Vitality Watch

Dossier pubmed observational 30-may-2026
Human cohort, case-control, or epidemiological association

Development of Machine Learning-Driven Dual-Task Gait Test Model for Cognitive Impairment Screening.

Key claim: An AI-aided dual-task gait test achieved high screening accuracy for dementia (AUC 0.883) and cognitive impairment (AUC 0.895) in community-dwelling adults aged 60 and older, with gait speed and stride variability as leading predictors.

Abstract

OBJECTIVE: To develop an artificial intelligence (AI)-aided dual-task gait test model for scalable, high-throughput cognitive impairment screening. DESIGN: A diagnostic case-control study between 2022 and 2023, community-dwelling adults aged ≥60 years capable of unaided walking were enrolled. SETTING(S): Primary care. PARTICIPANTS: A total of 201 participants were classified into 3 groups: dementia (n=26), mild cognitive impairment (n=124), and cognitively intact (n=51). All participants underwent 3 gait assessments: single-task, dual-task (serial subtraction by 7), and dual-task (animal-naming) gait test. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: A total of 45 gait parameter metrics were extracted using AI-assisted gait analysis. Cognition was adjudicated by an expert panel using cognitive tests, functional status, and clinical records. Logistic regression models were used to analyze the association between gait parameters and cognition, with area under the curve (AUC) used to evaluate diagnostic performance. Four machine learning models were employed for model optimization. Subgroup analyses were conducted to assess model fairness. RESULTS: Multiple gait parameters showed significant differences between groups, with gait speed and stride variability emerging as the strongest predictors. After adjustment for covariates, the logistic regression models demonstrated notable screening accuracy for dementia (AUC=0.883; 95% CI, 0.811-0.955) and for cognitive impairment (AUC=0.895; 95% CI, 0.850-0.941). Machine learning-based optimization further validated these findings, with the support vector machine model achieving an AUC of 0.786 (sensitivity: 83.54%, specificity: 78.18%) for dementia and an AUC of 0.808 (sensitivity: 85.54%, specificity: 76.46%) for cognitive impairment. Subgroup analyses were exploratory and constrained by small sample sizes, with results showing considerable variability. CONCLUSIONS: The AI-aided dual-task gait paradigm is a scalable, cost-effective, and highly accurate tool for population-level cognitive impairment screening in older adults. Its transformative potential lies in enabling early detection in community settings, establishing a cornerstone for proactive dementia prevention through timely interventions.

Why this matters

Dual-task gait slowing has long been linked to cognitive decline, but this is a concrete attempt to turn that association into a scalable primary-care screening tool using machine learning, with reported AUCs in the ~0.88–0.90 range. It is a trend data point for AI entering community cognitive triage via cheap functional measures rather than imaging or fluid biomarkers — the evidence remains single-cohort and observational, well short of a validated clinical screener.

Source

Link

Briefing card

Development of Machine Learning-Driven Dual-Task Gait Test Model for Cognitive Impairment Screening.
Stage: observational · pubmed · 30-may-2026

If you are 60+ and gait slowing or variability shows up on a dual-task walking test, that pattern is being used in research models to flag cognitive impairment, not only mobility risk.

An AI-aided dual-task gait test achieved high screening accuracy for dementia (AUC 0.883) and cognitive impairment (AUC 0.895) in community-dwelling adults aged 60 and older, with gait speed and stride variability as leading predictors.

Cross-references (0)

None recorded — doctrine links and citations appear here as scans and citation sweeps find them.

External: doi:10.1016/j.apmr.2026.05.013

build build 880 · 2bcef25 · 2026-11-18