Effects of acute sleep deprivation and recovery sleep on cognitive performance in depressed individuals

Abstract

Acute sleep deprivation (SD) can lead to rapid albeit transient antidepressant effects for some individuals with clinical depression. However, SD also has adverse effects on cognitive performance. As investigations into the mechanisms underlying the antidepressant effects of SD take place, it will also be important to fully characterize effects on other aspects of mental function. Here, we investigated relationships between depression and performance on a battery = assessing a range of cognitive domains at baseline, during acute SD, and after recovery sleep. Thirty-six individuals with current depression and 10 euthymic controls underwent a baseline night of sleep followed by 36 h of total SD and one night of recovery sleep in the laboratory. Participants completed the Cognition battery and a self-report survey of wellbeing twice after each protocol night. Depressed individuals had similar – if not faster – performance on Cognition subtests relative to healthy controls at baseline and reported worse wellbeing. SD had pronounced effects on both speed and accuracy across Cognition subtests, with all participants becoming slower, less accurate, and less efficient overall; no differences between depressed and healthy controls were observed. Performance returned to pre-deprivation levels after recovery sleep. These results suggest that currently depressed individuals exhibit the same decrements in cognitive performance after acute SD as non-depressed individuals, which is a critical consideration for future research aimed at elucidating the mechanisms that underlie SD’s antidepressant effects and potential therapeutic applications.

1. Introduction

A single night of acute sleep deprivation (SD) has been shown to lead to rapid and pronounced improvement in mood symptoms for up to 50 % of individuals with clinical depression (Boland et al., 2017). These antidepressant effects emerge within hours, in stark contrast to traditional treatments – such as medications or psychotherapy – that typically require weeks to achieve therapeutic benefits (Cuijpers et al., 2013; Gelenberg and Chesen, 2000). However, these fast-acting benefits are also fleeting: depressive symptoms return rapidly after subsequent sleep (Wu and Bunney, 1990). While the transience of its antidepressant effects renders SD a non-viable long-term treatment option, there is growing interest in elucidating the mechanisms underlying this phenomenon.

The antidepressant response of SD occurs despite the well-documented and adverse effects of acute SD on cognitive performance in individuals without depression. Acute SD leads to impairments across a wide variety of cognitive domains, including attention, processing speed, and working memory (Killgore, 2010; Lim and Dinges, 2010). Moreover, individuals with clinical depression have been shown to exhibit cognitive impairments in some of these very same cognitive domains that can persist even during periods of remission (Rock et al., 2014). Despite the well-characterized effects of SD on mood in depressed individuals and on cognition in healthy individuals, relatively little is known about how SD affects cognition specifically in depression. A few prior studies on the antidepressant effects of SD have included broad cognitive measures to ensure participant safety (e.g., the Montreal Cognitive Assessment in Xu et al., 2024), but none have systematically examined which specific cognitive domains are most impacted by interactions between SD and depression. Addressing this gap is important, as it may inform future efforts to harness the mood-boosting effects of SD while minimizing possible cognitive consequences. In this study, we investigated the relationship between depression and cognitive performance at baseline, following acute SD, and after recovery sleep using a cognitive battery consisting of 10 tests that assess a range of cognitive domains (Basner et al., 2015). Our group has previously published the mood findings from this study, noting that the antidepressant effect of acute SD was reduced in our highly controlled laboratory environment (Goldschmied et al., 2023). Our goal here was to characterize the cognitive effects of total SD and recovery sleep in this sample.

2. Methods

Participants.

The present study was approved by the University of Pennsylvania IRB. Thirty-six individuals with depression (69.4 % female; 33 ± 9.5 years) and 10 healthy controls with no history of disordered mood (40 % female; 35 ± 9.5 years) participated. Following the National Institute of Mental Health’s Research Domain Criteria (RDoC) project, which emphasizes dimensional rather than categorical approaches to psychopathology, patients were not required to meet DSM-5 criteria for a current depressive episode. Rather, current depression was defined more broadly as a score ≥18 on the 17-item Hamilton Rating Scale for Depression (HRSD) at an initial diagnostic visit (Hamilton, 1960; Williams, 1988), allowing us to capture clinically meaningful levels of depressive symptoms in individuals who may not meet strict diagnostic criteria. Participants on antidepressants were allowed to participate following a 2–4 week washout period supervised by a study psychiatrist. All participants also met the following inclusion criteria: 1) age between 21 and 50 years; 2) body mass index within the normal range (18.5–30); 3) no shift work, trans-meridian travel or irregular sleep/wake routine in past 60 days; 4) no current sleep disorder other than insomnia; 5) no substance abuse in the past year or current tobacco use; 6) no acute or chronic debilitating medical conditions (e.g., epilepsy); and 7) no other psychiatric disorders that would preclude participation (e.g., bipolar). Participants underwent one night of at-home screening for sleep-disordered breathing using an ambulatory recording device (Embletta) and were excluded if they had an apnea-hypopnea index >15 events/hour.

In-laboratory protocol.

Participants completed a 5-day, 4-night in-laboratory sleep deprivation protocol (Fig. 1). All participants received two baseline nights of sleep (B1, B2) that consisted of a 9 h time in bed (TIB) sleep opportunity (22:00–7:00). The baseline nights were followed by a night of total sleep deprivation (SD). During the SD period, participants were kept awake from 7:00 until 19:00 the following day for a total of 36 h of continuous wakefulness. After SD, participants were allowed a recovery night (REC) of 12 h TIB (19:00–7:00) before leaving the laboratory. While in the laboratory, participants were continuously monitored by trained staff to ensure participant safety and adherence.

Fig. 1.

In-laboratory protocol.

Note: B1 = Baseline 1, B2 = Baseline 2, SD = Sleep Deprivation, Rec = Recovery.

Mood ratings.

The HRSD-NOW – a modified version of the HRSD that removes four items with timeframes inappropriate for daily ratings – was administered to all participants around 10:00 after each baseline night, after SD, and after REC (Fig. 1). To estimate current depression severity in sensitivity analyses, scores on the HRSD-NOW were averaged across the two in-laboratory baseline assessments for each participant.

Cognition.

Testing was conducted using Cognition software (version 3; version 2 Emotion Recognition Task (ERT) with 40 stimuli) on Dell laptops. Each participant completed the Cognition battery and survey 13 times in the lab (Fig. 1): once for familiarization (20:30), three times per baseline day (10:30, 14:30, 18:30), once pre-SD (22:30), three times after SD (11:15, 14:15, 18:15), and twice post-REC (11:15, 14:15).

The following description of the Cognition battery has been modified from Basner and colleagues 2021 (Basner et al., 2021). The Cognition battery includes 10 tests that assess a range of cognitive domains (Basner et al., 2015). It has 15 unique stimulus sets that allow for repeated administration. Analyses focused on one main speed and one main accuracy outcome for each Cognition test; precise calculation methods are summarized in Supplementary Table S1. Scores were corrected for practice and stimulus set difficulty effects (Basner et al., 2020) and were z-transformed based on the average and standard deviation of baseline performance scores (i.e., all administrations on B1 and B2) across all participants to facilitate comparisons across tasks. Summary scores for speed and accuracy were calculated by averaging across z-transformed scores within the speed and accuracy domain, respectively (although risk taking propensity on the Balloon Analogue Risk Task (BART) did not contribute to accuracy summary scores). To standardize the directionality of performance scores, speed scores were multiplied by −1 so that higher scores reflected faster speed. An efficiency summary score was calculated by averaging the accuracy and speed summary z-scores.

Full descriptions of each of the 10 Cognition tests can be found in Basner and colleagues 2020 (Basner et al., 2020). Table 1. summarizes the cognitive domains assessed by each test and Supplementary Fig. 1 depicts screenshots of each test. The tests were always completed in the order shown in Table 1.

Table 1.

The 10 tests of the Cognition battery and the cognitive domains assessed by each.

Abbreviation	Full name	Cognitive domain(s) assessed
MP	Motor praxis	Sensorimotor speed
VOLT	Visual object learning task	Visuospatial learning and memory
F2B	Fractal 2-back	Working memory
AM	Abstract matching	Abstraction, concept formation
LOT	Line orientation task	Spatial orientation
ERT	Emotion recognition task	Emotion identification
MRT	Matrix reasoning task	Abstract reasoning
DSST	Digit symbol substitution test	Complex scanning, working memory
BART	Balloon analogue risk task	Risk decision making
PVT	Psychomotor vigilance test	Vigilant attention

Self-report survey.

Before each battery, participants completed a brief survey with 13 items rated on 11-point Likert scales (anchors in parentheses; midpoint labeled “neutral”): (1) sleep quality (good–poor); (2) today’s workload (very high–very low); (3) current sleepiness (not sleepy at all–very sleepy); (4) mood (happy–unhappy); (5) health (sick–healthy); (6) physical energy (energetic–exhausted); (7) mental sharpness (sharp–fatigued); (8) stress (not stressed–very stressed); (9) tiredness (tired–fresh); (10) depression (very depressed–not at all); (11) boredom (very bored–not at all); (12) loneliness (not lonely–very lonely); (13) daily monotony (very monotonous–not at all). Items 2, 5, 9, 10, 11, and 13 were reverse scored so higher values indicated more negative responses.

Statistical approach.

All Cognition tests were screened for participant nonadherence prior to analysis. Analyses were conducted in R (RCoreTeam, 2022). Code, output, and data are available via the Open Science Foundation: https://osf.io/y8vht/. P-values were adjusted using the false discovery rate method (Benjamini and Hochberg, 1995) for 23 Cognition outcomes or 13 survey items.

Baseline effects of depression, demographics, and time of day.

To assess how depression status, age, gender, education, and time of day influenced baseline Cognition and survey responses, mixed-effects models were run using data from six test sessions conducted at 10:30, 14:30, and 18:30 on baseline days (B1 and B2). Each model included one standardized Cognition or unstandardized survey outcome as the dependent variable and fixed effects for depression status (Depressed; Control = reference), age, gender (Female = reference; Male; Other), educational attainment (Some College or Less = reference; College or More), and time of day (Morning = reference; Afternoon; Evening), with a random effect for participant and adjustment for protocol day (B1 = reference; B2). A sensitivity analysis used mean HDRS-NOW scores on B1/B2 as a continuous depression index.

Effects of depression, sleep deprivation (SD), and recovery sleep (REC).

To assess the impact of depression, SD, and REC on Cognition performance and survey responses, we used two complementary approaches to (1) isolate the effect of depression status on each protocol day and (2) examine performance and survey trajectories across protocol days. First, we aimed to isolate the effect of depression status on each sleep-manipulated protocol day – SD and REC – by running two separate sets of mixed effects models. The SD models included post-SD sessions (at 10:30, 14:30, and 18:30); REC models included post-REC sessions (at 11:15 and 14:15). All models included depression status as a fixed effect, participant as a random effect, and controlled for age, gender, education, time of day, and prior performance (average B2 performance for SD models; average SD performance for REC models). This allowed us to assess group differences after each sleep manipulation while accounting for individual differences in baseline performance (for SD models) and in sensitivity to SD (for REC models). Second, we modeled performance and survey responses across consecutive protocol days to examine trajectories of change. For these analyses, six sessions were used: two on B2, two post-SD, and two post-REC (all matched for time of day at ~11:00 and ~14:15). Models included fixed effects for depression status, protocol day (B2 = reference; SD; REC), and their interaction, with the same covariates and random participant effects specified in previous analyses. Non-significant interactions were removed to focus on main effects.

3. Results

Participant demographic and clinical data are summarized in Table 2. Out of a total of 5940 administered Cognition tests, 197 (3.3 %) were excluded from analysis due to participant non-adherence. All p-values reported below reflect adjustment for multiple testing according to the false discovery rate method for the 23 Cognition outcomes or 13 self-report survey responses.

Table 2.

Demographic and baseline characteristics by group and for the full sample.

	Full Sample	Depressed	Control	p
N	46	36	10
Age in years	33.4 (9.4)	32.9 (9.5)	34.9 (9.5)	0.54
Gender				0.09
Female (%)	63.0	69.4	40.0
Male (%)	32.6	25.0	60.0
Other (%)	4.4	5.6	0.0
Race				0.56
Black (%)	39.1	36.1	50.0
White (%)	52.2	55.6	40.0
Asian (%)	2.2	2.8	0.0
Not reported (%)	6.5	5.6	10.0
Ethnicity				0.84
Hispanic (%)	8.7	8.3	10.0
Non-Hispanic (%)	76.1	77.8	70.0
Not reported (%)	15.2	13.9	20.0
Highest level of education attained				0.10
High school/GED (%)	17.4	16.7	20.0
Business/trade/vocational school (%)	4.4	0.0	20.0
1–3 years of college (%)	30.4	30.6	30.0
College (%)	32.6	33.3	30.0
Graduate school (%)	15.2	19.4	0.0
HRSD-17 score		16.9 (6.2)	0.9 (1.3)	<0.0001
HRSD-17 clinical categorization
No depression (%)		5.6	100.0
Mild depression (%)		44.4	0.0
Moderate depression (%)		38.9	0.0
Severe depression (%)		11.1	0.0
BDI score		24.9 (9.8)	0.8 (1.9)	<0.0001
BDI clinical categorization
No depression (%)		2.78	100.0
Borderline depression (%)		25.0	0.0
Mild depression (%)		16.7	0.0
Moderate depression (%)		27.8	0.0
Severe depression (%)		22.2	0.0
Extreme depression (%)		5.6	0.0

Note: Values in X (Y) format represent mean (standard deviation). Reported scores on the HRSD-17 and BDI were obtained during the initial diagnostic visit. HRSD = Hamilton Rating Scale for Depression; BDI = Beck Depression Inventory. P-values represent results of t-tests (for continuous normally distributed variables), Wilcoxon tests (for continuous non-normally distributed variables), or Fisher’s exact tests (for categorical variables) comparing the two groups.

Baseline effects of depression, demographics, and time of day.

At baseline, there were no significant effects of depression on any measure of speed or accuracy after adjusting for multiple comparisons (Fig. 2A and B; Supplementary Table S2). However, increasing age was significantly associated with reduced overall speed (β = −0.03 SDs/year, p = 0.003), accuracy (β = −0.03 SDs/year, p = 0.002), and efficiency (β = −0.03 SDs/year, p = 0.0005). When examining individual tests, increasing age was significantly associated with reduced speed on the MP (β = −0.07 SDs/year, p = 0.0001), the ERT (β = −0.04 SDs/year, p = 0.007), the DSST (β = −0.06 SDs/year, p = 0.0002), and the BART (β = −0.04 SDs/year, p = 0.04) and with reduced accuracy on the MP (β = −0.03/year, p = 0.03), the VOLT (β = −0.05/year, p = 0.001), the F2B (β= −0.06 SDs/year, p = 0.0003), the LOT (β = −0.03 SDs/year, p = 0.02), and the MRT (β = −0.04 SDs/year, p = 0.002). The only exception to this trend was found on the DSST accuracy metric, where increasing age was significantly associated with increased accuracy (β = 0.04 SDs/year, p = 0.002). No effects of gender, educational attainment, or time of day survived correction for multiple comparisons (Supplementary Table S2). Using average baseline HRSD-NOW scores as a continuous measure of depression did not significantly alter results (Supplementary Table S3).

Fig. 2.

Baseline Cognition performance and survey responses by group.

Note: Estimates in Panels A and C reflect marginal means (of z-scores based on the mean and standard deviation of baseline performance for Cognition outcomes in Panel A; of scores based on an 11-point scale for survey responses in Panel C) from mixed effects models; error bars reflect unadjusted 95 % confidence intervals (CIs). Fixed effects included depression status (Depression, Control = reference). A random effect over participant was included. All models were adjusted for age, gender, educational attainment, study day, and time of day. Effect sizes and contrast estimates in Panels B and D reflect the differences in marginal means (of z-scores in Panel B; of 11-point scale scores in Panel D). Estimates greater than 0 reflect better performance or more self-reported distress among depressed participants relative to controls; estimates less than 0 reflect worse performance or less self-reported distress among depressed participants relative to controls. Effect sizes magnitudes represented by shading in Panel B follow Cohen’s d criteria, with 0.2–0.5 = small, 0.5–0.8 = medium, and ≥0.8 = large. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 after correcting for multiple comparisons (for 23 Cognition outcomes or 13 survey items). MP = Motor Praxis Task; VOLT = Visual Object Learning Test; F2B = Fractal 2-Back; AM = Abstract Matching Test; LOT = Line Orientation Test; ERT = Emotion Recognition Task; MRT = Matrix Reasoning Test; DSST = Digit-Symbol Substitution Task; BART = Balloon Analog Risk Test; PVT = Psychomotor Vigilance Test.

At baseline, depressed participants reported feeling significantly worse than controls on 12 of 13 self-report items (Fig. 2C/D; Supplementary Table S2), including poor sleep (3.00 points, p = 0.002), sleepiness (1.50 points, p = 0.02), unhappiness (2.21 points, p = 0.001), sickness (1.81 points, p = 0.03), exhaustion (1.95 points, p = 0.003), mental fatigue (2.31 points, p = 0.002), stress (3.55 points, p = 0.002), tiredness (2.21 points, p = 0.002), depression (4.27 points, p < 0.0001), boredom (2.19 points, p = 0.01), loneliness (3.15 points, p = 0.002), and monotony (2.10 points, p = 0.02). The only non-significant item was perceived workload (p = 0.66). Time-of-day effects were also found for workload, tiredness, and boredom (Supplementary Table S2). Sensitivity analyses using HRSD-NOW scores did not alter results (Supplementary Table S3).

Effects of depression, sleep deprivation (SD), and recovery sleep (REC).

While there were no significant differences in performance between depressed and non-depressed participants on any of the 23 Cognition performance metrics at baseline (Fig. 2; Supplementary Table S2), we examined whether performance differed between the two groups after acute SD and after recovery sleep, respectively. After acute SD, depression status remained non-significant across all metrics when controlling for baseline performance (all ps > 0.05; Supplementary Table S4), with baseline scores being the strongest predictor of SD performance (all ps < 0.001). Similarly, post-recovery performance was not significantly affected by depression status, while SD performance strongly predicted recovery outcomes (all ps < 0.02; Supplementary Table S5).

We then modeled performance metrics across protocol days to examine trajectories of change and whether those trajectories differed based on depression status. There were no significant interaction effects between depression status and study day on any of our 23 Cognition outcomes after adjusting for multiple comparisons (all ps ≥ 0.16, Supplementary Table S6); therefore, interaction terms were dropped. Standardized estimates and contrasts for all 23 Cognition outcomes across days of the protocol are shown in Fig. 3; p-values for all type-III fixed effects and contrast tests with confidence intervals can be found in Supplementary Table S7.

Fig. 3.

Contrast estimates on Cognition outcomes by protocol day.

Note: Effect size estimates reflect differences in marginal means (of z-scores based on the mean and standard deviation of baseline performance) from mixed effects models; error bars reflect unadjusted 95 % confidence intervals. All models included fixed effects of study day (Baseline; SD; Recovery) and depression status (Depressed; Control) and were adjusted for age, gender, education, and time of day; a random effect over participant was included. Shaded regions of the plot represent the magnitude of the effect. Estimates greater than 0 reflect better performance on the first day relative to the second day as indicated by the legend; estimates less than 0 reflect worse performance on the first day relative to the second day as indicated by the legend. Effect sizes magnitudes represented by shading follow Cohen’s d criteria, with 0.2–0.5 = small, 0.5–0.8 = medium, and ≥0.8 = large. SD-Baseline and Rec-Baseline contrast estimates also reflect estimated marginal means for the SD and Rec protocol days, as baseline performance was transformed to reflect a value of 0 and was used as the reference in mixed effects models. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 after correcting for multiple comparisons. MP = Motor Praxis Task; VOLT = Visual Object Learning Test; F2B = Fractal 2-Back; AM = Abstract Matching Test; LOT = Line Orientation Test; ERT = Emotion Recognition Task; MRT = Matrix Reasoning Test; DSST = Digit-Symbol Substitution Task; BART = Balloon Analog Risk Test; PVT = Psychomotor Vigilance Test.

There were no significant effects of depression on any Cognition outcomes after correcting for multiple comparisons (all ps ≥ 0.37; Supplementary Table S7). Protocol day was a significant fixed effect in 16 out of 23 models after adjusting for multiple comparisons (all ps ≤ 0.03). There was a robust effect of SD on overall speed, accuracy, and efficiency: participants demonstrated significantly slower, less accurate, and less efficient performance after SD relative to both performance at baseline (Overall Speed: −0.25 SDs, p < 0.0001; Overall Accuracy: −0.36 SDs, p < 0.0001; Overall Efficiency: −0.30 SDs, p < 0.0001) and post-recovery performance (Overall Speed: −0.25 SDs, p < 0.0001; Overall Accuracy: −0.31 SDs, p < 0.0001; Overall Efficiency: −0.28 SDs, p < 0.0001). Post-recovery performance did not significantly differ from baseline performance on any Cognition outcome.

A closer look at the individual subtests revealed that the most robust effects of SD were observed on the psychomotor vigilance task (PVT) and the working memory fractal 2-back task (F2B). On the PVT, performance was slower and less accurate after SD as compared to baseline performance (Speed: −0.67 SDs, p < 0.0001; Accuracy: −0.78 SDs, p < 0.0001) and post-recovery performance (Speed: −0.86 SDs, p < 0.0001; Accuracy: −1.02 SDs, p < 0.0001). The same pattern was observed on the F2B: participants were slower and less accurate after SD than at baseline (Speed: −0.37 SDs, p = 0.009; Accuracy: −0.37 SDs, p = 0.0005) and after recovery sleep (Speed; −0.35 SDs, p = 0.02; Accuracy: −0.32 SDs, p = 0.004).

On several subtests, acute SD appeared to differentially impact speed and accuracy metrics. Slower speeds after SD as compared to both baseline and recovery performance were observed on the ERT (SD-BL: −0.57 SDs, p < 0.0001; SD-Rec: −0.37 SDs, p = 0.01), the BART (SD-BL: −0.29 SDs, p = 0.0001; SD-Rec: −0.33 SDs, p < 0.0001), and the DSST (SD-BL: −0.45 SDs, p < 0.0001; SD-Rec: −0.52 SDs, p < 0.0001) without significantly reduced accuracy. On the AM, speeds were slower after SD relative to baseline (−0.26 SDs, p = 0.02) but not post-recovery performance (Speed: −0.16 SDs, p = 0.26) – again without a significant impact on accuracy. The opposite pattern emerged on the VOLT: performance was less accurate after SD relative to baseline performance (−0.55 SDs, p < 0.0001) and post-recovery performance (−0.61 SDs, p < 0.0001) without inducing a corresponding deficit in speed.

Because of baseline differences in survey responses between depressed participants and controls (Fig. 2C and D; Supplementary Table S2), we focused on models examining group differences on each protocol day rather than across protocol days. (Models that examine performance trajectories by group are reported in Supplementary Table S6 and S7.) After SD, there were no significant differences in survey responses between depressed participants and controls on any survey items after controlling for baseline survey responses (Fig. 4C and D; Supplementary Table S4). After recovery sleep, there were differences in survey responses between depressed and non-depressed participants on 5 out of 13 survey items (Fig. 4E and F; Supplementary Table S5).

Fig. 4.

Cognition survey responses by group on each protocol day.

Note: First row: Estimates reflect marginal means from three sets of mixed effects models (Panel A = B2, Panel C = SD; Panel E = REC); error bars reflect unadjusted 95 % confidence intervals. Estimates for survey responses reflect points on an 11-point scale; variables are listed by anchors for high values. All models included fixed effects of study day (Baseline; SD; Recovery) and depression status (Depressed; Control) and were adjusted for age, gender, educational attainment, time of day, and either average baseline survey response (for SD models) or average SD response (for REC models); a random effect over participant was also included. Second row: Contrast estimates for the Depressed group relative to the Control group on each protocol day. Estimates greater than 0 reflect higher point values (i.e., feeling worse) in the Depressed group relative to the control group. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 after correcting for multiple comparisons (N = 13 survey responses).

4. Discussion

In this study, we examined the effects of acute SD and recovery sleep on cognitive performance in depressed individuals. We found that depressed individuals did not perform significantly worse than controls on the Cognition battery at baseline. On the contrary, depressed individuals were faster than healthy controls on eight of our ten Cognition tests – without compromised accuracy – though the small-to-medium effect sizes were not significant (Fig. 2B). We may have been under-powered to detect significant effects due to our small sample size, especially in the control group. This result is not entirely consistent with the large body of literature on neurocognitive deficits in depression.

Although many of our depressed participants began experiencing mood improvements prior to the SD intervention (Goldschmied et al., 2023), current research suggests that individuals with current depression and a history of remitted depression still show cognitive deficits (Hemmeter et al., 2010; Kriesche et al., 2023; Rock et al., 2014). The reasons for this discrepancy in cognitive performance between our cohort and the prior literature are unclear. One possibility is that prior work suggests a relationship between depression severity or number of prior depressive episodes and neurocognitive function (McClintock et al., 2010), and our sample consisted primarily of individuals with moderate depression severity. However, sensitivity analysis using baseline depression severity as a continuous predictor (rather than depression status as a categorical grouping variable) did not significantly affect the results. It is also possible that depressed individuals experienced stronger demand characteristics (e.g., were more motivated to perform well given the study’s explicit focus on depression). Finally, it is important to acknowledge that this neurocognitive battery was not developed for clinical use but rather to test performance in highly skilled individuals (Basner et al., 2015). Many of the Cognition tests are more complex than those typically used in clinical assessments – such as the Cambridge Neuropsychological Test Automated Battery (CANTAB) – and may engage a broader range of cognitive processes. More complex tasks might reduce sensitivity to subtler deficits in cognitive functioning or interact differently with motivational factors. In addition, heterogeneity in cognitive tasks, outcome measures, and terminology (such as the broad definition of “executive function”) contributes to inconsistencies in the literature and makes direct comparison challenging.

At baseline, we did find that increasing age was associated with slower, less accurate, and less efficient performance overall, consistent with prior studies utilizing the Cognition battery (Lee et al., 2020; Moore et al., 2017). Depressed individuals also rated themselves as feeling worse than healthy controls at baseline – even though they had already begun experiencing mood improvements upon study entry as assessed by clinician ratings.

Moving beyond baseline differences, our key takeaway is that depressed individuals exhibited the same pattern of cognitive decrements following acute SD as non-depressed individuals. Speed, accuracy, and efficiency across all tests were worse after SD compared to after baseline and post-recovery sleep at small but significant effect sizes. There was some variability in which tests and which outcome measures were most affected by SD and recovery sleep. Unsurprisingly, performance on the Psychomotor Vigilance Test (PVT) – which measures vigilant attention and is highly sensitive to sleep loss (Basner and Dinges, 2011) – declined significantly in both the speed and accuracy domains at medium-to-large effect sizes. Similarly, performance on the fractal 2-back task – which relies on working memory capacity – worsened after SD relative to baseline and post-recovery sleep, consistent with prior literature (Drake et al., 2001; Lim and Dinges, 2010). In general, speed appeared to be more adversely affected by sleep deprivation (on 6/10 tests) than accuracy (4/10 tests), and at slightly larger effect sizes. In addition to slower performance on the PVT and F2B, participants also slowed down on tests measuring emotion recognition (ERT), abstraction (AM), complex scanning and working memory (DSST), and risk-taking (BART) after SD. On each of these tests, accuracy remained unimpaired, suggesting that participants may have slowed down to preserve accuracy.

The strongest predictor of cognitive performance after SD or recovery sleep was an individual’s performance the previous day. Our results align with previous findings from Galli and colleagues (Galli et al., 2022), which showed that the best predictor of PVT performance under sleep deprivation was an individual’s well-rested baseline performance. These findings are also consistent with prior research indicating that interindividual variability in cognitive resilience to sleep loss is stable and trait-like, meaning that some individuals consistently maintain performance under SD while others are more vulnerable (Van Dongen et al., 2004). Future studies should investigate whether these individual differences can be leveraged to develop personalized interventions that mitigate cognitive impairments following SD (e.g., adaptive sleep and task schedules).

There were no significant differences between baseline performance and post-recovery performance on any Cognition tasks that showed a significant effect of SD, suggesting that our recovery sleep opportunity was sufficient to restore cognitive performance to pre-deprivation levels. Previous studies have shown that one night of recovery sleep can restore performance on tasks assessing simple cognitive functions like response inhibition (Drummond et al., 2006) and visual attention (Kendall et al., 2006). However, there is also evidence that multiple nights of recovery sleep are necessary to restore vigilant attention to baseline levels (Lamond et al., 2007). Notably, research utilizing chronic sleep restriction paradigms suggests that the magnitude of recovery is influenced by the duration and severity of prior sleep restriction, with more prolonged restriction requiring additional nights of recovery sleep (Banks et al., 2024). Further research on recovery dynamics following sleep loss is warranted, especially in clinical populations.

After SD, there were no longer any significant group differences in survey responses. Statistically, this was driven by controls reporting feeling worse after SD rather than by depressed individuals feeling better. After recovery sleep, significant group differences reemerged in only five survey items (poor sleep, unhappiness, stress, depression, and boredom), compared to the 12 group differences we observed at baseline. This pattern is consistent with a general mood-destabilizing effect of SD in healthy individuals (Baum et al., 2014; Caldwell et al., 2004; Kahn-Greene et al., 2007).

Our findings should be interpreted in light of several limitations. As previously discussed, mood symptoms in our depressed group began to improve before the SD manipulation was introduced (Goldschmied et al., 2023). While our study was tightly controlled, the laboratory setting may have provided psychological and environmental supports that improved mood. This early improvement could have obscured baseline cognitive deficits and complicated the interpretation of SD effects. At the same time, the absence of baseline deficits remains informative, as meta-analyses suggest such impairments often persist even in remitted depression. This underscores how contextual factors – such as removal from stressful home environments, added structure of the laboratory setting, or increased interactions with support-providing staff – may moderate mood, and thereby the expression of cognitive difficulties in depression. Additionally, strict eligibility criteria (e.g., medication-free status, regular sleep schedule) likely resulted in the inclusion of a less severely depressed sample, limiting generalizability. While in line with the RDoC initiative, our inclusion of depressed patients without a DSM-5 diagnosis may mean that our results apply to depressed mood and not strictly to major depressive disorder. Finally, the relatively small sample size – particularly within the control group – may have limited statistical power and contributed to variability in our outcome measures, but effect sizes were reported here to provide an indication of the magnitude of observed differences and to inform future, larger-scale studies.

In conclusion, our findings highlight the complex interplay between sleep, cognition, and mood in depression. While depressed individuals did not show expected cognitive impairments at baseline, they exhibited similar cognitive declines after SD as controls, with effects that were largely reversed following recovery sleep. These results underscore the need for future research to explore strategies that optimize the cognitive and mood-related effects of SD while minimizing functional impairment in depressed patients.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jpsychires.2025.09.062.

Data availability

All code, output, and data are available via the Open Science Foundation: https://osf.io/y8vht/.