Effect of Dual-Task Training on Cognitive Function in Community-Dwelling Older Adults With Mild Cognitive Impairment: Sequential Multiple Assignment Randomized Trial

Background Nonpharmacological interventions are important prevention strategies for mild cognitive impairment (MCI), but effects vary significantly between individuals based on personal characteristics, while current practice relies on experience-based approaches lacking personalized, adaptive intervention strategies. Objective The objective of our study was to develop and evaluate evidence-based adaptive intervention strategies for optimizing cognitive function among older adults with MCI using a Sequential Multiple Assignment Randomized Trial (SMART) design, comparing the effectiveness of cognitive training (CT) combined with virtual reality Taichi (VRTC) versus offline Taichi (OffTC) versus control, and to identify baseline characteristics that predict treatment response for personalized intervention delivery. Methods We recruited 92 community-dwelling adults aged ≥60 years diagnosed with MCI from 3 districts in Shanghai, China. A 24-week SMART was conducted between April and December 2023. During the first stage (weeks 1-12), participants were randomly assigned to control (n=26) or intervention groups receiving CT combined with either OffTC (n=33) or VRTC (n=34). Nonresponders at week 12 were rerandomized to alternative or intensified interventions during the second stage (weeks 13-24). The primary outcome was the Memory Guard score (MGs) at 24 weeks. Dynamic treatment regimen analysis assessed optimal adaptive strategies using regression models. Results A total of 81 participants completed the trial. CT+VRTC demonstrated significantly superior cognitive improvement compared to control (5.10 MGs, 95% CI 2.93-7.27; Cohen d=1.425, 95% CI 0.785-2.060; P<.001) and CT+OffTC (3.61 MGs, 95% CI 1.71-5.51; Cohen d=1.009, 95% CI 0.461-1.560; P<.001). At 24 weeks, adjusted mean MGs were: CT+VRTC 32.9 (95% CI 31.3-34.5), CT+OffTC 29.3 (95% CI 27.7-30.9), and control 27.8 (95% CI 26.0-29.6). Dynamic treatment regimen analysis revealed VRTC-based adaptive strategies consistently outperformed static approaches, with VRTC responders achieving the highest effectiveness (5.40 MGs improvement, 95% CI 3.10-7.70; P<.001). Treatment intensification proved more effective than modality switching for nonresponders. Subgroup analyses revealed that younger participants (≤71 years), individuals with lower baseline cognitive function, and those with comorbid conditions demonstrated enhanced responsiveness, suggesting these populations may derive greater benefit from virtual reality–based approaches. Conclusions This SMART trial established the first evidence-based adaptive intervention framework for community MCI prevention, demonstrating that virtual reality–enhanced Taichi combined with CT produced superior outcomes compared with traditional exercise and control conditions. Treatment intensification proved more effective than modality switching for nonresponders. Diabetes showed a better response enabling personalized intervention selection. These findings provide clinicians with objective decision rules for treatment adaptation and identify high-benefit populations (younger adults, those with lower baseline cognition, and patients with metabolic comorbidities) for targeted intervention. The protocol can be implemented by community health workers within existing urban health care infrastructure, offering a scalable approach to precision-based cognitive health management in aging populations. Trial Registration Chinese Clinical Trial Registry ChiCTR2100042748; https://www.chictr.org.cn/showproj.html?proj=120941