Abstract
OBJECTIVE: To investigate which brain activity frequency of electroencephalogram (EEG)-neurofeedback training (NFT) was the most effective for enhancing working memory (WM) and episodic memory (EM) in healthy participants through network meta-analysis (NMA). METHODS: Searched PubMed, Embase, and Cochrane Library for studies published from January 1990 to January 2025. We performed Bayesian NMA, pooling continuous outcome data using the standardized mean difference effect size (ES). Global and local evaluations of inconsistency were conducted using the chi-square test, side-splitting, and loop-specific approaches. A consistency model was applied and the global approach to inconsistency showed no significance. Efficacy ranks were determined using the surface under the cumulative ranking curve (SUCRA) for each intervention. Publication bias was assessed using the comparison-adjusted funnel plot and Egger's test. Finally, sensitivity analysis confirmed our findings' robustness. RESULTS: Sixty studies were included, comprising 50 trials on WM and 24 trials on EM. While the global inconsistency analysis showed no significant inconsistency for WM (χ(2)(22) = 30.89, p = 0.10) and EM (χ(2)(10) = 13.48, p = 0.19), the consistency model exhibited the most significant difference between active control (AC) and alpha combined with working memory training (WMT) (ES of 6.64, p < 0.001) for WM, and between AC and alpha (ES of 0.84, p = 0.01) for EM. Alpha combined with WMT for WM (100%) and alpha NFT for EM (87.0%) also showed the highest efficacy according to the SUCRA. No publication bias was found for either type of memory. The sensitivity analysis for WM and EM aligns with the original results. CONCLUSION: Through NMA, alpha activity (7-13 Hz) may be a crucial frequency impacting memory. Brain activity combined with other training methods requires more robust studies for future investigation. This study registered with www.crd.york.ac.uk/prospero/ (CRD42024539656).