A comprehensive search was conducted in the following databases: PubMed, PsycINFO, Cochrane Library, Web of Science, Embase, Scopus, CINAHL, ProQuest, and China National Knowledge Infrastructure (CNKI). The search covered each database from inception through March 1, 2025. We additionally hand-searched relevant review articles and the reference lists of all included studies to identify further records.

Using PubMed as an example (full search strategies for all databases are provided in Supplementary file Search Strategy): [(“Music Therapy” OR “Music” OR “music-based intervention” OR “music intervention^*” OR “music therapy” OR “therapeutic music” OR “rhythmic auditory stimulation” OR “group singing” OR “choir sing^*” OR “music listening”)] AND [(“Cognition” OR cogniti^* OR memory OR “executive function^*” OR attention OR “processing speed”)] AND [(“Aged” OR “Aged, 80 and over” OR elder^* OR “older adult^*” OR senior^* OR geriatric^*)] AND [(randomized controlled trial OR controlled clinical trial OR random^* OR trial OR “clinical trial”)]. For the Chinese database (CNKI), the search string (in Chinese) was: AB=(“音乐” + “节律性听觉刺激” + “合唱” + “集体唱歌” + “音乐聆听”) and AB=(“老年” + “年长” + “高龄”) and AB=(“认知” + “记忆” + “注意” + “执行功能”).

Inclusion criteria: (1) study design: published randomized controlled trials (RCTs) with no language restrictions; (2) interventions: structured interventions in which musical elements constitute the core active component, such as music listening, music therapy, or music participation/engagement; (3) participants: older adults with intact baseline cognition (age ≥60 years; cognitively normal at baseline as determined by screening); (4) outcomes: cognitive function–related outcome measures pre-specified as endpoints (see the “Study selection and data extraction” section for details).

Exclusion criteria: (1) participants with dementia, psychiatric disorders, or other conditions associated with cognitive impairment; (2) duplicate publications arising from the same cohort; (3) studies with incomplete data reporting for which additional information could not be obtained from the study authors.

This review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines (Page et al., 2021). Two reviewers independently screened full texts and extracted data based on the predefined inclusion and exclusion criteria; disagreements were adjudicated by a third, independent reviewer. For each study, we extracted the following information: basic study characteristics, participant characteristics, study design, intervention(s), comparator(s), outcome measures, and participant adherence/compliance.

Following the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions on handling multi-arm trials (Higgins et al., 2024), if a study included multiple eligible intervention arms that contributed to the same pooled comparison, we combined the music arms into a single intervention group and compared it with the control group. When multiple control groups were available, we prioritized the no-intervention control; if unavailable, we selected a non-music active control.

Outcome measures included (Chiu et al., 2017; Li et al., 2011) global cognition, executive function, memory, and attention. Global cognition was operationalized as a composite integrating the Mini-Mental State Examination (MMSE) together with measures of executive function, memory, and attention. For memory, the preferred measure was the Wechsler Memory Scale—Logical Memory II (WMS–LM II), followed by the Verbal Learning and Memory Test (VLMT), the East Boston Memory test (immediate recall), and total Digit Span (Kowa et al., 2022; Witt et al., 2019; Dong et al., 2014). For executive function, the preferred measure was the Trail Making Test Part B (TMT-B), followed by verbal fluency—Category Switching (VF–Category Switching) and the Means–End Problem Solving test (MEPS) (Sánchez-Cubillo et al., 2009; Shao et al., 2014). For attention, the preferred measure was the Trail Making Test Part A (TMT-A), followed by the Stroop test—administered under three conditions (word, color, and word–color/interference)—and MixC_var, the mixed-cost index from a perceptual set-shifting task (Sánchez-Cubillo et al., 2009; Rubin and Meiran, 2005).

We assessed outcome-level risk of bias for the included randomized controlled trials using Cochrane's Risk of Bias 2.0 (RoB 2) tool (Sterne et al., 2019). Judgments were made across the five RoB 2 domains: (1) bias arising from the randomization process; (2) bias due to deviations from intended interventions; (3) bias due to missing outcome data; (4) bias in measurement of the outcome; and (5) bias in selection of the reported result. Assessments followed the tool's signaling questions and drew on the full text, Supplementary materials, and—where available—trial registry/protocol information. For each study, outcome-level judgments were issued for the pre-specified primary cognitive outcomes (e.g., executive function, task switching). Overall risk-of-bias ratings were synthesized per RoB 2 guidance (Sterne et al., 2019): Low risk if all domains were low risk; some concerns if at least one domain raised some concerns and none were high risk; and high risk if any domain was high risk, or if multiple domains cumulatively threatened the credibility of the same outcome (e.g., ≥2 domains with some concerns). Two reviewers performed assessments independently and reached consensus through discussion; disagreements were resolved by a third reviewer.

We conducted a meta-analysis using a random-effects model. For continuous outcomes, we calculated mean differences (MDs) with 95% confidence intervals (CIs). When different studies used non-identical scales, we pooled standardized mean differences (SMDs) with 95% CIs. All studies meeting the inclusion criteria were incorporated into the meta-analysis; studies were neither excluded nor differentially weighted based on risk-of-bias assessments. To evaluate the robustness of the pooled estimates and the potential impact of risk of bias, we performed sensitivity analyses. Primary outcomes were analyzed according to the intention-to-treat (ITT) principle, i.e., using participants as randomized to their initial allocation. If an included study did not report a valid ITT analysis, we extracted and used the analysis method as reported in the original article. For attrition or missing data, we preferentially adopted the handling approach described by the original study. Given the general lack of follow-up data among the included trials, we extracted effect sizes at post-intervention. For studies reporting multiple assessment time points, to minimize potential confounding from differences in intervention duration and to enhance temporal consistency, we applied pre-specified rules to select the assessment closest to the mean intervention duration across all included studies for the primary analysis. Heterogeneity was assessed using Cochran's Q test and the I² statistic; I² > 50% or p < 0.10 indicated substantial heterogeneity. In such cases, sensitivity analyses were undertaken to explore sources of heterogeneity. All analyses were performed in RevMan 5.3, and p < 0.05 was considered statistically significant.

A preliminary search identified 2,097 records. By tracing the reference lists of relevant reviews and of the included studies, 32 additional records were added manually. After removing 394 duplicates, titles and abstracts were screened and 1,691 records were excluded, leaving 44 articles for full-text assessment. Following full-text review, studies were further excluded for reasons including non-standard randomized controlled trial designs, absence of cognitive function–related measures, participants outside the target age range, enrollment of patient populations, and missing data. Ultimately, nine studies (Biasutti and Mangiacotti, 2017; Bugos et al., 2007; Bugos and Wang, 2022; Castillejos and Godoy-Izquierdo, 2020; Degé and Kerkovius, 2018; Mack et al., 2024; Marie et al., 2023; Noice and Noice, 2008; Shinada et al., 2025) were included in the systematic review. The study selection process and results are presented in Figure 1.

Nine studies were included in the systematic review, originating from the United States (n = 3), a dual-center study in Germany and Switzerland (n = 2), Germany (n = 1), Spain (n = 1), Italy (n = 1), and Japan (n = 1). The total sample comprised 625 participants, with 313 in intervention groups and 312 in control groups. All studies were randomized controlled trials. The mean intervention period was approximately 18 weeks, and the mean cumulative intervention duration across studies was 22.5 h. Detailed characteristics of the included studies are presented in Table 1.

Regarding the overall risk of bias, one study was rated Low risk, seven were rated some concerns, and one was rated high risk (see Table 2). Specifically, one study Noice and Noice, (2008) was judged high risk in the randomization process domain; additional issues included lack of preregistration, an open-label design, and reliance on subjective self-report scales as primary outcomes, thereby increasing the likelihood of expectancy/performance and reporting biases. For the remaining studies, the main problems clustered in the domains of deviations from intended interventions and selection of the reported result, indicating limited preregistration/analytic-plan transparency and shortcomings in reporting procedures. Two studies Bugos and Wang, (2022); Mack et al., (2024) employed intention-to-treat (ITT) analyses. Overall, the risk of bias across the included studies was predominantly some concerns, and the findings should be interpreted with caution.

The meta-analysis demonstrated a significant improvement in global cognitive performance among cognitively normal older adults receiving music-based interventions, consistent with our a priori hypothesis that such interventions enhance cognitive function. Two principal mechanisms may account for these benefits (Brancatisano et al., 2020): (1) neurobiological perspective. Music-based interventions activate widespread brain regions—including the temporal, frontal, parietal, and occipital cortices; the primary motor cortex; and subcortical structures such as the basal ganglia and the cerebellum—thereby enhancing neural flexibility and; (2) psychological perspective. Relative to other sensory stimuli, music more effectively captures attention because participants continuously track, perceive, and categorize dynamically changing features (e.g., rhythm, harmony, timbre, and meter) during intervention.

Music-based approaches can be categorized as active and passive. Active interventions require direct engagement in musical activities—such as singing, playing an instrument, or creating music—whereas passive interventions primarily involve listening to curated music to regulate psychophysiological states (Vink et al., 2004). In a 2010 study, Bugos (2010) contrasted these modalities and found that active music-based interventions markedly improved cognition in older adults, whereas passive approaches did not yield significant effects. This is broadly consistent with our findings: we observed an overall benefit for global cognition, and among the nine included trials, seven implemented active interventions (e.g., piano, electronic drums, bass, keyboard performance, singing, and drumming) and two used both modalities; studies employing active interventions accounted for >77% of the included trials. By contrast, the meta-analysis by Li et al. (2011) reported no clear cognitive improvement with active music therapy, likely because their review targeted older adults with dementia and assessed cognition solely with the Mini-Mental State Examination (MMSE). Mitchell and Malladi (2010) recommend using the MMSE in combination with additional cognitive measures to identify dementia, particularly beyond the earliest stage. In the present review, we incorporated a broader assessment battery (e.g., Digit Span, verbal fluency), which may detect changes attributable to active interventions; accordingly, active music-based engagement may positively influence global cognition in cognitively normal older adults.

This review further indicates that music-based interventions improve memory and executive function in cognitively normal older adults. Prior work comparing older musicians vs. non-musicians (Grassi et al., 2017) and studies in older adults with Parkinson's disease have reported similar conclusions (Thaut et al., 2009). In a study spanning multiple age groups, Jabès et al. (2021) likewise found that music-based interventions produced significant gains in working memory among older participants and explored potential links among memory capacity, aging, cognitive function, and changes in resting-state oscillatory activity. Watanabe et al. (2019) proposed two non-mutually exclusive mechanisms whereby music interventions may enhance memory in older adults: (i) when participants complete memory tasks, those experiencing a sense of task accomplishment show markedly increased neural activity in frontal and subcortical regions; and (ii) stimulation by musical melodies and verbal/lexical information may engage broader neural networks— including the basal ganglia, thalamus, and hypothalamus. These mechanisms may underlie, at least in part, the memory-enhancing effects of music interventions. Vandervert (2015) further reported that the central executive component of working memory is shaped by cortico-cerebellar system dynamics; accordingly, music interventions may strengthen central executive control and thereby augment volitional/self-regulatory control. Complementarily, Electroencephalography studies (Cheung et al., 2017) have shown that music interventions can modulate cortical synchronization within neural networks subserving verbal memory formation, which may also account for improvements in both executive function and memory.

This review found no significant effect of music-based interventions on attention in cognitively normal older adults. By contrast, several prior studies in healthy children have reported significant attention gains following music training (Strait et al., 2015; Dittinger et al., 2017), and extensive, long-term (multi-year) training has also improved attention in healthy adults (Strait et al., 2015; Wang et al., 2015). Three explanations may account for the null finding here. First, the music-intervention protocols used in the included trials may indeed exert little effect on attention in cognitively normal older adults; across the nine trials in the meta-analysis, the mean intervention duration was 18 weeks, and only one study (Mack et al., 2024) exceeded 6 months—importantly, that 1-year intervention did observe significant improvements in attention-related outcomes. Second, the small number of attention measures and limited sample sizes may have yielded small estimated effects and insufficient statistical power. Third, the attention measures employed may have been insufficiently sensitive to detect training-related changes in older adults. In the included literature, attention was assessed mainly with the Trail Making Test (TMT) or Digit Span Forward (DSF); drawing on additional tasks from the broader literature could improve detection. For example, Strait et al. (2015) asked participants to identify—and act out—one of two simultaneously presented stories delivered from left and right loudspeakers; Nan et al. (2018) used a masked-word target recognition task and reported enhanced executive attention following piano training in children. Consequently, future studies might increase sensitivity to change in older adults by combining multiple attention measures within a comprehensive assessment battery.

This study has several limitations. First, the available evidence is limited in quantity and lacks cultural and geographic representativeness. Trials of music-based interventions targeting cognitively normal older adults remain few, are mostly conducted in high-income countries, and focus predominantly on Western instrumental training (e.g., piano, percussion). There is insufficient attention to traditional musical practices representative of developing countries such as China (e.g., guqin, Chinese opera), constraining external validity and cultural applicability. Second, the nature of music interventions makes blinding difficult, introducing risks of performance and detection bias. Blinding participants and interventionists is typically infeasible; although some studies attempted to blind outcome assessors, most did not fully mitigate expectancy effects among participants. In addition, sample sizes were generally small: among the nine included studies, five enrolled fewer than 50 participants, which may reduce statistical power to detect small effects and may also inflate effect-size estimates. Finally, there was heterogeneity in intervention protocols. Although all interventions were music-centered, their specific formats, intensities, frequencies, and durations varied, potentially increasing between-study heterogeneity and complicating the identification of the most effective intervention model.

Based on this systematic review, we propose several recommendations and considerations for future research. First, while continuing to address the needs of older adults with cognitive impairment, future work should concurrently pursue preventive trials among those without overt cognitive decline to slow cognitive aging. Parallel studies in developing countries are particularly valuable to test cross-cultural generalizability and improve the representativeness of study samples. Second, methodological rigor should be strengthened by implementing robust allocation concealment; blinding outcome assessors; and adopting preregistered protocols with prespecified statistical analysis plans. Finally, factorial or adaptive designs should be used to compare different modalities (e.g., instrumental training, rhythmic interventions, choir/singing, receptive listening) and different doses (e.g., per-session duration, frequency, total hours), in order to identify the active ingredients that drive benefit (e.g., motor engagement, cognitive load, social participation).