by vulnerability group

			PRE		POST
			Mean	Std. Dev.	Mean	Std. Dev.
Vulnerability	Resilient	Bedtimea	23:41	0.86	00:04	1.08
		Wake timeb	07:47	0.91	08:01	1.40
		TSTc	8.08	0.62	7.90	0.91
		Bedtime variabilityd	35.41	20.52	63.95	53.79
		Wake time variabilitye	41.41	23.37	67.68	36.70
		TST variabilityf	0.78	0.41	1.35	0.69
	Intermediate	Bedtimea	23:50	0.90	00:27	1.20
		Wake timeb	07:54	0.91	08:24	1.12
		TSTc	8.06	0.69	7.92	1.02
		Bedtime variabilityd	37.82	24.40	59.26	36.61
		Wake time variabilitye	42.32	19.41	73.68	40.39
		TST variabilityf	0.81	0.45	1.45	0.74
	Vulnerable	Bedtimea	23:59	0.92	00:29	1.41
		Wake timeb	07:56	0.84	08:23	1.39
		TSTc	7.96	0.56	7.86	0.99
		Bedtime variabilityd	41.36	21.53	61.04	35.67
		Wake time variabilitye	46.54	24.65	77.81	43.38
		TST variabilityf	0.88	0.47	1.43	0.79

^aAverage bedtime (HH:MM) with Std. Dev. in minutes.

^bAverage wake time (HH:MM) with Std. Dev. in minutes.

^cTotal sleep time in hours.

^dStandard deviation of bedtimes across 7 days in minutes.

^eStandard deviation of wake times across 7 days in minutes.

^fStandard deviation in TST across 7 days in hours.

In the week following laboratory study, participants had an average 7.9 ± 1.0 h of TST per night, with an average bedtime of 00:20 ± 1.2 h and wake time of 08:16 ± 1.3 h. Additionally, participants had a standard deviation of 61 min in their bedtimes, 73 min in their wake times, and 1.4 h in TST. Age had a significant effect on habitual sleep–wake patterns, F(6,256) = 3.95, p = 0.001; Pillai’s V = 0.09; partial η² = .09, specifically on the average bedtime and wake time (both p < 0.001), with older adults going to bed and waking up earlier. There were no differences in actigraphy measures among vulnerability groups in the week following the laboratory study, F(12,514) = 1.0, p = 0.44; Pillai’s V = 0.05; partial η² = .02.

Change in sleep physiology across SR and vulnerability group

Across all participants, the experimental sleep manipulation resulted in a decline in TST from 513.1 ± 55.7 min at baseline to 229.5 ± 14.0 min at SR5. SOL decreased from 18.7 ± 19.6 min on B2 to 2.8 ± 4.2 min on SR5, and the NWAK decreased from 18 ± 12 on B2 to 3 ± 3 on SR5. There was also an increase in the proportion of TST spent in SWS to 30.2% ± 13.5% on SR5 from 14.2% ± 7.4% on B2. No change in the proportion of TST spent in REM sleep was observed. The means for polysomnographic measures by vulnerability group are presented for each protocol day in Figure 2 and Table 6. No differences in sleep outcomes were observed among vulnerability groups (all p > 0.05).

Figure 2.

Change in polysomnographic sleep measures across days in the study protocol (B2—baseline; SR1—first night of SR; SR5—fifth night of SR). Resilient (Type 1) subjects are shown by blue circles, Intermediate (Type 2) subjects are graphed in green triangles, and Vulnerable (Type 3) subjects are represented by red stars. (A) TST (min), (B) SOL (min), (C) NWAK, (D) proportion of TST spent in SWS, and (E) proportion of TST spent REM sleep.

Table 6.

Sleep PSG measures at baseline (B2) and following one (SR1) and five (SR5) nights of SR to 4 h TIB opportunity by vulnerability group

		Resilient		Intermediate		Vulnerable
		Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
B2	TSTa	509.7	50.3	512.1	59.1	517.6	57.7
	SOLb	17.6	19.1	22.2	23.6	16.3	14.8
	NWAKc	18.6	11.3	17.1	11.2	18.1	12.5
	% SWSd	14.4	7.5	14.6	7.1	13.7	7.6
	% REMe	24.7	3.8	24.6	5.2	24.3	6.0
SR1	TSTa	223.6	11.8	222.5	14.9	223.6	11.1
	SOLb	4.9	4.7	4.6	5.0	4.0	4.6
	NWAKc	5.4	3.6	5.1	4.2	4.9	4.2
	% SWSd	28.9	13.1	27.1	13.3	27.0	13.0
	% REMe	23.8	7.5	24.4	8.3	23.1	7.3
SR5	TSTa	231	8	227.3	20.7	230.6	9.5
	SOLb	3	5	3.1	4.7	2.1	2.9
	NWAKc	4	3	3.1	2.6	3.5	3.6
	% SWSd	31	14	31.5	13.0	28.1	12.8
	% REMe	25	7	25.0	7.0	26.7	8.7

^aTotal sleep time in minutes.

^bSleep onset latency in minutes.

^cNumber of awakenings after sleep onset.

^dProportion of TST spent in slow-wave sleep.

^eProportion of TST spent in REM sleep.

The main effect of protocol day was significant for TST (F(2,772) = 509.56, p < 0.001), SOL (F(2,772) = 19.92, p < 0.001), NWAK (F(2,772) = 3.56, p = 0.03), and proportion of time spent in SWS (F(2,772) = 32.23, p < 0.001). The proportion of time spent in REM sleep did not differ by protocol day (F(2,772) = 1.91, p = 0.15). There was no significant interaction effect of vulnerability group with protocol day for any of the PSG outcomes (all p > 0.30).

There was a significant effect of age on TST (F(1,772) = 33.38, p < 0.001), the NWAK (F(1,772) = 68.49, p < 0.001), proportion of time spent in SWS (F(1,772) = 100.81, p < 0.001), and the proportion of TST spent in REM sleep (F(1,772) = 42.42, p < 0.001). The effect of age on SOL (F(1,772) = 0.21, p = 0.65) was not significant. Older age was associated with shorter TST, more awakenings, less time spent in SWS, and more time spent in REM sleep. The interactions between protocol day, age, and vulnerability for PSG outcomes were not significant (all p > 0.50).

Discussion

The present analyses provide further evidence that the effects of sleep loss on vigilant attention are trait-like and that the strongest predictor of future deficits in vigilant attention is baseline PVT performance. Individuals with more lapses of attention on the PVT at baseline when rested exhibited greater deficits following chronic partial SR. Performance on the MWT was the only other factor that was significantly associated with change in PVT lapses across SR. None of the other factors assessed in the study, including demographic characteristics, subjective reports, personality, IQ, and chronotype, significantly contributed to the prediction of PVT lapses.

Interestingly, shorter sleep latency on the MWT at baseline was associated with greater PVT performance deficits following sleep loss. This finding raises the possibility of an inherent ability to maintain wakefulness in the face of mounting sleep pressure that may account for a portion of the interindividual differences in neurobehavioral responses to sleep loss. A growing body of research suggests that the ability to maintain wakefulness under conditions of sleep loss is trait-like in itself. A study by Rupp et al. reported significant interindividual variability in performance on the MWT during total sleep deprivation and SR that is stable within individuals across multiple exposures to sleep loss [83]. After controlling for baseline performance on the MWT, the intraclass correlation coefficient increased from 0.54 to 0.63 [83]. This is further supported by a recent independent study of SR and circadian misalignment in which MWT performance had an intraclass correlation coefficient of 0.76 that increased to 0.80 after accounting for baseline [84]. Taken together, these studies suggest a phenotypic ability to maintain wakefulness under conditions of sleep loss. However, data on the relationship between MWT and neurobehavioral performance are scarce. In one study, individuals classified as vulnerable to the effects of sleep loss fell asleep on average 8 min faster on the MWT than those classified as resilient following two nights of total sleep deprivation [28]. However, acknowledging significant limitations to that study regarding the duration of the MWT and the use of a modified version of the PVT, the findings should be interpreted with caution. The current study is the first to demonstrate a significant relationship between longer sleep latency during the MWT at rested baseline and resilience to the effects of chronic partial SR on PVT performance.

The present work also examined differences in baseline sleep physiology, markers of sleep homeostasis, and habitual sleep patterns in relation to individual differences in response to chronic SR. Results demonstrated that individuals who differ in vulnerability to the neurobehavioral effects of sleep loss did not differ in sleep physiology at baseline or in the rate of accumulation of homeostatic sleep pressure during chronic partial SR. All sleep-restricted subjects, regardless of their vulnerability type as classified by change in performance on the PVT from baseline, showed increased sleep pressure following SR as compared with baseline sleep, marked by faster sleep onset latencies, fewer nighttime awakenings, and increased proportion of time spent in SWS as a result of shorted TIB. Examination of habitual sleep patterns outside the laboratory using actigraphy data for the 7 days prior to study participation failed to find significant differences among vulnerability groups. Results did not differ when evaluating habitual sleep patterns prior to study entry or following completion of the SR protocol, when individuals were no longer constrained by study participation requirements.

Despite the strengths of the study, it is not without limitations. The analysis of polysomnographic data focused exclusively on sleep architecture and did not include power spectral analysis. Prior research has demonstrated trait-like patterns of response in EEG activity during total sleep deprivation and SR [85]. This study also focused on vigilant attention as a marker of interindividual differences in response to SR and although vigilant attention is the most sensitive metric of the neurobehavioral deficits induced by SR, future studies examining whether baseline factors are predictive of other neurocognitive responses to SR are needed [86]. It is possible that interindividual differences in deficits in vigilant attention or mean differences among vulnerability groups are present in EEG slow-wave activity during sleep or in EEG theta power during wakefulness, consistent with previous studies [40]. Additionally, recent findings have highlighted the effect of menstrual phase in females on cognitive performance during extended wakefulness. Women in the follicular phase showed greater lapses of attention on the PVT as compared with those in the luteal phase, with more than 60% of their responses exceeding 500 ms in RTs [87]. These findings are consistent with other research showing greater impairment on cognitive tasks during sleep deprivation in women in the follicular phase [88]. The present study did not include information on menstrual phase for the female participants, which may have contributed to the lack of findings on sex effects. Future studies that directly examine, rather than control for, the impact of menstrual phase on neurobehavioral performance across SR are needed.

In a large sample of healthy adults (N = 278) who completed a highly controlled in-laboratory study of chronic partial SR, there was no significant impact of personality, academic intelligence, subjective reports of chronotype, sleepiness and fatigue, performance on working memory, and or demographic factors such as sex, ethnicity, and BMI, on vulnerability to deficits in psychomotor vigilance across sleep loss. Observed interindividual differences in vulnerability to the effects of sleep loss were not accounted for by prior sleep history, habitual sleep patterns outside of the laboratory, baseline sleep architecture, or homeostatic sleep response across chronic partial SR. Only superior baseline performance on the PVT and the ability to maintain wakefulness on the MWT were associated with relative resilience to decrements in vigilant attention across SR.

Research has shown that individuals can be motivated to fall asleep faster or stay awake longer even while sleep deprived [89, 90]. While not much is known about the role of motivation in the mediation of sleep and wakefulness, animal literature has implicated hypocretin and dopamine in both arousal and motivation, suggesting that internal or external stressors such as hunger, predators, and mating opportunities may promote wakefulness despite homeostatic and circadian signals for sleep in the presence of sufficient motivation [91–96]. Eban-Rothschild et al. posited an updated theoretical model for sleep–wake modulation that includes an “integrator neuron” which weighs all inputs from internal and external environments to decide whether to initiate or terminate sleep [97]. In the context of this new model, interindividual differences in the sensitivity of the proposed system could partially account for differences in vulnerability to the effects of sleep loss. In this view, individuals vulnerable to the accumulation of deficits under sleep deprivation or restriction may be more sensitive to the accumulation of homeostatic sleep pressure that overrides competing interests such as neurobehavioral task requirements or external demands for wakefulness. Further research is warranted to investigate the possibility of a modulatory drive in addition to the two-process model of sleep regulation, as differences within an internal system integrating competing signals for wakefulness and sleep may help inform the present gap in the literature regarding neurobiological mechanisms driving the phenotypic response to sleep loss.

Funding

Supported by the National Institutes of Health (NIH) through grants NR004281, UL1TR001878, CTRC UL1 RR0241340, as well as the National Space Biomedical Research Institute through NASA NCC 9-58 (DFD). C.W.J. was supported by an NIH National Research Service Award (5T32HL007713).

Disclosure Statement

Financial disclosure: O. Galli is an employee of Jazz Pharmaceuticals who, in the course of her employment, has received stock options exercisable for, and other stock awards of, ordinary shares of Jazz Pharmaceuticals plc. The entirety of the present work was completed prior to the start of her employment. C.W. Jones, O. Larson, M. Basner, and D.F. Dinges have nothing to disclose. Nonfinancial disclosure: none.

Data Availability

The data underlying this article may be shared on reasonable request to the corresponding author.