Recently, a growing number of studies have focused on digital biomarkers for the early detection and intervention in preclinical stages of mild cognitive impairment (MCI) and dementia due to their advantages, such as low cost, time efficiency, and high accessibility [1,2]. Traditional detection methods include imaging assessments (eg, positron emission tomography) and biological evaluations (eg, cerebrospinal fluid analysis), but these methods have limitations such as being invasive, inconvenient, time-consuming, expensive, and difficult to access [3]. Furthermore, while clinical and neuropsychological assessments remain key standards, they also have limitations, including long testing intervals to prevent learning effects [4], reliance on self-reporting, and interevaluator variability [5]. In contrast, digital biomarkers are defined as objective and quantifiable physiological and behavioral data collected through digital devices, enabling frequent or continuous evaluation with minimal disruption to daily life. This allows for objective, ecologically valid, and long-term follow-up in preclinical detection processes [6-9]. Developing new digital biomarkers that fulfill these requirements is important.

Research on digital biomarkers based on behavioral data has primarily focused on gait or hand movement data [10]. In particular, hand movement-based studies are gaining attention due to their relative advantages compared to gait, such as requiring less space [11,12], ensuring physical safety for participants (ie, gait poses a risk of falls), and using low-cost data measurement devices. However, previous studies have either measured imitation accuracy of predefined static hand movements using artificial intelligence (AI) [13] or collected dynamic hand movement data similar to daily activities using virtual reality (VR) devices [14,15]. These methods face limitations in overcoming the drawbacks of traditional biomarkers. Challenges include the occurrence of learning effects from repetitive presentation of fixed movements in repeated measurement processes, the high cost of VR devices, and psychological barriers stemming from the use of unfamiliar equipment. As a result, these methods lack sufficient validity verification required for their application as digital biomarkers in cognitive ability assessments.

To address these limitations, we used a hand rotation-based digital biomarker inspired by the correlation between upper limb motor skill capability and cognitive function. Numerous prior studies have demonstrated that cognitive impairment leads to a decline in upper limb motor abilities [16-19]. The hand rotation-based movement reflects hand motion control ability that has minimal learning effects [13] and can be measured by images with an AI-based measurement system to collect quantitative data, thereby minimizing interevaluator variability [20]. To quantitatively measure hand rotation data, several measures can be used, such as total rotation count, total rotation time, and symmetry [21], as well as derived indicators, such as changes in amplitude and variations in performance time. Before conducting experiments on patients with MCI or dementia, we investigated if the hand rotation-based digital biomarker could identify different age groups, based on the understanding that healthy aging induces neurochemical and structural changes in older adults [22], which consequently impacts their motor performance, leading to reduced motor control and coordination [23,24].

To achieve this, our objective is to validate the proposed hand rotation-based digital biomarker to distinguish differences between a young adult group and an older adult group. We used a mixed design to investigate the effects of age, hand dominance, and repeated trials on motor performance. The significance of this research lies in overcoming the economic and spatial limitations of existing behavior-based digital biomarkers and developing a more accessible digital biomarker. Additionally, this study provides direction and insights for future digital biomarker development.

This study used a mixed-design method to investigate the effects of age, hand dominance, and repeated trials on motor performance. The experiment included both between-subject and within-subject factors. The between-subject factor was age group, with participants divided into 2 groups: younger adults (ages 20-29 years) and older adults (ages 65-80 years). This design allows for the examination of age-related differences in motor function. The within-subject factors included hand (right vs left) and trial (3 repeated trials). Each participant completed the task using both hands. The task was repeated 3 times (trial 1, trial 2, and trial 3) to assess potential changes in performance due to repetition, learning, or fatigue. Participants were asked to perform a hand rotation task, in which they rotated their hands for a total of 10 seconds. The outcome variables measured are explained in Table 1. These measures were collected across both hands and trials for each participant, providing a comprehensive assessment of motor function. By using a mixed design, this study allowed for the analysis of both between-subject effects (comparing younger and older participants) and within-subject effects (comparing performance across hands and trials), as well as their interactions. The design also enabled us to investigate whether performance improvements across trials, indicative of a learning effect, differed between the age groups.

Based on the video-captured data, we have defined 7 measurements and automatically calculated them by the system (Table 1). The single-hand test indicator assesses wrist flexibility through rotational movements. The time comparison indicator between rotations evaluates changes in hand movement over time, and the angle comparison indicator between rotations assesses changes in hand angle over time. The single-hand test indicators include total rotation count, total rotation time, and total rotation angle. The time comparison indicators between rotations include total rotation time change and the number of rotation angle changes, while the angle comparison indicators between rotations include total rotation angle change and the number of rotation angle changes. Here, the time of rotation change amount represents the total time variation for performing 1 full rotation. Simply put, if the time of rotation change amount is negative, the rotation speed increases as the number of rotations increases; if positive, the rotation speed decreases. The angle of rotation change amount indicates the unit of change in rotation angle. If negative, the rotation angle performed by the participant decreases as the number of rotations increases; if positive, the rotation angle gradually increases [29].

This study recruited a total of 70 participants (31 young adults and 39 older adults). In total, 2 young adults were excluded based on the PPT exclusion criteria, resulting in a final sample of 68 participants: 39 (57.35%) in the experimental group and 29 (42.65%) in the control group. The analysis results showed that the average age of the young adults (20-29 years old) was a mean of 22.75 (SD 1.83) years, and the average age of the older adults (65-80 years old) was a mean of 70.36 (SD 3.70) years. Based on the dominant hand, there were 6 left-handed individuals (20.69%) and 23 right-handed individuals (79.31%) among the young adults, while there was 1 (2.78%) left-handed individual and 44 (97.22%) right-handed individuals among the older adults. In terms of educational level, there were 3 (10.34%) highly educated individuals (university graduates or higher) among the young adults and 6 (15.38%) among the older adults (refer to Table 3). This demographic information provides a basis for evaluating the impact of variables such as age, dominant hand, and educational level on hand motion control ability.

A GEE model was used to assess the effects of age group (younger vs older), hand (right vs left), and trials (1, 2, and 3). Table 4 shows the representative statistics for all the measurements that were recorded to give an overview of the results. Among the measures we have defined, only 3 measures (ie, total rotation count, total rotation time, and total rotation angle) have shown a significant effect on the main factors that we have designed. The analysis results are explained for the 3 measurements, including all the interaction effects, are provided in the main text in the following sections.

For the total rotation count, the model revealed a significant main effect of age group. Younger participants performed significantly more full rotations compared to older participants (B=5.29, SE 1.67, z=3.16, P=.002). Specifically, younger participants completed 5.29 more rotations on average than their older counterparts, suggesting that age significantly influences task performance. A significant main effect of trial was also observed, indicating that performance improved across trials. Participants in trial 2 completed 1.77 more rotations compared to trial 1 (B=1.77, SE 0.59, z=3.00, P=.003), and participants in trial 3 completed 2.31 more rotations compared to trial 1 (B=2.31, SE 0.72, z=3.20, P=.001). These findings suggest a learning effect across trials, with participants demonstrating better performance in subsequent trials. The main effect of the hand was not significant (B=–1.00, SE 0.87, z=–1.15, P=.25), indicating that there was no significant difference between the performance of the right and left hands.

Finally, to explore the significant main effect of trial, post hoc pairwise comparisons were conducted between trial 1, trial 2, and trial 3 using Bonferroni-corrected P values to account for multiple comparisons. The comparison between trial 1 and trial 2 revealed a statistically significant difference in performance, with participants showing an improvement from trial 1 to trial 2 (t₆₇=–4.87, P<.0001, Bonferroni-adjusted P=.000021). Similarly, a significant difference was found between trial 1 and trial 3, indicating continued improvement (t₆₇=–4.85, P<.0001, Bonferroni-adjusted P=.000023). However, no significant difference was observed between trial 2 and trial 3 (t₆₇=–0.76, P=.45, Bonferroni-adjusted P>.99), suggesting that performance plateaued between these 2 trials. These results confirm a learning effect between trial 1 and the subsequent trials, with participants significantly improving their performance from the first to the second and third trials. However, no further improvement was observed between the second and third trials.

The model revealed a significant main effect of age group, with younger participants exhibiting a significantly greater range of motion compared to older participants (B=1334.37, SE 492.99, z=2.71, P=.007). On average, younger participants had a 1334-unit larger range of motion than older participants, indicating that age significantly influences performance in this task. There was no significant main effect of hand (B=–202.30, SE 140.47, z=–1.44, P=.15), indicating no substantial difference in range of motion between the right and left hands. The main effect of the trial showed a trend toward significance for trial 2 compared to trial 1 (B=193.69, SE 101.86, z=1.90, P=.06), although this effect did not reach conventional levels of statistical significance. However, there was a significant increase in range of motion in trial 3 compared to trial 1 (B=295.09, SE 129.29, z=2.28, P=.02), suggesting that participants exhibited a larger range of motion in the third trial. For interaction effects, several interaction effects between age group, hand, and trial were also examined. The interaction between age group and hand was marginally different (B=470.23, SE 248.58, z=1.89, P=.07), suggesting that the effect of hand may differ by age group, though this did not meet the threshold for statistical significance. The interaction between age group and trial was significant for both trial 2 (B=736.10, SE 263.73, z=2.79, P=.005) and trial 3 (B=644.74, SE 317.01, z=2.03, P=.042). These results indicate that the younger participants exhibited a significantly larger increase in range of motion across trials compared to older participants, suggesting a learning effect for the younger group as they improved more across trials. Posthoc pairwise comparisons were conducted to further explore the significant main effect of trial, with Bonferroni-corrected P-values to account for multiple comparisons. The comparison between trial 1 and trial 2 revealed a statistically significant improvement in performance, with participants showing better results in trial 2 compared to trial 1 (t₆₇=–3.68, P<.001, Bonferroni-adjusted P=.0014, Hedges g=–0.19). This indicates a meaningful improvement from trial 1 to trial 2. A significant difference was also found between trial 1 and trial 3, with participants continuing to improve (t₆₇=–3.37, P=.0013, Bonferroni-adjusted P=.0038, Hedges g=–0.21). This suggests that participants maintained better performance in trial 3 compared to trial 1.

However, no significant difference was observed between trial 2 and trial 3 (t₆₇=–0.41, P=.68, Bonferroni-adjusted P≥.99, Hedges g=–0.01), indicating that performance plateaued between these 2 trials. These results confirm a learning effect between trial 1 and the subsequent trials, where participants significantly improved from trial 1 to both trial 2 and trial 3. However, no further improvement was observed between trial 2 and trial 3.

The model revealed a significant main effect of age group, indicating that younger participants took significantly longer on the task compared to older participants. Specifically, younger participants had a mean increase of 1.40 seconds in total task time compared to older participants (B=1.40, SE 0.35, z=3.94, P<.001). This suggests that age significantly affects the time taken to complete the rotations.

There was no significant main effect of hand (B=–0.30, SE 0.23, z=–1.33, P=.18), indicating no meaningful difference in time between the right and left hands.

The main effect of the trial was marginally different for trial 2 compared to trial 1, with participants taking approximately 0.33 seconds longer in trial 2 (B=0.33, SE 0.19, z=1.69, P=.09). However, this effect did not reach the conventional level of statistical significance. Similarly, the effect of trial 3 compared to trial 1 was not significant (B=0.32, SE 0.25, z=1.28, P=.20), suggesting that the overall time taken did not significantly change between the trials.

Research on digital biomarkers based on behavioral data has primarily focused on gait and hand movement data [10]. Among these, hand movement tasks have gained attention due to their minimal spatial requirements [11,12], absence of fall risk, and the use of low-cost measurement devices. This study investigated whether hand rotation—a task with these advantages—could effectively distinguish age-related differences in motor function.

Accordingly, we hypothesized that there would be significant differences between the young adult group and the older adult group. The results from the GEE analysis demonstrated clear distinctions between younger and older participants across the 3 measures: total rotation count, total rotation angle, and total rotation time. Younger participants consistently outperformed older participants, completing more rotations and exhibiting a greater range of motion, which highlights significant age-related differences in motor efficiency and flexibility. Although younger participants took slightly longer to complete the task, this may reflect a more controlled approach to performing the rotations. Importantly, all 3 measures were effective in distinguishing age groups, proving especially valuable in capturing the declines in motor performance associated with aging. Together, these measures provide a comprehensive profile of motor function, with total rotation count reflecting speed and efficiency, total rotation angle highlighting flexibility, and total rotation time capturing pacing strategies, all of which vary significantly between younger and older adults.

The age-related differences observed in hand rotation metrics likely reflect a combination of cognitive and physical factors. This biomarker, inspired by the link between upper limb motor skills and cognitive function, aligns with prior findings showing that cognitive impairment often accompanies motor declines. Tasks like hand rotation, which require controlled, repetitive movements, depend on cognitive functions such as coordination, spatial awareness, and consistent adjustments—areas that can be affected by age-related cognitive decline. Reduced cognitive capacity could thus contribute to the lower motor performance observed in the older adults group. At the same time, physical fitness factors, including muscle strength and joint flexibility, also decline with age and impact performance on repetitive motor tasks. Although the hand rotation task is low in exertion, it demands a baseline level of physical capability, particularly for maintaining consistent movements across trials. Therefore, this biomarker likely captures influences from both cognitive and physical domains, with declines in either potentially diminishing performance. Future research might benefit from a dual approach, examining both physical fitness and cognitive performance measures in conjunction with this motor biomarker to clarify their relative contributions. This approach could help identify the extent to which cognitive versus physical decline drives the observed performance differences and potentially enhance the biomarker's specificity for early detection of age-related cognitive impairment.

The results clearly demonstrated a significant learning effect across trials, particularly for younger participants, who showed substantial improvement from trial 1 to trial 2 and trial 3 in both full rotations and range of motion. The lack of a significant difference between trial 2 and trial 3 suggests that participants reached a performance plateau by the second trial. This indicates that trial 1 likely served as a familiarization or practice phase, during which participants adapted to the task demands. To improve the accuracy of future analyses and minimize the confounding effect of learning, it would be advisable to treat the first trial as a practice trial and focus analysis on trial 2 and trial 3 as representative of stable performance. By doing so, researchers can better isolate genuine performance differences between age groups, unconfounded by initial adaptation or learning.

This study has several limitations. First, the sample size used in the study was limited, which may restrict the generalizability of the results. Future research should increase the sample size to enhance the reliability of the findings. Additionally, this study was cross-sectional, analyzing data at a single point in time. To more accurately understand changes associated with age, longitudinal studies that track changes over time are needed. As the first step in validating digital biomarkers, this study conducted experiments on younger and older adults through hand rotation movements prior to MCI and dementia diagnoses. Although we informed that we were recruiting participants without MCI, we did not conduct an additional screening test. In subsequent phases, we plan to collaborate with Dong-eui University Oriental Medicine Hospital to conduct cognitive tests on older adults diagnosed with MCI and dementia, as well as older adults without cognitive impairment. Through this, we aim to explore and analyze differences in physical, physiological, and gender factors. Furthermore, future research will increase the sample size to study differences between genders and age groups and continue to explore the potential of hand rotation movements as digital biomarkers.

This study confirmed that hand rotation metrics can effectively distinguish motor performance differences between younger and older adults. The older group showed statistically significant decreases in rotation frequency, angle, and duration compared to the younger group, indicating an age-related decline in motor control abilities. Additionally, a learning effect was observed during repeated trials, suggesting that excluding the first trial may enhance measurement stability in future assessments.

These findings demonstrate the potential of hand rotation as a digital biomarker for capturing age-related changes in motor performance and further suggest its applicability as an early detection tool for MCI and other early cognitive declines. This method can be implemented across various platforms—including webcams, wearable devices, and VR systems—enabling broad applications such as clinical diagnostic support, home-based self-assessment kiosks, and community health screening. In particular, it holds promise as an important assessment tool for the early identification of neurodegenerative diseases and for the development of tailored intervention programs.

Future studies should aim to further refine this hand rotation-based digital biomarker for broader applicability. This study has some limitations, including a relatively small and demographically homogeneous sample, which may limit the generalizability of the findings. Follow-up research involving larger and more diverse populations is essential to improve the generalizability of the findings. Moreover, applying this method to individuals with MCI and other cognitive impairments will help evaluate its effectiveness in detecting early signs of cognitive decline. Expanding the framework to include other motor tasks and integrating multimodal indicators—including cognitive, physical, and emotional factors—could contribute to building a comprehensive and personalized digital health monitoring system for early diagnosis and preventive care.