An LLM-supported just-in-time adaptive intervention delivered by a socially assistive robot for cognitive training in...

Mild cognitive impairment (MCI) is a clinically important stage at which non-pharmacological interventions may help preserve cognitive function and daily functioning. Existing digital cognitive training programs are often limited by static task progression, declining engagement, and insufficient responsiveness to moment-to-moment cognitive and emotional states.

This study protocol describes a single-center, parallel-group, assessor-blinded randomized controlled trial evaluating an adaptive cognitive training intervention delivered by a socially assistive robot and supported by a large language model (LLM).

A total of 126 adults aged 60 years or older with MCI will be randomized 1:1 to either an LLM-supported robot-assisted adaptive cognitive training intervention (n = 63) or a matched non-adaptive computerized cognitive training control condition (n = 63). Both groups will receive three individually delivered sessions per week for 24 weeks, with each session lasting approximately 45-60 min, followed by a 12-week post-intervention follow-up to assess durability. The intervention group will receive multi-domain cognitive training through a socially assistive robot, with the LLM used as a bounded communication module that generates supportive language for controller-selected task adaptation, instruction style, feedback, and motivational support according to prespecified performance and affective indicators. The primary outcome is change in Montreal Cognitive Assessment (MoCA) score from baseline to week 24. Maintenance of cognitive change at week 36 will be analyzed as a secondary exploratory durability outcome. Secondary outcomes include domain-specific cognitive performance, emotional symptoms, motivation, engagement, adherence, usability, acceptance, expectancy and novelty effects, and safety. A nested exploratory imaging-biomarker sub-cohort will undergo resting-state functional magnetic resonance imaging and plasma biomarker assessment to explore changes in functional connectivity and neurobiological markers. Analyses will follow the intention-to-treat principle using mixed-effects models for repeated measures.

The protocol has been approved by the Ethical Committee of the School of Nursing and Rehabilitation, Shandong University (Approval No. 2026-R-072; approved March 10, 2026). Written informed consent will be obtained before enrollment.

https://www.chictr.org.cn/historyversionpubEN.html?regno=ChiCTR2600125870, identifier ChiCTR2600125870.