Process-level eye-movement dynamics enhance sensitivity to early Alzheimer's disease-related cognitive vulnerability

Abstract

Background

Early Alzheimer's disease (AD)-related cognitive vulnerability is characterized by subtle disruptions in attentional and executive processing that may precede overt declines in global cognitive scores. Identifying process-sensitive behavioral markers capable of capturing these early alterations remains a major challenge in psychogeriatric assessment.

Objective

To examine whether eye-movement–derived process-level measures, particularly fixation-shift dynamics, are associated with early cognitive vulnerability in older adults and whether they provide incremental value beyond conventional cognitive screening scores in an AD-related context.

Methods

This cross-sectional study enrolled 109 adults aged ≥65 years, categorized as cognitively normal (CN, n = 33), mild cognitive impairment (MCI, n = 35), and dementia due to AD (ADD, n = 41). Participants completed a standardized online interactive cognitive task while eye movements were recorded using a non–head-fixed, desktop-based eye-tracking system. Cognitive-only, eye-movement–only, and combined multimodal models were compared using receiver operating characteristic (ROC) analysis and area under the curve (AUC).

Results

The combined multimodal model consistently outperformed single-modality approaches. Discriminative performance was high for CN versus ADD (AUC = 0.925) and for CN versus cognitively impaired participants overall (AUC = 0.897). Fixation-shift dynamics emerged as a robust process-level marker, demonstrating particular sensitivity in distinguishing CN individuals from those with MCI.

Conclusions

Fixation-shift eye-movement dynamics capture early AD-related cognitive vulnerability beyond conventional total cognitive scores. This multimodal framework may support psychogeriatric screening and clinical triage by detecting subtle cognitive inefficiencies that precede measurable declines on standard cognitive tests.

Keywords

Alzheimer's disease cognitive vulnerability eye-movement measures mild cognitive impairment psychogeriatric assessment

Introduction

Dementia represents a major public health challenge driven by global population aging. According to the World Health Organization, more than 55 million people worldwide currently live with dementia, a number projected to increase to 78 million by 2030 and 139 million by 2050.¹ In Taiwan, epidemiological surveys report a dementia prevalence of 7.99% among adults aged 65 years and older, while an additional 17.99%, over 730,000 individuals, meet criteria for mild cognitive impairment (MCI), a transitional stage associated with increased risk of dementia, including Alzheimer's disease (AD).^2,3 AD accounts for approximately 60%–70% of dementia cases, with prevalence rising sharply with advancing age. Together, these trends underscore the need for scalable and clinically meaningful approaches to identify early cognitive vulnerability in aging populations before irreversible functional decline occurs.^4,5

Early identification of MCI and AD-related cognitive impairment commonly relies on brief cognitive instruments such as the Mini-Mental State Examination (MMSE), the Clinical Dementia Rating (CDR), and the Montreal Cognitive Assessment (MoCA). Among these, the MoCA has demonstrated improved sensitivity for distinguishing MCI from normal cognitive aging.⁶ Nevertheless, important limitations persist. Global cognitive scores provide relatively coarse summaries of working memory, visuospatial processing, and executive control, and test performance may be influenced by educational background, language proficiency, and cultural context.^7,8 Consequently, subtle disease-related alterations in cognitive processing efficiency, particularly those affecting attentional and executive control rather than overt task accuracy, may precede measurable declines in total scores and remain undetected by conventional screening approaches.

Eye-tracking technology offers a noninvasive and objective means of quantifying oculomotor behavior associated with attention, memory, and visuospatial–executive processing. Core eye-movement measures, including fixations, saccades, and smooth pursuit, reflect underlying cognitive processes that are vulnerable in the earliest stages of neurodegeneration. Prior studies have shown that eye-movement features can differentiate individuals with MCI or early AD from cognitively normal older adults, reveal subtle impairments not captured by conventional neuropsychological assessments, and exhibit relative independence from language and educational confounders.^9–11 These characteristics suggest that eye-movement metrics may provide complementary insight into early cognitive vulnerability in older adults.

Despite these promising findings, it remains unclear whether eye-movement–derived process-level measures provide incremental value when assessed within a standardized cognitive task context, particularly during transitional stages such as MCI. Conventional approaches relying on end-point cognitive scores may lack sensitivity to early inefficiencies in cognitive processing, reflecting a broader measurement gap between observable task performance and underlying disease-related cognitive dynamics.

Disruptions in fixation-shift dynamics, which index the efficiency and flexibility of visual attention allocation, may reflect early dysfunction of frontoparietal attentional and executive networks implicated in prodromal AD. Such process-level inefficiencies may emerge prior to overt performance failure on conventional cognitive tests, thereby providing a window into early AD-related cognitive vulnerability.

Accordingly, the present study examined whether eye-movement–derived process-level measures, particularly fixation-shift dynamics, are associated with early cognitive vulnerability in older adults across cognitively normal individuals, those with MCI, and those with dementia. By evaluating the incremental value of these behavioral markers beyond conventional cognitive screening scores, we aimed to clarify their potential relevance for psychogeriatric assessment, early risk stratification, and clinical monitoring in outpatient and community-based settings.

Methods

Setting/study population

A total of 112 individuals were purposively recruited from neurology and geriatric outpatient clinics, community senior centers, and adult day care facilities in northern and central Taiwan between May and September 2025. Three participants were excluded due to incomplete assessments or withdrawal, resulting in a final analytic sample of 109 participants (cognitively normal [CN], n = 33; mild cognitive impairment [MCI], n = 35; dementia due to AD [ADD], n = 41).

Eligible participants were aged 65 years or older, provided informed consent, and met CDR criteria for their respective cognitive groups: CN (CDR = 0), MCI (CDR = 0.5 with preserved activities of daily living or clinician-confirmed MCI), and ADD (CDR ≥ 0.5 with a physician-confirmed diagnosis). Exclusion criteria included acute medical or psychiatric illness, major psychiatric disorders, significant visual impairment, severe ocular disease, or inability to complete the computerized cognitive or eye-tracking assessments. Participants were allowed to wear their habitual corrective lenses (e.g., glasses) during the assessment to ensure adequate visual performance. Individuals with severe dementia were not included, as they were unable to reliably complete the computerized and eye-tracking tasks.

The mean age of participants was 74.8 ± 7.6 years (range 65–96), with CN participants younger (70.5 ± 7.0) than MCI (76.7 ± 8.2) and dementia (76.6 ± 6.2) groups. Women comprised 64.2% of the sample. Regarding education, 59.6% had completed junior high school or higher, with higher proportions in CN (78.8%) and MCI (65.7%) groups, whereas most dementia participants (61.0%) had elementary school education or below. The dementia group included 41 participants able to follow test instructions, with 13 (31.7%) having a CDR of 0.5 and 28 (68.3%) a CDR of 1. Participant characteristics are summarized in Table 1.

Table 1.

Demographic and clinical characteristics of cognitively normal, MCI, and dementia participants.

Characteristics	Total(n = 109)	Cognitively normal(n = 33)	Mild cognitive impairment(n = 35)	Dementia(n = 41)
Age (years)	74.76 ± 7.62	70.45 ± 6.96	76.66 ± 8.20	76.61 ± 6.24
Age group
65–79 years	80 (73.4%)	28 (84.8%)	23 (65.7%)	29 (70.7%)
≥80 years	29 (26.6%)	5 (15.2%)	12 (34.3%)	12 (29.3%)
Gender
Male	39 (35.8%)	12 (36.4%)	14 (40.0%)	13 (31.7%)
Female	70 (64.2%)	21 (63.6%)	21 (60.0%)	28 (68.3%)
Educational attainment
≤ Elementary school	44 (40.4%)	7 (21.2%)	12 (34.3%)	25 (61.0%)
≥ Junior high school	65 (59.6%)	26 (78.8%)	23 (65.7%)	16 (39.0%)
Clinical Dementia Rating (CDR)
0		33 (100.0%)
0.5			35 (100.0%)	13 (31.7%)
1				28 (68.3%)
Cognitive Performance (MMSE)	23.21 ± 4.73	27.94 ± 2.19	23.60 ± 2.45	19.07 ± 3.92
Orientation	8.14 ± 1.96	9.73 ± 0.63	8.37 ± 1.50	6.66 ± 1.94
Attention and calculation	6.23 ± 1.47	7.06 ± 1.17	6.43 ± 0.85	5.39 ± 1.67
Memory	1.50 ± 1.29	2.94 ± 0.24	1.46 ± 1.07	0.39 ± 0.70
Language	6.62 ± 1.29	7.30 ± 0.85	6.63 ± 0.88	6.07 ± 1.60
Visuospatial construction	0.72 ± 0.77	0.91 ± 0.29	0.71 ± 0.46	0.56 ± 1.14

Continuous variables are presented as mean (SD); categorical variables are presented as n (%).

SD: standard deviation.

Instruments/device

Demographic and clinical data

Demographic and clinical data were obtained from participants’ electronic medical records and documentation provided by hospital or adult day care centers. Collected variables included age, sex, and education level. Cognitive status and dementia severity were assessed using the CDR scale and the MMSE.

The CDR is a semi-structured, interview-based instrument administered to participants and their informants to assess cognitive and functional performance across six domains: memory, orientation, judgment and problem solving, community affairs, home and hobbies, and personal care.¹² Based on the global CDR score, participants were classified into five levels of cognitive impairment, ranging from no dementia to severe dementia (0, 0.5, 1, 2, and 3). The MMSE is a brief cognitive screening tool consisting of 11 items that assess orientation, attention and calculation, memory, language, and visuospatial construction.¹³ Total MMSE scores range from 0 to 30, with lower scores indicating greater cognitive impairment; scores of 21–26 indicate mild impairment, 10–20 moderate impairment, and <10 severe impairment.

Customized online interactive cognitive screening tool

This study employed a customized online interactive cognitive screening tool developed by the research team. Task selection and targeted cognitive domains were informed by the cognitive constructs assessed in the MoCA, a widely used screening instrument with established reliability and validity.^14,15 Rather than replicating or replacing the MoCA, the tool was designed as a standardized digital adaptation to approximate key cognitive domains while providing a controlled task context for capturing process-level behavioral dynamics during task execution. Design principles from prior work on interactive computer-based paradigms for early detection of AD guided both the response processes and the user interface.¹⁶

To enhance feasibility in an online setting, the assessment excluded tasks requiring paper-and-pencil drawing or verbal responses. All items were administered using mouse-click responses, including multiple-choice formats and interactive selection mechanisms. The final version comprised 13 items with a maximum total score of 17 points, covering major cognitive domains typically evaluated by the MoCA, with the exception of language-related items. This design aimed to preserve broad cognitive coverage while ensuring compatibility with online, self-paced administration. Detailed descriptions of individual item formats and their corresponding cognitive domains are provided in the Supplemental Material.

Because this instrument represents a customized online adaptation rather than the original MoCA, established MoCA cutoff scores and total-score interpretations were not applied. Classification performance and optimal cutoff values were instead determined empirically using receiver operating characteristic (ROC) analyses, sensitivity, and specificity, consistent with its intended use as a screening and research tool rather than a diagnostic substitute.

For analytic purposes, test items were further categorized into three task types based on their cognitive and visual processing demands. Search tasks required participants to visually locate and identify a target stimulus prior to response selection. Memory tasks emphasized immediate encoding and delayed retrieval processes. Cognitive processing tasks involved rule-based judgment or decision-making prior to response execution. The mapping of individual items to cognitive domains and task categories is detailed in the Supplemental Material.

Although certain MoCA subdomains, such as clock drawing, language, and abstract reasoning, were not administered in their original formats, the underlying cognitive processes associated with these domains were engaged through the interactive task structure. Specifically, visuospatial planning, semantic discrimination, and rule-based abstraction were elicited by tasks requiring visual search, decision-making, or response selection (e.g., search and cognitive processing tasks). Eye-movement measures were applied only to tasks involving visual exploration, capturing the efficiency and organization of cognitive processing at the level of task execution rather than overt verbal or motor output. Tasks primarily assessing orientation or memory without explicit visual exploration demands were included in the cognitive score analyses but were not used for eye-movement feature extraction.

During testing, all items were presented sequentially on the computer screen, and participants completed the assessment using mouse-click responses. This design allowed for both standard cognitive scoring and selective integration of eye-movement metrics for process-level analyses.

Eye-tracking device

Eye movements were recorded using the Gazepoint GP3 eye tracker (Gazepoint Research Inc.), a desktop-based, infrared eye-tracking system that operates without wearable sensors or head stabilization. The system was configured in a non–head-fixed setup designed to approximate natural computer use, with participants seated comfortably in front of a laptop computer under standard ambient lighting conditions. Participants were instructed to maintain a viewing distance of approximately 55–65 cm, consistent with the manufacturer's recommended operating range (Figure 1).

Figure 1.

Schematic illustration of the test setup. Participants were instructed to maintain a viewing distance of approximately 55–65 cm.

Prior to task initiation, a standard calibration procedure was performed for each participant to establish gaze–screen correspondence. Calibration was repeated if initial calibration quality did not meet the manufacturer's acceptance criteria. Eye-tracking data were collected at a sampling rate of 60 Hz while participants completed the customized online interactive cognitive screening tasks. The test interface was displayed on a laptop computer with a screen resolution of 1920 × 1080 pixels and a display area of approximately 34.4 × 19.4 cm. All device placement, calibration, and data acquisition procedures adhered to the manufacturer's guidelines.¹⁷

During task performance, the Gazepoint GP3 continuously captured multiple eye-movement parameters using infrared pupil-tracking technology, including fixation shifts, fixation error, target fixation position, and cursor position. To enhance robustness under non–head-fixed conditions, subsequent analyses focused on relative and aggregate eye-movement metrics rather than absolute gaze coordinates. Fixation shifts were defined as transitions between successive fixation points and were used to characterize patterns of visual attention allocation and information-processing efficiency.¹⁸ Fixation error was defined as the spatial discrepancy between the estimated fixation location and the target stimulus and was used as an index of visual localization accuracy and oculomotor control.^16,19

Basic data quality control procedures were applied prior to analysis. Eye-tracking segments exhibiting evident tracking loss, such as prolonged absence of valid gaze samples or unstable pupil detection, were excluded. Participants with insufficient usable eye-tracking data across tasks were not included in eye-movement analyses. These procedures were implemented to reduce the influence of signal noise while preserving ecological validity in a non–head-fixed testing environment. Only eye-movement features directly related to visual search and spatial localization, specifically fixation shifts and fixation error, were retained for inclusion in the analytic models.

Procedure

After obtaining written informed consent, participants’ medical records and demographic data were reviewed, including CDR and MMSE scores. Eligibility criteria were reconfirmed prior to testing. Participants subsequently completed a series of online interactive cognitive screening tasks in a quiet, well-lit environment. During task performance, eye movements were continuously recorded using a Gazepoint GP3 infrared eye tracker positioned below the laptop screen, with a viewing distance maintained between 55 and 65 cm. The tasks were administered sequentially, and participants provided responses via mouse clicks.

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of Taoyuan General Hospital, Ministry of Health and Welfare, Taiwan (IRB No. TYGH113107). The study was conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Written informed consent was obtained from all participants or their legally authorized representatives prior to participation, and participants with dementia provided assent whenever possible. All data were anonymized and handled confidentially in accordance with institutional regulations.

Statistical analysis

All analyses were performed using SPSS (version 26.0). Descriptive statistics were used to summarize participant characteristics, cognitive screening scores, and eye-tracking measures. Continuous variables are presented as means ± standard deviations, and categorical variables as frequencies and percentages. Three predictive models were constructed to assess the discriminative performance of an online interactive cognitive screening tool and eye-tracking measures: (1) a scale-only model based on the total screening score, (2) an eye-tracking–only model using gaze features, and (3) a combined model incorporating both measures. Four binary classification tasks were evaluated: CN versus cognitively impaired (MCI + dementia), CN versus MCI, CN versus dementia, and MCI versus dementia.

Model performance was evaluated using ROC curve analysis, with the area under the curve (AUC) as the primary measure of discriminative validity. The optimal cutoff score for the screening test was determined using Youden's Index to maximize the balance between sensitivity and specificity.²⁰ AUCs were compared across models to examine the incremental diagnostic value of eye-tracking measures. All statistical tests were two-tailed, with statistical significance set at p < 0.05.

Results

Discriminative performance of the online interactive cognitive test

ROC curve analysis showed that the total score of the online interactive cognitive test was able to discriminate cognitively normal older adults from cognitively impaired individuals (including MCI and dementia), with an area under the curve (AUC) of 0.877 (p < 0.001). The optimal cutoff score was 14 points (threshold = 14.5), corresponding to a sensitivity of 75.0% and a specificity of 84.8%.

In the comparison between cognitively normal older adults and individuals with MCI, the AUC was 0.834 (p < 0.001). The optimal cutoff score was 15 points (threshold = 15.5), yielding a sensitivity of 88.6% and a specificity of 66.7%. For the comparison between cognitively normal older adults and individuals with dementia, the test demonstrated an AUC of 0.914 (p < 0.001). Using the same cutoff score of 14 points (threshold = 14.5), sensitivity and specificity were 92.7% and 66.7%, respectively.

In the comparison between individuals with MCI and those with dementia, the AUC was 0.672 (p = 0.010). The optimal cutoff score of 14 points (threshold = 14.5) resulted in a sensitivity of 82.9% and a specificity of 65.7% (Table 2).

Table 2.

Discriminative performance of the online interactive cognitive test (ROC analysis).

Comparison group	AUC (95% CI)	p	Optimal cutoff	Sensitivity	Specificity
CN versus cognitively impaired	0.877(0.805–0.950)	<0.001	14	74.0	84.8
CN versus MCI	0.834(0.735–0.933)	<0.001	15	88.6	66.7
CN versus dementia	0.914(0.852–0.977)	<0.001	14	92.7	66.7
MCI versus dementia	0.672 (0.549–0.796)	0.010	14	82.9	65.7

Continuous data are shown as area under the curve (AUC) with 95% confidence intervals, and sensitivity and specificity are presented as percentages.

CN: cognitively normal; MCI: mild cognitive impairment.

Cognitively impaired refers to participants with either MCI or dementia.

Discriminative performance of eye-movement features

Overall task

In the overall task condition, fixation shifts demonstrated discriminative ability in distinguishing cognitively normal older adults from cognitively impaired individuals (including MCI and dementia), with an area under the curve (AUC) of 0.797 (p < 0.001). The highest discriminative performance was observed in the comparison between cognitively normal individuals and those with MCI (AUC = 0.825), with a sensitivity of 88.6% and a specificity of 69.7% (Table 3).

Table 3.

Discriminative validity of fixation shifts and fixation error in the overall task (ROC analysis).

Eye-movement feature	Comparison group	AUC (95% CI)	p	Optimal cutoff	Sensitivity	Specificity
Fixation shifts	CN versus cognitively impaired	0.797 (0.702–0.891)	<0.001	63.53	81.6	72.7
	CN versus MCI	0.825 (0.719–0.913)	<0.001	61.4	88.6	69.7
	CN versus dementia	0.772 (0.662–0.883)	<0.001	63.53	80.5	72.7
	MCI versus dementia	0.481 (0.350–0.613)	0.778
Fixation error	CN versus cognitively impaired	0.641 (0.520–0.763)	0.020	79.15	72.4	57.6
	CN versus MCI	0.610 (0.474–0.745)	0.121
	CN versus dementia	0.668 (0.539–0.798)	0.013	79.15	80.5	57.6
	MCI versus dementia	0.578 (0.446–0.710)	0.245

Values represent AUC estimates from ROC analysis. Optimal cutoff, sensitivity, and specificity are shown only for statistically significant models.

CN: cognitively normal; MCI: mild cognitive impairment; AUC: area under the curve; CI: confidence interval.

Cognitively impaired refers to individuals with either MCI or dementia.

Search task

In the search task, fixation shifts continued to discriminate cognitively normal individuals from those with MCI, yielding an AUC of 0.814 (p < 0.001). In addition, gaze deviation showed discriminative ability in distinguishing cognitively normal individuals from cognitively impaired participants, with an AUC of 0.704 (p = 0.001) (Table 4).

Table 4.

Discriminative validity of fixation shifts and fixation error in the search task (ROC analysis).

Eye-movement feature	Comparison group	AUC (95% CI)	p	Optimal cutoff	Sensitivity	Specificity
Fixation shifts	CN versus cognitively impaired	0.772 (0.679–0.866)	<0.001	74.89	77.6	66.7
	CN versus MCI	0.814 (0.709–0.920)	<0.001	74.89	85.7	66.7
	CN versus dementia	0.736 (0.622–0.851)	0.001	71.36	78.0	60.6
	MCI versus dementia	0.493 (0.361–0.626)	0.921
Fixation error	CN versus cognitively impaired	0.704 (0.582–0.826)	0.001	55.56	84.2	54.5
	CN versus MCI	0.672 (0.538–0.806)	0.015	55.59	80.0	54.5
	CN versus dementia	0.731 (0.608–0.854)	0.001	55.62	87.8	54.5
	MCI versus dementia	0.550 (0.418–0.682)	0.456

Values represent AUC estimates from ROC analysis. Optimal cutoff, sensitivity, and specificity are shown only for statistically significant models.

CN: cognitively normal; MCI: mild cognitive impairment; AUC: area under the curve; CI: confidence interval.

Cognitively impaired refers to individuals with either MCI or dementia.

Memory task

In the memory task, fixation shifts demonstrated discriminative validity in differentiating cognitively normal individuals from those with MCI, with an AUC of 0.733 (p = 0.001) (Table 5).

Table 5.

Discriminative validity of fixation shifts in the memory task (ROC analysis).

Comparison group	AUC (95% CI)	p	Optimal cutoff	Sensitivity	Specificity
CN versus MCI	0.733 (0.609–0.857)	0.001	54.17	74.3	69.7

Values represent AUC estimates from ROC analysis.

CN: cognitively normal; MCI: mild cognitive impairment; AUC: area under the curve; CI: confidence interval.

Cognitive processing task

In the cognitive processing task, fixation shifts also differentiated cognitively normal individuals from those with MCI, yielding an AUC of 0.791 (p < 0.001) (Table 6).

Table 6.

Discriminative validity of fixation shifts in the cognitive processing task (ROC analysis).

Comparison group	AUC (95% CI)	p	Optimal cutoff	Sensitivity	Specificity
CN versus cognitively impaired	0.767 (0.666–0.868)	<0.001	46.70	82.7	66.7
CN versus MCI	0.791 (0.675–0.908)	<0.001	46.0	88.6	66.7
CN versus dementia	0.745 (0.628–0.863)	<0.001	49.75	77.5	72.7
MCI versus dementia	0.489 (0.675–0.908)	0.869

Values represent AUC estimates from ROC analysis. Optimal cutoff, sensitivity, and specificity are shown only for statistically significant models.

CN: cognitively normal; MCI: mild cognitive impairment; AUC: area under the curve; CI: confidence interval.

Cognitively impaired refers to individuals with either MCI or dementia.

Discriminative performance of the combined model integrating the online cognitive test and eye-movement features

The screening performance of the combined model integrating the online interactive cognitive test with eye-movement features across cognitive-status comparisons is presented in Table 7.

Table 7.

Classification performance of the combined online interactive test and eye-movement features across cognitive status groups.

Comparison group	Significant predictors	p	Model fit (Nagelkerke R²)	AUC
CN versus cognitively impaired	Fixation shifts ↑, fixation error ↑, online interactive test score ↓	<0.001	0.588	0.897
CN versus MCI	Fixation shifts ↑	<0.001	0.529	0.868
CN versus dementia	Fixation shifts ↑, fixation error ↑, online interactive test score ↓	<0.001	0.666	0.925
MCI versus dementia	None	0.088	0.110	0.675

Arrows indicate the direction of association with a higher likelihood of cognitive impairment (↑ = positive association; ↓ = negative association).

CN: cognitively normal; MCI: mild cognitive impairment; AUC: area under the curve.

Cognitively impaired refers to individuals with either MCI or dementia.

Cognitively normal versus cognitively impaired

The overall model was statistically significant (p < 0.001). In logistic regression, fixation shifts (overall task) and gaze deviation (search task) were positively associated with cognitive impairment, whereas higher cognitive test scores were negatively associated. ROC analysis showed AUCs of 0.867, 0.877, and 0.897 for the cognitive-test–only, eye-movement–only, and combined models, respectively.

Cognitively normal versus mild cognitive impairment

This model was statistically significant (p < 0.001). Fixation shifts in the overall task were significantly associated with cognitive status (p = 0.006). AUCs were 0.834 for the cognitive-test–only model, 0.862 for the eye-tracking–only model, and 0.868 for the combined model.

Cognitively normal versus dementia

The model reached significance (p < 0.001). The cognitive test score was negatively associated with cognitive status (p = 0.002). AUCs were 0.914 (cognitive test), 0.864 (eye tracking), and 0.925 (combined).

Mild cognitive impairment versus dementia

This comparison did not reach statistical significance (p = 0.088). AUCs were 0.672, 0.597, and 0.675 for the cognitive-test–only, eye-tracking–only, and combined models, respectively.

Discussion

In this study, we demonstrated that integrating eye-movement measures with an online interactive cognitive task improves sensitivity to early cognitive vulnerability in older adults. The combined multimodal model outperformed single-modality approaches, underscoring the value of complementing conventional outcome-based cognitive performance with process-level eye-movement markers. These findings suggest that eye-movement–derived measures capture subtle alterations in cognitive processing that may not be fully reflected by end-point test scores alone. Importantly, this approach is intended to support psychogeriatric cognitive screening and clinical triage by providing complementary process-level information, rather than to replace comprehensive neuropsychological evaluation or biomarker-based diagnostic procedures.

The online interactive cognitive task alone effectively distinguished cognitively normal older adults from those with cognitive impairment, consistent with prior evidence supporting the utility of computerized cognitive assessments for early cognitive evaluation.²¹ However, its reduced discriminative performance in differentiating MCI from dementia suggests that reliance on a single global performance score may be insufficient to capture the heterogeneity and continuum of cognitive impairment in older adults. This finding highlights the limited sensitivity of end-point performance measures to underlying process-level cognitive inefficiencies, particularly during transitional stages of impairment.²² Importantly, this pattern is likely to be clinically meaningful rather than purely methodological. As cognitive impairment progresses beyond the prodromal stage, overlapping neurodegenerative changes may attenuate the ability of both performance-based scores and process-level markers to further differentiate disease severity. Furthermore, the modest specificity observed here reflects the inherent trade-off in screening tools, which prioritize high sensitivity to minimize the risk of missing early-stage decline. While inter-individual variability in education level or sensory factors may also influence performance, these process-level markers remain highly sensitive to initial cognitive shifts. Accordingly, eye-movement–derived measures appear most informative for early-stage characterization, particularly in distinguishing cognitively normal aging from MCI.

Eye-movement analyses focused on tasks requiring visual search and decision-making, thereby capturing process-level cognitive engagement relevant to attentional control and executive functioning. Among the examined parameters, fixation-shift dynamics emerged as the most sensitive and stable indicator of early cognitive vulnerability, reflecting subtle disruptions in the efficiency and organization of visual exploration during task performance. These alterations likely reflect reduced flexibility in visual–attentional sampling and executive control, cognitive processes supported by frontoparietal networks that have been implicated in prior neuroimaging studies and are known to be vulnerable during early cognitive decline in aging.

In contrast, fixation error appeared to be more strongly influenced by task demands and cognitive load, potentially reflecting later-stage impairments in attentional control and spatial targeting under increased processing demands.^23–25 Although fixation error demonstrated discriminative ability in visually demanding search tasks, its performance was less consistent across task types, suggesting greater susceptibility to task demands and cognitive load rather than stable early-stage cognitive vulnerability. Tasks that did not involve explicit visual exploration were therefore excluded to ensure that eye-movement metrics captured meaningful process-level cognitive behavior rather than task-specific or motor-related artifacts.

Importantly, integrating cognitive performance with eye-movement features improved discriminative performance beyond either modality alone, particularly in distinguishing cognitively normal individuals from those with MCI and from cognitively impaired participants overall.^26–28 This complementarity highlights the value of combining outcome-based cognitive performance with process-level behavioral indicators to improve sensitivity to early cognitive vulnerability, which may not be fully reflected by end-point test scores alone.

From a clinical perspective, eye-movement–derived process markers offer several advantages. These measures are noninvasive, objective, and relatively independent of language and educational background, addressing key limitations of traditional cognitive assessments. While not intended to replace established instruments, integrating standardized cognitive tasks with concurrent eye-movement recording may augment existing approaches by supporting psychogeriatric assessment, clinical characterization, early risk stratification, and monitoring in outpatient and community-based settings.^16,19,29

Future research should prioritize longitudinal validation to determine whether eye-movement–derived process-level measures are associated with cognitive trajectories or predict progression from MCI to dementia. Further optimization of task design and analytic strategies may enhance robustness, interpretability, and scalability in real-world psychogeriatric clinical and research settings. By bridging outcome-based cognitive performance with process-level behavioral assessment, this approach aligns with emerging calls to move beyond total scores toward more mechanistic markers of early AD-related cognitive vulnerability.

Limitations

Several limitations warrant consideration. First, certain components of the online cognitive screening task may have limited sensitivity to the earliest stages of cognitive decline. For example, the naming task primarily employed highly familiar stimuli (e.g., common animals), which may allow individuals with MCI to maintain high accuracy, thereby limiting sensitivity to subtle impairments in early semantic retrieval. In addition, the fixed structure of the visual search paradigm may have constrained variability in eye-movement behavior, potentially reducing sensitivity to subtle process-level inefficiencies. This design may have limited participants’ visual exploration strategies and may partially explain why some eye-movement parameters did not show significant group differences. Future studies may improve sensitivity by incorporating adaptive and difficulty-adjusted task designs. Task complexity could be systematically manipulated—for example, by introducing concurrent working memory demands during visual search, increasing target uncertainty, adding distractors, or implementing multi-target search conditions. Such modifications may enhance variability in eye-movement patterns and improve the detection of early cognitive vulnerability. These limitations are particularly relevant in early-stage conditions, where performance accuracy may remain preserved despite underlying process-level inefficiencies.

Second, cognitively normal participants were younger and more highly educated than those with MCI or dementia. Although eye-movement measures are generally less influenced by educational background than conventional cognitive scores, residual confounding by age and education cannot be fully excluded and should be addressed through age-matched designs or statistical adjustment in future work. Age and education were not included as covariates in the primary ROC analyses, which focused on overall discriminative performance rather than etiologic inference. Future studies incorporating age-matched samples or covariate-adjusted models are warranted to further clarify the independent contribution of eye-movement measures.

Third, participants with severe dementia were not included, limiting generalizability to later disease stages. Finally, the cross-sectional design precludes inference regarding cognitive trajectories or disease progression.

These factors, together with the inherent overlap in cognitive and neurodegenerative processes across early disease stages, particularly between cognitively normal aging and MCI, may contribute to the relatively lower specificity observed in some classification tasks, which is consistent with the intended role of the model as a screening and triage tool that prioritizes sensitivity to early-stage cognitive vulnerability.

Despite these limitations, the present findings provide clinically meaningful proof-of-concept evidence that integrating eye-movement measures into interactive cognitive screening improves sensitivity to early cognitive vulnerability in older adults beyond conventional performance-based scores. This multimodal approach may be particularly relevant for psychogeriatric screening and triage, where subtle alterations in cognitive processing may precede measurable declines in total scores. With further longitudinal validation, such frameworks may support scalable screening, early risk stratification, and monitoring in outpatient and community-based settings, including primary care contexts.

Conclusion

In conclusion, fixation-shift–derived eye-movement metrics, when integrated with computerized cognitive assessment, improve sensitivity to early cognitive vulnerability in older adults by bridging outcome-based performance with process-level behavioral assessment. This complementary approach provides objective insight into cognitive processes that may not be captured by total scores alone. While not intended to replace established instruments, it may augment existing assessment strategies by supporting psychogeriatric screening, early risk stratification, and monitoring in clinical and community-based settings, particularly where access to biomarker-based assessment is limited.

Supplemental Material

sj-docx-1-alz-10.1177_13872877261457344 - Supplemental material for Process-level eye-movement dynamics enhance sensitivity to early Alzheimer's disease-related cognitive vulnerability

Supplemental material, sj-docx-1-alz-10.1177_13872877261457344 for Process-level eye-movement dynamics enhance sensitivity to early Alzheimer's disease-related cognitive vulnerability by Ping-Yu Chen, Wann-Yun Shieh, Li-Cheng Weng, Hsiu-Hsin Tsai, Wei-Xiang Wang and Chia-Yih Liu in Journal of Alzheimer's Disease

Footnotes

Acknowledgements

We thank all participants and their families for their time and cooperation. We also acknowledge the assistance of the staff at participating regional hospitals with neurology or geriatric outpatient services, as well as community-based settings providing older adult day care or community support services for their support in participant recruitment and data collection.

ORCID iDs

Ping-Yu Chen

Hsiu-Hsin Tsai

Ethical considerations

This study was approved by the Institutional Review Board of Taoyuan General Hospital, Ministry of Health and Welfare, Taiwan (IRB No. TYGH113107).

Consent to participate

The study was conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Written informed consent was obtained from all participants or their legally authorized representatives prior to participation, and participants with dementia provided assent whenever possible. All data were anonymized and handled confidentially in accordance with institutional regulations.

Consent for publication

Not applicable.

Author contribution(s)

Ping-Yu Chen: Conceptualization; Data curation; Formal analysis; Methodology; Project administration; Software; Writing – original draft; Writing – review & editing.

Wann-Yun Shieh: Conceptualization; Data curation; Formal analysis; Methodology; Supervision; Writing – review & editing.

Li-Cheng Weng: Conceptualization; Methodology; Writing – review & editing.

Hsiu-Hsin Tsai: Conceptualization; Funding acquisition; Methodology; Supervision; Writing – review & editing.

Wei-Xiang Wang: Conceptualization; Methodology; Writing – review & editing.

Chia-Yih Liu: Methodology; Supervision; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Science and Technology Council (grant number NSTC 113-2314-B-182-035-MY3) and the Chang Gung Memorial Hospital (grant number BMRP849).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data that support the findings of this study are not publicly available due to ethical and privacy restrictions but are available from the corresponding author upon reasonable request.

Supplemental material

Supplemental material for this article is available online.

References

World Health Organization. Global status report on the public health response to dementia. https://www.who.int/publications/i/item/9789240033245 (2021, accessed 26 April 2025).

Ministry of Health and Welfare (Taiwan). National community-based epidemiological survey of dementia in Taiwan, 2024. (in Chinese), https://www.mohw.gov.tw/cp-6653-78102-1.html (2024, accessed 26 April 2025).

Taiwan Alzheimer’s Disease Association. Estimated number of people with dementia in Taiwan, 2023. (in Chinese), https://www.tada2002.org.tw/Messages/Content?Id=1207 (2023, accessed 26 April 2025).

Panegyres

Berry

Burchell

. Early dementia screening. Diagnostics 2016; 6: 6.

Sabbagh

Boada

Borson

, et al. Early detection of mild cognitive impairment (MCI) in an at-home setting. J Prev Alzheimers Dis 2020; 7: 171–178.

Chun

Seward

Patterson

, et al. Evaluation of available cognitive tools used to measure mild cognitive decline: a scoping review. Nutrients 2021; 13: 3974.

Trzepacz

Hochstetler

Wang

, et al. Relationship between the Montreal Cognitive Assessment and Mini-mental State Examination for assessment of mild cognitive impairment in older adults. BMC Geriatr 2015; 15: 07.

Cova

Nicotra

Maestri

, et al. Translations and cultural adaptations of the Montreal cognitive assessment: a systematic and qualitative review. Neurol Sci 2022; 43: 113–124.

Hannonen

Andberg

Kärkkäinen

, et al. Shortening of saccades as a possible easy-to-use biomarker to detect risk of Alzheimer's disease. J Alzheimers Dis 2022; 88: 609–618.

10.

Tadokoro

Yamashita

Fukui

, et al. Early detection of cognitive decline in mild cognitive impairment and Alzheimer's disease with a novel eye tracking test. J Neurol Sci 2021; 427: 117529.

11.

Tsai

Lee

Wang

, et al. Psychometrics of the Montreal Cognitive Assessment (MoCA) and its subscales: validation of the Taiwanese version of the MoCA and an item response theory analysis. Int Psychogeriatr 2012; 24: 651–658.

12.

Morris

. The clinical dementia rating (CDR): current version and scoring rules. Neurology 1993; 43: 2412–2414.

13.

Folstein

McHugh

. "Mini-mental state.” A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12: 189–198.

14.

Nasreddine

Phillips

Bédirian

, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc 2005; 53: 695–699.

15.

Zhang

Wang

, et al. Development of a computerized tool for the Chinese version of the Montreal cognitive assessment for screening mild cognitive impairment. Int Psychogeriatr 2015; 27: 213–219.

16.

Tokushige

Matsumoto

Matsuda

, et al. Early detection of cognitive decline in Alzheimer's disease using eye tracking. Front Aging Neurosci 2023; 15: 1123456.

17.

Gazepoint. Gazepoint Control User Manual. 2021.

18.

Rayner

. Eye movements in Reading and information processing: 20 years of research. Psychol Bull 1998; 124: 372–422.

19.

Rane

Dash

Dutt

, et al. Distinctive visual tasks for characterizing mild cognitive impairment and dementia using oculomotor behavior. Front Aging Neurosci 2023; 15: 1125651.

20.

Youden

. Index for rating diagnostic tests. Cancer 1950; 3: 32–35.

21.

Cho

Yang

Kim

, et al. The validity and reliability of a Computerized Dementia Screening Test developed in Korea. J Neurol Sci 2002; 203-204: 109–114.

22.

Ashford

Schmitt

Bergeron

, et al. Now is the time to improve cognitive screening and assessment for clinical and research advancement. J Alzheimers Dis 2022; 87: 305–315.

23.

Liu

Yang

, et al. The effectiveness of eye tracking in the diagnosis of cognitive disorders: a systematic review and meta-analysis. PLoS One 2021; 16: e0254059.

24.

Lian

Guo

, et al. Abnormal eye movements: relationship with clinical symptoms and predictive value for Alzheimer’s disease. Front Aging Neurosci 2024; 16: 1471698.

25.

Shah

Khalil

Andberg

, et al. Eye tracking based detection of mild cognitive impairment: a review. Inf Fus 2025; 122: 103202.

26.

Garcia-Alvarez

Gomar

Sousa

, et al. Breadth and depth of working memory and executive function compromises in mild cognitive impairment and their relationships to frontal lobe morphometry and functional competence. Alzheimers Dement (Amst) 2019; 11: 170–179.

27.

Johansson

. Look here, eye movements play a functional role in memory retrieval. Psychol Sci 2014; 25: 236–242.

28.

Crutcher

Calhoun-Haney

Manzanares

, et al. Eye tracking during a visual paired comparison task as a predictor of early dementia. Am J Alzheimers Dis Other Demen 2009; 24: 258–266.

29.

Wolf

Tripanpitak

Umeda

, et al. Eye-tracking paradigms for the assessment of mild cognitive impairment: a systematic review. Front Psychol 2023; 14: 1197567.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB