A systematic review and meta-analysis on dual-task sensor-based motion analysis for dementia detection.

Introduction: Early diagnosis of dementia may be improved by objective, scalable tests that capture how cognitive tasks interfere with movement. This study examined the use of instrumented dual-task paradigms for dementia detection and characterisation. Methods: We performed a PRISMA-guided systematic review and meta-analysis of peer-reviewed studies that used dual-task paradigms in adults with clinically defined dementia and an appropriate comparator. We extracted primary motor tasks, secondary cognitive or motor loads, sensor modalities, and analytic approaches. Walking outcomes were meta-analysed using inverse-variance weighted random-effects models, including subgroup analyses for single-task versus dual-task conditions and for arithmetic versus memory and verbal fluency assessments. Results: The literature was dominated by cognitive-motor dual-task paradigms in Alzheimer's disease cohorts. Inertial measurement units and force plates were the most common instruments, and most studies used classical statistics, with fewer applying machine learning. Pooled effects showed consistent group differences; compared with controls, people with dementia walked more slowly, took shorter steps, and showed less steady timing. Although heterogeneity was substantial across studies, the direction of effects was stable, and dual-task conditions generally amplified group differences relative to single-task performance. Arithmetic loads tended to accentuate changes linked to speed and cadence, whereas memory and verbal fluency assessments tended to prolong timing measures. Balance, turning, and some upper-limb outcomes also differentiated groups. Discussion: Instrumented dual-task assessments appear to enhance detection of cognitive-motor impairment in dementia and may complement existing evaluations. To support clinical translation, future work should extend beyond Alzheimer's disease, standardise task instructions and reporting, and evaluate multi-modal, validated analytic approaches across different dementia subtypes. Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251114199, CRD420251114199.