Neurophysiological measures and correlates of cognitive load in attention‐deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD) and dyslexia: A scoping review and research recommendations

Working memory is integral to a range of critical cognitive functions such as reasoning and decision‐making. Although alterations in working memory have been observed in neurodivergent populations, there has been no review mapping how cognitive load is measured in common neurodevelopmental conditions such as attention‐deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD) and dyslexia. This scoping review explores the neurophysiological measures used to study cognitive load in these specific populations. Our findings highlight that electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) are the most frequently used methods, with a limited number of studies employing functional near‐infrared spectroscopy (fNIRs), magnetoencephalography (MEG) or eye‐tracking. Notably, eye‐related measures are less commonly used, despite their prominence in cognitive load research among neurotypical individuals. The review also highlights potential correlates of cognitive load, such as neural oscillations in the theta and alpha ranges for EEG studies, blood oxygenation level‐dependent (BOLD) responses in lateral and medial frontal brain regions for fMRI and fNIRS studies and eye‐related measures such as pupil dilation and blink rate. Finally, critical issues for future studies are discussed, including the technical challenges associated with multimodal approaches, the possible impact of atypical features on cognitive load measures and balancing data richness with participant well‐being. These insights contribute to a more nuanced understanding of cognitive load measurement in neurodivergent populations and point to important methodological considerations for future neuroscientific research in this area.

Working memory is a cognitive system that allows for the manipulation of temporarily stored information (Baddeley & Hitch, 1974;Cowan, 1999;Miyake & Shah, 1997).Because of its role in reasoning, decisionmaking and many cognitive functions central to literacy and numeracy, working memory is considered crucial for learning and academic attainment (Alloway & Alloway, 2010;Cowan, 2014;Diamond, 2013).An individual's working memory is defined by two key aspects: its capacity and its load (Barrouillet et al., 2007;Chai et al., 2018).Working memory capacity pertains to the individual variability in the amount of information that can be stored in working memory (Engle et al., 1999;Miller, 1956).It linearly develops across childhood, reaching a stable level during adolescence (Bathelt et al., 2018;Gathercole, 1998;Gathercole et al., 2004).Working memory capacity is commonly measured using a range of cognitive tasks such as the dual-task paradigm, which combines a memory span measure-such as a counting span, operation span or reading span taskwith a concurrent processing task such as solving simple mathematical operations (Conway et al., 2005;Wilhelm et al., 2013).Cognitive load, also known as working memory load or mental workload, refers to the amount of working memory resources used when engaged in a cognitive task (Sweller et al., 1998).Cognitive load has historically been measured with subjective self-report instruments (Paas et al., 2016).However, recent technological advances have allowed for the development of objective measures of cognitive load (Korbach et al., 2017;Martin, 2015).
Such methods of measurement include eye-tracking, electroencephalography (EEG), magnetoencephalography (MEG), functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS) (Antonenko et al., 2010;Fishburn et al., 2014;Hu et al., 2019;Ranchet et al., 2017;Whelan, 2007;Zarjam et al., 2015).For instance, pupil diameter increases with cognitive load and has been found to experimentally differentiate between high versus low cognitive load with an accuracy of up to 75% (Hogervorst et al., 2014).EEG has been used to quantify cognitive load through eventrelated potentials (ERPs), and an inverse relationship has been found between the amplitude of ERPs and the cognitive load experienced during the completion of a working memory task (Ortiz et al., 2020;Tamanna & Parvez, 2021).EEG frequency bands can also classify high versus low cognitive load, with a reported accuracy of about 84.5% (Sarailoo et al., 2022).In addition to providing electrophysiological information, MEG yields a better spatial resolution of source localisation compared with EEG (Baillet, 2017).For example, one MEG study found that prolonged periods of demanding cognitive activity were associated with an increase in the betafrequency band power in the right inferior and middle frontal gyri (Tanaka et al., 2014).Research using fMRI and fNIRS to measure cognitive load relies on blood oxygenation level-dependent (BOLD) contrast that results from metabolic changes during cognitive tasks (Miri Ashtiani & Daliri, 2023;Whelan, 2007).For instance, interregional correlations of BOLD fluctuations in the working memory and default mode networks have been found to be modulated by changes in cognitive load (Newton et al., 2011).These neurophysiological methods are sometimes combined to obtain a more complete picture such as using fMRI and pupillometry (Fietz et al., 2022), EEG and eye-related measures (Hogervorst et al., 2014;Scharinger et al., 2020) or fNIRS and EEG (Aghajani et al., 2017).
Working memory has been linked to the ability to focus attention on task-relevant information and to ignore distractions (Zanto & Gazzaley, 2009).There is strong evidence showing that working memory impairments are associated with low learning outcomes and represent a risk factor for poor academic performance (Alloway et al., 2009;Gathercole & Pickering, 2000).Individuals with working memory impairments find it more difficult to maintain task goals in working memory (Unsworth et al., 2004).Working memory also affects the speed with which individuals process information and switch attention between tasks or sub-elements of tasks while learning (Unsworth et al., 2004).Such working memory impairments are particularly relevant for neurodivergent students, as they are common in several neurodevelopmental conditions.Attention-deficit/ hyperactivity disorder (ADHD), which affects between 6.8% and 10.2% of the population, is most clearly associated with deficits in working memory (Hong et al., 2022;Kofler et al., 2020;Roodenrys, 2012;Song et al., 2021;Xu et al., 2018).Working memory is also often impaired in individuals with autism spectrum disorders (ASD), which has a global prevalence of 0.6%, though researchers have argued that the design of the working memory test itself can strongly affect the results for autistic participants (Nakahachi et al., 2006;Salari et al., 2022;Wang et al., 2017).Working memory impairments in dyslexiawhere most estimates of prevalence fall in the range of 3%-7% but have been found to be as high as 17.4% (Wagner et al., 2020)-are well documented, and the deficits are most evident regarding phonological measures of working memory, where dyslexia is consistently associated with lower scores (Jeffries & Everatt, 2004;Smith-Spark & Fisk, 2007;Wang et al., 2022).Multiple metaanalyses also support a strong association between working memory impairments and neurodiversity (Alderson et al., 2013;Habib et al., 2019;Peng & Fuchs, 2016).In fact, some studies suggest that working memory capacity can be a stronger predictor of academic progress than IQ in students with learning difficulties (Alloway, 2009;Alloway & Alloway, 2010).However, although working memory capacity is dependent on factors outside of an instructor's control such as the students' prior knowledge and individual differences in cognitive ability (Brady et al., 2016;Conway et al., 2003), cognitive load can be modulated through instructional design (Paas et al., 2003).As such, researchers argue that cognitive load in neurotypical populations varies based on the task, the environment and the subject characteristics such as cognitive abilities (Kirschner, 2002).For neurodivergent populations, which are often characterised by working memory deficits, the interplay between task demands, environmental factors and individual factors may have important implications to delineate the taskinvariant and task-specific aspects of cognitive load in these groups.
Several reviews have been carried out to investigate the neurophysiological measures of cognitive load in neurotypical people (Brünken et al., 2003;Paas et al., 2016), and one review enquired into neurophysiological measures of cognitive load in younger adults, patients with mild cognitive impairment (MCI) and patients with Alzheimer's disease (AD) (Ranchet et al., 2017).However, despite the strong evidence pointing at a relationship between working memory and neurodiversity, there has been no review to date investigating which neurophysiological measures have been typically used to investigate cognitive load in some of the most common neurodevelopmental conditions that impact learning and academic achievement.In this review, we aim to map neurophysiological measures of cognitive load used in ADHD, ASD and dyslexia research.We describe the scope of research in the field and discuss research recommendations.

| METHODS
A scoping review method was selected, as it is considered a rigorous approach to knowledge synthesis when reviewing topics that are not conducive to systematic reviews, such as when addressing questions beyond those related to the effectiveness of an intervention (Munn et al., 2018;Peters et al., 2015).In particular, scoping reviews are recommended to investigate the conduct of research on a certain topic so as to inform the design of future research studies (Lockwood et al., 2019), which is the case in the present review.This review adheres to the reporting guidelines outlined in the PRISMA extension for scoping reviews (Tricco et al., 2018).The following five steps were followed: identifying the research question; identifying relevant studies; selecting eligible studies; charting the data; and collating, summarising and reporting the results (Arksey & O'Malley, 2005).The protocol was registered on the Open Science Framework (Le Cunff et al., 2022).

| Identification of relevant studies
A comprehensive electronic search for relevant literature was performed in the following databases: Web of Science, Academic Search Complete, Scopus, Medical Literature Analysis and Retrieval System Online (MEDLINE) and PubMed Central Psychological Information Database (PsycINFO).The search strategy focused on neurophysiological measures and correlates of cognitive load, in conjunction with terms relevant to the neurodevelopmental conditions considered in this review-ADHD, ASD and dyslexia.In order to balance sensitivity and precision for this scoping review, we opted to use specific terms from the DSM-V.These include the search terms 'ADHD', 'attention deficit hyperactivity disorder', 'ASD' and 'autism spectrum disorder', while excluding alternative terms such as 'hyperkinetic disorder' and 'Asperger syndrome'.Despite not having a separate entry in the DSM-V, dyslexia is recognised as a commonly used term for a specific learning disorder with impairment in reading.As with other specific learning disorders, the DSM-V indicates that this specific pattern of reading difficulties should not be attributable to uncorrected visual or hearing impairments, psychological adversity, intellectual disabilities, other mental or neurological conditions, inadequate proficiency in the language used for academic instruction or inadequate educational instruction (American Psychiatric Association, 2013).To maintain a focus on literature that specifically addresses this condition, as opposed to reading issues caused by other factors, we did not include terms such as 'reading difficulties' in our search.We will address the potential limitations of this approach in the discussion section.The search strategy was piloted to check the appropriateness of keywords in the Web of Science, which is one of the largest and most widely used multidisciplinary databases of research publications and citations (Birkle et al., 2020).The search strategy is provided in Table 1.No date range limits were applied.Only peer-reviewed literature was included in the searches.
Finally, backward and forward citation tracking was performed to identify supplementary studies relevant to neurophysiological measures of cognitive load in ADHD, ASD and dyslexia.All eligible articles were uploaded to the reference manager Mendeley (Elsevier, 2019), and duplicates were automatically identified and removed.

| Screening and eligibility
First, title and abstract screening were performed using Screenatron, a dedicated screening tool (Clark et al., 2020), and guided by the following inclusion criteria: (1) any empirical study written in English using a neurophysiological measure of cognitive load namely fMRI, fNIRS, EEG, MEG and/or eye-tracking on their own or in combination with another technique; (2) participants should have declared or been tested for ADHD, ASD and/or dyslexia; (3) no restriction regarding country, age, gender or ethnicity.We excluded: (1) review studies, abstract-only papers as preceding papers and opinion papers; (2) studies focusing on neurodevelopmental conditions that are out-of-scope for this review, such as dysgraphia, schizophrenia and Tourette Syndrome; (3) studies that investigate cognitive load in ADHD, ASD and/or dyslexia using only cognitive tasks or questionnaires.

| Data charting
The following data were extracted from the full-text version of studies included in the review: information about the study (authors, year of publication, country of origin), aims/purpose, population (age, gender, conditions under study), methodology (sample size, neurophysiological measures and correlates of cognitive load) and key findings that relate to the scoping review.For ease of reading, the data were tabulated, and an ID number was assigned to each study.

| Data analysis
A narrative report was produced to summarise the extracted data and identify the most common neurophysiological measures of cognitive load in ADHD, ASD and dyslexia.Research trends were discussed, and implications for neurodiversity research focusing on cognitive load are suggested.

| RESULTS
The searches of the five databases retrieved 2301 records.An additional 52 studies were identified via citation searches, giving a total of 2353 records before deduplication and 1363 after deduplication.A total of 1249 records were excluded based on titles and abstract screening.The full text of the remaining 114 records was obtained and screened using a PICO-based taxonomy to classify each reason to exclude a record from the review (Edinger & Cohen, 2013).The most common reason for exclusion was a wrong outcome measure (n = 37), as many studies administered a working memory task but did not investigate any neurophysiological correlates of cognitive load.A total of 67 studies were included in the review (Suppl.A1).For a further description of the screening process, see the PRISMA study flow diagram (Figure 1).
The breakdown of neurodevelopmental conditions was as follows: 61.2% of studies focused on ADHD (n = 41), 20.9% on ASD (n = 14) and 17.9% on dyslexia (n = 12).The majority of studies (n = 40) used a diagnostic classification system such as the Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association) or the International Classification of Diseases (World Health Organization).Despite our search strategy focusing on the DSM-V criteria, most studies used the DSM-IV criteria (n = 28) (American Psychiatric Association, 1994), followed by the DSM-V criteria (n = 10) (American Psychiatric Association, 2013).Many studies (n = 27) did not specify a diagnostic classification system.Whether they used a diagnostic classification system or not, half of the studies (n = 35) relied on a prior diagnosis, for instance by recruiting from a larger genetics study, from a clinic, from an existing university-based clinical research registry or by requiring a documented history of the condition under study.In many cases (n = 20), the prior diagnosis was confirmed by the researchers with one or more additional diagnostic tools such as a rating scale or a structured interview.Studies predominantly employed only one diagnostic tool (n = 42), but a substantial proportion combined two or more diagnostic tools (n = 24).A detailed breakdown of neurodevelopmental conditions, diagnostic criteria and diagnostic tools used for each study included in this review can be found in the supplementary materials (Suppl.A2).
In terms of neurophysiological measures, more than half of the studies used EEG (n = 35), followed by fMRI (n = 22).Other studies used fNIRS (n = 6), eye-tracking (n = 4) or MEG (n = 2).In some cases, EEG was combined with eye-tracking (Zhang et al., 2017) or with fMRI (Lenartowicz et al., 2016).There was a high number of studies using EEG and/or fMRI across all conditions: 90.2% of ADHD studies (n = 37), 64.3% of ASD studies (n = 9) and 83.3% of dyslexia studies (n = 10) used EEG and/or fMRI to measure correlates of cognitive load (Suppl.A2). Figure 2 provides a visual overview of the studies included in this review, stratified by neurophysiological measures and neurodevelopmental conditions under study.This review was not suitable for a meta-analysis because of the heterogeneity of the included studies.As such, the variability in statistically significant and nonsignificant results reported below does not necessarily indicate conflicting findings, as studies with similar effect sizes can differ in significance based on sample size alone (Amrhein, Greenland, & McShane, 2019;Amrhein, Trafimow, & Greenland, 2019).Rather than providing a quantitative synthesis, this review reports on the variety of research practices when studying cognitive load in ADHD, ASD and dyslexia.
Most EEG studies analysed neural oscillations as a measure of cognitive load (n = 17), though the specific brain waves associated with cognitive load differed across studies.The majority of studies looking at neural oscillations considered both the theta range (4-8 Hz) and alpha range (8-12 Hz) powers (n = 8), followed by the alpha range only (n = 4) and theta range only (n = 2).In neurotypical participants, theta spectral power has been found to increase with increased cognitive load, and a suppression of alpha waves has been linked to increased levels of cognitive load (Antonenko et al., 2010;Mazher et al., 2017).Considering both the theta and alpha range powers was an approach applied across ADHD (n = 5), ASD (n = 2) and dyslexia (n = 2).
In ADHD, results were inconsistent.One study found no significant differences between ADHD and non-ADHD groups when manipulating load, neither for theta power nor for alpha power (Gomarus et al., 2009a).The increase in both occipital-alpha and frontal-theta power relative to baseline was found to be more pronounced in ADHD relative to controls regardless of load during working memory maintenance in another study (Lenartowicz et al., 2014).A study that aimed to replicate these results in an independent sample also reported occipital-alpha power during maintenance to be significantly greater in ADHD but failed to replicate the previously reported effects in frontal-theta power (Lenartowicz et al., 2019).The ADHD group in another study exhibited an increased alpha power and reduced theta power compared with the control group (Jang et al., 2020).Three studies, all conducted with ADHD participants, considered alpha band power only as a correlate of cognitive load.One study found lower posterior alpha power in ADHD participants compared with controls during a working memory task, but only at a trend level (Liu et al., 2016).The other two studies reported that neurofeedback training-in which participants learn to respond to a display of their own brain waves (Kamiya, 1962)-could help increase alpha band power in ADHD with an associated positive impact on several measures such as working memory and learning (Escolano et al., 2014;Wang, 2017).Two studies, both in ADHD, also analysed the beta range (12-30 Hz) power and found a weaker power across the beta frequency range compared with controls (Fabio et al., 2018;Zammit & Muscat, 2019).
Only two studies of cognitive load in ASD considered both the theta and alpha range powers, but they both found impaired theta and alpha modulations in this population.The first one concluded that, controls displayed a modulation of theta and alpha power according to cognitive load, whereas ASD participants did not demonstrate such modulation (Larrain-Valenzuela et al., 2017).The second study examined theta and alpha power in ASD (Chrisilla et al., 2020).It is worth noting that according to the guidelines outlined by Munn et al. (2018) for choosing between systematic and scoping reviews, this particular study would be excluded from a systematic review because of its small sample and lack of detailed statistical reporting, which would limit its usefulness to provide concrete guidance for policy or clinical practice.In contrast to systematic reviews, scoping reviews do not require an assessment of the quality of evidence (Arksey & O'Malley, 2005;Gottlieb et al., 2021).Therefore, we included this study in our scoping review to offer a broad overview of the research landscape and, in this case, the variety of studies considering the theta and alpha range powers.However, as several limitations prohibit drawing definitive conclusions from the data, we have opted not to mention any comparative results from this study and to only focus on the experimental design and relevant measures.
Studies focused on dyslexia found that theta activity was shifted to the right hemisphere in dyslexic participants during working memory tasks, whereas controls seem to rely on a more selective left hemispheric processing (Klimesch et al., 2001;Spironelli et al., 2006).
The findings paint a complex picture, as measures of spectral power revealed inconsistent results across conditions.In neurotypical individuals, increased cognitive load was associated with heightened theta spectral power and suppressed alpha waves, but in ADHD, the results were inconsistent, with some studies noting a similar trend but others presenting contrary findings.In contrast, the limited studies on ASD suggest a lack of the expected modulations in theta and alpha ranges that are typically associated with changes in cognitive load, whereas dyslexic individuals appeared to exhibit a unique pattern with a right hemisphere shift in theta activity during working memory tasks.Collectively, these inconsistent results highlight the need to account for the heterogeneity within and across these conditions when using neural oscillations as a measure of cognitive load.
Rather than looking at brain oscillations, 16 studies focused on ERPs (averaged brain waves locked to an event) component analysis (n = 16).By far, the most studied ERP was the P3 wave (n = 12).P3 amplitude has been observed to decrease with increasing cognitive load in neurotypical participants (Gevins & Cutillo, 1993).The P3 wave was used in studies investigating ADHD (n = 8) and dyslexia (n = 4), with no studies investigating cognitive load in ASD using the P3 wave as a correlate.In ADHD, most of these studies (n = 7) found a smaller P3 amplitude in ADHD compared with the control group (Breitling-Ziegler et al., 2020;Keage et al., 2008;Kim et al., 2014;Kim & Kim, 2016;Lintas et al., 2019;Ortega et al., 2020;Stroux et al., 2016).Although not significant, one EEG study found high effect sizes of cognitive load on the P3 wave in ADHD participants for both relevant and irrelevant stimuli, suggesting a potential effect of cognitive load on selective attention (Gomarus et al., 2009b).In dyslexia, most studies (n = 3) also found a smaller P3 amplitude in dyslexic participants versus the non-dyslexic comparison group during a working memory task (Horowitz-Kraus, 2014;Lotfi et al., 2022;Shiran & Breznitz, 2011).The one remaining P3-based study used a within-subjects design-with only dyslexic participants and no controls-and found a larger P3 amplitude after cognitive training (Lotfi et al., 2020).
The second most studied ERP component was the contralateral delay activity (CDA) (n = 3).The CDA is a negative slow wave whose amplitude can be used to track fluctuations in working memory performance in neurotypical participants (Adam et al., 2018).All three EEG studies using CDA to investigate cognitive load focused on ADHD, with inconsistent results.Spronk et al. (2013) found a smaller CDA amplitude only for the low-load condition in their ADHD sample, whereas Luo, Guo, et al. (2019) found a smaller CDA amplitude for ADHD participants in working memory tasks in both high-and low-load conditions.Gu et al. (2018) found that the CDA of participants with ADHD did not distinguish between high and low cognitive load conditions, and they dedicated a similar amount of neural resources to the high and low cognitive load conditions, which could explain working memory impairments in ADHD.
The analysis of ERP components, particularly the P3 and CDA waves, provides unique insights into the cognitive load responses of neurodivergent individuals.Individuals with ADHD and dyslexia exhibit a similar decrease in P3 amplitude as do neurotypical individuals in response to an increase in cognitive load.However, limited results suggest that the influence of cognitive load on the P3 wave may extend to affect selective attention in ADHD, implying a nuanced role for this ERP component beyond being merely a marker of cognitive load.In dyslexia, cognitive training shows promise for augmenting the amplitude of the P3 wave, hinting at a potential intervention pathway.Lastly, the studies on CDA in ADHD also yielded mixed results, with studies indicating altered neural resource allocation under different cognitive load conditions, but no consensus as to the exact relationship between CDA amplitude and cognitive load.A summary of EEG correlates of cognitive load in ADHD, ASD and dyslexia can be found in Table 3.
When analysing EEG data, it is essential to consider whether differences in frequency bands may reflect variations in ERP morphology rather than distinct oscillatory activity.Some of the included studies using time-frequency decompositions of EEG data did not clearly distinguish between evoked activity from induced oscillations, leaving the origins of observed effects unclear (e.g., Missonnier et al., 2013;Stroux et al., 2016).
Additionally, the studies employed a diverse range of experimental paradigms, including n-back (Jang et al., 2020;Kim et al., 2014), Sternberg (Lenartowicz et al., 2014), oddball (Gomarus et al., 2009b), reading (Klimesch et al., 2001) and other working memory tasks.The specificity of the experimental tasks is an important factor when interpreting EEG results, especially in light of the heterogeneous findings across studies.Several studies, for instance, analysed neural oscillations during working memory maintenance periods (Jang et al., 2020;Lenartowicz et al., 2014), whereas others focused on encoding-related activity (Kim et al., 2014;Zammit & Muscat, 2019).The ERP findings were also dependent on task specifics.Some studies showed reduced P3 amplitude in ADHD and dyslexia during working memory paradigms (Keage et al., 2008;Kim & Kim, 2016), whereas another study using an oddball task found enhanced, not reduced, P3 in ADHD (Gomarus et al., 2009b).Some studies investigated P3 amplitude during congruent trials (Ortega et al., 2020), whereas others focused on incongruent conditions (Stroux et al., 2016).These methodological differences likely contributed to the heterogeneous EEG outcomes for both neural oscillations and ERPs as measures of cognitive load in ADHD, ASD and dyslexia.
These EEG studies in ADHD, ASD and dyslexia present a complex yet insightful picture of how cognitive load may vary in neurodivergent populations.Neural oscillations, specifically in the theta and alpha range powers, seem sensitive to cognitive load manipulations, but their modulation patterns are inconsistent across and within conditions.Neurotypical individuals typically show increased theta and suppressed alpha waves with higher cognitive load, but such responses vary significantly in ADHD, show a lack of modulation in ASD and reveal a unique right hemispheric shift in theta activity in dyslexia.ERP components, namely the P3 and CDA waves, also highlight differences in cognitive load responses across these groups, with nuanced implications for selective attention in ADHD and potential intervention pathways in dyslexia.These findings suggest that EEG measurements of cognitive load may require more nuanced, taskspecific and condition-specific interpretations.

| BOLD-based measures of cognitive load
After EEG measures, BOLD-based measures were most commonly used to study cognitive load in ADHD, ASD and dyslexia, with 28 studies (42.6%) using fMRI (n = 22) or fNIRS (n = 6) (see Table 4).
The majority of fMRI studies performed whole-brain analysis to investigate working memory-related (WM-related) areas (n = 21).Only one study used anatomically defined regions of interest in the prefrontal cortex and primary motor cortex (Sheridan et al., 2007).Researchers performed whole-brain analysis in two fNIRS studies, whereas the other fNIRS studies (n = 4) focused on the BOLD response in the prefrontal cortex only.Most studies using BOLD-based measures of cognitive load found increased activation in the frontal lobe, sometimes on its own, but more often in combination with parietal, temporal and/or occipital areas activity (Table 5), and in particular in cortical areas such as the dorsolateral prefrontal cortex, the dorsal cingulate cortex and the posterior parietal cortex.In addition to those cortical areas, some studies (n = 9) also found activation in subcortical areas such as the cerebellum, striatum and hippocampus in their analysis.
In ADHD, several studies found an interaction effect between diagnosis and cognitive load on brain activation, such that both ADHD and control groups showed increased activation of WM-related brain regions such as the dorsolateral prefrontal cortex, ventrolateral prefrontal cortex and striatum under high cognitive load, but the increase in activation was significantly lower in the ADHD group compared with controls, pointing to a brain activation deficit in ADHD during increased-load working memory tasks (Burgess et al., 2010;Ehlis et al., 2008;Ko et al., 2013;Mukherjee et al., 2021;Schecklmann et al., 2013;Stevens et al., 2016).Only one of these studies included an intervention and observed an increase in WM-related ADHD brain activity in several frontal, parietal and temporal lobe regions following working memory training (Stevens et al., 2016).Though both ADHD and non-ADHD groups showed significantly increased activation under higher cognitive load in widespread WM-related brain areas, one study reported a significant diagnosis-by-load interaction effect, with decreased activation in ADHD participants compared with controls under cognitive load specifically in the right inferior parietal cortex (Bayerl et al., 2010).One study (Mattfeld et al., 2016) found that ADHD participants with unimpaired working memory displayed a similar activation pattern as healthy controls, whereas ADHD participants with impaired working memory showed the same pattern of hypoactivation of WMrelated brain regions found in the other studies.
Not all studies discovered this pattern of hypoactivation to cognitive load in ADHD.For instance, one fNIRS study found that although both ADHD and control groups showed increased activation in the amygdala, paracingulate gyrus and dorsolateral prefrontal cortex during a working memory task compared with baseline, there was no significant interaction between these activation patterns and group (Kaiser et al., 2022).Another one found no significant differences in cortical prefrontal activation between ADHD and control groups (Schecklmann et al., 2010).Only one study investigated the effect of sex on WM-related brain activation in ADHD: the results suggested that adults with ADHD show less activity under cognitive load than controls in prefrontal regions, but sex-by-group analyses revealed that male ADHD participants display significantly less activity in WM-related regions relative to male controls, whereas female ADHD participants show no differences from female controls (Valera et al., 2010).
Despite prevailing patterns of hypoactivation in WMrelated brain regions, the findings from these studies point towards a more intricate reality in which individuals with ADHD may not universally exhibit this trait.A subset of individuals with ADHD, specifically those with unimpaired working memory, exhibit activation patterns T A B L E 5 Number of studies according to WM-related brain regions included across neurodevelopmental conditions.

Cortical areas
Frontal (n = 10) 6 2 2 Fronto-temporal Hippocampus (n = 2) 1 1 -similar to those of individuals without ADHD, suggesting heterogeneity within the ADHD population when it comes to cognitive load (Mattfeld et al., 2016).Additionally, the results hint at the role of other factors such as gender in modulating brain activation associated with varying levels of cognitive load (Valera et al., 2010).
Those findings suggest a nuanced interplay between cognitive load, working memory and brain activation patterns in ADHD.
In ASD, similar difficulties in recruiting WM-related brain regions are reported in several studies: controls show increasing recruitment of these regions, whereas ASD participants do not show such modulation of increasing brain activation with increasing load (Luna et al., 2002;Vogan et al., 2014Vogan et al., , 2018)).In addition, ASD participants exhibit more right-lateralised activation in WM-related brain regions (Han et al., 2022;Koshino et al., 2005;Yeung et al., 2019).
Similarly, studies of dyslexia found that compared with controls, dyslexic participants show a significant hypoactivation of WM-related areas with increased cognitive load (Beneventi et al., 2010;Winn et al., 2006).One study also found that dyslexic participants exhibited significantly more activation than controls with increasing cognitive load in the left superior frontal gyrus and the inferior frontal gyrus and less activation in the middle frontal gyrus and the superior parietal cortex (Vasic et al., 2008).
BOLD-based measures have contributed to our increased understanding of the patterns of brain activation observed under varying levels of cognitive load in neurodivergent populations such as those with ADHD, ASD or dyslexia.Studies using fMRI or fNIRS have successfully identified differential activation patterns across the spectrum of conditions, revealing both general and condition-specific insights.In ADHD, for instance, BOLD-based measures have revealed a general trend of reduced activation under increased cognitive load (Burgess et al., 2010;Ehlis et al., 2008;Ko et al., 2013;Mukherjee et al., 2021;Schecklmann et al., 2013;Stevens et al., 2016), yet uncovering a subset of individuals with unimpaired working memory who exhibit similar activation patterns to controls (Mattfeld et al., 2016).This heterogeneity extends to gender-specific effects, with male ADHD participants displaying lower WM-related activity than their female counterparts (Valera et al., 2010).ASD was associated with more right-lateralised activation in WM-related regions, suggesting the use of distinct processing strategies in individuals with this condition (Han et al., 2022;Koshino et al., 2005;Yeung et al., 2019).There was a significant hypoactivation of WM-related areas in dyslexia under increased cognitive load (Beneventi et al., 2010;Winn et al., 2006), but also an overactivation in specific regions such as the left superior and inferior frontal gyri (Vasic et al., 2008).Although these findings attest to the utility of BOLD-based measures in studying cognitive load in ADHD, ASD and dyslexia, the diversity in findings, demonstrated even within the same condition such as ADHD (Mattfeld et al., 2016;Valera et al., 2010), highlights the complexity and heterogeneity of these conditions and calls for caution in the data collection and analysis of BOLD-based measures for the study of cognitive load.See Supporting Information for a breakdown of neurodevelopmental conditions, BOLD-based measures and activated brain regions considered in each study (Suppl.A3).

| Other single and combined neurophysiological measures of cognitive load
Several studies have used a variety of other neurophysiological measures to explore cognitive load in ADHD, ASD and dyslexia.These measures include MEG, eyetracking and combinations of multiple measures.
Two studies used MEG to explore cognitive load in ASD.MEG is a non-invasive technique that captures the subtle magnetic fields produced by neural activity and allows researchers to track brain activation patterns with high temporal resolution (Singh, 2014).The first MEG study focused on theta and alpha frequencies and found differences in interregional theta and alpha brain synchronisation during a working memory task between the ASD group and the comparison group: in the theta frequency band, each group used different networks during maintenance, and in the alpha frequency band, each group recruited distinct networks across encoding, maintenance and recognition, with little overlap (Audrain et al., 2020).The second study found reduced WM-related activation for participants with ASD during a high-load task in the left insula and mid-cingulate gyrus and stronger WM-related activation of the left angular gyrus and of the left precuneus (Urbain et al., 2015).
Eye-tracking was employed in four studies spanning ADHD, ASD and dyslexia.Eye-tracking, also a noninvasive method, allows researchers to monitor eye movements, pupil dilation and blink rate, providing indirect markers of various cognitive processes (Eckstein et al., 2017).Pupil dilation, which increases as the demand on working memory increases, is a wellvalidated measure of cognitive load (Beatty, 1982;Just et al., 2003;Kramer, 1991).From a neuroscientific standpoint, the norepinephrine system is thought to play a crucial role in the relationship between pupil dilation and executive function (Sara, 2009;Wahn et al., 2016).In particular, it has been suggested that the pupil dilation response is controlled by the locus coeruleus-norepinephrine (LC-NE) system and reflects levels of physiological arousal and cognitive effort.The LC-NE system receives signals related to greater task demands from brain regions associated with cognitive control, such as the anterior cingulate cortex (ACC), the frontal and parietal cortices and the superior colliculus (Aston-Jones & Cohen, 2005;Foote & Morrison, 1987;Nieuwenhuis et al., 2011).These brain regions are associated with attentional control, arousal and cognitive load (Van der Wel & van Steenbergen, 2018).However, several confounds can affect the recorded pupil size when the participant attends to a dynamic stimulus, such as watching a video or reading content on a website.First, pupil dilation is strongly affected by luminance, which can mask the pupillary responses related to cognitive load (Cherng et al., 2020).In addition, the gaze position can distort the pupil's apparent size, as rotations of the eyes change the angle at which the camera records the pupil, which is known as the 'pupil foreshortening error' (PFE) (Gagl et al., 2011;Hayes & Petrov, 2016).There have been some attempts at building eye models to correct this error, which can be larger than many cognitive pupillometric effects, but these models have only been tested on limited types of tasks (Brisson et al., 2013;Hayes & Petrov, 2016;Petersch & Dierkes, 2022).For these reasons, eye-tracking manufacturers often recommend only measuring pupil dilation during tasks with a constant luminance and a constant fixation location (Hayes & Petrov, 2016).
When dynamic stimuli are used, metrics such as blink rate help avoid the confounds of changes in gaze position (Zagermann et al., 2016).Eye blink rate has emerged as a highly discriminatory parameter for the assessment of cognitive load, with evidence that eye blink rate could be a predictor of WM task performance (Ayres et al., 2021;Ortega et al., 2022;Siegle et al., 2008).Cognitive load and eye blink rate appear to be inversely correlated in tasks involving visual attention: a higher cognitive load is associated with a lower blink rate (Ledger, 2013).In turn, a lower blink rate has been hypothesised to help minimise the chance of losing critical information when focusing on a task involving visual attention (Hoppe et al., 2018;Nakano et al., 2009;Ranti et al., 2020).The neurophysiological basis of eye blinks in relation to cognitive load is complex and not fully understood.Research suggests that blink rate tends to decrease during tasks requiring high cognitive load, possibly due to the brain's resource allocation mechanisms (Siegle et al., 2008).The basal ganglia, a group of nuclei involved in motor control and cognitive functions, are implicated in the control of blinking (Basso et al., 1993).Although spontaneous blink rate has been proposed as an indirect measure of dopaminergic activity, which is crucial for cognitive function (Karson, 1983), recent studies have questioned the reliability of this association, at least in healthy human adults (Dang et al., 2017).Therefore, although cognitive processes influence eye blinks, the exact underlying neurophysiological mechanisms-including the role of dopamine-remain an area of active research.
All four studies used a remote eye-tracking system placed under the computer monitor (screen-based), and no study used a wearable system (head-mounted).Pupil dilation was the most common eye-related measure of cognitive load across the four eye-tracking studies (Table 6).A study found that dyslexic readers demonstrated greater pupil dilation when reading compared with typical readers, indicating a higher cognitive load (Ozeri-Rotstain et al., 2020).Different pupil dilation patterns may be present in ADHD as well, with a larger pupil diameter for participants with ADHD than controls early in the working memory task (0 to 5 s), then larger for controls than for those with ADHD for the next 20 s and no pronounced differences anymore between groups after about 25 s (Mies et al., 2019).
The eye-tracking studies included in this review employed a range of experimental paradigms.As such, the oculomotor results should be interpreted with caution, as they may reflect differences in cognitive processes associated with varied tasks rather than consistent effects of cognitive load per se.As an example, the study reporting greater pupil dilation in dyslexic children used a reading task (Ozeri-Rotstain et al., 2020), whereas the study reporting different pupil dilation patterns with increasing task difficulty in ADHD used a working memory task (Mies et al., 2019).Task specificity is a crucial consideration for interpreting pupil dilation results, especially in longer tasks spanning multiple seconds (Sabatino DiCriscio et al., 2018;van der Wel & Van Steenbergen, 2018).For instance, as the pupil response is sluggish and pupil size is affected by trial timing, dilating and then constricting sluggishly following trial onsets, T A B L E 6 Eye-related measures of cognitive load across neurodevelopmental conditions.

ID Condition
Eye-related measures of cognitive load sustained fixation is optimal for measuring pupil changes during task performance (Burlingham et al., 2022).Overall, the eye-tracking studies in this review do not provide conclusive evidence as to which oculomotor measures are most appropriate to measure cognitive load, as the heterogeneous paradigms and tasks may be linked to different underlying cognitive mechanisms.
Finally, a few studies have employed a combination of neurophysiological measures to explore cognitive load in ADHD and ASD.This multimodal approach can provide a more comprehensive understanding of cognitive load by integrating various physiological markers (Chen et al., 2017).Two studies of cognitive load in ASD led by the same researcher combined eye-tracking (blink rate, pupil dilation, fixation and saccade duration) with EEG (theta, alpha, beta and gamma frequency bands) and other physiological measures: electrocardiogram (ECG), electromyography (EMG), respiration, skin temperature, photoplethysmogram (PPG) and galvanic skin response (GSR) (Zhang et al., 2015;Zhang et al., 2017).In the first study, only eye gaze data were analysed to measure cognitive load, and the results indicated a higher blink rate and larger pupil diameter under increased cognitive load (Zhang et al., 2015).The second study used machine learning methods to fuse the eye-tracking, EEG and other physiological data to measure cognitive load and found that one of their machine learning models based on the fused multimodal data could predict self-reported cognitive load measures with 84.43% accuracy in participants with ASD (Zhang et al., 2017).Using a concurrent EEG-fMRI protocol, the research team found that alpha event-related desynchronisation (ERD) during a working memory task was associated with both occipital activation and fronto-parieto-occipital functional connectivity in both ADHD and non-ADHD participants but that alpha ERD was associated less strongly with occipital activity and more strongly with fronto-parieto-occipital connectivity in ADHD compared with controls (Lenartowicz et al., 2016).As evidenced by a machine learning model's successful prediction of self-reported cognitive load in one study (Zhang et al., 2017), these multimodal approaches have the potential to increase the accuracy of cognitive load measurement.However, although these methods can offer valuable insights into the neural underpinnings of cognitive load in neurodivergent populations, their appropriateness may depend on the specific research question and condition under study.

| DISCUSSION
This scoping review set out to map the neurophysiological measures and correlates of cognitive load most commonly used in ADHD, ASD and dyslexia research.Scoping reviews provide a rigorous approach to synthesising knowledge when addressing questions beyond just intervention effectiveness, which allows for a broader scope and inclusion of diverse study designs compared with systematic reviews (Munn et al., 2018).However, a key distinction is that scoping reviews do not assess the quality of evidence of included studies (Arksey & O'Malley, 2005;Gottlieb et al., 2021).The following discussion of the results should be interpreted considering the strengths and limitations of the scoping review methodology.Overall, studies differed in what neurophysiological correlates were considered relevant to cognitive load in ADHD, ASD and dyslexia, but a few correlates emerged as potential candidates.
For EEG studies, neural oscillations in the theta and alpha ranges were often combined to create a neurophysiological index of cognitive load.This is in line with wider working memory research, where alpha power is thought to decrease and theta power to increase with higher cognitive load (Jensen & Tesche, 2002;Keil et al., 2006).The amplitude of the P3 wave, which is generally recognised as a valid correlate of cognitive load in typically developing participants (Antonenko et al., 2010;Kok, 2001), was also found to be a widely used indicator of cognitive load.For fMRI and fNIRS studies, lateral and medial frontal brain regions seemed to reliably show an increase in BOLD response under increased cognitive load, which is consistent with a large body of research suggesting that working memory relies strongly-though not exclusively-on frontal brain areas, and in particular on the dorsolateral prefrontal cortex (Courtney et al., 1998;Curtis & D'Esposito, 2003;Kane & Engle, 2002).Lastly, pupil dilation and blink rate were the most commonly used eye-related measures of cognitive load in this review.Overall, EEG and fMRI emerged as the most favoured techniques across all three conditions and even more clearly in ADHD where 90% of studies used one or both of those two techniques (Suppl.A2).
It is worth noting that few studies in this review combined several measures of cognitive load.A possible explanation for this might be that it is currently challenging to reconcile the diverse range of analysis parameters and brain-functional assumptions used for different measures, as well as characterising multi-scale interactions in neurophysiological activity (Cohen & Gulbinaite, 2014).Moreover, the statistical analyses of high-dimensional data present further methodological difficulties such as choosing the best approach to combine numerous variables (Weuve et al., 2015).For example, one study in this review used machine learning models to fuse the eyetracking, EEG and other physiological data of cognitive load that had been collected (Zhang et al., 2017).
Although machine learning has been applied in EEG research for many years (Hosseini et al., 2020), recent advances in multivariate analyses and artificial intelligence algorithms provide opportunities to uncover patterns in increasingly large, multimodal datasets (Vortmann et al., 2022).However, significant challenges remain in reconciling the diverse analytic parameters and brain-functional assumptions underlying different neurophysiological measures, as well as in characterising their multi-scale interactions (Cohen & Gulbinaite, 2014).The appropriateness of such computationally intensive approaches depends on the specific research question and condition being studied.Overall, although combining several neurophysiological measures can offer valuable insights into cognitive load, multimodal approaches require careful consideration of the added complexity.
Despite its widespread use as a neurophysiological measure of cognitive load with neurotypical participants (Mahanama et al., 2022;Zagermann et al., 2016), this review revealed that eye-tracking is conspicuously underutilised in studies of cognitive load in ADHD, ASD and dyslexia.There may be a few explanations for this discrepancy.First, there are specific difficulties in using eyetracking in people with ADHD, ASD and dyslexia because of atypical features in oculomotor processes, ophthalmological status and visual attention (Cs akv ari & Gyori, 2015).Those atypical features may lead to higher rates of missing data.For example, one study in this review that included eye-related measures of cognitive load reported that 31 samples had to be removed because of unwanted movements of participants with ASD during the experiments, which represents 14% of the 216 samples they had collected at 120 Hz (Zhang et al., 2015).However, although missing data can complicate the analysis, this can be mitigated by using a high sampling rate and interpolating the pupil dilation signal (Andersson et al., 2010;Mack et al., 2017).Another potential explanation is that medication, in particular psychostimulant treatments for ADHD, can reduce the extent of pupillary fluctuation and has shown a tendency towards decreasing pupil diameter (Nagyov et al., 2007), which may complicate its use as a valid measure of cognitive load.Finally, individual calibration is indispensable for eyetracking, but the process can be tedious for participants who struggle to maintain concentration and can even be unsuccessful in young participants with ADHD (Blignaut, 2017).Finally, difficulties in the inhibition and control of eye movements are also commonly accepted as a feature of ADHD and dyslexia (Hatch & Faao, 2020;Maron et al., 2021;Pavlidis, 1985), which may represent a confound in eye-related measures of cognitive load in those populations, although a recent review reports inconclusive evidence regarding the presence of various oculomotor deficits in ADHD (Sherigar et al., 2022).Considering its widespread use in cognitive load research, we suggest that it is worth including eye-related measures in future studies of cognitive load in ADHD, ASD and dyslexia, bearing in mind these caveats.
Although the aim of this scoping review was to map the measures used to investigate the neurophysiological correlates of cognitive load in these three conditions, we will briefly discuss how some of the results support or contradict prevailing theories of ADHD, ASD and dyslexia.For instance, the working memory model of ADHD suggests that working memory deficits are a fundamental feature of the disorder, upstream of traditionally recognised core symptoms such as inattention, hyperactivity and impulsivity (Rapport et al., 2001).Consistent with this theoretical framework, results from several EEG studies support the notion that ADHD is characterised by deficits in working memory processing, evidenced by altered ERP amplitudes and EEG oscillatory activity during cognitive load tasks (Missonnier et al., 2013;Szuromi et al., 2011).BOLD-based studies also corroborated the link between working memory and ADHD, as ADHD participants consistently exhibited altered activation patterns in WM-related brain regions during high cognitive load tasks (Bayerl et al., 2010;Burgess et al., 2010;Ehlis et al., 2008;Ko et al., 2013;Mattfeld et al., 2016;Mukherjee et al., 2021;Roman-Urrestarazu et al., 2016;Schecklmann et al., 2013;Stevens et al., 2016).These findings provide converging evidence for the working memory model of ADHD, indicating that individuals with ADHD have altered neural activity in brain regions associated with working memory, attention and cognitive control during tasks with a high cognitive load.However, our review also includes findings that do not align with the working memory model of ADHD.For instance, one study found no interaction between patterns of WM-related brain activation and cognitive load (Kaiser et al., 2022).Another study found no significant differences in cortical prefrontal activation between ADHD and control groups during cognitive load tasks (Schecklmann et al., 2010).Finally, Valera et al. (2010) found that female ADHD participants showed no differences from female controls in WM-related brain regions during cognitive load tasks, whereas male ADHD participants displayed significantly less activity.Those results suggest that many factors may play a role in the neural underpinnings of cognitive load differences in ADHD.
Several studies found that, as cognitive load increased, individuals with ASD showed reduced recruitment and altered modulation of brain regions involved in working memory compared to neurotypical controls (Luna et al., 2002;Vogan et al., 2014Vogan et al., , 2018)).Other studies also found evidence of atypical lateralisation in ASD (Han et al., 2022;Koshino et al., 2005;Yeung et al., 2019).These neuroimaging findings align with behavioural research findings indicating that individuals with ASD perform poorly on working memory tasks, especially as cognitive load increases, and that ASD is associated with differences in working memory (Barendse et al., 2013;Habib et al., 2019;Steele et al., 2007;Wadhera & Kakkar, 2020).The impaired modulation of WM-related brain activity with increasing cognitive load provides neurophysiological evidence that working memory deficits may contribute to ASD symptomatology.However, the results were mixed regarding EEG oscillatory activity, with one study finding impaired theta and alpha modulation in ASD (Larrain-Valenzuela et al., 2017) and another reporting increased theta and alpha power compared with controls, albeit with a small sample size (Chrisilla et al., 2020).This suggests that further research is needed to clarify the relationship between EEG frequency bands and cognitive load in ASD.Overall, the neurophysiological correlates revealed working memory alterations that align with behavioural research, though the heterogeneity of findings highlights the need for continued study into the nuanced neural underpinnings of cognitive load in ASD.
The findings related to dyslexia align with some existing theories regarding the neural bases of this condition.Studies using EEG found that compared with controls, individuals with dyslexia exhibited reduced P3 amplitude during working memory tasks, suggesting alterations in underlying cognitive processes (Horowitz-Kraus, 2014;Lotfi et al., 2022;Shiran & Breznitz, 2011).EEG also revealed a hemispheric processing difference, with a right-lateralised shift in theta oscillations during working memory tasks in dyslexia (Klimesch et al., 2001;Spironelli et al., 2006).These results support the magnocellular theory of dyslexia, which proposes right hemispheric dominance as a cause of reading impairment (Stein, 2001).fMRI studies further corroborated working memory differences, showing hypoactivation of working memory regions in dyslexia as cognitive load increased (Beneventi et al., 2010;Winn et al., 2006).Although not a direct examination of phonological processing, the reduced recruitment of WM-related regions points to potential downstream effects on phonological processing, supporting the phonological deficit theory, which posits that reading difficulties in dyslexia stem from impairments in phonological processing (Vellutino et al., 2004).However, dyslexic individuals also exhibited overactivation compared with controls in certain regions, such as the left superior frontal gyrus, indicating mixed results (Vasic et al., 2008).Overall, the findings provide some neurophysiological support for the magnocellular and phonological theories of dyslexia.However, the variability in results highlights the need for further research to clarify the neurophysiological correlates of cognitive load in dyslexia across different paradigms, such as in a recent study using object substitution masking (Koffman et al., 2023).
Differences in experimental paradigms and task designs, particularly for the EEG and eye-tracking studies, could partially explain the heterogeneity of the results.The EEG studies employed a wide range of working memory and attention tasks, ranging from n-back to oddball and reading tasks, each evoking distinct cognitive processes.Similarly, the four eye-tracking studies included in this review used a variety of paradigms, such as working memory tasks, reading tasks and driving simulations.This diversity of experimental designs makes it challenging to identify consistent effects of cognitive load on neural oscillations, ERPs or pupil dilation.Even fMRI and fNIRS studies that predominantly used n-back tasks exhibited some variability in the type of stimuli and the levels of memory loads.Such methodological differences may lead to different brain activation patterns, contributing to the heterogeneous patterns of results across BOLD-based studies.Future research should aim for more standardised protocols to determine which effects are reliable indicators of cognitive load irrespective of the specifics of the tasks.This will require coordinated efforts to match not only the overarching paradigm (e.g., n-back) but also finer details such as stimulus characteristics, trial structure and levels of working memory loads.Such systematisation could help distinguish cognitive load measures from task-dependent effects.
It might also be challenging to separate evoked responses from oscillations when analysing EEG data using time-frequency analysis.This challenge was demonstrated by Rousselet et al. (2007), showing that early evoked brain activity decomposed using time-frequency transformations does not easily separate evoked responses from oscillatory activity.This makes it challenging to determine if observed effects in the timefrequency domain originated from evoked activity (phase-locked to the stimulus) or induced activity (related to the stimulus).As a result, differences across frequency bands discussed in this review can simply reflect differences in ERP shapes rather than distinct oscillatory activity.When using EEG to examine cognitive load, careful experimental design and analytical procedures are required to properly distinguish between the two.
Another important consideration when interpreting the variability in findings is the heterogeneity within each neurodevelopmental condition under study.Both genetic and phenotypic heterogeneity have been documented within ADHD, ASD and dyslexia populations (Lenroot & Yeung, 2013;Luo, Weibman, et al., 2019;Masi et al., 2017;McArthur et al., 2013;Zoubrinetzky et al., 2014).This heterogeneity likely underlies some of the variability in findings, as samples may differ in clinically relevant factors such as age of onset, symptom severity, cognitive profile and comorbidities.For instance, some EEG studies in ADHD yielded seemingly contradictory results regarding theta and alpha modulations as a function of cognitive load (Gomarus et al., 2009a;Jang et al., 2020;Lenartowicz et al., 2014).Such inconsistencies may stem from varying clinical presentations within ADHD samples.Similarly, the review included BOLD-based studies showing either hypoactivation or hyperactivation of WM-related brain areas in dyslexia compared with controls as cognitive load increased (Beneventi et al., 2010;Vasic et al., 2008;Winn et al., 2006).This suggests that treating neurodevelopmental conditions as homogeneous in cognitive load studies may fail to capture important within-group variability.Future research may benefit from directly assessing and accounting for heterogeneity by using stratified analysis approaches to examine cognitive load effects separately in sub-groups differentiated by factors such as age of onset, symptom profiles and comorbidities.In addition to stratified analysis, future research could employ approaches that directly examine relationships between neurophysiological measures of cognitive load and scores on validated symptom scales, such as correlational and regression analysis using measures such as the Adult ADHD Self-Report Scale (ASRS; Kessler et al., 2005), Autism Spectrum Quotient (AQ; Baron-Cohen et al., 2001) and Adult Reading History Questionnaire (ARHQ; Lefly & Pennington, 2000).
Additionally, neurophysiological measures of cognitive load in ADHD, ASD and dyslexia require special consideration to avoid spurious group differences.For instance, as noted in one of the fMRI studies, the correlation of motion with between-subject variability in ADHD can artificially increase local connectivity and decrease long-range connectivity, which has implications with respect to functional connectivity analyses (Lenartowicz et al., 2016).Similar care in dealing with subject motion should be given when using other neurophysiological measures of cognitive load.EEG, fNIRS, MEG and eyetracking are all prone to artefacts because of unwanted movement, though those motion effects can be attenuated with a motion correction approach and validation of group analyses of motion (Brigadoi et al., 2014;Medvedovsky et al., 2007;Mihajlovi c et al., 2014;Power et al., 2012).The experimental setup can also be designed to reduce unwanted movement, for example by using a fixed chin rest to hold a participant's head stable during eye-tracking experiments (Carter & Luke, 2020).Another option is to use eye-tracking glasses whose estimates of eye position and pupil diameter are not confounded by head motion (Cognolato et al., 2018).
In addition to these considerations, the comfort of participants with ADHD, ASD and/or dyslexia should be taken into account (Fletcher-Watson et al., 2021).In one fNIRS study, 34 out of 105 ADHD patients and three out of 55 controls did not complete the tasks because of discomfort and because the procedure was experienced as too strenuous (Schecklmann et al., 2013).Each measure of cognitive load comes with specific constraints that can impact the comfort of participants.For instance, atypical sensory processing is an important feature of ASD (Marco et al., 2011).As a result, the combination of sensitivity to sensory stimuli and MRI-related characteristics such as noise and narrowness can make fMRI so distressing for autistic people that sedation is often used in clinical settings to facilitate examination (Ahmed et al., 2014;Kamat et al., 2018;Stogiannos, Harvey-Lloyd, et al., 2022).Although improving the comfort of neurodivergent participants is an ongoing topic of discussion, using simple instructions to avoid sensory overload and adjusting radiographic practice during the scans were suggested as some of many potential ways to mitigate these challenges (Stogiannos, Carlier, et al., 2022).In addition, new fMRI techniques are progressively allowing for silent imaging (Damestani et al., 2021), with the potential to make the experience less stressful for neurodivergent participants.
When using EEG, researchers must carefully consider the trade-off between the depth of information required and the comfort of participants.The majority of EEG studies in this review used 40 or fewer channels.More channels do provide richer data but necessitate more time for electrode placement, which could be a concern for populations with specific conditions.For certain research questions and populations, fewer channels may be sufficient and more practical by making the experiments less labour-intensive and costly for researchers while also potentially increasing the comfort level of participants (Maher et al., 2023;Montoya-Martínez et al., 2021;Tacke et al., 2022).However, the choice of electrode number should be based on the specific research question at hand, and researchers should be cautious in interpreting the results accordingly.
This scoping review highlights the general lack of consideration for comorbidities between the examined neurodevelopmental disorders in the existing corpus of literature.For instance, it is estimated that 28%-44% of adults diagnosed with ASD also meet the criteria for ADHD, and some estimates are as high as 50%-70% (Hours et al., 2022;Polderman et al., 2014).Comorbidity rates as high as 25%-40% have been reported between conditions such as ADHD and dyslexia (Sexton et al., 2012;Willcutt & Pennington, 2000).Such high comorbidity rates make disentangling the unique neural correlates of each condition challenging (Astle et al., 2022;Craig et al., 2016).This overlap in prevalence and symptomatology has led to theories positing shared attentional working memory deficits in ADHD and dyslexia (Lonergan et al., 2019;Willcutt et al., 2005).This review cannot directly speak to these theories, as the studies largely examined each condition in isolation.However, the apparent similarities found in certain neurophysiological measures, such as reduced P3 amplitude during working memory tasks in both ADHD (Breitling-Ziegler et al., 2020;Keage et al., 2008;Kim et al., 2014;Kim & Kim, 2016;Lintas et al., 2019;Ortega et al., 2020;Stroux et al., 2016) and dyslexia (Horowitz-Kraus, 2014;Lotfi et al., 2022;Shiran & Breznitz, 2011), lend preliminary support to the idea of some shared neural mechanisms.Future research should actively recruit samples with co-occurring conditions and use study designs that can unpack both shared and unique neural correlates of cognitive load across these populations.This more inclusive approach may yield insights that are not readily apparent when conditions are studied in isolation.
Cognitive load is also dependent on factors such as participants' prior knowledge and baseline cognitive abilities (Conway et al., 2003;Cook, 2006).In addition, factors such as age of onset, symptom profile and medication status may affect working memory and its associated neural processes (Kofler et al., 2020;Sjöwall et al., 2013).Some studies in this review attempted to create more homogeneous samples by establishing diagnoses with multiple tools (e.g., Lenartowicz et al., 2014), confirming medication status (e.g., Bayerl et al., 2010) or grouping participants based on working memory capacity (e.g., Mattfeld et al., 2016).Future research would benefit from capturing relevant confounds such as prior knowledge, cognitive abilities and clinical symptoms and analysing results in sub-groups stratified by clinical and cognitive characteristics.
This scoping review has several methodological implications for future research.Given the heterogeneity of ADHD, ASD and dyslexia, future research should employ a cautious approach to task design.For instance, Lieder et al. (2019) suggest that autistic and dyslexic individuals have different perceptual biases: autistic participants showed slower updating of their internal model in response to new sensory evidence, whereas dyslexic participants rapidly discounted previous evidence and were biassed towards recent stimuli.In addition, there is evidence that ADHD is associated with heightened sensory-seeking behaviours, whereas ASD is associated with sensory sensitivity and avoidance (Little et al., 2018).Researchers should consider accounting for such differences in perception and sensory processing when designing tasks that aim to modulate cognitive load.Tasks could be adapted to include stimuli that are more relevant to the specific cognitive challenges faced by these populations.Relevant parameters include task duration, the type of stimuli used and the specific cognitive domains being targeted.Additionally, researchers should consider the use of multimodal neurophysiological measures to provide a more comprehensive understanding of cognitive load in ADHD, ASD and dyslexia, for instance, combining EEG and fMRI (e.g., Lenartowicz et al., 2016) or EEG and eye-tracking (e.g., Zhang et al., 2017).Integrating these measures could be particularly beneficial in neurodivergent populations, where a single modality may not fully capture the complexity and variability of cognitive load.Advanced computational methods can help integrate and interpret the resulting datasets, providing more nuanced insights into the variability and complexity of cognitive load in neurodivergent populations (Oppelt et al., 2022;Vulpe-Grigorasi, 2023).Lastly, ethical considerations should be at the forefront, ensuring that study designs are inclusive and sensitive to the needs of participants with ADHD, ASD and/or dyslexia.The review highlights the need for a careful trade-off between the required depth of information and the comfort of participants.To ensure the well-being of neurodivergent participants, researchers should consider adjusting the experimental setup to take into account cognitive and sensory differences in these conditions and, in particular, make efforts to minimise stress and sensory overload (Stogiannos, Carlier, et al., 2022;Stogiannos, Harvey-Lloyd, et al., 2022;Strömberg et al., 2022).
The results of this review are subject to certain limitations.Although our search strategy was designed to capture the most pertinent literature on neurophysiological measures and correlates of cognitive load in ADHD, ASD and dyslexia, it is important to acknowledge the potential limitations associated with the specificity of our keyword selection.By relying solely on DSM-V terminology, we may have excluded studies that use alternative terms.For instance, our decision to exclude terms such as 'hyperkinetic disorder', 'Asperger syndrome' and 'reading difficulties' may have resulted in the omission of relevant studies in conditions that symptomatically and neurologically overlap with ADHD, ASD and dyslexia, respectively.Consequently, the body of literature we reviewed may not represent the entirety of the literature exploring neurophysiological measures and correlates of cognitive load in these three conditions.Future systematic reviews may expand the purview of this scoping review by considering a wider array of terms to thoroughly investigate the neurophysiological underpinnings of cognitive load across these and related conditions.
As our aim was to map the state of research practices to inform the design of future studies, and not to provide a complete synthesis of current evidence, the scoping review method was considered most appropriate to address our research question.In addition, following the PRISMA guidelines for scoping reviews still ensured a high level of rigour in the production of this review (Tricco et al., 2018).Nevertheless, a limitation that is inherent to such reviews is the narrative nature of the synthesis, which leaves room for selective and subjective interpretation of the results.Of particular note, studies with similar effect sizes can produce results on opposite sides of the significance threshold because of differences in sample size alone, as Amrhein, Trafimow and Greenland (2019) and Amrhein, Greenland and McShane (2019) demonstrated.As the included studies were too heterogeneous for comparison, this scoping review was not amenable to a meta-analysis, and the reporting of statistical significance alone, without accompanying effect sizes and sample size considerations, is insufficient to determine if the findings across ADHD, ASD and dyslexia agree or disagree.Therefore, we could not determine if studies with conflicting statistical significance truly had disparate outcomes.In addition, although all studies were retained to map the extent of research, interpreting and generalising their results warrant caution because of the statistical issues around significance testing and effect size estimation with small samples (Button et al., 2013;Gelman, 2018;Gelman & Loken, 2016).Rather than providing a quantitative synthesis, this review reports on the variety of measures and correlates of cognitive load in these three conditions.Future studies could focus on a more homogeneous subset of studies to be able to perform a meta-analysis.
In addition, by its very nature, a scoping review is inclusive of studies with methodological limitations that may have been excluded from a systematic review, in order to provide a comprehensive overview of research practices (Arksey & O'Malley, 2005;Gottlieb et al., 2021;Munn et al., 2018).Because of broader inclusion criteria compared with a systematic review and the lack of assessment of the strength of evidence, one caveat in interpreting the results is the variable quality of the studies included in this review.For instance, we included a study on EEG correlates of cognitive load in ASD that did not report key statistical details and had a very small sample size (Chrisilla et al., 2020).Although scoping reviews allow the inclusion of such studies to map the extent of research, caution is warranted when interpreting and generalising the results.Despite these limitations, the insights gained from this review may be of assistance to researchers in deciding which neurophysiological measure(s) would be most appropriate for studies of cognitive load in ADHD, ASD and/or dyslexia.Its findings contribute to our understanding of the factors to consider when investigating cognitive load in these populations, including atypical features that may impact neurophysiological measures and practical aspects such as the comfort of participants.

| CONCLUSION
This scoping review mapped the landscape of neurophysiological measures used to study cognitive load in ADHD, ASD and dyslexia.The review has identified EEG and fMRI as the two most used neurophysiological measures, with fewer studies employing fNIRS, MEG or eyetracking.The underutilisation of eye-related measures is particularly surprising given their prominence for cognitive load research in neurotypical populations.The diversity in measures and correlates of cognitive load across these neurodevelopmental conditions poses challenges for synthesising findings and drawing definitive conclusions and brings to light several considerations for future research.For instance, the technical constraints of combining multiple neurophysiological measures such as EEG and fMRI necessitate careful task design and advanced analytical approaches, possibly leveraging machine learning techniques for data integration.Moreover, the review highlights the importance of considering the unique characteristics of neurodivergent participants and the potential impact of atypical features on cognitive load measures when designing studies.Lastly, ethical considerations are critical, particularly ensuring the comfort and well-being of participants with ADHD, ASD and/or dyslexia.Future studies have the potential to substantially advance our understanding of cognitive load in ADHD, ASD and dyslexia through careful task design, advanced analytical methods and thorough ethical considerations.

F
I G U R E 1 Scoping review PRISMA flow diagram process.T A B L E 2 Age range of participants in the included studies.
Search strategy for Web of Science.
Number of studies using specific EEG correlates of cognitive load across neurodevelopmental conditions.
T A B L E 3