Sleep-Like Slow Waves during Wakefulness Mediate Attention and Vigilance Difficulties in Adult Attention-Deficit/Hyperactivity Disorder

Abstract

Attention-deficit/hyperactivity disorder (ADHD) is characterized by behavioral variability and heightened inattention associated with increased mind wandering (MW) and mind blanking (MB). Individuals with ADHD frequently experience sleep disorders and excessive daytime sleepiness, suggesting interactions between attention and arousal systems. Research examining brain activity using electroencephalography (EEG) has demonstrated that sleep-like slow waves (SWs) during wakefulness are linked to inattention in neurotypical individuals following sleep deprivation, yet their role in ADHD remains unclear. This study investigated whether individuals with ADHD present with altered waking SW distribution compared with neurotypical controls and whether SWs explain attentional difficulties in ADHD. Adults with (n = 32) and without ADHD (n = 31) completed a sustained attention task while EEG recorded brain activity. Mental state probes (on-task, MW, MB) were embedded within the task. Sleep-like SWs reflect cortical slowing and were detected from EEG activity. Omission/commission errors, reaction time (RT), RT variability, mental state reports, and subjective sleepiness were analyzed. Mediation analysis examined whether SW density explained ADHD-related performance differences. Individuals with ADHD exhibited more commission errors, MW and MB, more theta oscillations over frontotemporal electrodes, and higher SW density (SW/minute) over parietotemporal electrodes. Increased SW density correlated with higher omission errors, slower RTs, greater RT variability, and elevated sleepiness ratings. On-task reports were negatively correlated with SW density. Mediation analysis revealed that SW density significantly accounted for ADHD-related attentional difficulties. Wake SW may explain attentional difficulties in ADHD, providing a potential mechanistic link between sleep disturbances and attentional fluctuations.

Significance Statement

We investigated whether slow waves (SWs) during wakefulness could explain attentional difficulties in ADHD by comparing neurotypical adults and medication-withdrawn adults with ADHD during a sustained attention task with embedded mental state probes. Using electroencephalography, we quantified SW activity and examined its relationship with objective performance measures and subjective reports of mind wandering and blanking. The ADHD group exhibited significantly higher SW activity which correlated with increased objective and subjective attentional lapses. Importantly, mediation analysis revealed that SW density accounted for performance differences between groups, suggesting that wake SWs represent a neurophysiological mechanism underlying attentional difficulties in ADHD. These findings bridge sleep and attention research in ADHD, offering new insights into ADHD's heterogeneous nature and potential intervention targets.

Introduction

Attention-deficit/hyperactivity disorder (ADHD) is a neurodevelopmental condition affecting ∼2.5% of adults and is characterized by excessive inattention, hyperactivity, and/or impulsivity (May et al., 2023; Faraone et al., 2024). Individuals with ADHD exhibit greater response time variability and higher commission and omission errors compared with neurotypical individuals, reflecting difficulty with top–down attentional control (Bellgrove et al., 2005; Tucha et al., 2017; Bozhilova et al., 2020, 2021). Additionally, individuals with ADHD report increased mind wandering (MW; operationally defined here as task-unrelated thought) through self-report questionnaires (Seli et al., 2015; Mowlem et al., 2019) and experience sampling (probes assessing current thoughts) during cognitive tasks (Madiouni et al., 2020; Bozhilova et al., 2021). This emphasizes the importance of considering subjective experience alongside cognitive performance in ADHD.

MW occurs in up to 50% of waking hours (Killingsworth and Gilbert, 2010) and is typically assessed by contrasting task-focused states based on the assumption that individuals continuously engage with conscious thought—either task-related or unrelated. However, people can also reliably report mind blanking (MB) (Andrillon et al., 2025), characterized by momentary absence of conscious thought (i.e., thinking about nothing). When given the option to report both MW and MB, children and adults with ADHD reported more frequent MB, at the expense of both MW and task-focused states (Van Den Driessche et al., 2017).

Both MW and MB correlate with decreased performance (e.g., more omission and/or commission errors) and represent subjective attentional lapses (Smallwood and Schooler, 2015). Yet MW and MB have distinct behavioral signatures: in a Sustained Attention to Response Task (SART; Robertson et al., 1997), MW is associated with impulsive responses [faster reaction times (RTs) and increased commission errors], while MB is associated with sluggish responses (slower RTs and increased omission errors; Andrillon et al., 2021). Additionally, both states correlate with lower arousal, measured via the pupil size or subjective ratings, suggesting a relationship between arousal and inattention over time. This has particular relevance in ADHD research, where dysregulated arousal and attentional variability are core features (May et al., 2023; Faraone et al., 2024).

Individuals with ADHD report higher rates of sleep disturbances, including insomnia, parasomnias, and increased daytime sleepiness (Bioulac et al., 2015; Hvolby, 2015; Bjorvatn et al., 2017), correlating with elevated frontal delta activity during rest (Helfer et al., 2020). Since delta activity is typically observed in sleep using electroencephalography (EEG), its presence during wake may reflect sleep-like activity within wakeful states, termed “local sleep” (Vyazovskiy et al., 2011; Andrillon et al., 2019). This manifests as high-amplitude slow waves (SWs) associated with decreased neuronal activity in humans and animals (Steriade, 2005; Vyazovskiy et al., 2011). These SWs increase with time spent awake or on-task and correlate with sustained attention fluctuations in the general population (Hung et al., 2013; Bernardi et al., 2015; Andrillon et al., 2021). SW behavioral effects vary by scalp location, associating with impulsivity and increased MW and MB (Freedman and Ibos, 2018; Andrillon et al., 2021; Pinggal et al., 2022). Thus, sleep-like SW during wakefulness could account for the varied cognitive profiles of individuals with ADHD (increased MW, MB, impulsive and sluggish responses, and excessive daytime sleepiness), resulting from associated sleep disturbances.

Building on these neurophysiological findings, this study investigated whether SW activity during wakefulness was more prevalent in individuals with ADHD and whether this accounted for their attentional fluctuations compared with neurotypical individuals. To examine attentional difficulties at behavioral and experiential levels, we used a SART with integrated thought probes, prompting participants to report their current mental state. Recordings were paired with high-density EEG to capture SW activity. We aimed to examine behavioral and neural differences between groups. Our primary hypotheses were that the ADHD group would: (1) demonstrate decreased task performance, (2) report more frequent MW and MB, and (3) exhibit increased sleep-like SW relative to the neurotypical group. We also hypothesized that increased SW activity would mediate the effect of ADHD on task performance and subjective experience.

Materials and Methods

Participants

Sixty-five (N = 65; 33 with ADHD and 32 neurotypical individuals) participants were recruited through online advertisements, word of mouth, and flyers around Monash University. Neurotypical participants were screened to ensure they had no history of neuropsychiatric conditions, and their t-scores on subscales G and H of the Conners’ Adult ADHD Rating Scales–Self-Report: Long Version (CAARS–S:L) were below 60 (indicating that they did not meet criteria for ADHD). Two participants were excluded from analyses: one participant due to data corruption and another due to misunderstanding task instructions. The final sample consisted of 63 participants (N = 32 ADHD and 31 neurotypical adults; see Table 1 for demographics and clinical characteristics). The protocol was approved by Monash University Human Research Ethics Committee (project number: 17839).

Table 1.

Demographics and questionnaire data

	ADHD (N = 32)	Neurotypical (N = 31)	p value
Age [years, mean (SE)]	23 (0.63)	23 (0.68)	0.736
Sex (% females)	93.8%	77.4%	0.064
Education level [years, mean (SE)]	15.8 (0.32)	15.5 (0.43)	0.592
Other lifetime psychiatric diagnoses
Depression	50%	3.2%
Any anxiety disorder	62.5%	0%
Personality disorder	3.13%	0%
Autism spectrum disorder	21.9%	0%
Post-traumatic stress disorder	3.13%	0%
Eating disorder	3.13%	0%
Obsessive compulsive disorder	3.13%	0%
Questionnaire data
Conners’ Adult ADHD Rating Scale (CAARS)
DSM-IV Inattentive Symptoms [Subscale E, mean (SE)]	80.3 (1.5)	50.8 (1.75)	<0.001
DSM-IV Hyperactive–Impulsive Symptoms [Subscale F, mean (SE)]	65.9 (2.0)	44.0 (1.42)	<0.001
DSM-IV Total ADHD Symptoms [Subscale G, mean (SE)]	77.0 (1.5)	47.6 (1.8)	<0.001
ADHD Index [Subscale H, mean (SE)]	72.1 (1.6)	45.5 (1.7)	<0.001
Epworth Sleepiness Scale [mean, (SE)]	8.7 (0.8)	6.9 (0.6)	0.079
ADHD presentation (DIVA diagnosis)
Combined	70.97%
Inattentive	29.03%
Hyperactive–impulsive	0%

ADHD: attention-deficit/hyperactivity disorder. Higher scores on the CAARS subscales indicate greater severity of ADHD-related symptoms. Higher scores on the ESS indicate a greater level of daytime sleepiness.

Experimental design

A between-subject study design was employed wherein adults with ADHD and without were recruited. Participants completed an online questionnaire battery that included demographic questions, the CAARS–S:L, Adult ADHD Self-Report Scale (ASRS-5), Wender Utah Rating Scale, Epworth Sleepiness Scale, and specific subscales from the Inventory of Depression and Anxiety Symptoms (IDAS-II) to assess traits of depression and anxiety (dysphoria, well-being, panic) and general depression. To establish diagnosis validity, all ADHD-diagnosed participants, with one exception, additionally underwent an online clinical interview using the Diagnostic Interview for ADHD in Adults (DIVA 2.0). Prior to the experimental session, participants with ADHD were required to cease their medication for a minimum of 72 h (details of participants’ ADHD medication are provided in Supplementary Table S1). Upon arrival at the laboratory, participants completed a second set of questionnaires, including Part 1 of the Imaginal Process Inventory, Morningness–Eveningness Questionnaire, Alcohol Use Disorders Identification Test, and Drug Use Disorders Identification Test.

The results from the DIVA, CAARS, and ESS questionnaires are presented in Table 1, while findings from the remaining measures are provided in Supplementary Table S2.

SART instructions were presented on both a tablet and on the computer screen before the task. Participants were seated in a dimly lit room with their chin positioned on a chin rest ∼60 cm from the computer screen and fitted with a 64-channel EEG cap. An EyeLink system (EyeLink 1000) monitored eye movements during the task; however, these data are not analyzed in the current paper.

Modified SART

Participants completed a modified digit SART (Fig. 1a), in which pseudorandom sequences of digits (1–9) were presented on a screen. Stimuli were presented via Psychtoolbox (version 3) in MATLAB (MathWorks) with an interstimulus interval of 1–1.25 s (random uniform jitter). Participants were instructed to respond with a button press (Go trials) every time they saw a new digit except for Digit 3 (No-Go trials).

Figure 1.

Experimental protocol. a, Participants completed four blocks of a modified SART where they were instructed to respond to every number except the number 3. b, Every 40–70 s, participants were asked a series of 3–4 questions about their mental state immediately prior to the interruption.

Every 40–70 s, participants were interrupted and asked a series of questions regarding their attentional focus (mental state) and alertness (Fig. 1b). Participants were first asked: “Where was your attention focused? On-Task: Press 1, Off-Task: Press 2, Blank: Press 3, Don't Remember: Press 4,” where the response options corresponded to (1) thinking about the SART, (2) thinking about something other than the SART, (3) thinking about nothing, or (4) don't remember. In cases where participants answered being “off-task” (Option 2), they were subsequently asked: “What distracted your attention from the task? Something: (1) in the room, (2) personal, (3) about the task.” All participants were then prompted to rate the intentionality of their mental state with the question: “Was your state of mind intentional? From 1: entirely intentional to 4: entirely unintentional.” Finally, participants rated their alertness using the question: “How alert have you been: Extremely alert: Press 1, Alert: Press 2, Sleepy: Press 3, Extremely sleepy: Press 4.”

The task lasted for 52 (±0.3 SEM across participants) minutes and consisted of four blocks of 12–15 min with 10 probes for each block (total: 40 probes and an average of 2,120 trials). The main task was preceded by a 27-trial training session, with three digit (1–9) pseudorandomized sequences.

Behavioral data analysis

For each trial, we defined (1) omission errors as the absence of a response during a Go trial, (2) commission errors as a response during a No-Go trial (Digit 3), (3) RTs as the time between the digit presentation and the participant's response on correctly detected Go trials, and (4) RT variability as the intraindividual coefficient of variation (ICVRT). ICVRT was computed as the standard deviation of RTs divided by the mean RT, multiplied by 100, providing a normalized measure of within-participant RT variability. Omission errors reflect a failure to respond to relevant visual stimuli, whereas commission errors indicate a failure to inhibit a response. Trials with RTs shorter than 150 ms were excluded from analyses as responses below this duration typically represent anticipatory responses rather than stimulus-driven behavior (Thorpe et al., 1996; Fabre-Thorpe et al., 2001; Kirchner and Thorpe, 2006). In addition, we computed the d′, a sensitivity index reflecting the ability to discriminate Go from No-Go trials, and criterion, a response bias measure indicating the tendency to favor Go or No-Go responses. d′ was calculated as the difference between the z-transformed hit rate (proportion of correct responses on Go trials) and z-transformed false alarm rate (proportion of commission errors on No-Go trials). Criterion was calculated as −0.5 times the sum of these two z-transformed rates.

From the integrated thought probes, we extracted the percentage of each reported attentional state (on-task, MW, MB, “don’t remember”), as well as average intentionality ratings for MW and average sleepiness ratings for each group.

We extracted behavioral performance metrics, attentional states, and sleepiness ratings from a 20 s window preceding each probe, as in Andrillon et al. (2021), ensuring the inclusion of two No-Go trials while capturing trials temporally close to probe presentation (Supp. Fig. S1). The behavioral patterns observed in these pre-probe windows were consistent with the results from our main behavioral analysis that included all trials (Fig. 2).

Figure 2.

SART performance results. Performance metrics included (a) omission errors, (b) commission errors, (c) mean RT, and (d) RT variability (ICVRT). The left panels show the mean performance of each group (neurotypical in orange and ADHD in blue) across blocks (N = 4 blocks) with error bars representing the SE of the performance variables. The right panels show the overall mean and SE by group. Stars denote the significance level of block effect on performance (left panels) and performance differences between groups (right panels) as determined by FDR-corrected LMEs (ns: nonsignificant; *p < 0.05; **p < 0.01; ***p < 0.001).

We also examined the behavioral performance, distribution of mental states, and self-report sleepiness ratings according to ADHD presentation (Supp. Fig. S2).

EEG

Data acquisition and preprocessing

Participants sat in a dim room, with their chin on a rest (∼60 cm from a screen). EEG data were recorded using 64 active scalp electrodes with a BrainAmp DC system (Brain Products), referenced to FCz, and sampled at 500 Hz using bandpass filters of 0.016–250 Hz. Eye movements were recorded (EyeLink 1000) but not analyzed in the current study to focus our analyses on EEG data.

EEG data were preprocessed using a custom MATLAB script focusing on probe-based segmentation. Continuous EEG recordings were segmented from −25 to +2 s relative to probe onset, prior to preprocessing. Using the FieldTrip toolbox (Oostenveld et al., 2011), data were demeaned, bandpass filtered (0.1–40 Hz) with a fourth-order Butterworth filter, notch filtered at 50 and 100 Hz to remove line noise, rereferenced to the average of all channels, and baseline corrected. For the independent component analysis (ICA), we applied a 1 Hz high-pass instead of 0.1 Hz to minimize slow drift contamination. This was done only with the ICA data and did not affect the analysis of SW activity in the δ–θ range (1–7 Hz), which was the focus of our subsequent analyses.

We implemented a data-driven approach to automatically identify and remove noisy EEG channels using statistical thresholds based on standard deviation (SD) and kurtosis metrics. For each probe, logarithmic values of SD and kurtosis were computed for each channel. Thresholds for identifying noisy channels were set at three SDs above the mean of nonfrontal channels. For frontal channels, thresholds were adjusted using frontal channel-specific means and standard deviations to account for ocular artifacts. These thresholds identified channels with excessive variability (SD) or abnormal distributions (kurtosis). Channels exceeding these thresholds in >50% of probes were identified as noisy (mean number of noisy channels detected per participant ± SEM, 0.22 ± 0.07) and interpolated using the weighted method from FieldTrip with neighboring channels. Probes were excluded from further analyses if >20% of their channels exceeded the statistical thresholds for SD and/or kurtosis (mean number of excluded probes per participant ± SEM, 0.25 ± 0.6).

The preprocessed data were then analyzed using ICA in EEGLAB (Delorme and Makeig, 2004) for automated classification of artifact components using the ICLabel plugin (Pion-Tonachini et al., 2019). Components were automatically removed if they showed a high correlation with eye movements (probability > 0.9) or cardiac activity (probability > 0.8). The artifact components identified from probe-epoched data were then applied to remove artifacts from trial-epoched data.

Trial-epoched data were segmented around stimulus onset, with epochs from −0.2 to +1.5 s after stimulus presentation.

We used probe-epoched data for power spectrum and SW analysis and trial-epoched data for event-related potential (ERP) analysis.

ERPs

Trial-epoched data were rereferenced to the average of TP9 and TP10 electrodes (located near the mastoid regions) and a baseline (−200 to 0 ms) correction was applied. We averaged these time-locked responses by condition to obtain ERPs for Go and No-Go trials separately.

Spectral analysis

The power spectral density (PSD) was computed for each electrode using the probe-epoched data (20 s before each probe). Spectral analysis was performed using FieldTrip's implementation of the multitaper frequency transform (ft_freqanalysis) with a smoothing frequency of 0.5 Hz across the 1–40 Hz range. Data were padded to the next power of 2 to optimize the fast Fourier transform computation and was log-transformed (base-10 log).

Sleep-like SWs

Sleep-like SWs were detected independently for each channel based on established algorithms to detect SWs in sleep (Riedner et al., 2007) and wakefulness (Andrillon et al., 2021; Pinggal et al., 2022). EEG signals were rereferenced to the average of TP9 and TP10 electrodes (located near the mastoid regions) and bandpass filtered between 1 and 10 Hz using a Chebyshev Type 2 filter. The start and end of individual waves were defined by consecutive negative and positive zero-crossings. Candidate waves were required to have durations between 0.1 and 1.0 s. For each wave, we extracted a series of parameters including the most negative and positive peaks and their difference (peak-to-peak amplitude; PTP), duration, and frequency (calculated as 1/duration). Waves were discarded if they occurred >20 s before a probe, had a frequency <1 Hz or >7 Hz (resulting in a final analyzed range of 1–7 Hz) and a positive peak exceeding 75 µV, or occurred within 1 s of a high-amplitude event (>150 µV PTP).

High-amplitude SWs were defined using electrode-specific thresholds derived from the neurotypical group. For each electrode and each neurotypical individual, we computed the 80th percentile of the PTP across all waves. These percentile values were then averaged across the neurotypical group (after excluding participants with values >5 standard deviations from the mean). SW thresholds were defined as the average of these individual percentiles for each electrode. Finally, we selected waves exceeding this threshold and classified them as high-amplitude “sleep-like” SWs. Consistent with previous work (Vyazovskiy et al., 2011; Hung et al., 2013; Bernardi et al., 2015; Quercia et al., 2018; Andrillon et al., 2021), we included waves in the δ–θ range ([1–7] Hz). For each probe, we computed the frequency and density (waves/minute) of wake SWs. To examine potential group differences in frequency distributions, we analyzed the proportion of waves across 1 Hz frequency bins (1–7 Hz) both before and after amplitude-based selection. The threshold selected here (80th percentile) is lower than the 90th percentile used in previous publications focusing on sleep-like SWs (Andrillon et al., 2021; Pinggal et al., 2022) and was based on the examination of the PTP distributions (Fig. 5a) to avoid being overly conservative for the selection of SWs. We replicated all our analyses with the 90th percentile (Figs. 5, 6), which provided the same results except for an absence of significant clusters in Figure 5d.

Figure 5.

Increased SW activity in the ADHD group compared with neurotypical controls (NT). a, Left panel: Distribution of SW amplitude (waves per minute) showing higher density in NT (orange) but smaller amplitude compared with ADHD (blue) participants. Right panel: Difference wave (ADHD minus NT) with shaded SE of the mean (SEM) showing amplitude ranges where participants with ADHD exhibited lower or higher amplitudes than NT, highlighting the characteristic shift toward higher-amplitude SWs in ADHD. b, The left panel depicts the mean SW density of each group (NT in orange and ADHD in blue) across blocks with error bars representing the SE. The right panel shows the overall mean SW density and SE by group. Stars denote the significance level of block effect on SW density (left panels) and differences between groups (right panels) as determined by LMEs (ns: nonsignificant; *p < 0.05; **p < 0.01; ***p < 0.001). c, Topographical maps of the SW density distribution across the scalp for the ADHD and NT groups. d, Statistical comparison maps showing the t values resulting from a linear mixed-effect model performed for each electrode for group main effect (ADHD vs NT, left) and group by block interaction (right). Gray dots indicate significant electrodes with uncorrected p < 0.05, while black dots represent electrodes with significant differences after cluster permutation (1,000 permutations; cluster α = 0.05).

Figure 6.

SW density correlates with and mediates task performance. a, Topographical maps showing t values from LMEs testing the association between SW density (SWD) and behavioral measures: omission errors, RT, RT variability (ICVRT), subjective sleepiness ratings, and on-task reports. Positive t values (red) indicate positive associations, while negative t values (blue) indicate negative associations. b, Conceptual diagram of the mediation analysis examining how SWs mediate the relationship between group status (neurotypicals in orange, ADHD in blue) and task performance. c, Topographical maps showing t values of the mediation effect for each performance measure. Significant electrodes are marked as black, bold dots (FDR-corrected p < threshold), or gray (uncorrected p < 0.05). FDR correction was applied jointly across analyses in a and c.

Statistical analyses

To compare demographics, we conducted t tests (age, education, CAARS, and ESS) and a χ² test (sex; p values in Table 1).

Linear mixed-effect models (LMEs) in MATLAB R2022b (“fitlme,” Statistics and Machine Learning Toolbox) assessed group effects (categorical fixed effect) on commission errors, omission errors, RTs (correct Go trials), RT variability, d′, mental states, sleepiness ratings, theta power, and SW density. Errors and RTs were computed across trials, RT variability, d′ and mental states across blocks, and sleepiness ratings and SW density across probes. Models included subject identity as a random effect and block and group as fixed effects, with some including a block-by-group interaction. For each dependent variable, likelihood ratio tests were used to compare two models: a base model with main effects of group and block [Dependent Variable ∼ Block + Group + (1|SubID)] and a model with a group × block interaction [Dependent Variable ∼ Block × Group + (1|SubID)]. The best-fitting model from this comparison was then tested with age added as a covariate [e.g., Dependent Variable ∼ Block + Group + Age + (1|SubID) or Dependent Variable ∼ Block × Group + Age + (1|SubID)]. The best-fitting model based on likelihood ratio tests was then used for inference. To estimate the effect of categorical variables, we ran an ANOVA on the model (“anova” function) and reported the F values along with the associated degrees of freedom and p value. We also report the models’ coefficients (β) and the associated standard errors (SE) to estimate the strength and direction of the effect of a specific variable. For performance and mental state analyses, p values were adjusted using false discovery rate (FDR) via the Benjamini–Hochberg method (α = 0.05).

χ² tests assessed group differences across sleepiness levels (1–4) and the association between sleepiness and mental state reports. For the latter and for each mental state/sleepiness pair, standardized residuals were inspected post hoc, and values greater than ±1.96 were considered as indicating significant over- or underrepresentation.

To estimate the effect size of the association between sleepiness and mental states (two categorical variables), we computed Cramer's V. To estimate the effect size of the group on each sleepiness rating or mental state (post hoc test), we used the Phi coefficient. Independent t tests (FDR Benjamini–Hochberg-corrected) examined the relationship between intentionality and MW, with corresponding Cohen's d effect sizes reported. Exploratory analyses assessed whether performance and mental state reports differed by ADHD presentation (Supp. Fig. S2).

Examining ERPs in the ADHD and neurotypical groups revealed a clear P300 component (400 ms poststimulus; Fig. 4) for No-Go trials compared with Go trials. To quantify the group differences, we focused on representative electrodes (Pz and Fz). For spectral analysis, we computed PSD across the 1–40 Hz range. We used ANOVAs to assess group differences over a specific time (ERPs) or frequency (PSD) window. We averaged power over the significant frequency cluster (3–8.7 Hz; corresponding to the theta band) and applied LMEs and a cluster permutation procedure (Maris and Oostenveld, 2007) to identify electrode clusters showing a significant group difference for that frequency band (1,000 permutations; cluster α = 0.05; Monte Carlo α = 0.05).

Figure 4.

ERPs and power spectrum in ADHD compared with neurotypical controls (NT). a, ERP waveforms computed on Pz comparing No-Go (continuous lines) and Go (dashed lines) conditions for NT (gray) and ADHD (green) groups. The shaded areas around each line represent SE of the mean (SEM). Significant but uncorrected group differences in this No-Go–Go difference were observed between 0.6 and 1.5 s, indicated by the gray shaded region. b, Power spectrum density measured at Fz comparing NT (orange) and ADHD (blue) groups across 1–40 Hz. The shaded areas around each line represent SEM. Significant but uncorrected group differences in spectral power were observed between 3 and 8.7 Hz, indicated by the gray shaded region. c, Topographical maps of the power distribution across the scalp for the ADHD (left) and NT (middle) groups. The map on the right shows the statistical comparison between the two groups, with t values from linear mixed models performed at each electrode. Gray dots indicate significant electrodes with uncorrected p < 0.05, while black dots represent electrodes with significant differences after cluster permutation (1,000 permutations; cluster α = 0.05).

For SW density, we fitted LMEs with age, group, block, and the group × block interaction as fixed effects and participant ID as a random effect [SW Density ∼ Age + Block × Group + (1|SubID)]. For electrode-level analysis, a cluster-based permutation approach (Maris and Oostenveld, 2007) was used to correct for multiple comparisons (1,000 permutations; cluster α = 0.05; Monte Carlo α = 0.05).

We conducted a mediation analysis to test whether the relationship between group and task performance could be explained, at least in part, by individual differences in sleep-like SW activity. To assess whether performance was associated with SW density independently of the group, we examined their relationship after statistically controlling for the group (ADHD vs neurotypical) and age. Establishing that (1) group modulates performance, (2) group modulates SW density, and (3) SW density and performance are still correlated after statistically controlling for the influence of group and age can be interpreted as evidence for a mediation effect of the group on performance by SWs. In practice, we first fit separate LMEs for performance and SW density at each electrode, including age, group, block, and the group × block interaction as fixed effects and subject identity as a random intercept. Residuals from each model were extracted, representing variance unexplained by age, group, and block. We then computed correlations between the residuals of these models to determine whether performance variance was associated with SW density after regressing out the effect of the group and age. We restricted our mediation analyses to performance variables that showed a significant group effect (omission errors, RT, RT variability, sleepiness ratings, and on-task reports). For each of these variables and each electrode, evidence for mediation of the group effect on performance via SW density was defined as (1) a significant effect of group on SW density and (2) a significant correlation between the residuals of SW density and the residual of the performance variable. To control for multiple comparisons across electrodes and variables, FDR correction was applied: p values from tests examining SW effects on performance and from mediation analyses were aggregated for FDR correction.

Code accessibility

All code used for this study are available at https://github.com/elaine18rose/MWMB_ADHD.

Results

Behavior

The best-fitting models for RT, RT variability, d′, and criterion included group and block number as fixed effects and subject identity as a random effect. For commission errors, the best-fitting model included age, group, and block number as fixed effects and subject identity as a random effect. For omission errors, the best-fitting model included age as a fixed effect, a group × block interaction, and subject identity as a random effect. FDR correction was applied using the Benjamini–Hochberg procedure (α = 0.05).

Although there was no overall age (F_{(1, 118,740)} = 0.57; p = 0.520) or group effect for omission errors (F_{(1, 118,740)} = 0.76; p = 0.479), omission errors increased across blocks (F_{(1, 118,740)} = 25.50; p < 0.001), and there was a significant group × block interaction (F_{(1, 118,740)} = 21.92; p < 0.001; see Table 2 for details). This interaction was driven by a steeper increase in omission errors across blocks in the ADHD group (β = 0.0057; SE = 0.0007 per block) compared with the neurotypical group (β = 0.0025; SE = 0.0005 per block; Fig. 2). Commission errors were higher in the ADHD than the neurotypical group (F_{(1, 14,829)} = 7.32; p = 0.017) and increased over time across groups (F_{(1, 14,829)} = 78.83; p < 0.001), although there was no significant group × block interaction (χ²₍₁₎ = 0.30; p = 0.586) or age effect (F_{(1, 14,829)} = 1.05; p = 0.417).

Table 2.

LME fixed-effect statistics for SART performance

Performance		Estimate (β)	SE	t value	df	Uncorrected p value	FDR-corrected p value
Omission errors	Intercept	0.026	0.020	1.28	118,740	0.20
	Age	−0.00065	0.00086	−0.75	118,740	0.451	0.520
	Group (ADHD)	−0.0056	0.0064	−0.87	118,740	0.383	0.479
	Block	0.0025	0.00049	5.05	118,740	<0.001	<0.001
	Group × block	0.0032	0.00069	4.68	118,740	<0.001	<0.001
Commission errors	Intercept	0.38	0.13	2.83	14,829	0.005
	Age	−0.0058	0.0057	−1.02	14,829	0.306	0.417
	Group (ADHD)	0.11	0.041	2.71	14,829	0.007	0.017
	Block	0.029	0.0033	8.88	14,829	<0.001	<0.001
d′	Intercept	2.61	0.17	14.94	249	<0.001
	Group (ADHD)	−0.47	0.22	−2.13	249	0.034	0.057
	Block	−0.15	0.030	−5.18	249	<0.001	<0.001
Criterion	Intercept	−0.95	0.05	−20.40	249	<0.001
	Group (ADHD)	−0.12	0.05	−2.36	249	0.019	0.036
	Block	−0.004	0.01	−0.38	249	0.704	0.755
RT	Intercept	0.39	0.0088	44.16	112,970	<0.001
	Group (ADHD)	0.0005	0.012	0.04	112,970	0.968	0.968
	Block	0.0006	0.00034	1.73	112,970	0.084	0.126
RT variability (ICVRT)	Intercept	0.268	0.021	12.92	249	<0.001
	Group (ADHD)	0.067	0.028	2.43	249	0.016	0.034
	Block	0.015	0.0025	5.70	249	<0.001	<0.001

Bold: significant main effect after p values were FDR-corrected using the Benjamini–Hochberg method. ADHD: attention-deficit hyperactivity disorder; ICVRT: intraindividual coefficient of variation of reaction time.

Signal detection analyses revealed a trend for lower d′ in the ADHD group compared with the neurotypical group (F_{(1, 249)} = 4.54; p = 0.057). Additionally, the ADHD group had a significantly lower response criterion compared with the neurotypical group (F_{(1, 249)} = 5.56; p = 0.036), reflecting a more liberal response strategy in ADHD and a relatively more conservative strategy in the neurotypical group.

RT analyses showed no significant group (F_{(1, 112,970)} = 0.0016; p = 0.968) or block effects (F_{(1, 112,970)} = 2.99; p = 0.126), indicating stable response speeds over time. However, RT variability (ICVRT) was higher in the ADHD group compared with the neurotypical group (F_{(1, 249)} = 5.90; p = 0.034). Additionally, RT variability increased over time regardless of the group (F_{(1, 249)} = 32.50; p < 0.001), suggesting greater overall response inconsistency, with ADHD participants exhibiting higher variability throughout.

Mental states

We examined mental state reports (on-task, MW, MB, and “don't remember”) using block-level LMEs (Fig. 3). Likelihood ratio tests indicated that the best-fitting model included group and block as fixed effects and subject identity as a random effect. For “don't remember” reports, the best-fitting model included a group × block interaction. FDR correction was applied using the Benjamini–Hochberg procedure (α = 0.05).

Figure 3.

Individuals with ADHD report more MW, less intentionality, and higher sleepiness than neurotypical controls (NT). a, Distribution of reported mental states across participants and groups. b, Distribution of reported intentionality for MW episodes. c, Distribution of sleepiness ratings. All panels show mean percentages across probes with error bars representing the SE of the mean for each group (NT in orange and ADHD in blue). Stars denote the significance level of differences between groups for each category as determined by LMEs (ns: nonsignificant; *p < 0.05; **p < 0.01; ***p < 0.001).

There was a significant main effect of the group on on-task reports (F_{(1, 249)} = 28.56; p < 0.001), with the ADHD group reporting fewer on-task states than the neurotypical group (see Table 3 for details). Additionally, in both groups, on-task reports decreased over time (F_{(1, 249)} = 37.81; p < 0.001). The significant main effect of the group on MW (F_{(1, 249)} = 8.68; p = 0.004) and MB (F_{(1, 249)} = 5.91; p = 0.016) was driven by higher reports in the ADHD group than the neurotypical group. There was a significant main effect of block on MW (F_{(1, 249)} = 27.69; p < 0.001), but not MB (F_{(1, 249)} = 1.10; p = 0.295). “Don't remember” responses had nonsignificant group (F_{(1, 248)} = 0.03; p = 0.864) and block effects (F_{(1, 248)} = 0.22; p = 0.642). However, the group × block interaction was significant (F_{(1, 248)} = 5.44; p = 0.02), indicating that “don't remember” reports increased more across blocks in the ADHD group compared with the neurotypical group.

Table 3.

LME fixed-effect statistics for mental state reports

Mental states		Estimate (β)	SE	t value	df	Uncorrected p value	FDR-corrected p value
On-task	Intercept	0.63	0.042	15.09	249	<0.001
	Group (ADHD)	−0.26	0.049	−5.34	249	<0.001	<0.001
	Block	−0.055	0.009	−6.15	249	<0.001	<0.001
MW	Intercept	0.25	0.044	5.67	249	<0.001
	Group (ADHD)	0.15	0.051	2.95	249	0.004	0.008
	Block	0.052	0.010	5.26	249	<0.001	<0.001
MB	Intercept	0.125	0.029	4.28	249	<0.001
	Group (ADHD)	0.076	0.03	2.43	249	0.016	0.028
	Block	−0.0079	0.0076	−1.05	249	0.295	0.379
“Don't remember”	Intercept	0.016	0.015	1.07	248	0.285
	Group (ADHD)	−0.0036	0.021	−0.17	248	0.864	0.864
	Block	0.0023	0.0049	0.47	248	0.642	0.722
	Group × block	0.016	0.0068	2.33	248	0.021	0.031

Bold: significant main effect after p values were FDR-corrected using the Benjamini–Hochberg method.

For MW intentionality (independent t tests, FDR Benjamini–Hochberg correction α = 0.05), neurotypical controls reported more instances of “somewhat intentional” MW (t₍₆₀₎ = 3.43; p = 0.006; Cohen's d = 0.86), whereas participants with ADHD reported more “entirely unintentional” episodes (t₍₆₀₎ = −3.37; p = 0.006; Cohen's d = −0.86; Fig. 3). The trend for group differences in “entirely intentional” (t₍₆₀₎ = 2.42; p = 0.05; Cohen's d = 0.61) did not survive correction. No significant group differences were observed for “somewhat unintentional” MW (t₍₆₀₎ = 0.06; p = 1.98; Cohen's d = 0.02).

Self-reported sleepiness ratings

To assess self-reported sleepiness ratings, the best-fitting LME model, as determined from likelihood ratio tests, included group and block as fixed effects and subject identity as a random effect. There was a significant main effect of group (F_{(1, 2,517)} = 17.12; p < 0.001) with participants with ADHD reporting higher sleepiness overall compared with neurotypical individuals (see Table 4 for detailed LME results). There was also a significant main effect of block (F_{(1, 2,517)} = 288.05; p < 0.001) with sleepiness increasing over time.

Table 4.

LME fixed-effect statistics for self-reported sleepiness ratings

		Estimate (β)	SE	t value	df	Uncorrected p value	FDR-corrected p value
Sleepiness ratings	Intercept	2.08	0.09	23.47	2,517	<0.001
	Group (ADHD)	0.49	0.12	4.14	2,517	<0.001	<0.001
	Block	0.19	0.01	16.97	2,517	<0.001	<0.001

Bold: significant main effect after p values were FDR-corrected using the Benjamini–Hochberg method.

Post hoc χ² tests revealed significant group differences across all sleepiness levels (Fig. 3). Neurotypical individuals reported higher proportions of alert ratings, particularly at Level 1 (extremely alert: χ²_{(1, 2,520)} = 59.82; p < 0.001; ϕ = 0.15) and Level 2 (alert: χ²_{(1, 2,520)} = 143.59; p < 0.001; ϕ = 0.24). Conversely, the ADHD group reported more sleepiness ratings with significant group differences at Level 3 (sleepy: χ²_{(1, 2,520)} = 67.82; p < 0.001; ϕ = 0.16) and Level 4 (extremely sleepy: χ²_{(1, 2,520)} = 60.53; p < 0.001; ϕ = 0.15). All effects remained significant after FDR correction (α = 0.05). Effect sizes indicate small-to-moderate associations between group and sleepiness ratings across levels (ϕ = 0.15–0.24).

To assess the association between sleepiness ratings and mental state reports, a χ² test of independence was conducted. A significant association was found (χ²_{(9, 2,560)} = 569.83; p < 0.001), with a moderate effect size (Cramer's V = 0.27). Post hoc analysis of standardized residuals identified specific cell-level contributions: on-task reports were significantly overrepresented when participants reported being extremely alert and alert, and underrepresented when extremely sleepy and sleepy, whereas MW and MB were overrepresented at higher sleepiness levels. Table 5 presents the percentage distribution of mental states across sleepiness ratings.

Table 5.

Percentage distribution of mental state reports by self-reported sleepiness ratings

	N	On-task	MW	MB	Don't remember
Extremely alert	155	90.97%	7.74%	1.29%	0%
Alert	832	54.81%	38.46%	4.93%	1.80%
Sleepy	998	28.46%	48.40%	19.34%	3.81%
Extremely sleepy	575	10.09%	59.30%	22.26%	8.35%

N indicates the number of probes within each sleepiness rating. Bolded percentages indicate statistically significant over- or underrepresentation (standardized residuals ≥ ±1.96) based on post hoc χ² residual analysis. Percentages reflect the distribution of mental state reports within each sleepiness category.

ERP and power spectrum

As in previous studies using the SART, we observed that No-Go trials elicited a strong P300 component compared with Go trials. We thus examined group differences in ERPs for No-Go and Go trials at Pz using ANOVAs for each time point. Uncorrected tests showed differences in the 0.63–1.5 s window (Fig. 4a), on a time window following the P300 (see Discussion). However, these differences did not survive a cluster-based permutation correction (1,000 permutations, α = 0.05).

For spectral analysis, we compared PSD (1–40 Hz) at Fz using the same approach. While no group differences survived correction, uncorrected results indicated differences in the 3–8.7 Hz range (Fig. 4b). Exploratory analyses across all electrodes (Fig. 4c) identified significant clusters primarily in frontal and temporal regions, which survived a cluster permutation approach (see Materials and Methods).

SWs in wakefulness

Prior to thresholding the waves, the distribution of all the waves detected revealed that the ADHD group had fewer low-amplitude SWs but more higher-amplitude waves compared with neurotypicals (Fig. 5a). This shift between the two groups is observed at moderate amplitude (∼15 µV), which prompted us to avoid being too conservative when selecting SWs based on their amplitude (see Materials and Methods). Analysis of the frequency distribution of all detected waves (before amplitude selection) showed no significant differences between groups, indicating that group differences were primarily driven by amplitude rather than frequency characteristics. We then selected the highest amplitude SWs based on amplitude thresholds computed at the electrode and subject level for the neurotypical group and averaged across subjects (see Materials and Methods).

Examination of the frequency distribution of high-amplitude SWs revealed similar patterns between groups, with both peaking at ∼3 Hz (frequency bin effect, β = −0.022; p < 0.001) and no significant group × frequency bin interaction (β = −0.003; p = 0.511). This indicates that both groups showed similar frequency patterns, with individuals with ADHD showing elevated SW density uniformly across the 1–7 Hz range.

The global SW density (averaged across electrodes) indicated a sustained and significant increase in the ADHD group compared with the neurotypical group (F_{(1, 154,740)} = 5.75; p = 0.016) after controlling for age (Fig. 5b). There was also a significant main effect of block (F_{(3, 154,740)} = 5.88; p < 0.001) with SW density increasing over time and a significant group × block interaction (F_{(3, 154,740)} = 14.25, p < 0.001), indicating that the pattern of SW density changes across blocks differed between groups (Fig. 5b). Age was also a significant covariate (F_{(1, 154,740)} = 5.14; p = 0.023), with SW density decreasing with age across both groups (β = −0.81; SE = 0.36).

Topographies illustrate an increase in SW density across the scalp in the ADHD group compared with the neurotypical group (Fig. 5c). We then examined group differences in SW density by fitting LMEs at each electrode. To control for multiple comparisons, we completed a cluster permutation (1,000 permutations, cluster α = 0.05; see Materials and Methods for further details). SW density was modeled as the dependent variable, with age, group, block, and group × block interaction as fixed effects and subject identity as a random effect. Results are reported as topographies in Figure 5d and revealed higher SW density in the ADHD group, particularly in parietotemporal electrodes compared with neurotypicals. The significant electrodes found for the group-by-block interaction did not survive the cluster permutation analysis.

To investigate how SW activity in wakefulness relates to SART performance, LMEs were implemented separately for each electrode. Dependent variables included performance measures (commission errors, omission errors, mean RT, and RT variability) and probe responses (MW, on-task reports, and sleepiness ratings). For each model, fixed effects included age, block, and SW density, with subject identity as a random effect.

Results are reported as topographies in Figure 6a which showed that higher global SW density was associated with more omission errors, slower mean RTs, and greater RT variability, while commission errors showed no relationship. Increased SW density in frontal and occipital regions was also predictive of higher sleepiness ratings (Fig. 6a). While MW showed no association with SW density, lower SW density in temporal regions was associated with more on-task reports.

Mediation analysis

We conducted electrode-wise mediation analyses to examine if SW density in wake mediated the relationship between ADHD and task performance, controlling for block and age effects. As detailed in the Materials and Methods section, the mediation of the effect of group on performance by the density of SWs in electrode k is tested by verifying that there is a (1) a significant effect of group on behavior, (2) a significant effect of group on SW density for electrode k, and (3) a significant correlation between the SW density and performance after removing the variance explained by the group effect (correlation between residuals; Fig. 6). This is to show that SW density and performance are associated even after the removal of the group effect, suggesting that the correlation between the two variables is not simply driven by the group effect. Significant mediation effects (Fig. 6c) indicated that group differences in SW density mediated the link between ADHD and performance, particularly for omission errors, mean RT, and RT variability, with effects most prominent in frontal and centroparietal regions (FDR-corrected p < threshold). Sleepiness ratings also showed significant mediation in right centroparietal electrodes, while on-task reports showed minimal mediation, with few electrodes reaching significance.

Discussion

Attentional difficulties in ADHD: behavioral and cognitive manifestations

The SART findings highlight cognitive challenges in adults with ADHD, with higher commission errors reflecting elevated impulsivity and increased RT variability suggesting worsening attentional control over time (Kofler et al., 2013). These findings extend previous research (Van Den Driessche et al., 2017; Bozhilova et al., 2020; Madiouni et al., 2020), providing insights into time-on-task effects reflecting increased fatigue susceptibility and possible trait-like hypoarousal in ADHD.

Irrespective of the group, commission errors increased across blocks indicating declining executive control (Gunzelmann et al., 2011; Tucha et al., 2017). While omission errors did not differ between groups, the ADHD group showed steeper increases, consistent with greater cognitive fatigability. However, both groups maintained comparable mean RTs (Bozhilova et al., 2020; Madiouni et al., 2020), though the ADHD group showed greater RT variability.

Thought probe analyses revealed distinct subjective experiences (Fig. 3). Neurotypical individuals reported more on-task focus, while those with ADHD reported more MW (Seli et al., 2015; Madiouni et al., 2020). Participants with ADHD primarily experienced “entirely unintentional” MW, whereas neurotypical participants reported more “somewhat intentional” episodes, suggesting differences in attentional control mechanisms (Seli et al., 2015). Moreover, the ADHD group reported more MB, potentially reflecting executive disengagement (Van Den Driessche et al., 2017; Madiouni et al., 2020). Individuals with ADHD also accumulated more “don't remember” reports over time, indicating declining meta-awareness. Consistent with this, Franklin et al. (2017) found neurotypical individuals with low ADHD symptoms strategically allocate attentional resources—an ability absent in those with high ADHD symptoms. This parallels our ADHD group's greater MW frequency and lower intentionality, suggesting executive function difficulties in regulating attentional shifts.

Across all participants, mental states varied with subjective sleepiness. On-task reports were more frequent when alert, while MW and MB increased with sleepiness. Although nearly half of “sleepy” probes were associated with MW (48.4%), this reflected typical attentional disengagement under moderate fatigue. In contrast, extremely sleepy ratings indicated disproportionate increases in off-task states, suggesting increasing sleepiness further compromises sustained attention.

Task-based sleepiness ratings revealed participants with ADHD reported higher sleepiness and fewer alert ratings than neurotypical participants, contrasting with nonsignificant ESS group differences. This suggests subjective sleepiness fluctuates more dynamically in ADHD, consistent with Helfer et al. (2020) who observed elevated daytime sleepiness during a SART in ADHD, though trait-level sleepiness did not differ between groups.

SWs in wakefulness: a neurophysiological mechanism of attentional difficulties in ADHD?

EEG analyses revealed distinct SW activity patterns in ADHD, characterized by higher amplitude and density in parietotemporal electrodes (Fig. 5). Emerging research shows sleep-like SWs can intrude into specific brain regions during wakefulness (Vyazovskiy et al., 2011; Andrillon et al., 2021; Pinggal et al., 2022), and our findings support the “sleepier brains” hypothesis in ADHD (Andrillon et al., 2019), where increased wake SWs may neurophysiologically explain attentional challenges.

Global SW density increases correlated with performance decrements: higher omission errors, slower RT, and greater RT variability (Fig. 6). Effects were particularly evident in central electrodes where increased SWs appeared to impair response preparation and execution, corroborating research showing wakeful SWs disrupt cognitive processing (Andrillon et al., 2021; Pinggal et al., 2022). However, we found no correlation between SWs and commission errors. Notably, across our combined ADHD and neurotypical sample, we also observed no SW–MW association, contrasting with prior findings in neurotypical adults (Andrillon et al., 2021), suggesting this relationship may not generalize to heterogeneous samples. Yet, elevated SW activity in ADHD provides evidence it may underlie some characteristic attentional difficulties. Furthermore, global SW density increases were associated with elevated subjective sleepiness, supporting that SWs reflect accumulating sleep pressure, affecting both objective performance and subjective experience. Lastly, elevated local sleep correlated with fewer “on-task” reports, highlighting that increased SW density impacts objective performance and task engagement.

Electrode-wide mediation analysis explored whether SW density mediated relationships between ADHD diagnosis and SART performance. The strongest mediation effects were observed for omission errors, RT, and RT variability, with significant electrodes primarily clustered in frontal and centroparietal regions (Fig. 6c). These findings suggest ADHD affects behavior through specific SW alterations. While mediation effects for “on-task” reports were weaker, consistency across objective measures suggests SW activity may serve as a neurophysiological bridge between ADHD and its behavioral manifestations.

This highlights sleep-like SWs as promising intervention targets. Common ADHD medications, such as methylphenidate and atomoxetine, act on dopaminergic and noradrenergic arousal systems (Bymaster, 2002; Chamberlain et al., 2009; Koda et al., 2010; Kowalczyk et al., 2019), improving performance and reducing sleep-like SWs (Pinggal et al., 2022). By identifying local sleep as a mediator, these findings advance understanding of ADHD's neural mechanisms and highlight a potential biomarker for guiding interventions. Beyond pharmacological approaches, behavioral interventions like Cognitive Behavioral Therapy for Insomnia may reduce wakeful SWs. Improving sleep quality and regulation could, in turn, support daytime arousal and attentional stability.

Disentangling trait and state arousal impairments in ADHD

The observed hypoarousal in ADHD may stem from chronic reduced arousal, indicated by greater SW activity and higher self-reported sleepiness. Alternatively, it could arise from use-dependent mechanisms, with individuals with ADHD experiencing faster fatigue onset during sustained effort, supported by steeper omission error increases across blocks. These accounts are not mutually exclusive, and future research should disentangle trait-like versus task-induced hypoarousal in ADHD.

Supplementary neural signatures: ERP and spectral perspectives

Contrasting with previous studies (Szuromi et al., 2011), we did not observe P300 reduction. However, individuals with ADHD showed enhanced late positive potential (Jaswal et al., 2019), potentially reflecting enhanced postperceptual or error-related cognitive processes, though these differences did not reach significance after correction.

Similarly, spectral analyses at Fz revealed no significant group differences after cluster-based correction, though uncorrected statistics showed increased delta/theta power, consistent with our SW findings. Focusing on this band (3–8.7 Hz), a frontotemporal cluster showed significant increases in the ADHD group (Fig. 4c), with a topography different from SW density increases (Fig. 5d). Importantly, theta activity (4–8 Hz) has been associated with sustained attention, cognitive control (Cavanagh and Frank, 2014; Tan et al., 2024), and sleepiness (Snipes et al., 2022). Increased theta power in ADHD may reflect compensatory engagement of attentional control networks. However, the widespread topography does not fit typical frontal “cognitive control” theta distribution; thus, we hypothesize it may instead reflect the same phenomenon as sleep-like SWs. The absence of significant group differences in overall power may reflect differential SW modulation in ADHD: while high-amplitude SWs increase, low-amplitude SWs decrease (Fig. 5a). Consequently, conventional power analysis, which aggregates high- and low-amplitude waves, may obscure these opposing effects, yielding only modest overall power increases. This suggests sleep-like SWs may serve as more sensitive and specific neural markers for distinguishing ADHD from neurotypical individuals.

Limitations and future directions

Despite a 72 h washout period, limited research exists on lingering stimulant effects on cognitive performance. Additionally, our predominantly female cohort contrasts with typical male-favored ADHD research. By focusing on this underrepresented group, however, our study addresses critical gaps regarding ADHD manifestations in females, who often present with more inattentive symptoms and better masking strategies (Hinshaw et al., 2022; Faraone et al., 2024). Future studies should aim for gender-balanced cohorts to determine generalizability and potential gender differences in local sleep manifestations. Moreover, participant sleep behavior was not measured before laboratory visits. Thus, we cannot exclude influences of habitual or recent sleep patterns from our findings.

Our ADHD sample showed substantial comorbidity (50% depression, 62.5% anxiety), consistent with clinical populations where 70–80% of individuals with ADHD have at least one comorbidity (Faraone et al., 2024). While this limits specificity to ADHD, it strengthens ecological validity and mirrors typical presentation. Larger samples could examine whether specific comorbidity profiles exhibit distinct SW patterns.

As a cross-sectional study, we cannot determine whether local sleep represents a stable trait marker or transient task-induced state. While SW density showed modest age-related decline (β = −0.81/year), group differences persisted after controlling for age, with no evidence of diminishing ADHD–neurotypical differences. However, the narrow age range (M = 23 ± 3.5 years) limits examining broader maturational patterns. Consequently, whether elevated SW activity in ADHD reflects delayed cortical maturation (Shaw et al., 2007) or persistent neurophysiological differences remains unclear. Longitudinal research spanning wider developmental periods and incorporating diverse cognitive paradigms could clarify stability and specificity of these neural patterns. Since sleep and vigilance are controlled by time awake and circadian position (Borbély, 2022), future studies should also control for task timing and circadian profiles. Finally, given the heterogeneity within ADHD presentations, future research can investigate whether wake SWs differ between presentations.

Conclusion

This study investigates neurophysiological underpinnings of sustained attention difficulties in ADHD, with novel focus on local sleep-like activity during wakefulness. Participants with ADHD demonstrated greater attentional lapses, evidenced by task performance and MW reports, and exhibited higher SW density, particularly in parietotemporal electrodes. Mediation analyses revealed SWs predict and mediate relationships between ADHD and behavioral variability and sleepiness. Increased wake SWs may represent a key neural mechanism underlying attentional fluctuations in ADHD, linking neurophysiological differences to behavior across clinical ADHD and neurotypical populations.