Resting-state EEG correlates of sustained attention in healthy ageing: Cross-sectional findings from the LEISURE study

While structural and biochemical brain changes are well-documented in ageing, functional neuronal network differences, as indicated by electrophysiological markers, are less clear. Moreover, age-related changes in sustained attention and their associated electrophysiological correlates are still poorly understood. To address this, we analysed cross-sectional baseline electroencephalography (EEG) and cognitive data from the Lifestyle Intervention Study for Dementia Risk Reduction (LEISURE). Participants were 96 healthy older adults, aged 50 – 84. We examined resting-state EEG periodic (individual alpha frequency [IAF], aperiodic-adjusted individual alpha power [aIAP]) and aperiodic (exponent and offset) activity, and their associations with age and sustained attention. Results showed associations between older age and slower IAF, but not aIAP or global aperiodic exponent and offset. Additionally, hierarchical linear regression revealed that after controlling for demographic variables, faster IAF was associated with better Sustained Attention to Response Task performance, and mediation analysis confirmed IAF as a mediator between age and sustained attention performance. These findings indicate that IAF may be an important marker of ageing, and a slower IAF may signal diminished cognitive processing capacity for sustained attention in older adults.


Introduction
Older age is commonly associated with alterations in neuronal brain signals, potentially leading to a decline in cognitive abilities (Gaál et al., 2010;Meunier et al., 2014;Rossini et al., 2007).A common feature of age-related cognitive decline is heightened vulnerability to distractions due to irrelevant information, leading to difficulties in sustaining attention (Amer and Hasher, 2014;Kramer et al., 1999;Lustig et al., 2007;Weeks and Hasher, 2014).Notably, a recent meta-analysis by Vallesi et al. (2021) highlights variability in sustained attention abilities during ageing, noting that older adults sometimes outperform younger adults in some aspects, such as making fewer inattention errors, albeit with typically slower response times.Furthermore, Staub et al. (2014) observed that, unlike younger adults, older adults did not exhibit declines in sustained attention performance over time.Collectively, these results underscore the diverse influences of ageing on sustained attention performance.
A possible factor influencing sustained attention capabilities in older adults could be the extensively documented neurophysiological alterations associated with the ageing process.Among these changes, neural oscillations are particularly noteworthy.Characterised by the synchronised firing of pyramidal neurons, which facilitate the generation of rhythmic fluctuations in the brain across multiple frequencies, neural oscillations represent a key feature studied in scalp EEG recordings (Biasiucci et al., 2019;Buzsáki and Draguhn, 2004).Specifically, alterations in the alpha frequency band (~8-12 Hz), most prominent over occipital and parietal brain regions during eyes closed waking rest (Mierau et al., 2017), have been associated with cognitive processes (Foxe and Snyder, 2011;Klimesch, 1999;Palva and Palva, 2007).During periods of rest, alpha power has been suggested to act as a filter mechanism, capable of inhibiting distractors (Bonnefond and Jensen, 2012;Händel et al., 2011;MacLean et al., 2012), a mechanism that aligns with the earlier suggestions by Klimesch et al. (2007) who posited that alpha power reflects top-down attention control, which is key for attention regulation.The importance of these dynamics is further underscored by findings that suggest a connection between alpha power and sustained attention mechanisms (Clayton et al., 2015).Higher alpha power has been linked with better performance when a task requires sustained attention (Braboszcz and Delorme, 2011;Dockree et al., 2007;Sadaghiani and Kleinschmidt, 2016).Specifically, a study by Dockree et al. (2007) demonstrated that higher background (i.e., tonic) alpha power in the parieto-occipital region was associated with better performance during a modified Sustained Attention to Response Task (SART).
In addition to alpha power, individual alpha peak frequency (IAF), defined as an individual's discrete frequency with the highest power value in the alpha oscillation range (Klimesch, 1999), may also play a key role in sustained attention.IAF variability has been observed during tasks that increase cognitive load and during memory consolidation (Haegens et al., 2014;Klimesch et al., 1993) and has been associated with a variety of cognitive and motor functions, such as processing speed (Klimesch et al., 1996), working memory (Angelakis et al., 2004;Richard Clark et al., 2004) motor control (Hülsdünker et al., 2016) and preparedness for focusing attention on a stimulus (Min and Herrmann, 2007).Notably, a study by Jann et al., (2010) demonstrated faster resting-state IAF was associated with increased regional cerebral blood flow in a network of brain areas crucial for attention (specifically the inferior frontal gyrus and insular cortex).However, the relationship between IAF and sustained attention performance, as measured with a sustained attention task, and how ageing affects this dynamic, remains unknown.
While the link between IAF and age is consistent in the literature, such that IAF increases from childhood to adolescence (Cragg et al., 2011;Klimesch, 1999;Tröndle et al., 2022), followed by a notable decline from adulthood to older age (Cesnaite et al., 2023;Finley et al., 2022;Ishii et al., 2018;Knyazeva et al., 2018;Mizukami and Katada, 2018;Smith et al., 2023), the relationship between alpha power and age remains less clear.Previous studies have found alpha power to decline with age (Babiloni et al., 2006;Lodder and van Putten, 2011;Polich, 1997;Rossini et al., 2007;Scally et al., 2018;Vysata et al., 2012).However, recent studies have questioned the validity of these findings due to the confounding nature of using rigidly pre-defined frequency bands that do not account for centre frequency shifts that occur with age and have suggested basing alpha power estimations on a participant's IAF for a more accurate and individualised measure (Cesnaite et al., 2023;Donoghue et al., 2022Donoghue et al., , 2020;;Tröndle et al., 2023).Additionally, it has been shown that alpha power measures can be confounded by the amplitude mixing between periodic (oscillatory) and aperiodic (non-oscillatory) activity.Aperiodic activity, commonly referred to as background noise, is marked by a 1/f-like distribution, in which signal power decreases exponentially as a function of frequency.This activity is defined by two key parameters: the offset, which indicates a uniform shift in power across different frequencies, and the exponent, corresponding to the negative slope of the power spectrum in a log-log plot.Notably, several recent studies have shown that when controlling for aperiodic activity, no age-related alterations are seen in alpha power, particularly in older adults (Cesnaite et al., 2023;Merkin et al., 2022).
Studies have demonstrated that older adults tend to display a flatter aperiodic exponent (Dave et al., 2018;Merkin et al., 2022;Tran et al., 2020;Tröndle et al., 2023;Voytek et al., 2015) and reduced offset (Merkin et al., 2022;Tröndle et al., 2023) relative to younger adults.Additionally, Thuwal et al. (2021) also found a flatter exponent with increasing age across the lifespan (18-88 years).However, this association within older adults has yielded conflicting findings.A study by Cesnaite et al. (2023) found a flatter aperiodic exponent with increasing age (60-79 years) and Finley et al. (2022) found both a flatter exponent and reduced offset with increasing age (36-84 years).Conversely, a study by Merkin et al. (2022) found no association between age and the exponent and offset (older adult group, 50-86 years), and Smith et al., (2023) found no association with the exponent (50-80 years).Furthermore, studies have further suggested that age-related changes in aperiodic activity are a global phenomenon when measured with scalp electrodes (Cellier et al., 2021;Hill et al., 2022).Such age-related changes are thought to reflect shifts in the balance between synaptic excitation and inhibition (Gao et al., 2017), as well as a rise in asynchronous firing among neuronal populations (Voytek et al., 2015;Voytek and Knight, 2015).In addition, recent studies have identified links between aperiodic activity and cognition, specifically cognitive speed (Ouyang et al., 2020), lexical prediction (Dave et al., 2018) working memory (Donoghue et al., 2020;Thuwal et al., 2021;Voytek et al., 2015) and interference (Kałamała et al., 2024).Several recent studies have investigated the association between resting-state aperiodic activity and cognition within the adult lifespan and within older adults, yielding conflicting findings.While recent studies have reported the aperiodic exponent to be related to cognitive scores within older adults (Smith et al., 2023) and across the adult lifespan (Finley et al., 2024), others have reported no association (Cesnaite et al., 2023;Tröndle et al., 2023).
Interestingly, a recent study by Finley et al. (2024) examined the relationship between IAF and the aperiodic exponent on cognition over the span of 10 years in older adults.The study found that a higher aperiodic exponent combined with a lower IAF predicted greater cognitive decline, potentially suggesting a suboptimal balance between arousal and the ability to gate external stimuli and highlighting the importance of considering the interaction between these measures.Notably, no previous research has explored the relationships between IAF, aperiodic activity, and sustained attention in the context of ageing.
Given these findings, in the present study, we aimed to investigate relationships between periodic and aperiodic resting-state EEG parameters, sustained attention and healthy ageing in older adults.We utilised eyes-closed resting-state EEG data, given its proven test-retest reliability and known correlations with cognitive performance (Grandy et al., 2013b).Specifically, our first aim was to corroborate recent findings regarding the age-related associations between resting-state periodic and aperiodic activities in older adults (Cesnaite et al., 2023).To this end, we evaluated the bivariate associations between age and both resting-state aperiodic-adjusted individual alpha power (aIAP) and IAF in the parieto-occipital region, as well as global aperiodic activity (encompassing both exponent and offset).We hypothesised a negative correlation between IAF and age, and between aperiodic activity and age in our older adult sample.However, due to our methodological adjustments for IAF and periodic/aperiodic component mixing, we hypothesised no association between age and aIAP.
Our second aim was to investigate the neurophysiological mechanisms underlying sustained attention performance in healthy ageing.We examined how key periodic and aperiodic EEG markers associate with performance on two sustained attention tasks, specifically: the Rapid Visual Information Processing (RVP) test (Sahakian and Owen, 1992) and the Sustained Attention to Response Task (SART; Robertson et al., 1997).We explored whether the EEG measures mediate the relationship between performance and age.Additionally, we investigated potential interactions between age and EEG measures, as well as between IAF and the aperiodic exponent, on task performance.We hypothesised that after controlling for age, gender, and education, aIAP, IAF, and aperiodic activity would significantly associate with sustained attention performance.Furthermore, we predicted that these EEG measures would serve as mediators in the relationship between age and sustained attention performance.

Participants
Of the 99 EEG datasets available from the baseline timepoint of the Lifestyle Intervention Study for Dementia Risk Reduction (LEISURE), one participant was removed from further analysis due to missing cognitive data, and a further two participants were removed during spectral analysis (as described in more detail in 2.5.2Spectral analysis), leaving a final sample size of 96 healthy participants (76 F, M age = 65.3 ± 8.3 years, M education = 14.6 ± 2.8 years) to be included in this crosssectional study.The LEISURE study is currently ongoing, with data collection occurring at the Thompson Institute, University of the Sunshine Coast (UniSC).The complete protocol of the LEISURE study is published elsewhere (Treacy et al., 2023).In brief, participants were English-speaking, generally healthy community-dwelling older adults (50-85 years), without a diagnosis of any psychiatric disorder, major neurological condition, mild cognitive impairment, no history of prior head injury (loss of consciousness>60 min), stroke or epilepsy, or current usage of psychotropic medications.The study protocol has been prospectively registered with the Australian New Zealand Clinical Trials Registry (ACTRN12620000054910) and has since been approved by the Human Research Ethics Committee of the UniSC, Australia (A191301).All participants provided written, informed consent prior to commencing the study.

Sustained attention to response task
Participants completed the Sustained Attention to Response Task (SART) using E-Prime 2.0.10 software (Psychology Software Tools, Pittsburgh, PA, USA) on a desktop computer.The task involved the presentation of black digits (luminance: 45 cd/m 2 ), ranging from 1 to 9, against a grey background (luminance: 60 cd/m 2 ).These digits were displayed in varying sizes of Arial font (100, 120, 140, 160, 180 points) at the screen's centre.Each digit appeared for 200 ms and was succeeded by a mask-a yellow fixation cross-with a randomly determined duration between 1000 and 2000 ms.Participants were instructed to press a button in response to frequent non-target digits (1,2,4,5,6,7,8,9) and to refrain from responding to the infrequent target digit (3), the no-go target.
The initial phase of the task included a practice block of 45 trials (5 containing no-go signals) with performance feedback to ensure participants understood the task requirements.Subsequently, the experimental phase consisted of 540 trials, divided into 60 blocks of 9 trials each.Within each block, digits were selected randomly without replacement, ensuring that the no-go target digit appeared in 11.1 % of the total trials.
Performance on the SART was evaluated using the discrimination index (d'), calculated as the standardised difference between the hit rate (h) for signal-present trials and the false alarm rate (f) for signal-absent trials, thus d' = Z(h) -Z(f).A higher d' value signifies a greater ability to distinguish targets from non-targets, indicating superior signal detection.A log-linear correction was applied across the dataset to accommodate extreme values in the data (e.g., h = 1 or f = 0) to resolve issues stemming from extreme values in signal detection theory metrics.

Rapid visual information processing task
The Rapid Visual Information Processing (RVP) task, part of the CANTAB suite (Cambridge Cognition, 2019), was conducted on an iPad.In this task, digits 2 through 9 appeared in a pseudorandom order at a rate of 100 digits per minute for seven minutes.Each digit was shown one at a time in white text inside a central box against a black backdrop, displayed for 600 ms with no inter-stimulus interval, so a new digit appeared every 600 ms.Participants were required to monitor the digits for any of three specified target sequences ("3-5-7", "2-4-6", "4-6-8"), responding by pressing the touchscreen upon identifying a sequence.These target sequences were consistently interspersed, occurring four times every 30 s RVP task performance was assessed using the sensitivity index (A'), a nonparametric indicator calculated from the probabilities of correctly identifying targets (hits) and incorrectly recognising non-targets (false alarms).The formula for A' is defined as: A' values range from 0 to 1, with a score of 1 denoting perfect target detection.This measure evaluates the participant's proficiency in distinguishing target sequences within the continuous flow of digits, without being influenced by their response tendency.

Electrophysiological recordings
Four minutes of resting-state EEG (rsEEG) acquired from an eyesclosed condition were collected for each participant using a 32-channel Active-Two BioSemi system (BioSemi Active-Two, V.O.F., Amsterdam, Netherlands) with Ag/AgCl electrodes, digitised at a sampling rate of 1024 Hz.The electrodes were embedded in an elastic cap arranged according to the extended 10-20 coordinate system.For each session, the DC offset level for each electrode was checked (±40 mV) prior to data collection.Horizontal and vertical electrooculograms (EOG) were recorded using two pairs of active electrodes placed near the two outer canthi and above and below the left eye, respectively.A pair of dedicated mastoid electrodes were also placed behind the ears.Participants were informed that EEG would be recorded while they rested with their eyes closed.Instructions for the task were presented on the computer screen, and a research assistant was present to answer any questions from the participants.Participants were instructed to relax, remain still, and avoid movements for the duration of the task, which would conclude when they heard a buzzer.

Experimental procedure
Participants first completed the CANTAB battery (including the RVP).The rsEEG assessment was generally conducted the next day or, if not possible, within the same week, followed by the SART in the same session.

Electroencephalographic analysis 2.5.1. Signal pre-processing
Raw EEG signals were referenced offline to the common average reference and filtered with a 0.5-30 Hz band-pass filter (FIR with Hamming window, zero-phase, non-causal, one-pass, order 6578).EOG artefacts were removed by computing signal-space projection vectors using horizontal and vertical EOG channels acquired during EEG recording.These projections were then applied to the EEG signal to remove ocular artifacts.The aforementioned steps were completed using the MNE package for Python (Gramfort et al., 2013).The time series were divided into equal-sized, consecutive 5-second epochs without overlap.Epochs were cleaned of artefacts in all channels using the Python package Autoreject (AR; Jas et al., 2017Jas et al., , 2016)).Initially, global AR was used to find a single threshold which when exceeded, the epoch was instantly removed.If ≥24 epochs out of the total 48 (i.e., more than half) were removed, the participant was dropped without further processing.No participants were dropped in this stage.Next, local AR was applied for each epoch with consensus level of 13 and maximum of 4 channels to interpolate (default setting recommended by AR).As with global AR, the participant was dropped if the remaining epochs fell below 24 after local AR.No participants were dropped in this stage.The resulting EEG signal was visually inspected such that each participant's global field power plot displayed power magnitude in a similar scale throughout the whole signal, to ensure signals contaminated by body movements (which are several orders of magnitude larger than the EEG signals) were filtered out.

Spectral analysis
Spectral analyses involved calculating the power spectrum density (PSD) of the EEG signals and then examining the aperiodic component of the PSD.Firstly, the pre-processed EEG signals were transformed into the frequency domain by estimating the PSDs using Welch's method (Welch, 1967) for 1-30 Hz frequency range, with a 2.5-second Hamming window (50 % overlap, 97.5-second zero-padding).Zero-padding of 97.5 s was also applied to achieve more interpolation points in the frequency domain, facilitating a finer estimation of individual peak alpha frequency (IAF) resulting in an accuracy of 0.1 Hz, while recognising this does not increase the actual frequency resolution.The average PSD value (µV 2 /Hz) for each participant was then calculated globally (i.e., all 32 channels) and within the parieto-occipital region (i.e., channels P3/4, P7/8, Pz, PO3/4, O1/2, and Oz).
Next, using an open-access toolbox 'specparam' (Donoghue et al., 2020), the regional PSDs were log-transformed and parametrised to isolate the aperiodic component of the spectra (Fig 1).Fitting was performed using the 'fixed' aperiodic mode, as visual inspection of the output in log-log space revealed no distinct 'knee' in the power spectrum.Algorithm settings were set as: peak width limits: 1-12 Hz; max number of peaks: infinite; minimum peak height: 0.225 dB; peak threshold: 2 standard deviations; and aperiodic mode: fixed.Power spectra were parameterized across the frequency range of 1-30 Hz.
To ensure the quality of specparam models for further analysis, we visually inspected all models and flagged certain ones for closer inspection based on their goodness of fit metrics provided by the algorithm (Ostlund et al., 2022).Models were flagged if their explained variance (R 2 ) was ≤ 0.6 or Mean Absolute Error (MAE) was < 0.025 (indicating overfit) or > 0.100 (indicating underfit).One model was flagged due to having an R 2 ≤ 0.6 in the global region.However, after visual inspection, we determined that the aperiodic fit was reasonable, and this participant was retained for subsequent analyses.Conversely, of the remaining 98 participants with EEG data, two were removed due to poor aperiodic fit in both the global and parieto-occipital regions, as determined by visual inspection (i.e., fit includes significant omissions of signal).
Furthermore, three participants were excluded from further alpha activity analysis due to lacking a discernible alpha peak in the parietooccipital region, however, were retained for aperiodic activity analysis.The final sample size was 96 participants for aperiodic activity analysis in the global region (mean R 2 =0.96 ± 0.03, MAE=0.05 ± 0.01) and 93 participants for alpha frequency and power analysis in the parieto-occipital region prior to spectrum flattening (mean R 2 =0.98 ± 0.02, MAE=0.05 ± 0.01).
Finally, the spectrum was flattened by subtracting the aperiodic component from the original spectrum (Fig 1).From the flattened spectrum, a maximum power peak was detected within the range of 7-14 Hz and the IAF extracted.Furthermore, aIAP was calculated by using IAF as a centre point within a bandwidth of 6 Hz (e.g., if IAF = 9.73 Hz, then BW = 6.73-12.73Hz), rather than a pre-defined fixed bandwidth, in order to account for individual differences in IAF that may occur with ageing.For subsequent analyses, absolute aIAP, derived from the bandwidth around the IAF in the flattened spectrum (i.e., aperiodic adjusted), was specifically utilised.
Initially, gender was dummy-coded (0=male and 1=female) and Shapiro-Wilk tests were run to evaluate the normality of all variables.Following this, outlier corrections were made using the z-score standard deviation transformation method.Specifically, values with z-scores greater than 3.29 or less than − 3.29 were modified to one unit above or below the nearest acceptable value.In this sample, an outlier correction was applied to a single data point in the education variable.The data were then summarised using means and standard deviations for descriptive analysis.
To investigate the bivariate associations among resting-state EEG metrics and age, two-tailed Spearman's rho was utilised and p-values were corrected for multiple comparisons using false discovery rate (FDR) at 0.05 (Benjamini and Hochberg, 1995).Spearman's rho was used over Pearson's correlation coefficient due to the skewness observed in the education and aIAP variables in our data (strongly negative and strongly positive, respectively).
Further, the investigation into the links between EEG measures and sustained attention performance associated with healthy ageing was conducted through independent hierarchical linear regressions.These regressions examined the impact of four specific resting-state EEG measures (the exponent and offset globally, as well as aIAP and IAF in the parieto-occipital region), the interaction between the EEG measure and age, and the interaction between IAF and the aperiodic exponent on sustained attention performance.Each regression model treated an EEG measure as the main predictor and each sustained attention performance metric, RVP_A' and SART_d', as the dependent variables.In constructing the hierarchical regression models, age (at the time of EEG recording), gender, and years of education were controlled for in the first step.The EEG measure of interest was introduced in the second step, while the third step included the interaction between age and the EEG measure of interest.Separate regression models were conducted to specifically investigate the interaction between IAF and the exponent on each sustained attention performance metric, RVP_A' and SART_d'.Age, gender, and years of education were controlled for in the first step, with IAF and the exponent introduced in the second step, and the interaction between IAF and the exponent included in the third step.Tests were assigned significance at an alpha level of less than 0.05.
To further explore the relationship between EEG measures and sustained attention performance in healthy ageing, mediation analyses were conducted on significant relationships identified through independent hierarchical linear regressions.The model 4 mediation analysis utilized 5000 bootstrapped samples.Age was the independent variable, an EEG measure was the mediator, and a sustained attention performance measure was the dependent variable.The models controlled for gender and years of education.Mediation was confirmed if the indirect effects were significant (i.e., t-values > 1.96 and confidence intervals did not cross zero).

Descriptive information
Demographic and sample characteristics can be found in Supplementary Material 1.

Bivariate correlations of demographics, sustained attention, and resting-state EEG measures
Spearman's rho correlations were performed primarily to evaluate the associations between age and resting-state EEG metrics.
As shown in Fig 2, higher age was associated with a slower IAF (p<0.05),however, not with aIAP.Initially, both the aperiodic exponent and offset showed a significant association with age, however, this association did not retain significance after correction for multiple comparisons.Age was significantly negatively correlated with both RVP_A' (p<0.01) and SART_d' (p<0.05), and as such, both were incorporated into independent HLR models.Education was positively correlated with RVP_A' (p<0.05) and SART_d' (p<0.05)performance.Gender was not significantly related to any of the variables, as per an eta correlation analysis.A significant positive correlation was noted between SART_d' and RVP_A' (p<0.05),suggesting that improved performance on one sustained attention task is associated with enhanced performance on the other.Education was significantly negatively correlated with IAF (p<0.05), but not with other EEG measures.Significant positive associations were observed between aIAP, and both the aperiodic exponent (p<0.05) and offset (p<0.001).Additionally, the aperiodic exponent and offset were significantly associated (p<0.001).No significant associations were observed between IAF and the aperiodic exponent, offset or aIAP.A significant positive correlation was observed between IAF and SART_d' (p<0.01),however, this relationship did not extend to RVP_A'.Notably, the aperiodic exponent and aIAP did not show a significant association with either sustained attention metric.Initially, a significant positive correlation was observed between aperiodic offset and both RVP_A' and SART_d' however, these associations did not retain significance after correction for multiple comparisons.

Hierarchical linear regression of resting-state EEG measures and sustained attention
After accounting for age, gender, and education (step one), IAF (entered at step 2) was positively associated with SART_d' (p=0.010;see Fig 3) and resulted in a significant increase in explained variance (see Table 1), such that individuals with a faster IAF had better performance on the SART task (R 2 Δ=0.069).This inclusion of IAF in the model at step 2 resulted in a significant change in the model's F-value (FΔ=6.975,p=0.010), thereby enhancing the overall significance of the model.However, neither the interaction between age and IAF, nor the interaction between IAF and the aperiodic exponent, were significantly associated with SART_d' when entered in step 3 in their respective models.
Conversely, IAF, its interaction with age, and its interaction with the aperiodic exponent did not show a significant association with RVP_A'.Similarly, neither aIAP nor its interaction with age was significantly associated with either SART_d' or RVP_A'.Additionally, neither the aperiodic exponent nor its interaction with age or IAF exhibited a significant association with performance on SART_d' or RVP_A'.Likewise, the aperiodic offset, along with its interaction with age, was not significantly associated with either SART_d' or RVP_A'.For detailed results on all independent models and scatter plots of residuals, see Supplementary Material 1.

Mediation analysis
Next, we investigated the mediating role of IAF in the relationship between age and SART_d' performance, controlling for education and gender, prompted by the significant associations identified in the hierarchical linear regression analysis.The analysis revealed significant indirect and total effects, while no significant direct effects were observed (refer to Table 2

for effect model and Fig 4 for the path diagram).
The model controlled for gender and education.Individual alpha peak frequency (IAF); Independent variable (X); mediator variable (M); dependent variable (Y).Total effect (c), the relationship between the independent and dependent variables; Total effect (c), the relationship between X and Y; Direct effect (c'); the relationship between X and Y in the presence of M; Indirect effect (a*b); the pathway from X through M to Y. Confidence intervals, standard errors (SE), and estimates are provided as unstandardised values.SART_d', Sustained attention to response task d-prime.

IAF association with age and attention
Our findings suggest a prominent slowing of IAF with increasing age, confirming the existing literature (Cesnaite et al., 2023;Ishii et al., 2018;Knyazeva et al., 2018;Mizukami and Katada, 2018;Smith et al., 2023), suggesting that a slowing in peak frequency may reflect biological ageing processes.Additionally, a reduction in IAF has been associated with Alzheimer's disease patients when compared to healthy controls (Babiloni et al., 2004;Bennys et al., 2001;Benwell et al., 2020;Brenner et al., 1986;Coben et al., 1983;Moretti et al., 2004).However, a more recent study did not find this effect after controlling for aperiodic activity (Kopčanová et al., 2023;bioRxiv).
The observed age-related slowing of IAF may be explained by its susceptibility to conduction delays in thalamo-cortical pathways (Roberts and Robinson, 2008;Robinson et al., 2001) and link to white matter structure (Valdés-Hernández et al., 2010), known to degenerate in ageing (Bartzokis, 2004;Marner et al., 2003).While this hypothesis offers one explanation for our findings, future research employing multimodal neuroimaging techniques and longitudinal designs is essential to test this hypothesis.
We also observed that a faster parieto-occipital IAF was associated with a better d' score in the SART beyond the contributions of age, gender and education.This aligns with existing literature, emphasising the functional role of IAF in cognition (Angelakis et al., 2004;Grandy et al., 2013a;Haegens et al., 2014;Hülsdünker et al., 2016;Klimesch et al., 1993;Min and Herrmann, 2007;Richard Clark et al., 2004).Importantly, our study is the first to confirm an association between IAF and sustained attention.
The d' measure in the SART is a critical index of an individual's ability to discriminate between target and non-target stimuli, reflecting the accuracy and sensitivity in sustained attention tasks (Robertson et al., 1997).Higher d' scores signify superior attentional control, crucial for maintaining focus and minimising false alarms, with research demonstrating a diminished ability with older age to ignore and inhibit distracting information (Li et al., 2001), additionally aligning with our finding of a negative association between age and d' scores.The linkage between a faster IAF and enhanced d' scores may be understood through the framework of cognitive preparedness, reflecting the brain's capacity for optimal cognitive performance (Angelakis et al., 2004) at baseline, particularly in the parieto-occipital cortex, which is integral to visual and spatial information processing (Klimesch, 1999).This region's involvement in the dorsal attention network underscores its role in modulating attention and processing sensory inputs (Corbetta and Shulman, 2002).A faster IAF in this area may signify a more optimal neural oscillatory environment, facilitating efficient neural processing and thus improving the signal-to-noise ratio, thereby enhancing the

Table 2
Mediation effect model investigating the role of IAF as a mediator in the relationship between age and SART_d'.discrimination of relevant stimuli (Jensen and Mazaheri, 2010).Building on this understanding, the concept of an optimal excitation-inhibition ratio, previously associated with a faster IAF (Nelli et al., 2017;Samaha and Postle, 2015), further elucidates the underlying mechanisms of an optimal level of cognitive preparedness at baseline.The balance between excitation and inhibition is essential for maintaining the dynamic range of neural responses (Froemke, 2015;Vogels et al., 2011), a fundamental aspect that underpins cognitive adaptability and the efficiency of attentional networks.
Critically, the IAF-SART performance association did not extend to the RVP task.This difference may highlight how various cognitive functions, overlap and interact in sustained attention tasks (Sarter et al., 2001).The observed association between IAF and the SART, which emphasises inhibitory control, contrasted with the absence of such a relationship in the RVP task, which incorporates a significant working memory component, suggests that IAF may be more closely tied to the inhibitory aspects of sustained attention rather than its working memory facets.These findings underscore the complexity of accurately measuring sustained attention and may explain the conflicting results produced by previous research in this area.
Our results showed significant mediation effects of IAF on the relationship between age and SART performance.Specifically, the findings revealed a significant indirect effect and a non-significant direct effect of age on SART performance when mediated by IAF, suggesting that the impact of age on SART performance operates entirely through its influence on IAF.This indicates that IAF plays a crucial role in sustained attention variations associated with ageing and highlights the central role of neural oscillatory patterns in explaining the variability in sustained attention abilities observed in older adults.
Overall, our findings highlight the potential of resting-state IAF as an age-related neural marker of sustained attention, suggesting that changes in IAF are a crucial factor in the decline of sustained attention with ageing.

Aperiodic activity not associated with age or attention
Surprisingly, after controlling for FDR, we did not find an age-related association with aperiodic activity.Notably, studies have documented an age-related shift in aperiodic activity across broad age range comparisons (Dave et al., 2018;Merkin et al., 2022;Thuwal et al., 2021;Tran et al., 2020;Tröndle et al., 2023;Voytek et al., 2015), and more recently, from middle to late life (Finley et al., 2022) and within an older age range (Cesnaite et al., 2023).However, while the study by Merkin et al. (2022) primarily looked at differences between the aperiodic exponent and offset between younger and older adults, notably, they did not find an association with age within their older adult group (50-86 years).Similarly, Smith et al. (2023) did not find an association between age and the aperiodic exponent in their older adult group (50-80 years).Both findings consistent with our result.One possible explanation for our finding could be our sample size.Cesnaite et al. (2023) had a sample of 1703 participants, and Finley et al. (2022) had 268, whereas our study included 96 participants (50-84 years), similar to the sample sizes in the studies by Merkin et al. (2022) and Smith et al. (2023) who also reported no age-related associations in aperiodic activity within their older adult groups.This may suggest that while the aperiodic exponent significantly flattens and the offset is reduced over the course of early and middle adulthood, the changes during late life are not as pronounced, and the associations with age may be too subtle to detect in a smaller sample, highlighting the need for larger sample sizes to accurately assess this relationship.
Moreover, observed changes in aperiodic activity, particularly in the exponent, have been proposed to indicate shifts in the balance between synaptic excitation and inhibition (Gao et al., 2017) and this balance is crucial for short-term dynamics within neuronal synapses, as these affect the generation of oscillations and the efficiency of information transfer (Zhou and Yu, 2018).Our investigation found no significant relationship between either sustained attention measure and resting-state EEG aperiodic activity in older adults, diverging from recent studies suggesting aperiodic activity's predictive value for cognitive performance, specifically in older adults (Smith et al., 2023) or across the adult lifespan (Finley et al., 2024).Similar to our finding, other studies have also reported no association with cognition (Cesnaite et al., 2023;Tröndle et al., 2023), indicating that further research to clarify these conflicting findings.Interestingly, a study by Waschke et al. (2021) demonstrated that the aperiodic exponent is associated with visual attention in occipital regions, with opposite trends observed in more frontal areas.However, it is important to note that this study employed task-based EEG and consisted of undergraduate students (21 ± 3 years), with ageing not a focus of the study.Given that our study utilised global aperiodic activity measures, future research should explore region-specific aperiodic activity in relation to sustained attention performance, both at rest and during task-based conditions, to provide a more nuanced understanding of these dynamics in ageing.
Additionally, we found no significant interaction effects between the aperiodic exponent and IAF associated with either sustained attention measure, despite a recent study by Finley et al. (2024) finding that a higher aperiodic exponent coupled with a lower IAF was associated with greater cognitive decline.However, this study collected repeated cognitive scores over 10 years, with this interaction not significant on a cross-sectional level, and had a sample size of 235, suggesting that longitudinal data and larger sample sizes may be necessary to detect such interaction effects.Overall, our results highlight the need for long-term studies with larger cohorts to fully understand the complex relationships between these neural markers, age and sustained attention in an older adult cohort.

Aperiodic-adjusted individual alpha power not associated with ageing or attention
In contrast with IAF, our study revealed no associations between aIAP and age in our older adult sample.Our approach and subsequent finding is supported and aligns with recent studies that demonstrate when aperiodic activity is accounted for (by means of decomposing the PSD into separate periodic and aperiodic components), age-related associations with alpha power within older adults disappear (Cesnaite et al., 2023;Merkin et al., 2022), suggesting that previous observations that report a decrease in alpha power with ageing, specifically in older adults (Lodder and van Putten, 2011), may have been confounded by changes in the aperiodic background.This evidence underscores the importance of distinguishing true alpha oscillatory signals from the aperiodic background to accurately assess alpha power, free from the influence of non-oscillatory variations, which notably shift with ageing.The stability of alpha power with age in our older adult sample may reflect high cognitive reserve in participants within our sample-the brain's flexibility and adaptability of cognitive/brain networks to compensate for age-related changes in cognitive ability (Park and Reuter-Lorenz, 2009;Stern, 2002), possibly even maintained by the high educational level in our sample (M education = 14.6 ± 2.8 years).Griffa et al. (2021) corroborate this, showing that alpha power relates to cognitive reserve, such that higher alpha power was associated with greater cognitive reserve, considering age and education's effects.Furthermore, another explanation for the difference between our findings and previous literature may be reflective of a key methodological difference, in which we individually determined an individual's alpha band range, centred on their IAF.This aligns with recent practices that accommodate individual variations in frequency to provide a more precise measure of alpha power (Cesnaite et al., 2023;Donoghue et al., 2022;Tröndle et al., 2023).Given the evidence that supports the influence of age on both IAF and aperiodic activity, these methodological updates ensure that alpha power measurements accurately represent an individual's true alpha activity, addressing potential biases caused by age-related changes in peak frequency and the mixing of oscillatory and non-oscillatory activities.
Finally, our study was unable to find an association between individual aIAP and sustained attention, despite previous literature linking alpha power with sustained attention mechanisms (Clayton et al., 2015) and higher alpha power with better sustained attention performance on the SART (Dockree et al., 2007).This discrepancy between findings might stem from the methodological differences in measuring alpha power, (highlighted in the previous section).Specifically, the study conducted by Dockree et al. (2007) found that during a modified SART, tonic alpha power predicted ERP components associated with sustained attention performance, indicating a complex and nuanced relationship between alpha power and task performance that might not be evident during rest, as measured in our study.Previous literature suggests that alpha oscillations play a dual role in enhancing task-relevant processes and inhibiting irrelevant ones, such that alpha power has been shown to increase in task-irrelevant brain areas (Klimesch, 1999;Mazaheri et al., 2009;Pfurtscheller, 2003) and decrease in task-relevant areas (Händel et al., 2011;Jensen and Mazaheri, 2010;Thut et al., 2006).Such dynamics, pivotal during task performance, may not be as observable or relevant when measured at rest, highlighting the need for further research to understand the role of resting-state alpha power role in sustained attention, particularly in ageing.

Limitations
There are several limitations to our study that require consideration.Firstly, the cross-sectional nature of our data collection precludes the determination of causal relationships or the observation of temporal patterns and individual variations over time.Next, the gender composition of our sample, with a predominance of females (76 F, 79.2 %), limits analysis regarding gender differences and generalisations therein.Our cohort was a relatively uniform group of highly functional and welleducated older adults and as such may not accurately reflect the broader population, potentially limiting the applicability of our results to cohorts that include diagnoses such as MCI or dementia.

Conclusion
In summary, in an older adult sample (50-84 years), we demonstrated key associations between age, resting-state IAF and sustained attention performance in our sample of older adults.These findings suggest a neurophysiological basis for the efficiency of attentional processes in ageing, highlighting the potential of resting-state EEG measures as indicators of optimal cognitive processing capacity or decline.Together, these results highlight the potential of resting-state EEG parameters as biomarkers for cognition and ageing, enhance the knowledge of the neurophysiological foundations of sustained attention, and advance our understanding of the cognitive ageing process.

Declaration of Generative AI and AI-assisted technologies in the writing process
During the preparation of this work, the author(s) used ChatGPT to improve readability and language.After utilising this tool, the authors meticulously reviewed and edited the content as necessary and assumes full responsibility for the content of the publication.

Declaration of Competing Interest
None.

Fig. 1 .
Fig. 1.Process of spectral parametrisation i.e., (i) estimation of aperiodic activity parameters (exponent and offset) with 'specparam' and (ii) subtracting the aperiodic model from the spectrum (i.e., flattening) and estimating IAF and individual alpha power.

Fig. 3 .
Fig. 3. IAF residuals and SART_d' residuals, with age, gender and years of education held constant.The red line indicates the line of best fit and red shading reflects 95 % confidence interval.SART_d', Sustained attention to response task d-prime; IAF, Individual alpha peak frequency.

Fig. 4 .
Fig. 4. Path diagram for the mediation analysis between age and SART_d' (Sustained attention to response task d-prime) with IAF (Individual alpha peak frequency) as the mediator.Significant pathways are depicted with red arrows.The path coefficients are presented as standardised beta coefficients (β).* p<0.05, **p<0.01,***p<0.001.

Table 1
Hierarchical linear regression model predicting sustained attention performance from IAF (significant model only).