Efficacy of different computerized cognitive training modalities in improving global cognitive function in patients with mild cognitive impairment: a network meta-analysis

Objectives To compare the relative effects of different computerized cognitive training (CCT) modalities on global cognitive function in patients with mild cognitive impairment (MCI). Methods CNKI, VIP, Wanfang Data, CBMdisc/SinoMed, PubMed, Web of Science, Embase, and the Cochrane Library were searched from inception to April 16, 2025, for randomized controlled trials. Random-effects network meta-analyses were conducted separately for outcomes assessed using the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA). Results Twenty-three randomized controlled trials were included, of which 16 contributed to the MMSE network and 13 to the MoCA network. The MMSE network showed substantial heterogeneity (τ = 4.15, τ 2 = 17.23); none of the intervention categories showed a clear statistical advantage over UC, and the 95% prediction intervals for all major comparisons crossed the null value. In the MoCA network, the network effects of CAT, TCTT, and VR + E versus UC were MD = 3.98 (95% CI 3.07–4.90), 2.73 (2.02 to 3.44), and 2.29 (0.98 to 3.60), respectively. The 95% prediction intervals for CAT versus UC and TCTT versus UC did not cross the null value, whereas that for VR + E versus UC did. The MoCA SUCRA ranking was CAT, TCTT, VR, VR + E, TCT, and UC. Sensitivity analyses using alternative pre–post-correlations did not materially change the direction or statistical interpretation of the main relative effects. The overall certainty of evidence according to GRADE ranged from low to very low. Conclusions The MMSE network was too heterogeneous to support a reliable ranking of interventions. In the MoCA network, CAT and TCTT showed relatively stable score improvements at the end of their respective intervention periods. The average effect of VR + E was statistically significant, although the true effect in future studies remains uncertain. All rankings were exploratory and represented average effects of intervention-package categories rather than effects attributable to any single component.