Aging & Longevity

Multidimensional modeling of biological aging: integrating gait, eye movement, rest-state functional connectivity, and plasma biomarkers in non-dementia older adults.

TL;DR

Integrating physiological, neurological, and biomarker data substantially improves biological age modeling, with a combined multimodal model achieving R2 = 0.814 and MAE = 1.902 years compared to the best single-domain model (eye movements: R2 = 0.606; MAE = 3.060).

Key Findings

A combined multimodal model integrating gait, eye movements, resting-state functional connectivity, and plasma biomarkers achieved markedly higher biological age prediction accuracy than any single-domain model.

  • The multimodal model achieved R2 = 0.814 and MAE = 1.902 years.
  • The study included 908 non-dementia older adults over 60 years of age.
  • Data modalities included gait, eye movements, resting-state functional connectivity (rs-FC), and plasma biomarkers (NfL and GFAP).
  • The multimodal model substantially outperformed the best single-domain model (eye movements: R2 = 0.606; MAE = 3.060).

Eye movement features showed the strongest predictive performance among all single-domain models for biological age estimation.

  • Eye movement single-domain model achieved R2 = 0.606 and MAE = 3.060 years.
  • Two eye movement features were significantly correlated with age (p < 0.05).
  • Eye movements outperformed gait, resting-state functional connectivity, and plasma biomarker models when used individually.

Fourteen gait features were significantly correlated with chronological age in non-dementia older adults.

  • Significance threshold was p < 0.05.
  • Gait data were collected from 908 non-dementia older adults aged over 60 years.
  • Gait features contributed to the multimodal biological age model.

Nineteen resting-state functional connectivity (rs-FC) features were significantly correlated with chronological age.

  • Significance threshold was p < 0.05.
  • rs-FC features were among the four data modalities integrated into the multimodal model.
  • rs-FC data were collected as part of the neurological assessment in the 908-participant cohort.

Plasma GFAP levels, but not NfL, were significantly correlated with chronological age.

  • GFAP (glial fibrillary acidic protein) was significantly correlated with age (p < 0.05).
  • NfL (neurofilament light chain) was not listed among the features significantly correlated with age.
  • Plasma GFAP was included as a contributing biomarker in the multimodal model.
  • Both NfL and GFAP were collected as plasma biomarkers from the 908-participant sample.

The study enrolled 908 non-dementia older adults to develop and evaluate biological age models across multiple physiological and neurological domains.

  • Participants were older adults aged greater than 60 years.
  • All participants were classified as non-dementia.
  • Data collection spanned gait, eye movements, resting-state functional connectivity, and plasma biomarkers.
  • A total of 14 gait features, 2 eye movement features, 19 rs-FC features, and plasma GFAP were significantly correlated with age (p < 0.05).

What This Means

This research suggests that biological age — how old a person's body and brain appear to be based on measurable indicators — can be estimated much more accurately when multiple types of health data are combined. The researchers collected information from 908 older adults (all over age 60 and without dementia), measuring how they walk, how their eyes move, how different brain regions communicate at rest (using functional brain imaging), and levels of two proteins in the blood associated with brain health. When all of these data types were combined into one model, it could predict biological age with a mean error of less than 2 years, compared to about 3 years when using eye movement data alone — which was the best single measure. Among the individual measures, eye movement patterns were surprisingly the most informative single predictor of biological age, outperforming gait, brain connectivity, and blood biomarkers on their own. In total, 14 walking features, 2 eye movement features, 19 brain connectivity patterns, and blood levels of a protein called GFAP were all meaningfully linked to how old a person appeared biologically. Notably, another blood protein called NfL was not significantly associated with age in this non-dementia group. This research suggests that combining different types of easily measurable physical and neurological data — rather than relying on any single test — could lead to better tools for assessing how a person is aging. Such tools could potentially help identify individuals at higher risk for age-related diseases before symptoms appear, enabling more timely and targeted preventive strategies for healthy aging.

Have a question about this study?

Citation

Lin J, Liu Y, Ouyang Z, Yang X, Zhang S, Xu T, et al.. (2026). Multidimensional modeling of biological aging: integrating gait, eye movement, rest-state functional connectivity, and plasma biomarkers in non-dementia older adults.. The journal of prevention of Alzheimer's disease. https://doi.org/10.1016/j.tjpad.2026.100566