Behavioral variability and preparatory ERPs as neurobehavioral markers of mild cognitive impairment risk in older adults

Abstract

Mild cognitive impairment (MCI) represents a transitional stage between normal aging and dementia, yet sensitive behavioral and neural markers of early cognitive decline remain to be fully characterized. Increasing evidence suggests that instability of proactive control mechanisms, rather than generalized slowing, may play a critical role in early cognitive impairment. The present study investigated whether behavioral variability and preparatory neural activity constitute neurobehavioral markers of global cognitive functioning in older adults. Ninety‑two older adults (65–89 years) completed a discriminative response task while electroencephalography was recorded. Behavioral performance was quantified using mean response time, error rate, and the intra‑individual coefficient of variation of response time (ICV). Preparatory neural activity was indexed by the Bereitschaftspotential (BP) and prefrontal negativity (pN) measured in the pre‑stimulus interval. Global cognition was assessed using the Montreal Cognitive Assessment (MoCA), analyzed both categorically (MCI‑range vs. cognitively normal) and continuously. Group comparisons, correlational analyses, regression models, and mediation analyses were conducted. Compared with cognitively normal participants, individuals in the MCI range exhibited significantly greater behavioral variability (higher ICV) and reduced preparatory neural activity (attenuated BP and pN amplitudes), while mean response time did not differ between groups. Across the entire sample, higher ICV was associated with lower MoCA scores, whereas larger BP and pN amplitudes were associated with better global cognition. Behavioral instability and preparatory neural activity independently predicted MoCA scores in continuous models. Mediation analyses indicated that ICV and preparatory ERP measures partially mediated the relationship between age and cognitive performance. The present findings demonstrate that behavioral instability and reduced preparatory neural engagement are closely linked to global cognitive performance and MCI-range in older adults. By integrating behavioral and pre‑stimulus electrophysiological measures, this study supports a neurobehavioral framework in which proactive control instability is associated with early cognitive decline. Preparatory processes emerge as sensitive and informative markers linked with variability in cognitive functioning in aging, with potential relevance for understanding early neurobehavioral changes.

Introduction

Mild cognitive impairment (MCI) represents a critical transitional stage between healthy aging and dementia, characterized by objective cognitive decline in the absence of marked functional impairment (e.g., [1]. Although MCI is often conceptualized as a prodromal condition, its clinical expression and longitudinal trajectory are highly heterogeneous, with substantial inter‑individual variability in both the rate and extent of cognitive decline. This heterogeneity poses a major challenge for early identification and mechanistic understanding of vulnerability to dementia, underscoring the need for functional markers that capture early disruptions in cognitive control mechanisms rather than downstream cognitive outcomes alone.

A growing consensus in cognitive aging research suggests that cognitive decline is not adequately described by mean‑level reductions in performance alone (e.g., [2]. Instead, dynamic features of behavior have emerged as particularly sensitive indicators of cognitive instability. Intra‑individual variability in response times reflects moment‑to‑moment fluctuations in attentional engagement, executive control, and neural efficiency, and has been consistently linked to adverse cognitive outcomes in aging (e.g., [3, 4]. Elevated response‑time variability has been associated with poorer global cognition, reduced executive functioning, and increased risk of cognitive deterioration, supporting the view that behavioral instability captures early functional perturbations in control systems that precede overt performance deficits (e.g., [5].

Converging neurophysiological evidence further indicates that age‑related cognitive decline is accompanied by alterations in preparatory neural activity occurring prior to action initiation. Preparatory control mechanisms support proactive regulation of perception and action by configuring task‑relevant neural networks in advance of anticipated demands. Within the dual mechanisms of control framework [6], proactive control is particularly vulnerable to aging, whereas reactive control tends to be relatively preserved (e.g., [7]. Disruptions in proactive control have been observed across adulthood and are thought to underlie age‑related declines in goal maintenance and anticipatory regulation (e.g., [8, 9].

Electrophysiological indices of preparatory neural activity provide temporally precise measures of these anticipatory processes. The Bereitschaftspotential (BP), first described by [10], reflects a slow cortical potential preceding voluntary action and has been linked to motor preparation, internal timing, and higher‑order control processes. Subsequent work has further clarified the functional significance of the BP as an index of anticipatory neural engagement supporting efficient action and cognition [11, 12]. Attenuation of BP and related preparatory signals has been documented in aging (see [13] for review) and clinical populations (e.g., [14], suggesting a progressive breakdown of proactive neural control mechanisms.

In addition to the centro‑parietal BP, a distinct pre‑stimulus event-related potential (ERP) component reflecting preparatory control has been identified over prefrontal scalp regions. This component, termed prefrontal negativity (pN), was originally described by the Di Russo research group in a series of ERP studies using discriminative response (DRT) tasks [15, 16]. The pN is characterized by a slow‑rising negative deflection emerging several hundred milliseconds before stimulus onset and showing a prefrontal distribution. Crucially, source‑localization and multimodal evidence demonstrated that the pN originates from the inferior frontal gyrus, particularly the pars opercularis, rather than from medial premotor regions associated with the Bereitschaftspotential (e.g., [16, 17]. This anatomical dissociation supports the interpretation of the pN as a marker of prefrontal proactive control, distinct from motor preparation per se. Unlike the contingent negative variation, the pN emerges in the absence of explicit warning cues and appears specifically related to anticipatory control processes engaged to prevent premature or inappropriate responses under conditions of response uncertainty. Importantly, the pN shows robust modulation with aging, exhibiting earlier onset and larger amplitude in older adults, a pattern interpreted as compensatory recruitment of prefrontal control mechanisms to preserve behavioral accuracy despite declining cognitive efficiency [15, 18]. Within this framework, reduced efficiency of prefrontal proactive control indexed by the pN is expected to translate into increased trial‑to‑trial behavioral instability, providing a functional link between preparatory neural activity and intra‑individual variability of response times.

Despite strong parallel evidence implicating both behavioral variability and preparatory neural activity in aging, their joint contribution to cognitive performance in MCI remains insufficiently characterized. Specifically, it is unclear whether increased behavioral instability and reduced preparatory neural activity reflect redundant expressions of a common underlying deficit or whether they represent partially independent mechanisms that converge on global cognitive dysfunction. Addressing this question is critical for advancing integrative models of MCI that link behavioral dynamics with neural control processes.

Moreover, much of the existing literature relies on categorical group‑based comparisons between MCI and cognitively preserved individuals. While informative, such approaches may obscure meaningful inter‑individual variability in cognitive functioning that spans diagnostic boundaries. Dimensional approaches that treat cognition as a continuous construct offer increased sensitivity for detecting subtle neurobehavioral alterations and may provide complementary insights when integrated with categorical classifications. To address these issues, the present study adopts an integrative neurobehavioral framework to investigate preparatory control mechanisms in older adults with MCI. Behavioral stability, indexed by intra‑individual response‑time variability, and preparatory neural activity, indexed by the BP and pN amplitudes, were examined in relation to global cognitive performance assessed with the Montreal Cognitive Assessment (MoCA). Cognitive functioning was evaluated using complementary analytical strategies, including group‑based comparisons, continuous prediction models of cognitive performance, classification analyses, and mediation models linking age, neurobehavioral markers, and cognition.

By jointly examining behavioral and electrophysiological indices of preparatory control within both categorical and dimensional frameworks, this study aims to clarify the functional relevance of anticipatory brain mechanisms in MCI. More broadly, this integrative approach seeks to elucidate how age‑related disruptions in proactive control contribute to cognitive decline, offering a unified framework that links behavioral variability, preparatory neural dynamics, and global cognitive performance. First, we hypothesized that increased intra-individual response-time variability would be associated with lower global cognitive performance, reflecting reduced stability of executive control processes in MCI-range participants. Second, we hypothesized that both BP and pN amplitudes would be reduced in individuals in the MCI range, reflecting a disruption of anticipatory control mechanisms. The reduced preparatory activities would be further associated with increased behavioral variability. Last, we hypothesized that both behavioral variability and anticipatory neural markers would predict global cognitive performance along a continuous spectrum, supporting a dimensional model of cognitive decline.

Methods

Participants

The present study included 92 older adults drawn from an institutional database maintained by the Cognitive and Action Neuroscience Laboratory at the University of Rome “Foro Italico,” which comprises behavioral, electrophysiological, and cognitive data collected within ongoing research protocols on cognitive and action control in aging, containing 102 participants. Selected participants (see criteria below) had a mean age of 75.4 ± 5.2 years (range 65–89 years) and included 65 women and 27 men. Education level (years of formal education) was 9.8 ± 3.6 years (range 5–13 years). All participants had complete behavioral, electrophysiological, and cognitive datasets and were therefore retained for analysis.

Inclusion criteria were age ≥ 65 years; availability of valid behavioral data during the discriminative response task (DRT); EEG recordings of sufficient quality for event-related potential (ERP) analysis; and availability of Montreal Cognitive Assessment (MoCA) scores. Exclusion criteria included a history of major neurological disorders, psychiatric illness, traumatic brain injury, or unstable systemic medical conditions affecting cognition. All participants were right-handed, had normal or corrected-to-normal vision, and provided written informed consent. The study was approved by the local Ethics Committee (approval #47-23, date Nov. 15, 2023) and conducted in accordance with the Declaration of Helsinki.

A priori and sensitivity power analysis indicated that the available sample size (N = 92) provided adequate statistical power for the planned analyses. Assuming medium effect sizes (r = 0.30), the study achieved power greater than 0.80 for correlational analyses at α = 0.05. The sample size also exceeds recommended ratios for multivariable regression and mediation analyses, ensuring stable parameter estimation. For predictive classification analyses, the sample is sufficient to detect meaningful differences in areas under the ROC curve when adding neural predictors to demographic and behavioral models.

Cognitive assessment

Global cognitive functioning was assessed using the Montreal Cognitive Assessment (MoCA), a widely validated cognitive screening tool sensitive to mild cognitive impairment [19]. MoCA total score was used both as a continuous measure of global cognitive performance and, for stratified analyses, as a categorical variable to define MCI-range (MoCA < 24) versus cognitively normal status (MoCA ≥ 24), in line with recent meta‑analytic evidence indicating that this cutoff provides an optimal balance between sensitivity and specificity for mild cognitive impairment in older adults compared to the original cutoff of 26, although variability across samples and demographic factors should be acknowledged [20, 21].

Task design and behavioral measures

Participants performed a DRT with a structure identical to that described by [17]. Visual stimuli were presented centrally for 250 ms with a variable inter-stimulus interval ranging between 1 and 2 s. Stimuli were divided into target and non‑target categories. Participants responded as quickly and accurately as possible to targets with a button press, while withholding responses to non‑targets. Each participant completed more than 300 trials. Behavioral performance was quantified using mean response time (RT) for correct target trials and error rate defined as the sum of omission errors and false‑alarm errors. Omission errors corresponded to missed responses to target stimuli, whereas false alarms corresponded to responses erroneously emitted to non‑target stimuli. The intra-individual coefficient of variation (ICV) of RT was computed as the standard deviation of RT divided by the mean RT and used as an index of behavioral instability (e.g., [5].

EEG recording and analysis

EEG activity was recorded using two 32-channel BrainAmp amplifiers, 64 active scalp electrodes mounted on an ActiCap system, and BrainVision Recorder (version 1.2) software. Offline analyses were conducted using BrainVision Analyzer (version 2.3). All equipment was provided by Brain Products GmbH (Gilching, Germany). Electrodes were positioned according to the 10–10 International System and referenced to the average of the mastoids.

Eye movements and blinks were monitored using an additional BrainAmp amplifier and two pairs of bipolar electro‑oculogram (EOG) electrodes: one pair placed at the outer canthi of both eyes and the other above and below the left eye. Electrode impedances were kept below 10 kΩ. EEG signals were digitized at 250 Hz, amplified with an online band‑pass filter of 0.01–80 Hz (including a 50 Hz notch filter), and stored for offline processing. Eye‑movement artifacts were corrected using a regression‑based procedure. Subsequently, artifact rejection was applied to exclude epochs contaminated by voltage fluctuations exceeding ± 60 µV, ensuring the removal of residual non‑ocular artifacts. On average, each participant contributed 468 ± 31 artifact-free trials, with no significant differences between MoCA groups.

Following [17], EEG data were segmented into epochs from − 1100 to + 200 ms relative to stimulus onset and baseline-corrected using the − 1100 to − 900 ms interval. Only correct trials were included. ERP analyses focused on the BP and pN components. Both components were evaluated within the last 400 ms preceding stimulus onset. The BP was analyzed over a centro-parietal electrode cluster (C1, Cz, C2, CP1, CPz, CP2), whereas the pN was analyzed over a bilateral prefrontal cluster (Fp1, Fpz, Fp2, AF3, AFz, AF4). For each component, the mean amplitude across the analysis interval and selected electrodes was computed and used for subsequent statistical analyses.

Figure 1 shows the grand-average pre-stimulus ERP activity computed across the entire sample during the discriminative response task. The waveform highlights two preparatory components developing before stimulus onset: a slow negative drift over centro-parietal sites consistent with the BP, and a concomitant slow negative activity over bilateral prefrontal sites consistent with the pN. Both components are quantified in the final 400 ms before stimulus onset and reflect anticipatory engagement of motor–parietal (BP) and prefrontal proactive-control (pN) systems before stimulus processing. The accompanying scalp topography summarizes the spatial distribution of this pre-stimulus preparatory activity across electrodes, emphasizing the distinct centro-parietal vs. prefrontal maxima associated with BP and pN, respectively.

Fig. 1.

Grand-average pre-stimulus ERP waveforms and scalp topography (top-flat view) on the last 400 ms (whole sample). The Bereitschaftspotential (BP) is indexed over a centro-parietal cluster and represented by the red trace. The prefrontal negativity (pN) is indexed over a bilateral prefrontal cluster and represented by the blue trace. Electrode clusters are highlighted in the left inset

Statistical analysis

Statistical analyses were conducted to examine neurobehavioral markers of preparatory control and their relationship with global cognitive performance. Analyses were performed in four steps: group comparisons, correlational analyses, predictive models of cognitive performance, and mediation analyses. To control for potential confounding effects, additional analyses were conducted, including age, sex, and years of education as covariates. All adjusted analyses included these variables to account for their potential influence on MoCA performance and neurobehavioral measures.

Group differences between participants classified as MCI and cognitively preserved controls were assessed using independent‑samples t-tests. Effect sizes were quantified using Cohen’s d. When assumptions of normality were violated, non-parametric tests were applied. For this analysis, participants were stratified into an MCI-range group (MoCA < 24; n = 48) and a cognitively normal group (MoCA ≥ 24; n = 44). Group comparisons were performed to assess differences in behavioral and electrophysiological measures between cognitive status groups. The cut-off of 24 was used since evidence suggests that it provides a more balanced level of sensitivity and specificity than the original threshold of 26 [20, 22, 23]. Associations between age, behavioral measures, preparatory neural activity, and global cognition were examined using Spearman rank correlations, given mild deviations from normality in response‑time variability measures.

To assess whether neural markers predict cognitive performance, along a continuous dimension, multiple linear regression models were conducted with MoCA score as a continuous dependent variable. ICV, BP, and pN amplitudes were entered as simultaneous predictors. Regression coefficients were standardized to allow comparison of the relative contribution of individual predictors.

Classification analyses were performed using logistic regression models to distinguish MCI from non‑MCI participants. MCI status was defined using a MoCA cut‑off score of < 24. Model performance was evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC). Nested regression models were compared to assess the incremental contribution of behavioral and neural predictors beyond age. Changes in explained variance (ΔR²) were calculated for continuous models, while changes in classification performance (ΔAUC) were estimated for logistic models. These analyses were conducted on the same dataset and therefore reflect in-sample performance, which may overestimate generalizability. No external validation or cross-validation procedures were applied.

Finally, mediation analyses were conducted to test whether neurobehavioral markers mediated the relationship between age and global cognition. Intra‑individual response‑time variability, BP amplitude, and pN amplitude were entered simultaneously as mediators. Standardized coefficients were estimated, and indirect effects were computed as the product of the age‑to‑mediator and mediator‑to‑outcome paths. These analyses were interpreted as statistical models of indirect association and do not imply causal relationships, as all variables were measured cross-sectionally.

All analyses were conducted using R (version 4.4.1) statistical software, with significance set at p < 0.05 (two‑tailed). Indirect effects were estimated using bootstrapping procedures (5000 resamples) to obtain bias-corrected confidence intervals.

Results

Group differences

As expected, the two groups differed significantly in MoCA scores by definition. Age was modestly higher in the MCI‑range group (t₍₉₀₎ = 2.7, p < 0.05, d = 0.47). Although the overall sample included a higher proportion of women than men, sex distribution was comparable between cognitively normal and MCI‑range groups (χ²₍₁₎ = 0.32, p = 0.57), indicating that sex did not differ as a function of cognitive status.

Behavioral measures

Participants in the MCI‑range group exhibited greater response‑time variability (M = 30.8%, SD = 5.9), as indexed by the ICV (t₍₉₀₎ = 4.32, p < 0.01, 95% CI = 2.6, 7.2, d = 0.70), compared with cognitively normal participants (M = 25.9%, SD = 5.0, Fig. 2C). In contrast, mean response time (Normal = 565 ms SD = 45; MCI = 585 ms SD = 54) did not significantly differ between groups (t₍₉₀₎ = 1.89, 95% CI = 11, 40, p = 0.06), indicating that group differences were driven by trial‑to‑trial instability rather than overall slowing (Fig. 2A). The error rate was significantly higher in the MCI‑range group (t₍₉₀₎ = 2.29, p < 0.05, 95% CI = 1.5, 6.5, d = 0.28). Specifically, participants in the MCI‑range group showed higher error rates (M = 32.8%, SD = 9.8) compared with cognitively normal participants (M = 28.3%, SD = 9.6, Fig. 2B).

Fig. 2.

Behavioral group differences by cognitive status. Participants were classified as MCI-range (MoCA < 24; n = 48) or cognitively normal (MoCA ≥ 24; n = 44). A The mean response time for correct target trials. B The error rate. C Response‑time variability indexed by the intra-individual coefficient of variation. Bars represent the group median, the 25–75%, and the total data range. *p < 0.05, **p < 0.01

ERP measures

Electrophysiological analyses revealed marked group differences in pre‑stimulus preparatory neural activity. The MCI‑range group showed significantly reduced (t₍₉₀₎ = 4.34, p < 0.01, 95% CI = 0.39, 1.05, d = 0.70) BP amplitudes over centro‑parietal electrodes (M = − 2.62 μV, SD = 0.84) relative to cognitively normal participants (M = − 3.34 μV, SD = 0.75, Fig. 3A). Similarly, the pN amplitudes were significantly attenuated (t₍₉₀₎ = 4.45, p < 0.01, 95% CI = 0.52, 1.35, d = 0.70) in the MCI-range group (M = − 2.04 μV, SD = 1.04) over bilateral prefrontal regions compared to the normal group (M = − 2.97 μV, SD = 0.96, Fig. 3B). Together, these findings indicate a disruption of preparatory control mechanisms before stimulus processing in participants with lower cognitive status.

Fig. 3.

Neural group differences based on cognitive status (MoCA < 24 vs. MoCA ≥ 24). Mean pre-stimulus amplitudes (− 400 to 0 ms) for (A) the Bereitschaftspotential (BP) over a centro-parietal cluster and (B) the prefrontal negativity (pN) over a prefrontal cluster. Bars represent the group median, the 25–75%, and the total data range. **p < 0.01

For both behavioral and ERP measures, group differences remained significant after controlling for age, sex, and education. Specifically, the MCI-range group showed higher response-time variability (ICV: F_{(1, 88)} = 17.05, p < 0.01), as well as reduced preparatory neural activity, reflected in attenuated BP (F_{(1, 88)} = 28.15, p < 0.01) and pN amplitudes (F_{(1, 88)} = 21.65, p < 0.01). These findings indicate that the observed neurobehavioral differences are robust and cannot be explained by demographic factors alone.

Associations between neurobehavioral measures and global cognition

Associations between behavioral variability, preparatory neural activity, and global cognitive performance were examined using Spearman rank‑order correlations, given mild deviations from normality for reaction‑time–derived measures (see Methods). Response-time variability (ICV) was significantly and negatively associated with the MoCA score. Higher ICV values were related to poorer global cognition (ρ =  − 0.41, p < 0.01). This relationship indicates that increased trial‑to‑trial behavioral instability is robustly linked to lower cognitive performance (Fig. 4A). Both preparatory ERP components showed significant positive associations with MoCA scores. Specifically, more pronounced (i.e., more negative) BP amplitudes were associated with higher MoCA scores (ρ = 0.38, p < 0.01, Fig. 4B). Similarly, the pN amplitudes were positively correlated with MoCA score (ρ = 0.34, p < 0.01, Fig. 4C), indicating that stronger pre‑stimulus prefrontal engagement is linked to better global cognitive functioning. Importantly, both BP and pN amplitudes were negatively correlated with ICV (p < 0 0.01), indicating that reduced preparatory neural activity was associated with increased behavioral instability. Examination of correlation matrices revealed no collinearity issues (all ρ < 0.60). No correlation exceeded conventional thresholds for multicollinearity, and the observed associations remained stable when controlling for age and education, confirming that the relationships were not driven by age and education differences alone.

Fig. 4.

Scatterplots illustrate associations between MoCA score and (A) response‑time variability (ICV), B the BP amplitude, and (C) the pN amplitude. Lines indicate fitted regression trends for visualization. Higher behavioral variability was associated with poorer cognition, whereas stronger preparatory neural activity was associated with higher MoCA scores

Added predictive value

Hierarchical regression models were used to test the incremental contribution of neurobehavioral markers to global cognition. Age alone explained 19% of the variance in MoCA scores. Adding response‑time variability increased explained variance to 38%, while the inclusion of preparatory neural markers (BP and pN) further increased explained variance to 46% (ΔR² = 0.08). A similar pattern was observed in classification analyses, where neural markers increased discrimination between MCI and non‑MCI individuals from AUC = 0.86 to AUC = 0.91. In the full model, response‑time variability showed the largest standardized effect on MoCA performance, while both BP and pN amplitudes independently contributed to explained variance, indicating non‑redundant neural effects. In this final model, ICV showed the strongest association with MoCA scores (β = − 0.36, 95% CI =  − 0.54, − 0.18], p < 0.01), followed by pN (β = − 0.26, 95% CI = − 0.47, − 0.04, p = 0.02) and BP amplitudes (β = − 0.17, 95% CI = − 0.40, 0.05, p = 0.13).

Prediction of MCI status

To assess the clinical relevance of neurobehavioral markers, we evaluated their ability to discriminate individuals with MCI from cognitively preserved participants using a receiver operating characteristic (ROC) analysis. A multivariate logistic regression model was computed, including ICV, BP amplitude, and pN amplitude as predictors. The combined model showed good discriminative accuracy, yielding an AUC of 0.78, indicating that neurobehavioral markers provide substantial information for MCI classification beyond chance level. Inspection of the ROC profile suggested a balanced trade‑off between sensitivity and specificity, supporting the potential utility of these measures for identifying individuals with early cognitive impairment (Fig. 5). These estimates reflect in-sample classification performance and should be interpreted cautiously.

Fig. 5.

ROC curve illustrating the performance of a multivariate logistic regression model including ICV, BP, and pN amplitudes. The model shows good discriminative ability between MCI-range patients and controls (AUC = 0.78). The dashed diagonal line represents chance-level performance

Continuous models of cognitive performance

To characterize inter‑individual variability in cognitive functioning, global cognition was examined using a continuous modeling approach. MoCA score was treated as a continuous outcome and predicted through a multivariate linear regression model including ICV, BP, and pN amplitudes. In this model, higher ICV was significantly associated with lower MoCA scores (β = − 0.36, 95% CI = − 0.54, − 0.18, p < 0.01), indicating that increased behavioral variability was linked to poorer cognitive performance. Preparatory neural activity also contributed to cognitive performance, although the association was stronger and statistically more robust for pN than for BP in the multivariable model: BP (β = − 0.17, 95% CI = − 0.40, 0.05, p = 0.13) and pN amplitude (β = − 0.26, 95% CI = − 0.47, − 0.04, p = 0.02) were associated with MoCA scores. The combined model explained a substantial proportion of variance in global cognitive performance, accounting for 40% of the variance in MoCA scores (R² = 0.40, 95% CI = 0.26, 0.52, p < 0.01). This effect size indicates a meaningful contribution of neurobehavioral markers to individual differences in cognition, particularly given the multifactorial nature of cognitive decline in aging. Figure 6 illustrates the relationship between observed MoCA scores and values predicted by the continuous model. Predicted and observed scores showed a close correspondence, with data points distributed around the identity line, indicating that the model captured systematic variability in cognitive performance rather than random noise. Importantly, the continuous model revealed a graded association between preparatory control mechanisms and cognition, supporting a dimensional view of cognitive impairment. Instead of reflecting purely categorical differences, alterations in behavioral stability and preparatory neural activity scaled with global cognitive performance across individuals.

Fig. 6.

Scatterplot showing the relationship between observed MoCA scores and values predicted by a continuous linear model including ICV, BP, and pN amplitudes. The dashed line represents the identity line, indicating perfect correspondence between predicted and observed cognitive performance

Mediation analysis

Mediation analyses were conducted to formally test whether neurobehavioral markers mediated the relationship between age and global cognitive performance. Age showed a strong total effect on MoCA score (β = − 0.44, 95% CI = − 0.58, − 0.30, p < 0.01). Increasing age was associated with greater response‑time variability and reduced preparatory neural activity, particularly reflected by lower BP and pN amplitudes. In turn, higher behavioral instability and reduced preparatory neural activity significantly predicted lower MoCA scores. Significant indirect effects of age on MoCA scores were observed through preparatory neural activity. The indirect effect via ICV was modest and not statistically significant (β = 0.03, 95% CI = − 0.01, 0.07, p = 0.22), whereas significant indirect effects were observed for BP (β = 0.04, 95% CI = 0.003, 0.09, p = 0.038) and pN amplitudes (β = 0.04, 95% CI = 0.002, 0.08, p = 0.035). These results indicate that preparatory neural activity contributed more strongly to the association between age and cognitive performance than behavioral variability.

When all mediators were included simultaneously, the direct effect of age on MoCA score was reduced but remained significant (β = − 0.28, 95% CI = − 0.42, − 0.15, p < 0.01), indicating partial mediation. Overall, these results support a neurobehavioral mediation model in which age-related cognitive decline is statistically associated with alterations in preparatory control mechanisms operating before stimulus onset. The relative strength of indirect effects was larger for preparatory neural markers (BP and pN) than for behavioral variability, indicating that age-related cognitive decline is preferentially mediated by disruptions in preparatory neural mechanisms. These results reflect statistical mediation patterns and do not imply causal relationships, given the cross-sectional design.

Summary of findings

Across categorical, continuous, predictive, and mediation analyses, converging evidence indicates that MCI-range cognitive performance is characterized by increased behavioral instability and reduced preparatory neural activity. Importantly, these neurobehavioral markers not only differentiate cognitive groups but also explain individual differences in global cognition and contribute to the prediction of MCI status.

Discussion

The present study investigated neurobehavioral markers of proactive control in older adults, with the aim of identifying behavioral and neural mechanisms associated with global cognitive functioning and mild cognitive impairment. By integrating indices of behavioral variability with pre-stimulus ERP measures, the present findings provide converging evidence that early cognitive decline is characterized by instability of anticipatory control mechanisms, rather than by generalized slowing of responses.

A key finding of the present study is the robust association between response-time variability, indexed by the intra-individual coefficient of variation (ICV), and global cognitive performance. Participants in the MCI-range group exhibited significantly higher ICV compared with cognitively normal older adults, whereas mean response‑time did not differ between groups. This dissociation indicates that trial-to-trial instability, rather than average processing speed, is a sensitive behavioral signature of early cognitive impairment. These results are consistent with a large body of literature showing that increased intra-individual variability reflects reduced efficiency of attentional and executive control systems and predicts cognitive decline, MCI, and dementia onset (e.g., [24–26]. Importantly, in the present study, ICV was also strongly and negatively associated with MoCA score when cognition was modeled as a continuous variable, supporting the interpretation of behavioral variability as a dimensionally meaningful indicator of global cognition integrity.

At the neural level, the present findings demonstrate that both the BP and the pN were significantly reduced in the MCI-range group. The BP has traditionally been linked to motor preparation but is now recognized as reflecting broader preparatory processes that support cognitive and action control (e.g., [27]. Similarly, the pN has been consistently associated with prefrontal mechanisms of proactive inhibition and control, reflecting the neural implementation of anticipatory “braking” processes (e.g., [16, 28]. The attenuation of both components in MCI-range participants, therefore, is consistent with a disruption of proactive control and related preparatory processes. Crucially, BP and pN amplitudes were positively associated with MoCA score across the entire sample, suggesting a relationship between anticipatory neural engagement and global cognitive performance. These findings align with electrophysiological studies showing age-related alterations in pre-stimulus activity and proactive control mechanisms in older adults (e.g., [29, 30]. Further, our results are consistent with evidence showing that individual differences in cognitive reserve modulate premotor preparatory activity during visuo-motor tasks in older adults, suggesting that preserved anticipatory neural recruitment may support cognitive resilience in aging [31].

A central contribution of the present study lies in demonstrating a tight neurobehavioral coupling between preparatory neural activity and behavioral stability. Both BP and pN amplitudes were negatively correlated with ICV, indicating that individuals with reduced anticipatory neural engagement also showed greater behavioral variability. This coupling supports theoretical models of cognitive aging that emphasize reduced stability of large-scale control networks, particularly within frontal–parietal systems, as an important correlate of cognitive decline [24, 25]. Rather than reflecting isolated deficits, increased variability and reduced preparatory activity may reflect alterations in network efficiency and temporal coordination. Specifically, reduced BP and pN amplitudes might reflect an inability to suppress background neural noise, which may in turn relate to greater behavioral variability as measured by the ICV [32, 33]. This formulation effectively links the prefrontal mechanisms of proactive inhibition control and behavioral metrics under a unified neurophysiological framework in the aging brain [33].

Modeling MoCA score as a continuous outcome further strengthened the present conclusions. Behavioral instability (ICV) and preparatory ERP measures explained variance in global cognition beyond age, with the pN showing the more robust independent association in multivariable models, indicating that these markers capture graded variations in cognitive functioning across the aging spectrum. This dimensional perspective is consistent with contemporary views that conceptualize MCI as part of a continuum of cognitive decline, rather than as a discrete categorical state [20, 34]. Within this framework, proactive control instability appears associated with aging-related differences in neural function and cognitive performance.

The finding that preparatory ERP measures provided incremental value beyond demographic and behavioral predictors underscores their potential utility as early neurophysiological markers of cognitive impairment. Moreover, mediation analyses indicated that behavioral instability and preparatory neural activity partially mediated the relationship between age and cognition, suggesting that age-related cognitive decline is indirectly expressed through degradation of anticipatory control processes. These results are in line with theoretical frameworks proposing proactive control as a key locus of vulnerability in aging, especially in tasks requiring advance preparation and regulation of upcoming actions (e.g., [16, 28]. Although both ERP components were associated with cognitive functioning at the correlational level, only the pN retained a statistically significant independent association in the combined multivariable model, whereas the BP effects were attenuated after accounting for shared variance among predictors.

From a methodological perspective, the combination of simple behavioral variability metrics with pre-stimulus ERP measures offers a powerful approach to capture early neurocognitive alterations. Preparatory ERPs provide a unique window into internally generated control processes that may be particularly sensitive to subtle network dysfunction, beyond what can be detected by post-stimulus measures alone. Clinically, these neurobehavioral markers may complement cognitive screening tools such as the MoCA, which (despite its sensitivity) exhibits ceiling effects and limited reliability at the domain level in cognitively normal individuals (e.g., [35]. The continuous associations observed here may offer a promising framework for future research on longitudinal monitoring and risk stratification, pending external validation.

Limitations and future directions

Several limitations should be considered. Because cognitive status was defined using a screening tool rather than a clinical diagnosis, findings should be interpreted as reflecting variability in cognitive performance rather than established MCI. The cross-sectional design does not allow causal inferences regarding the temporal progression of neurobehavioral instability. Longitudinal studies are needed to determine whether alterations in preparatory neural activity and behavioral variability prospectively predict cognitive decline. Furthermore, replication in independent cohorts and clinically characterized MCI samples will be essential. Future research should also investigate whether interventions targeting cognitive control (such as cognitive-motor training or neuromodulation) can enhance preparatory neural activity and reduce behavioral variability, potentially mitigating cognitive decline. Future studies should include cross-validation or independent cohorts to assess the robustness and generalizability of these findings. A further limitation is that no source localization analysis was performed in the present study. Therefore, the functional interpretation of BP and pN as preparatory/proactive-control markers remains indirect and cannot exclude contributions from broader motor and attentional processes. In general, interpretations regarding potential mechanistic relationships between preparatory control processes and cognitive decline should remain cautious.

Conclusions

In conclusion, the present study demonstrates that increased behavioral instability and reduced preparatory neural activity are tightly linked to global cognitive performance and mild cognitive impairment in older adults. By integrating behavioral and electrophysiological measures, the findings support a neurobehavioral framework linking proactive control instability with early cognitive decline. Among preparatory ERP measures, prefrontal anticipatory activity indexed by the pN emerged as the most robust independent neural correlate of cognitive performance. Emerging as sensitive markers of cognitive health and promising candidates for future investigation in longitudinal and clinically characterized samples.

Author contribution

FDR, SP, and RB designed the study and developed the methodology. LB, VB, and MA performed the data analysis. MA and LB carried out the experiments and collected the data. FDR. provided resources and supervised the project. FDR drafted the manuscript. All authors discussed the results, contributed to the interpretation, and wrote the final manuscript.

Funding

This work was funded by the European Union NextGenerationEU program, grant number CDR2.PNRR2023_DM737/2021DRF (CUP: H53D23005450006) to Prof. Di Russo.

Data availability

The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.